Initial vendor packages

Signed-off-by: Valentin Popov <valentin@popov.link>

vendor/image/CHANGES.md

@@ -0,0 +1,582 @@
+# Release Notes
+
+## Known issues
+- Many decoders will panic on malicous input. In most cases, this is caused by
+  not enforcing memory limits, though other panics have been seen from fuzzing.
+- The color space information of pixels is not clearly communicated.
+
+## Changes
+
+### Unreleased
+
+- More convenient to use buffers will be added in the future. In particular,
+  improving initialization, passing of output buffers, and adding a more
+  complete representation for layouts. The plan is for these to interact with
+  the rest of the library through a byte-based interface similar to
+  `ImageDecoder`.
+  See ongoing work on [`image-canvas`](https://github.com/image-rs/canvas) if
+  you want to participate.
+
+### Version 0.24.7
+
+New features:
+- Added `{ImageBuffer, DynamicImage}::write_with_encoder` to simplify writing
+  images with custom settings.
+- Expose ICC profiles stored in tiff and wepb files.
+- Added option to set the background color of animated webp images.
+- New methods for sampling and interpolation of `GenericImageView`s
+
+Bug fixes:
+- Fix panic on empty dxt.
+- Fix several panics in webp decoder.
+- Allow unknown chunks at the end of webp files.
+
+### Version 0.24.6
+
+- Add support for QOI.
+- ImageDecoders now expose ICC profiles on supported formats.
+- Add support for BMPs without a file header.
+- Improved AVIF encoder.
+- WebP decoding fixes.
+
+### Version 0.24.5
+
+Structural changes:
+- Increased the minimum supported Rust version (MSRV) to 1.61.
+- Increased the version requirement for the `tiff` crate to 0.8.0.
+- Increased the version requirement for the `jpeg` crate to 0.3.0.
+
+Bug fixes:
+- The `as_rgb32f` function of `DynamicImage` is now correctly documented.
+- Fixed a crash when decoding ICO images. Added a regression test.
+- Fixed a panic when transforming webp images. Added a regression test.
+- Added a check to prevent integer overflow when calculating file size for BMP
+  images. The missing check could panic in debug mode or else set an incorrect
+  file size in release mode.
+- Upgraded the PNG image encoder to use the newer `PngEncoder::write_image`
+  instead of the deprecated `PngEncoder::encode` which did not account for byte
+  order and could result in images with incorrect colors.
+- Fixed `InsufficientMemory` error when trying to decode a PNG image.
+- Fix warnings and CI issues.
+- Typos and links in the documentation have been corrected.
+
+Performance:
+- Added check for dynamic image dimensions before resizing. This improves
+  performance in cases where the image does not need to be resized or has
+  already been resized.
+
+### Version 0.24.4
+
+New Features:
+- Encoding for `webp` is now available with the native library. This needs to
+  be activate explicitly with the `web-encoder` feature.
+- `exr` decoding has gained basic limit support.
+
+Bug fixes:
+- The `Iterator::size_hint` implementation of pixel iterators has been fixed to
+  return the current length indicated by its `ExactSizeIterator` hint.
+- Typos and bad references in the documentation have been removed.
+
+Performance:
+- `ImageBuffer::get_pixel{,_mut}` is now marked inline.
+- `resize` now short-circuits when image dimensions are unchanged.
+
+### Version 0.24.3
+
+New Features:
+- `TiffDecoder` now supports setting resource limits.
+
+Bug fixes:
+- Fix compile issues on little endian systems.
+- Various panics discovered by fuzzing.
+
+### Version 0.24.2
+
+Structural changes:
+- CI now runs `cargo-deny`, checking dependent crates to an OSS license list
+  and against RUSTSEC advisories.
+
+New Features:
+- The WebP decoder recognizes and decodes images with `VP8X` header.
+- The DDS decoder recognizes and decodes images with `DX10` headers.
+
+Bug fixes:
+- Calling `DynamicImage`/`ImageBuffer`'s methods `write_to` and `save` will now
+  work properly even if the backing container is larger than the image layout
+  requires. Only the relevant slice of pixel data is passed to the encoder.
+- Fixed a OOM-panic caused by malformed images in the `gif` decoder.
+
+### Version 0.24.1
+
+Bug Fixes:
+- ImageBuffer::get_pixel_checked would sometimes return the incorrect pixel.
+- PNG encoding would sometimes not recognize unsupported color.
+
+### Version 0.24.0
+
+Breaking changes
+
+Structural changes:
+- Minimum Rust version is now `1.56` and may change in minor versions until
+  further notice. It is now tracked in the library's `Cargo.toml`, instead, by
+  the standard `[package.rust-version]` field. Note: this applies _to the
+  library itself_. You may need different version resolutions for dependencies
+  when using a non-stable version of Rust.
+- The `math::utils::{nq, utils}` modules have been removed. These are better
+  served through the `color_quant` crate and the standard library respectively.
+- All codecs are now available through `image::codecs`, no longer top-level.
+- `ExtendedColorType` and `DynamicImage` have been made `#[non_exhaustive]`,
+  providing more methods instead of exhaustive matching.
+- Reading images through the generic `io::Reader`, as well as generic
+  convenience interfaces, now requires the underlying reader to be `BufRead +
+  Seek`. This allows more efficient support more formats. Similarly, writing
+  now requires writers to be `Write + Seek`.
+- The `Bgra*` variants of buffers, which were only half-supported, have been
+  removed. The owning buffer types `ImageBuffer` and `DynamicImage`
+  fundamentally already make a choice in supported pixel representations. This
+  allows for more consistent internal behavior. Callers are expected to convert
+  formats when using those buffers, which they are required to do in any case
+  already, and which is routinely performed by decoders.
+
+Trait reworks:
+- The `Pixel` trait is no longer implemented quite as liberally for structs
+  defined in the crate. Instead, it is now restricted to a set of known channel
+  which ensures accuracy in computations involving those channels.
+- The `ImageDecoderExt` trait has been renamed to `ImageDecoderRect`, according
+  to its actual functionality.
+- The `Pixel` trait and its `Subpixel` field no longer require (or provide) a
+  `'static` lifetime bound.
+- The `Pixel` trait no longer requires specifying an associated, constant
+  `ColorType`. This was of little relevance to computation but made it much
+  harder to implement and extend correctly. Instead, the _private_
+  `PixelWithColorType` extension is added for interfaces that require a
+  properly known variant.
+- Reworked how `SubImage` interacts with the `GenericImage` trait. It is now a
+  default implementation. Note that `SubImage` now has _inherent_ methods that
+  avoid double-indirection, the trait's method will no longer avoid this.
+- The `Primitive` trait now requires implementations to provide a minimum and
+  maximum logical bound for the purpose of converting to other primitive
+  representations.
+
+Additions
+
+Image formats:
+- Reading lossless WebP is now supported.
+- The OpenEXR format is now supported.
+- The `jpeg` decoder has been upgraded to Lossless JPEG.
+- The `AvifEncoder` now correctly handles alpha-less images. Some additional
+  color formats are converted to RGBA as well.
+- The `Bmp` codec now decodes more valid images. It can decode a raw image
+  without performing the palette mapping. It provides a method to access the
+  palette. The encoder provides the inverse capabilities.
+- `Tiff` is now an output format.
+
+Buffers and Operations:
+- The channel / primitive type `f32` is now supported. Currently only the
+  OpenEXR codec makes full use of it but this is expected to change.
+- `ImageBuffer::{get_pixel_checked, get_pixel_mut_checked}` provide panic-free
+  access to pixels and channels by returning `Option<&P>` and `Option<&mut P>`.
+- `ImageBuffer::write_to` has been added, encoding the buffer to a writer. This
+  method already existed on `DynamicImage`.
+- `DynamicImage` now implements `From<_>` for all supported buffer types.
+- `DynamicImage` now implements `Default`, an empty `Rgba8` image.
+- `imageops::overlay` now takes coordinates as `i64`.
+
+Limits:
+- Added `Limits` and `LimitSupport`, utilized in `io::Reader`. These can be
+  configured for rudimentary protection against resource exhaustion (images
+  pretending to require a very large buffer). These types are not yet
+  exhaustive by design, and more and stricter limits may be added in the
+  future.
+- Encoders that do provide inherent support for limits, or reserve a
+  significant amount of internal memory, are urged to implement the
+  `set_limits` extension to `ImageDecoder`. Some strict limit are opt-in, which
+  may cause decoding to fail if not supported.
+
+Miscellaneous:
+- `PNMSubtype` has been renamed to `PnmSubtype`, by Rust's naming scheme.
+- Several incorrectly capitalized `PNM*` aliases have been removed.
+- Several `enum` types that had previously used a hidden variant now use the
+  official `#[non_exhaustive]` attribute instead.
+
+### Version 0.23.14
+
+- Unified gif blending in different decode methods, fixing out-of-bounds checks
+  in a number of weirdly positioned frames.
+- Hardened TGA decoder against a number of malicious inputs.
+- Fix forward incompatible usage of the panic macro.
+- Fix load_rect for gif reaching `unreachable!()` code.
+
+- Added `ExtendedColorType::A8`.
+- Allow TGA to load alpha-only images.
+- Optimized load_rect to avoid unnecessary seeks.
+
+### Version 0.23.13
+
+- Fix an inconsistency in supported formats of different methods for encoding
+  an image.
+- Fix `thumbnail` choosing an empty image. It now always prefer non-empty image
+  dimensions.
+- Fix integer overflow in calculating requires bytes for decoded image buffers
+  for farbfeld, hdr, and pnm decoders. These will now error early.
+- Fix a panic decoding certain `jpeg` image without frames or meta data.
+- Optimized the `jpeg` encoder.
+- Optimized `GenericImage::copy_from` default impl in various cases.
+
+- Add `avif` decoders. You must enable it explicitly and it is not covered by
+  our usual MSRV policy of Rust 1.34. Instead, only latest stable is supported.
+- Add `ImageFormat::{can_read, can_write}`
+- Add `Frame::buffer_mut`
+- Add speed and quality options on `avif` encoder.
+- Add speed parameter to `gif` encoder.
+- Expose control over sequence repeat to the `gif` encoder.
+- Add `{contrast,brighten,huerotate}_in_place` functions in imageproc.
+
+- Relax `Default` impl of `ImageBuffer`, removing the bound on the color type.
+- Derive Debug, Hash, PartialEq, Eq for DynamicImage
+
+### Version 0.23.12
+
+- Fix a soundness issue affecting the impls of `Pixel::from_slice_mut`. This
+  would previously reborrow the mutable input reference as a shared one but
+  then proceed to construct the mutable result reference from it. While UB
+  according to Rust's memory model, we're fairly certain that no miscompilation
+  can happen with the LLVM codegen in practice.
+  See 5cbe1e6767d11aff3f14c7ad69a06b04e8d583c7 for more details.
+- Fix `imageops::blur` panicking when `sigma = 0.0`. It now defaults to `1.0`
+  as all negative values.
+- Fix re-exporting `png::{CompressionType, FilterType}` to maintain SemVer
+  compatibility with the `0.23` releases.
+
+- Add `ImageFormat::from_extension`
+- Add copyless DynamicImage to byte slice/vec conversion.
+- Add bit-depth specific `into_` and `to_` DynamicImage conversion methods.
+
+
+### Version 0.23.11
+
+- The `NeuQuant` implementation is now supplied by `color_quant`. Use of the
+  type defined by this library is discouraged.
+- The `jpeg` decoder can now downscale images that are decoded by 1,2,4,8.
+- Optimized the jpeg encoding ~5-15%.
+- Deprecated the `clamp` function. Use `num-traits` instead.
+- The ICO decoder now accepts an empty mask.
+- Fixed an overflow in ICO mask decoding potentially leading to panic.
+- Added `ImageOutputFormat` for `AVIF`
+- Updated `tiff` to `0.6` with lzw performance improvements.
+
+### Version 0.23.10
+
+- Added AVIF encoding capabilities using the `ravif` crate. Please note that
+  the feature targets the latest stable compiler and is not enabled by default.
+- Added `ImageBuffer::as_raw` to inspect the underlying container.
+- Updated `gif` to `0.11` with large performance improvements.
+
+### Version 0.23.9
+
+- Introduced correctly capitalized aliases for some scream case types
+- Introduced `imageops::{vertical_gradient, horizontal_gradient}` for writing
+  simple color gradients into an image.
+- Sped up methods iterating over `Pixels`, `PixelsMut`, etc. by using exact
+  chunks internally. This should auto-vectorize `ImageBuffer::from_pixel`.
+- Adjusted `Clone` impls of iterators to not require a bound on the pixel.
+- Add `Debug` impls for iterators where the pixel's channel implements it.
+- Add comparison impls for `FilterType`
+
+### Version 0.23.8
+
+- `flat::Error` now implements the standard `Error` trait
+- The type parameter of `Map` has been relaxed to `?Sized`
+- Added the `imageops::tile` function that repeats one image across another
+
+### Version 0.23.7
+
+- Iterators over immutable pixels of `ImageBuffer` can now be cloned
+- Added a `tga` encoder
+- Added `ColorMap::lookup`, an optional reversal of the map
+- The `EncodableLayout` trait is now exported
+
+### Version 0.23.6
+
+- Added `png::ApngDecoder`, an adapter decoding the animation in an APNG.
+- Fixed a bug in `jpeg` encoding that would darken output colors.
+- Added a utility constructor `FlatSamples::with_monocolor`.
+- Added `ImageBuffer::as_flat_samples_mut` which is a mutable variant of the
+  existing ffi-helper `ImageBuffer::as_flat_samples`.
+
+### Version 0.23.5
+
+- The `png` encoder now allows configuring compression and filter type. The
+  output is not part of stability guarantees, see its documentation.
+- The `jpeg` encoder now accepts any implementor of `GenericImageView`. This
+  allows images that are only partially present in memory to be encoded.
+- `ImageBuffer` now derives `Hash`, `PartialEq`, `Eq`.
+- The `Pixels`/`PixelsMut` iterator no longer yields out-of-bounds pixels when
+  the underlying buffer is larger than required.
+- The `pbm` decoder correctly decodes ascii data again, fixing a regression
+  where it would use the sample value `1` as white instead of `255`.
+- Fix encoding of RGBA data in `gif` frames.
+- Constructing a `Rows`/`RowsMut` iterator no longer panics when the image has
+  a width or height of `0`.
+
+### Version 0.23.4
+
+- Improved the performance of decoding animated gifs
+- Added `crop_imm` which functions like `crop` but on a shared reference
+- The gif `DisposalMethod::Any` is treated as `Keep`, consistent with browsers
+- Most errors no longer allocate a string, instead implement Display.
+- Add some implementations of `Error::source`
+
+### Version 0.23.3
+
+- Added `ColorType::has_alpha` to facilitate lossless conversion
+- Recognize extended WebP formats for decoding
+- Added decoding and encoding for the `farbfeld` format
+- Export named iterator types created from various `ImageBuffer` methods
+- Error in jpeg encoder for images larger than 65536 pixels, fixes panic
+
+### Version 0.23.2
+
+- The dependency on `jpeg-decoder` now reflects minimum requirements.
+
+### Version 0.23.1
+
+- Fix cmyk_to_rgb (jpeg) causing off by one rounding errors.
+- A number of performance improvements for jpeg (encode and decode), bmp, vp8
+- Added more details to errors for many formats
+
+### Version 0.23.0
+
+This major release intends to improve the interface with regards to handling of
+color format data and errors for both decoding and encoding. This necessitated
+many breaking changes anyways so it was used to improve the compliance to the
+interface guidelines such as outstanding renaming.
+
+It is not yet perfect with regards to color spaces but it was designed mainly
+as an improvement over the current interface with regards to in-memory color
+formats, first. We'll get to color spaces in a later major version.
+
+- Heavily reworked `ColorType`:
+  - This type is now used for denoting formats for which we support operations
+      on buffers in these memory representations. Particularly, all channels in
+      pixel types are assumed to be an integer number of bytes (In terms of the
+      Rust type system, these are `Sized` and one can crate slices of channel
+      values).
+  - An `ExtendedColorType` is used to express more generic color formats for
+      which the library has limited support but can be converted/scaled/mapped
+      into a `ColorType` buffer. This operation might be fallible but, for
+      example, includes sources with 1/2/4-bit components.
+  - Both types are non-exhaustive to add more formats in a minor release.
+  - A work-in-progress (#1085) will further separate the color model from the
+      specific channel instantiation, e.g. both `8-bit RGB` and `16-bit BGR`
+      are instantiations of `RGB` color model.
+- Heavily rework `ImageError`:
+  - The top-level enum type now serves to differentiate cause with multiple
+      opaque representations for the actual error. These are no longer simple
+      Strings but contains useful types. Third-party decoders that have no
+      variant in `ImageFormat` have also been considered.
+  - Support for `Error::source` that can be downcast to an error from a
+      matching version of the underlying decoders. Note that the version is not
+      part of the stable interface guarantees, this should not be relied upon
+      for correctness and only be used as an optimization.
+  - Added image format indications to errors.
+  - The error values produced by decoder will be upgraded incrementally. See
+      something that still produces plain old String messages? Feel free to
+      send a PR.
+- Reworked the `ImageDecoder` trait:
+  - `read_image` takes an output buffer argument instead of allocating all
+      memory on its own.
+  - The return type of `dimensions` now aligns with `GenericImage` sizes.
+  - The `colortype` method was renamed to `color_type` for conformity.
+- The enums `ColorType`, `DynamicImage`, `imageops::FilterType`, `ImageFormat`
+  no longer re-export all of their variants in the top-level of the crate. This
+  removes the growing pollution in the documentation and usage. You can still
+  insert the equivalent statement on your own:
+  `use image::ImageFormat::{self, *};`
+- The result of `encode` operations is now uniformly an `ImageResult<()>`.
+- Removed public converters from some `tiff`, `png`, `gif`, `jpeg` types,
+  mainly such as error conversion. This allows upgrading the dependency across
+  major versions without a major release in `image` itself.
+- On that note, the public interface of `gif` encoder no longer takes a
+  `gif::Frame` but rather deals with `image::Frame` only. If you require to
+  specify the disposal method, transparency, etc. then you may want to wait
+  with upgrading but (see next change).
+- The `gif` encoder now errors on invalid dimensions or unsupported color
+  formats. It would previously silently reinterpret bytes as RGB/RGBA.
+- The capitalization of  `ImageFormat` and other enum variants has been
+  adjusted to adhere to the API guidelines. These variants are now spelled
+  `Gif`, `Png`, etc. The same change has been made to the name of types such as
+  `HDRDecoder`.
+- The `Progress` type has finally received public accessor method. Strange that
+  no one reported them missing.
+- Introduced `PixelDensity` and `PixelDensityUnit` to store DPI information in
+  formats that support encoding this form of meta data (e.g. in `jpeg`).
+
+### Version 0.22.5
+
+- Added `GenericImage::copy_within`, specialized for `ImageBuffer`
+- Fixed decoding of interlaced `gif` files
+- Prepare for future compatibility of array `IntoIterator` in example code
+
+### Version 0.22.4
+
+- Added in-place variants for flip and rotate operations.
+- The bmp encoder now checks if dimensions are valid for the format. It would
+  previously write a subset or panic.
+- Removed deprecated implementations of `Error::description`
+- Added `DynamicImage::into_*` which convert without an additional allocation.
+- The PNG encoder errors on unsupported color types where it had previously
+  silently swapped color channels.
+- Enabled saving images as `gif` with `save_buffer`.
+
+### Version 0.22.3
+
+- Added a new module `io` containing a configurable `Reader`. It can replace
+  the bunch of free functions: `image::{load_*, open, image_dimensions}` while
+  enabling new combinations such as `open` but with format deduced from content
+  instead of file path.
+- Fixed `const_err` lint in the macro expanded implementations of `Pixel`. This
+  can only affect your crate if `image` is used as a path dependency.
+
+### Version 0.22.2
+
+- Undeprecate `unsafe` trait accessors. Further evaluation showed that their
+  deprecation should be delayed until trait `impl` specialization is available.
+- Fixed magic bytes used to detect `tiff` images.
+- Added `DynamicImage::from_decoder`.
+- Fixed a bug in the `PNGReader` that caused an infinite loop.
+- Added `ColorType::{bits_per_pixel, num_components}`.
+- Added `ImageFormat::from_path`, same format deduction as the `open` method.
+- Fixed a panic in the gif decoder.
+- Aligned background color handling of `gif` to web browser implementations.
+- Fixed handling of partial frames in animated `gif`.
+- Removed unused direct `lzw` dependency, an indirect dependency in `tiff`.
+
+### Version 0.22.1
+
+- Fixed build without no features enabled
+
+### Version 0.22
+
+- The required Rust version is now `1.34.2`.
+- Note the website and blog: [image-rs.org][1] and [blog.image-rs.org][2]
+- `PixelMut` now only on `ImageBuffer` and removed from `GenericImage`
+  interface. Prefer iterating manually in the generic case.
+- Replaced an unsafe interface in the hdr decoder with a safe variant.
+- Support loading 2-bit BMP images
+- Add method to save an `ImageBuffer`/`DynamicImage` with specified format
+- Update tiff to `0.3` with a writer
+- Update png to `0.15`, fixes reading of interlaced sub-byte pixels
+- Always use custom struct for `ImageDecoder::Reader`
+- Added `apply_without_alpha` and `map_without_alpha` to `Pixel` trait
+- Pixel information now with associated constants instead of static methods
+- Changed color structs to tuple types with single component. Improves
+  ergonomics of destructuring assignment and construction.
+- Add lifetime parameter on `ImageDecoder` trait.
+- Remove unnecessary `'static` bounds on affine operations
+- Add function to retrieve image dimensions without loading full image
+- Allow different image types in overlay and replace
+- Iterators over rows of `ImageBuffer`, mutable variants
+
+[1]: https://www.image-rs.org
+[2]: https://blog.image-rs.org
+
+### Version 0.21.2
+
+- Fixed a variety of crashes and opaque errors in webp
+- Updated the png limits to be less restrictive
+- Reworked even more `unsafe` operations into safe alternatives
+- Derived Debug on FilterType and Deref on Pixel
+- Removed a restriction on DXT to always require power of two dimensions
+- Change the encoding of RGBA in bmp using bitfields
+- Corrected various urls
+
+### Version 0.21.1
+
+- A fairly important bugfix backport
+- Fixed a potentially memory safety issue in the hdr and tiff decoders, see #885
+- See [the full advisory](docs/2019-04-23-memory-unsafety.md) for an analysis
+- Fixes `ImageBuffer` index calculation for very, very large images
+- Fix some crashes while parsing specific incomplete pnm images
+- Added comprehensive fuzzing for the pam image types
+
+### Version 0.21
+
+- Updated README to use `GenericImageView`
+- Removed outdated version number from CHANGES
+- Compiles now with wasm-unknown-emscripten target
+- Restructured `ImageDecoder` trait
+- Updated README with a more colorful example for the Julia fractal
+- Use Rust 1.24.1 as minimum supported version
+- Support for loading GIF frames one at a time with `animation::Frames`
+- The TGA decoder now recognizes 32 bpp as RGBA(8)
+- Fixed `to_bgra` document comment
+- Added release test script
+- Removed unsafe code blocks several places
+- Fixed overlay overflow bug issues with documented proofs
+
+### Version 0.20
+
+- Clippy lint pass
+- Updated num-rational dependency
+- Added BGRA and BGR color types
+- Improved performance of image resizing
+- Improved PBM decoding
+- PNM P4 decoding now returns bits instead of bytes
+- Fixed move of overlapping buffers in BMP decoder
+- Fixed some document comments
+- `GenericImage` and `GenericImageView` is now object-safe
+- Moved TIFF code to its own library
+- Fixed README examples
+- Fixed ordering of interpolated parameters in TIFF decode error string
+- Thumbnail now handles upscaling
+- GIF encoding for multiple frames
+- Improved subimages API
+- Cargo fmt fixes
+
+### Version 0.19
+
+- Fixed panic when blending with alpha zero.
+- Made `save` consistent.
+- Consistent size calculation.
+- Fixed bug in `apply_with_alpha`.
+- Implemented `TGADecoder::read_scanline`.
+- Use deprecated attribute for `pixels_mut`.
+- Fixed bug in JPEG grayscale encoding.
+- Fixed multi image TIFF.
+- PNM encoder.
+- Added `#[derive(Hash)]` for `ColorType`.
+- Use `num-derive` for `#[derive(FromPrimitive)]`.
+- Added `into_frames` implementation for GIF.
+- Made rayon an optional dependency.
+- Fixed issue where resizing image did not give exact width/height.
+- Improved downscale.
+- Added a way to expose options when saving files.
+- Fixed some compiler warnings.
+- Switched to lzw crate instead of using built-in version.
+- Added `ExactSizeIterator` implementations to buffer structs.
+- Added `resize_to_fill` method.
+- DXT encoding support.
+- Applied clippy suggestions.
+
+### Version 0.4
+ - Various improvements.
+ - Additional supported image formats (BMP and ICO).
+ - GIF and PNG codec moved into separate crates.
+
+### Version 0.3
+ - Replace `std::old_io` with `std::io`.
+
+### Version 0.2
+ - Support for interlaced PNG images.
+ - Writing support for GIF images (full color and paletted).
+ - Color quantizer that converts 32bit images to paletted including the alpha channel.
+ - Initial support for reading TGA images.
+ - Reading support for TIFF images (packbits and FAX compression not supported).
+ - Various bug fixes and improvements.
+
+### Version 0.1
+- Initial release
+- Basic reading support for png, jpeg, gif, ppm and webp.
+- Basic writing support for png and jpeg.
+- A collection of basic imaging processing function like `blur` or `invert`

vendor/image/Cargo.toml

@@ -0,0 +1,188 @@
+# THIS FILE IS AUTOMATICALLY GENERATED BY CARGO
+#
+# When uploading crates to the registry Cargo will automatically
+# "normalize" Cargo.toml files for maximal compatibility
+# with all versions of Cargo and also rewrite `path` dependencies
+# to registry (e.g., crates.io) dependencies.
+#
+# If you are reading this file be aware that the original Cargo.toml
+# will likely look very different (and much more reasonable).
+# See Cargo.toml.orig for the original contents.
+
+[package]
+edition = "2018"
+rust-version = "1.61.0"
+name = "image"
+version = "0.24.7"
+authors = ["The image-rs Developers"]
+exclude = [
+    "src/png/testdata/*",
+    "examples/*",
+    "tests/*",
+]
+description = "Imaging library. Provides basic image processing and encoders/decoders for common image formats."
+homepage = "https://github.com/image-rs/image"
+documentation = "https://docs.rs/image"
+readme = "README.md"
+categories = [
+    "multimedia::images",
+    "multimedia::encoding",
+]
+license = "MIT"
+repository = "https://github.com/image-rs/image"
+resolver = "2"
+
+[lib]
+name = "image"
+path = "./src/lib.rs"
+
+[[bench]]
+name = "decode"
+path = "benches/decode.rs"
+harness = false
+
+[[bench]]
+name = "encode"
+path = "benches/encode.rs"
+harness = false
+
+[[bench]]
+name = "copy_from"
+harness = false
+
+[dependencies.bytemuck]
+version = "1.7.0"
+features = ["extern_crate_alloc"]
+
+[dependencies.byteorder]
+version = "1.3.2"
+
+[dependencies.color_quant]
+version = "1.1"
+
+[dependencies.dav1d]
+version = "0.6.0"
+optional = true
+
+[dependencies.dcv-color-primitives]
+version = "0.4.0"
+optional = true
+
+[dependencies.exr]
+version = "1.5.0"
+optional = true
+
+[dependencies.gif]
+version = "0.12"
+optional = true
+
+[dependencies.jpeg]
+version = "0.3.0"
+optional = true
+default-features = false
+package = "jpeg-decoder"
+
+[dependencies.libwebp]
+version = "0.2.2"
+optional = true
+default-features = false
+package = "webp"
+
+[dependencies.mp4parse]
+version = "0.17.0"
+optional = true
+
+[dependencies.num-rational]
+version = "0.4"
+default-features = false
+
+[dependencies.num-traits]
+version = "0.2.0"
+
+[dependencies.png]
+version = "0.17.6"
+optional = true
+
+[dependencies.qoi]
+version = "0.4"
+optional = true
+
+[dependencies.ravif]
+version = "0.11.0"
+optional = true
+
+[dependencies.rgb]
+version = "0.8.25"
+optional = true
+
+[dependencies.tiff]
+version = "0.9.0"
+optional = true
+
+[dev-dependencies.crc32fast]
+version = "1.2.0"
+
+[dev-dependencies.criterion]
+version = "0.4"
+
+[dev-dependencies.glob]
+version = "0.3"
+
+[dev-dependencies.jpeg]
+version = "0.3.0"
+features = ["platform_independent"]
+default-features = false
+package = "jpeg-decoder"
+
+[dev-dependencies.num-complex]
+version = "0.4"
+
+[dev-dependencies.quickcheck]
+version = "1"
+
+[features]
+avif = ["avif-encoder"]
+avif-decoder = [
+    "mp4parse",
+    "dcv-color-primitives",
+    "dav1d",
+]
+avif-encoder = [
+    "ravif",
+    "rgb",
+]
+benchmarks = []
+bmp = []
+dds = ["dxt"]
+default = [
+    "gif",
+    "jpeg",
+    "ico",
+    "png",
+    "pnm",
+    "tga",
+    "tiff",
+    "webp",
+    "bmp",
+    "hdr",
+    "dxt",
+    "dds",
+    "farbfeld",
+    "jpeg_rayon",
+    "openexr",
+    "qoi",
+]
+dxt = []
+farbfeld = []
+hdr = []
+ico = [
+    "bmp",
+    "png",
+]
+jpeg_rayon = ["jpeg/rayon"]
+openexr = ["exr"]
+pnm = []
+qoi = ["dep:qoi"]
+tga = []
+webp = []
+webp-encoder = ["libwebp"]

vendor/image/Cargo.toml.public-private-dependencies

@@ -0,0 +1,98 @@
+cargo-features = ["public-dependency"]
+
+[package]
+name = "image"
+version = "0.24.0-alpha"
+edition = "2018"
+rust-version = "1.56"
+
+license = "MIT"
+description = "Imaging library written in Rust. Provides basic filters and decoders for the most common image formats."
+authors = ["The image-rs Developers"]
+readme = "README.md"
+
+# crates.io metadata
+documentation = "https://docs.rs/image"
+repository = "https://github.com/image-rs/image"
+homepage = "https://github.com/image-rs/image"
+categories = ["multimedia::images", "multimedia::encoding"]
+
+# Crate build related
+exclude = [
+    "src/png/testdata/*",
+    "examples/*",
+    "tests/*",
+]
+
+[lib]
+name = "image"
+path = "./src/lib.rs"
+
+[dependencies]
+bytemuck = { version = "1.7.0", features = ["extern_crate_alloc"] } # includes cast_vec
+byteorder = "1.3.2"
+num-iter = "0.1.32"
+num-rational = { version = "0.4", default-features = false }
+num-traits = { version = "0.2.0", public = true }
+gif = { version = "0.11.1", optional = true }
+jpeg = { package = "jpeg-decoder", version = "0.2.1", default-features = false, optional = true }
+png = { version = "0.17.0", optional = true }
+tiff = { version = "0.9.0", optional = true }
+ravif = { version = "0.8.0", optional = true }
+rgb = { version = "0.8.25", optional = true }
+mp4parse = { version = "0.12.0", optional = true }
+dav1d = { version = "0.6.0", optional = true }
+dcv-color-primitives = { version = "0.4.0", optional = true }
+exr = { version = "1.4.1", optional = true }
+color_quant = { version = "1.1", public = true }
+
+[dev-dependencies]
+crc32fast = "1.2.0"
+num-complex = "0.4"
+glob = "0.3"
+quickcheck = "1"
+criterion = "0.3"
+
+[features]
+# TODO: Add "avif" to this list while preparing for 0.24.0
+default = ["gif", "jpeg", "ico", "png", "pnm", "tga", "tiff", "webp", "bmp", "hdr", "dxt", "dds", "farbfeld", "jpeg_rayon", "openexr"]
+
+ico = ["bmp", "png"]
+pnm = []
+tga = []
+webp = []
+bmp = []
+hdr = []
+dxt = []
+dds = ["dxt"]
+farbfeld = []
+openexr = ["exr"]
+
+# Enables multi-threading.
+# Requires latest stable Rust.
+jpeg_rayon = ["jpeg/rayon"]
+# Non-default, enables avif support.
+# Requires latest stable Rust.
+avif = ["avif-encoder"]
+# Requires latest stable Rust and recent nasm (>= 2.14).
+avif-encoder = ["ravif", "rgb"]
+# Non-default, even in `avif`. Requires stable Rust and native dependency libdav1d.
+avif-decoder = ["mp4parse", "dcv-color-primitives", "dav1d"]
+
+# Build some inline benchmarks. Useful only during development.
+# Requires rustc nightly for feature test.
+benchmarks = []
+
+[[bench]]
+path = "benches/decode.rs"
+name = "decode"
+harness = false
+
+[[bench]]
+path = "benches/encode.rs"
+name = "encode"
+harness = false
+
+[[bench]]
+name = "copy_from"
+harness = false

vendor/image/LICENSE

@@ -0,0 +1,21 @@
+The MIT License (MIT)
+
+Copyright (c) 2014 PistonDevelopers
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

vendor/image/README.md

@@ -0,0 +1,250 @@
+# Image
+[![crates.io](https://img.shields.io/crates/v/image.svg)](https://crates.io/crates/image)
+[![Documentation](https://docs.rs/image/badge.svg)](https://docs.rs/image)
+[![Build Status](https://github.com/image-rs/image/workflows/Rust%20CI/badge.svg)](https://github.com/image-rs/image/actions)
+[![Gitter](https://badges.gitter.im/image-rs/image.svg)](https://gitter.im/image-rs/image?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge)
+
+Maintainers: [@HeroicKatora](https://github.com/HeroicKatora), [@fintelia](https://github.com/fintelia)
+
+[How to contribute](https://github.com/image-rs/organization/blob/master/CONTRIBUTING.md)
+
+## An Image Processing Library
+
+This crate provides basic image processing functions and methods for converting to and from various image formats.
+
+All image processing functions provided operate on types that implement the `GenericImageView` and `GenericImage` traits and return an `ImageBuffer`.
+
+## Supported Image Formats
+
+`image` provides implementations of common image format encoders and decoders.
+
+<!--- NOTE: Make sure to keep this table in sync with the one in src/lib.rs -->
+
+| Format | Decoding | Encoding |
+| ------ | -------- | -------- |
+| AVIF   | Only 8-bit \*\* | Lossy |
+| BMP    | Yes | Rgb8, Rgba8, Gray8, GrayA8 |
+| DDS    | DXT1, DXT3, DXT5 | No |
+| Farbfeld | Yes | Yes |
+| GIF    | Yes | Yes |
+| ICO    | Yes | Yes |
+| JPEG   | Baseline and progressive | Baseline JPEG |
+| OpenEXR  | Rgb32F, Rgba32F (no dwa compression) | Rgb32F, Rgba32F (no dwa compression) |
+| PNG    | All supported color types | Same as decoding |
+| PNM    | PBM, PGM, PPM, standard PAM | Yes |
+| QOI    | Yes | Yes |
+| TGA    | Yes | Rgb8, Rgba8, Bgr8, Bgra8, Gray8, GrayA8 |
+| TIFF   | Baseline(no fax support) + LZW + PackBits | Rgb8, Rgba8, Gray8 |
+| WebP   | Yes | Rgb8, Rgba8 \* |
+
+- \* Requires the `webp-encoder` feature, uses the libwebp C library.
+- \*\* Requires the `avif-decoder` feature, uses the libdav1d C library.
+
+### The [`ImageDecoder`](https://docs.rs/image/*/image/trait.ImageDecoder.html) and [`ImageDecoderRect`](https://docs.rs/image/*/image/trait.ImageDecoderRect.html) Traits
+
+All image format decoders implement the `ImageDecoder` trait which provide
+basic methods for getting image metadata and decoding images. Some formats
+additionally provide `ImageDecoderRect` implementations which allow for
+decoding only part of an image at once.
+
+The most important methods for decoders are...
++ **dimensions**: Return a tuple containing the width and height of the image.
++ **color_type**: Return the color type of the image data produced by this decoder.
++ **read_image**: Decode the entire image into a slice of bytes.
+
+## Pixels
+
+`image` provides the following pixel types:
++ **Rgb**: RGB pixel
++ **Rgba**: RGB with alpha (RGBA pixel)
++ **Luma**: Grayscale pixel
++ **LumaA**: Grayscale with alpha
+
+All pixels are parameterised by their component type.
+
+## Images
+Individual pixels within images are indexed with (0,0) at the top left corner. 
+### The [`GenericImageView`](https://docs.rs/image/*/image/trait.GenericImageView.html) and [`GenericImage`](https://docs.rs/image/*/image/trait.GenericImage.html) Traits
+
+Traits that provide methods for inspecting (`GenericImageView`) and manipulating (`GenericImage`) images, parameterised over the image's pixel type.
+
+Some of these methods for `GenericImageView` are...
++ **dimensions**: Return a tuple containing the width and height of the image.
++ **get_pixel**: Returns the pixel located at (x, y).
++ **pixels**: Returns an Iterator over the pixels of this image.
+
+While some of the methods for `GenericImage` are...
++ **put_pixel**: Put a pixel at location (x, y).
++ **copy_from**: Copies all of the pixels from another image into this image.
+
+### Representation of Images
+`image` provides two main ways of representing image data:
+
+#### [`ImageBuffer`](https://docs.rs/image/*/image/struct.ImageBuffer.html)
+An image parameterised by its Pixel types, represented by a width and height and a vector of pixels. It provides direct access to its pixels and implements the `GenericImageView` and `GenericImage` traits.
+
+```rust
+use image::{GenericImage, GenericImageView, ImageBuffer, RgbImage};
+
+// Construct a new RGB ImageBuffer with the specified width and height.
+let img: RgbImage = ImageBuffer::new(512, 512);
+
+// Construct a new by repeated calls to the supplied closure.
+let mut img = ImageBuffer::from_fn(512, 512, |x, y| {
+    if x % 2 == 0 {
+        image::Luma([0u8])
+    } else {
+        image::Luma([255u8])
+    }
+});
+
+// Obtain the image's width and height.
+let (width, height) = img.dimensions();
+
+// Access the pixel at coordinate (100, 100).
+let pixel = img[(100, 100)];
+
+// Or use the `get_pixel` method from the `GenericImage` trait.
+let pixel = *img.get_pixel(100, 100);
+
+// Put a pixel at coordinate (100, 100).
+img.put_pixel(100, 100, pixel);
+
+// Iterate over all pixels in the image.
+for pixel in img.pixels() {
+    // Do something with pixel.
+}
+```
+
+#### [`DynamicImage`](https://docs.rs/image/*/image/enum.DynamicImage.html)
+A `DynamicImage` is an enumeration over all supported `ImageBuffer<P>` types.
+Its exact image type is determined at runtime. It is the type returned when opening an image.
+For convenience `DynamicImage` reimplements all image processing functions.
+
+`DynamicImage` implement the `GenericImageView` and `GenericImage` traits for RGBA pixels.
+
+#### [`SubImage`](https://docs.rs/image/*/image/struct.SubImage.html)
+A view into another image, delimited by the coordinates of a rectangle.
+The coordinates given set the position of the top left corner of the rectangle.
+This is used to perform image processing functions on a subregion of an image.
+
+```rust
+use image::{GenericImageView, ImageBuffer, RgbImage, imageops};
+
+let mut img: RgbImage = ImageBuffer::new(512, 512);
+let subimg = imageops::crop(&mut img, 0, 0, 100, 100);
+
+assert!(subimg.dimensions() == (100, 100));
+```
+
+## Image Processing Functions
+These are the functions defined in the `imageops` module. All functions operate on types that implement the `GenericImage` trait.
+Note that some of the functions are very slow in debug mode. Make sure to use release mode if you experience any performance issues.
+
++ **blur**: Performs a Gaussian blur on the supplied image.
++ **brighten**: Brighten the supplied image.
++ **huerotate**: Hue rotate the supplied image by degrees.
++ **contrast**: Adjust the contrast of the supplied image.
++ **crop**: Return a mutable view into an image.
++ **filter3x3**: Perform a 3x3 box filter on the supplied image.
++ **flip_horizontal**: Flip an image horizontally.
++ **flip_vertical**: Flip an image vertically.
++ **grayscale**: Convert the supplied image to grayscale.
++ **invert**: Invert each pixel within the supplied image This function operates in place.
++ **resize**: Resize the supplied image to the specified dimensions.
++ **rotate180**: Rotate an image 180 degrees clockwise.
++ **rotate270**: Rotate an image 270 degrees clockwise.
++ **rotate90**: Rotate an image 90 degrees clockwise.
++ **unsharpen**: Performs an unsharpen mask on the supplied image.
+
+For more options, see the [`imageproc`](https://crates.io/crates/imageproc) crate.
+
+## Examples
+### Opening and Saving Images
+
+`image` provides the `open` function for opening images from a path.  The image
+format is determined from the path's file extension. An `io` module provides a
+reader which offer some more control.
+
+```rust,no_run
+use image::GenericImageView;
+
+fn main() {
+    // Use the open function to load an image from a Path.
+    // `open` returns a `DynamicImage` on success.
+    let img = image::open("tests/images/jpg/progressive/cat.jpg").unwrap();
+
+    // The dimensions method returns the images width and height.
+    println!("dimensions {:?}", img.dimensions());
+
+    // The color method returns the image's `ColorType`.
+    println!("{:?}", img.color());
+
+    // Write the contents of this image to the Writer in PNG format.
+    img.save("test.png").unwrap();
+}
+```
+
+### Generating Fractals
+
+```rust,no_run
+//! An example of generating julia fractals.
+fn main() {
+    let imgx = 800;
+    let imgy = 800;
+
+    let scalex = 3.0 / imgx as f32;
+    let scaley = 3.0 / imgy as f32;
+
+    // Create a new ImgBuf with width: imgx and height: imgy
+    let mut imgbuf = image::ImageBuffer::new(imgx, imgy);
+
+    // Iterate over the coordinates and pixels of the image
+    for (x, y, pixel) in imgbuf.enumerate_pixels_mut() {
+        let r = (0.3 * x as f32) as u8;
+        let b = (0.3 * y as f32) as u8;
+        *pixel = image::Rgb([r, 0, b]);
+    }
+
+    // A redundant loop to demonstrate reading image data
+    for x in 0..imgx {
+        for y in 0..imgy {
+            let cx = y as f32 * scalex - 1.5;
+            let cy = x as f32 * scaley - 1.5;
+
+            let c = num_complex::Complex::new(-0.4, 0.6);
+            let mut z = num_complex::Complex::new(cx, cy);
+
+            let mut i = 0;
+            while i < 255 && z.norm() <= 2.0 {
+                z = z * z + c;
+                i += 1;
+            }
+
+            let pixel = imgbuf.get_pixel_mut(x, y);
+            let image::Rgb(data) = *pixel;
+            *pixel = image::Rgb([data[0], i as u8, data[2]]);
+        }
+    }
+
+    // Save the image as “fractal.png”, the format is deduced from the path
+    imgbuf.save("fractal.png").unwrap();
+}
+```
+
+Example output:
+
+<img src="examples/fractal.png" alt="A Julia Fractal, c: -0.4 + 0.6i" width="500" />
+
+### Writing raw buffers
+If the high level interface is not needed because the image was obtained by other means, `image` provides the function `save_buffer` to save a buffer to a file.
+
+```rust,no_run
+fn main() {
+
+    let buffer: &[u8] = unimplemented!(); // Generate the image data
+
+    // Save the buffer as "image.png"
+    image::save_buffer("image.png", buffer, 800, 600, image::ColorType::Rgb8).unwrap()
+}
+```

vendor/image/benches/README.md

@@ -0,0 +1,6 @@
+# Getting started with benchmarking
+
+To run the benchmarks you need a nightly rust toolchain.
+Then you launch it with
+
+    cargo +nightly bench --features=benchmarks

vendor/image/benches/copy_from.rs

@@ -0,0 +1,14 @@
+use criterion::{black_box, criterion_group, criterion_main, Criterion};
+use image::{GenericImage, ImageBuffer, Rgba};
+
+pub fn bench_copy_from(c: &mut Criterion) {
+    let src = ImageBuffer::from_pixel(2048, 2048, Rgba([255u8, 0, 0, 255]));
+    let mut dst = ImageBuffer::from_pixel(2048, 2048, Rgba([0u8, 0, 0, 255]));
+
+    c.bench_function("copy_from", |b| {
+        b.iter(|| dst.copy_from(black_box(&src), 0, 0))
+    });
+}
+
+criterion_group!(benches, bench_copy_from);
+criterion_main!(benches);

vendor/image/benches/decode.rs

@@ -0,0 +1,109 @@
+use std::{fs, iter, path};
+
+use criterion::{criterion_group, criterion_main, Criterion};
+use image::ImageFormat;
+
+#[derive(Clone, Copy)]
+struct BenchDef {
+    dir: &'static [&'static str],
+    files: &'static [&'static str],
+    format: ImageFormat,
+}
+
+fn load_all(c: &mut Criterion) {
+    const BENCH_DEFS: &'static [BenchDef] = &[
+        BenchDef {
+            dir: &["bmp", "images"],
+            files: &[
+                "Core_1_Bit.bmp",
+                "Core_4_Bit.bmp",
+                "Core_8_Bit.bmp",
+                "rgb16.bmp",
+                "rgb24.bmp",
+                "rgb32.bmp",
+                "pal4rle.bmp",
+                "pal8rle.bmp",
+                "rgb16-565.bmp",
+                "rgb32bf.bmp",
+            ],
+            format: ImageFormat::Bmp,
+        },
+        BenchDef {
+            dir: &["gif", "simple"],
+            files: &["alpha_gif_a.gif", "sample_1.gif"],
+            format: ImageFormat::Gif,
+        },
+        BenchDef {
+            dir: &["hdr", "images"],
+            files: &["image1.hdr", "rgbr4x4.hdr"],
+            format: ImageFormat::Hdr,
+        },
+        BenchDef {
+            dir: &["ico", "images"],
+            files: &[
+                "bmp-24bpp-mask.ico",
+                "bmp-32bpp-alpha.ico",
+                "png-32bpp-alpha.ico",
+                "smile.ico",
+            ],
+            format: ImageFormat::Ico,
+        },
+        BenchDef {
+            dir: &["jpg", "progressive"],
+            files: &["3.jpg", "cat.jpg", "test.jpg"],
+            format: ImageFormat::Jpeg,
+        },
+        // TODO: pnm
+        // TODO: png
+        BenchDef {
+            dir: &["tga", "testsuite"],
+            files: &["cbw8.tga", "ctc24.tga", "ubw8.tga", "utc24.tga"],
+            format: ImageFormat::Tga,
+        },
+        BenchDef {
+            dir: &["tiff", "testsuite"],
+            files: &[
+                "hpredict.tiff",
+                "hpredict_packbits.tiff",
+                "mandrill.tiff",
+                "rgb-3c-16b.tiff",
+            ],
+            format: ImageFormat::Tiff,
+        },
+        BenchDef {
+            dir: &["webp", "images"],
+            files: &[
+                "simple-gray.webp",
+                "simple-rgb.webp",
+                "vp8x-gray.webp",
+                "vp8x-rgb.webp",
+            ],
+            format: ImageFormat::WebP,
+        },
+    ];
+
+    for bench in BENCH_DEFS {
+        bench_load(c, bench);
+    }
+}
+
+criterion_group!(benches, load_all);
+criterion_main!(benches);
+
+fn bench_load(c: &mut Criterion, def: &BenchDef) {
+    let group_name = format!("load-{:?}", def.format);
+    let mut group = c.benchmark_group(&group_name);
+    let paths = IMAGE_DIR.iter().chain(def.dir);
+
+    for file_name in def.files {
+        let path: path::PathBuf = paths.clone().chain(iter::once(file_name)).collect();
+        let buf = fs::read(path).unwrap();
+        group.bench_function(file_name.to_owned(), |b| {
+            b.iter(|| {
+                image::load_from_memory_with_format(&buf, def.format).unwrap();
+            })
+        });
+    }
+}
+
+const IMAGE_DIR: [&'static str; 3] = [".", "tests", "images"];

vendor/image/benches/encode.rs

@@ -0,0 +1,134 @@
+extern crate criterion;
+
+use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion};
+use image::{codecs::bmp::BmpEncoder, codecs::jpeg::JpegEncoder, ColorType};
+
+use std::fs::File;
+use std::io::{BufWriter, Seek, SeekFrom, Write};
+
+trait Encoder {
+    fn encode_raw(&self, into: &mut Vec<u8>, im: &[u8], dims: u32, color: ColorType);
+    fn encode_bufvec(&self, into: &mut Vec<u8>, im: &[u8], dims: u32, color: ColorType);
+    fn encode_file(&self, file: &File, im: &[u8], dims: u32, color: ColorType);
+}
+
+#[derive(Clone, Copy)]
+struct BenchDef {
+    with: &'static dyn Encoder,
+    name: &'static str,
+    sizes: &'static [u32],
+    colors: &'static [ColorType],
+}
+
+fn encode_all(c: &mut Criterion) {
+    const BENCH_DEFS: &'static [BenchDef] = &[
+        BenchDef {
+            with: &Bmp,
+            name: "bmp",
+            sizes: &[100u32, 200, 400],
+            colors: &[ColorType::L8, ColorType::Rgb8, ColorType::Rgba8],
+        },
+        BenchDef {
+            with: &Jpeg,
+            name: "jpeg",
+            sizes: &[64u32, 128, 256],
+            colors: &[ColorType::L8, ColorType::Rgb8, ColorType::Rgba8],
+        },
+    ];
+
+    for definition in BENCH_DEFS {
+        encode_definition(c, definition)
+    }
+}
+
+criterion_group!(benches, encode_all);
+criterion_main!(benches);
+
+type BenchGroup<'a> = criterion::BenchmarkGroup<'a, criterion::measurement::WallTime>;
+
+/// Benchmarks encoding a zeroed image.
+///
+/// For compressed formats this is surely not representative of encoding a normal image but it's a
+/// start for benchmarking.
+fn encode_zeroed(group: &mut BenchGroup, with: &dyn Encoder, size: u32, color: ColorType) {
+    let bytes = size as usize * usize::from(color.bytes_per_pixel());
+    let im = vec![0; bytes * bytes];
+
+    group.bench_with_input(
+        BenchmarkId::new(format!("zero-{:?}-rawvec", color), size),
+        &im,
+        |b, image| {
+            let mut v = vec![];
+            with.encode_raw(&mut v, &im, size, color);
+            b.iter(|| with.encode_raw(&mut v, image, size, color));
+        },
+    );
+    group.bench_with_input(
+        BenchmarkId::new(format!("zero-{:?}-bufvec", color), size),
+        &im,
+        |b, image| {
+            let mut v = vec![];
+            with.encode_raw(&mut v, &im, size, color);
+            b.iter(|| with.encode_bufvec(&mut v, image, size, color));
+        },
+    );
+    group.bench_with_input(
+        BenchmarkId::new(format!("zero-{:?}-file", color), size),
+        &im,
+        |b, image| {
+            let file = File::create("temp.bmp").unwrap();
+            b.iter(|| with.encode_file(&file, image, size, color));
+        },
+    );
+}
+
+fn encode_definition(criterion: &mut Criterion, def: &BenchDef) {
+    let mut group = criterion.benchmark_group(format!("encode-{}", def.name));
+
+    for &color in def.colors {
+        for &size in def.sizes {
+            encode_zeroed(&mut group, def.with, size, color);
+        }
+    }
+}
+
+struct Bmp;
+
+struct Jpeg;
+
+trait EncoderBase {
+    fn encode(&self, into: impl Write, im: &[u8], dims: u32, color: ColorType);
+}
+
+impl<T: EncoderBase> Encoder for T {
+    fn encode_raw(&self, into: &mut Vec<u8>, im: &[u8], dims: u32, color: ColorType) {
+        into.clear();
+        self.encode(into, im, dims, color);
+    }
+
+    fn encode_bufvec(&self, into: &mut Vec<u8>, im: &[u8], dims: u32, color: ColorType) {
+        into.clear();
+        let buf = BufWriter::new(into);
+        self.encode(buf, im, dims, color);
+    }
+
+    fn encode_file(&self, mut file: &File, im: &[u8], dims: u32, color: ColorType) {
+        file.seek(SeekFrom::Start(0)).unwrap();
+        let buf = BufWriter::new(file);
+        self.encode(buf, im, dims, color);
+    }
+}
+
+impl EncoderBase for Bmp {
+    fn encode(&self, mut into: impl Write, im: &[u8], size: u32, color: ColorType) {
+        let mut x = BmpEncoder::new(&mut into);
+        x.encode(im, size, size, color).unwrap();
+    }
+}
+
+impl EncoderBase for Jpeg {
+    fn encode(&self, mut into: impl Write, im: &[u8], size: u32, color: ColorType) {
+        let mut x = JpegEncoder::new(&mut into);
+        x.encode(im, size, size, color).unwrap();
+    }
+}

vendor/image/deny.toml

@@ -0,0 +1,38 @@
+# https://embarkstudios.github.io/cargo-deny/
+
+targets = [
+    { triple = "aarch64-apple-darwin" },
+    { triple = "aarch64-linux-android" },
+    { triple = "x86_64-apple-darwin" },
+    { triple = "x86_64-pc-windows-msvc" },
+    { triple = "x86_64-unknown-linux-gnu" },
+    { triple = "x86_64-unknown-linux-musl" },
+]
+
+
+[advisories]
+vulnerability = "deny"
+unmaintained = "warn"
+yanked = "deny"
+ignore = []
+
+
+[bans]
+multiple-versions = "deny"
+wildcards = "allow" # at least until https://github.com/EmbarkStudios/cargo-deny/issues/241 is fixed
+deny = []
+skip = [
+    { name = "num-derive" } # ravif transatively depends on 0.3 and 0.4.
+]
+skip-tree = [
+    { name = "criterion" },  # dev-dependency
+    { name = "quickcheck" }, # dev-dependency
+    { name = "dav1d" }, # TODO: needs upgrade
+    { name = "clap" },
+]
+
+
+[licenses]
+unlicensed = "allow"
+allow-osi-fsf-free = "either"
+copyleft = "allow"

vendor/image/docs/2019-04-23-memory-unsafety.md

@@ -0,0 +1,54 @@
+# Advisory about potential memory unsafety issues
+
+[While reviewing][i885] some `unsafe Vec::from_raw_parts` operations within the
+library, trying to justify their existence with stronger reasoning, we noticed
+that they instead did not meet the required conditions set by the standard
+library. This unsoundness was quickly removed, but we noted that the same
+unjustified reasoning had been applied by a dependency introduced in `0.21`.
+
+For efficiency reasons, we had tried to reuse the allocations made by decoders
+for the buffer of the final image. However, that process is error prone. Most
+image decoding algorithms change the representation type of color samples to
+some degree. Notably, the output pixel type may have a different size and
+alignment than the type used in the temporary decoding buffer. In this specific
+instance, the `ImageBuffer` of the output expects a linear arrangement of `u8`
+samples while the implementation of the `hdr` decoder uses a pixel
+representation of `Rgb<u8>`, which has three times the size. One of the
+requirements of `Vec::from_raw_parts` reads:
+
+> ptr's T needs to have the same size and alignment as it was allocated with.
+
+This requirement is not present on slices `[T]`, as it is motivated by the
+allocator interface. The validity invariant of a reference and slice only
+requires the correct alignment here, which was considered in the design of
+`Rgb<_>` by giving it a well-defined representation, `#[repr(C)]`. But
+critically, this does not guarantee that we can reuse the existing allocation
+through effectively transmuting a `Vec<_>`!
+
+The actual impact of this issue is, in real world implementations, limited to
+allocators which handle allocations for types of size `1` and `3`/`4`
+differently. To the best of my knowledge, this does not apply to `jemalloc` and
+the `libc` allocator. However, we decided to proceed with caution here.
+
+## Lessons for the future
+
+New library dependencies will be under a stricter policy. Not only would they
+need to be justified by functionality but also require at least some level of
+reasoning how they solve that problem better than alternatives. Some appearance
+of maintenance, or the existence of `#[deny(unsafe)]`, will help. We'll
+additionally look into existing dependencies trying to identify similar issues
+and minimizing the potential surface for implementation risks.
+
+## Sound and safe buffer reuse
+
+It seems that the `Vec` representation is entirely unfit for buffer reuse in
+the style which an image library requires. In particular, using pixel types of
+different sizes is likely common to handle either whole (encoded) pixels or
+individual samples. Thus, we started a new sub-project to address this use
+case, [image-canvas][image-canvas]. Contributions and review of its safety are
+very welcome, we ask for the communities help here. The release of `v0.1` will
+not occur until at least one such review has occurred.
+
+
+[i885]: https://github.com/image-rs/image/pull/885
+[image-canvas]: https://github.com/image-rs/canvas

vendor/image/release.sh

@@ -0,0 +1,24 @@
+#!/bin/bash
+# Checks automatic preconditions for a release
+determine_new_version() {
+	grep "version = " Cargo.toml | sed -Ee 's/version = "(.*)"/\1/' | head -1
+}
+
+check_notexists_version() {
+	# Does the api information start with: '{"errors":'
+	[[ $(wget "https://crates.io/api/v1/crates/image/$1" -qO -) == "{\"errors\":"* ]]
+}
+
+check_release_description() {
+	major=${1%%.*}
+	minor_patch=${1#$major.}
+	minor=${minor_patch%%.*}
+	patch=${minor_patch#$minor.}
+	# We just need to find a fitting header line
+	grep -Eq "^### Version ${major}.${minor}$" CHANGES.md
+}
+
+version="$(determine_new_version)"
+check_release_description $version || { echo "Version does not have a release description"; exit 1; }
+check_notexists_version $version || { echo "Version $version appears already published"; exit 1; }
+

vendor/image/src/animation.rs

@@ -0,0 +1,342 @@
+use std::iter::Iterator;
+use std::time::Duration;
+
+use num_rational::Ratio;
+
+use crate::error::ImageResult;
+use crate::RgbaImage;
+
+/// An implementation dependent iterator, reading the frames as requested
+pub struct Frames<'a> {
+    iterator: Box<dyn Iterator<Item = ImageResult<Frame>> + 'a>,
+}
+
+impl<'a> Frames<'a> {
+    /// Creates a new `Frames` from an implementation specific iterator.
+    pub fn new(iterator: Box<dyn Iterator<Item = ImageResult<Frame>> + 'a>) -> Self {
+        Frames { iterator }
+    }
+
+    /// Steps through the iterator from the current frame until the end and pushes each frame into
+    /// a `Vec`.
+    /// If en error is encountered that error is returned instead.
+    ///
+    /// Note: This is equivalent to `Frames::collect::<ImageResult<Vec<Frame>>>()`
+    pub fn collect_frames(self) -> ImageResult<Vec<Frame>> {
+        self.collect()
+    }
+}
+
+impl<'a> Iterator for Frames<'a> {
+    type Item = ImageResult<Frame>;
+    fn next(&mut self) -> Option<ImageResult<Frame>> {
+        self.iterator.next()
+    }
+}
+
+/// A single animation frame
+#[derive(Clone)]
+pub struct Frame {
+    /// Delay between the frames in milliseconds
+    delay: Delay,
+    /// x offset
+    left: u32,
+    /// y offset
+    top: u32,
+    buffer: RgbaImage,
+}
+
+/// The delay of a frame relative to the previous one.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd)]
+pub struct Delay {
+    ratio: Ratio<u32>,
+}
+
+impl Frame {
+    /// Constructs a new frame without any delay.
+    pub fn new(buffer: RgbaImage) -> Frame {
+        Frame {
+            delay: Delay::from_ratio(Ratio::from_integer(0)),
+            left: 0,
+            top: 0,
+            buffer,
+        }
+    }
+
+    /// Constructs a new frame
+    pub fn from_parts(buffer: RgbaImage, left: u32, top: u32, delay: Delay) -> Frame {
+        Frame {
+            delay,
+            left,
+            top,
+            buffer,
+        }
+    }
+
+    /// Delay of this frame
+    pub fn delay(&self) -> Delay {
+        self.delay
+    }
+
+    /// Returns the image buffer
+    pub fn buffer(&self) -> &RgbaImage {
+        &self.buffer
+    }
+
+    /// Returns a mutable image buffer
+    pub fn buffer_mut(&mut self) -> &mut RgbaImage {
+        &mut self.buffer
+    }
+
+    /// Returns the image buffer
+    pub fn into_buffer(self) -> RgbaImage {
+        self.buffer
+    }
+
+    /// Returns the x offset
+    pub fn left(&self) -> u32 {
+        self.left
+    }
+
+    /// Returns the y offset
+    pub fn top(&self) -> u32 {
+        self.top
+    }
+}
+
+impl Delay {
+    /// Create a delay from a ratio of milliseconds.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use image::Delay;
+    /// let delay_10ms = Delay::from_numer_denom_ms(10, 1);
+    /// ```
+    pub fn from_numer_denom_ms(numerator: u32, denominator: u32) -> Self {
+        Delay {
+            ratio: Ratio::new_raw(numerator, denominator),
+        }
+    }
+
+    /// Convert from a duration, clamped between 0 and an implemented defined maximum.
+    ///
+    /// The maximum is *at least* `i32::MAX` milliseconds. It should be noted that the accuracy of
+    /// the result may be relative and very large delays have a coarse resolution.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use std::time::Duration;
+    /// use image::Delay;
+    ///
+    /// let duration = Duration::from_millis(20);
+    /// let delay = Delay::from_saturating_duration(duration);
+    /// ```
+    pub fn from_saturating_duration(duration: Duration) -> Self {
+        // A few notes: The largest number we can represent as a ratio is u32::MAX but we can
+        // sometimes represent much smaller numbers.
+        //
+        // We can represent duration as `millis+a/b` (where a < b, b > 0).
+        // We must thus bound b with `b·millis + (b-1) <= u32::MAX` or
+        // > `0 < b <= (u32::MAX + 1)/(millis + 1)`
+        // Corollary: millis <= u32::MAX
+
+        const MILLIS_BOUND: u128 = u32::max_value() as u128;
+
+        let millis = duration.as_millis().min(MILLIS_BOUND);
+        let submillis = (duration.as_nanos() % 1_000_000) as u32;
+
+        let max_b = if millis > 0 {
+            ((MILLIS_BOUND + 1) / (millis + 1)) as u32
+        } else {
+            MILLIS_BOUND as u32
+        };
+        let millis = millis as u32;
+
+        let (a, b) = Self::closest_bounded_fraction(max_b, submillis, 1_000_000);
+        Self::from_numer_denom_ms(a + b * millis, b)
+    }
+
+    /// The numerator and denominator of the delay in milliseconds.
+    ///
+    /// This is guaranteed to be an exact conversion if the `Delay` was previously created with the
+    /// `from_numer_denom_ms` constructor.
+    pub fn numer_denom_ms(self) -> (u32, u32) {
+        (*self.ratio.numer(), *self.ratio.denom())
+    }
+
+    pub(crate) fn from_ratio(ratio: Ratio<u32>) -> Self {
+        Delay { ratio }
+    }
+
+    pub(crate) fn into_ratio(self) -> Ratio<u32> {
+        self.ratio
+    }
+
+    /// Given some fraction, compute an approximation with denominator bounded.
+    ///
+    /// Note that `denom_bound` bounds nominator and denominator of all intermediate
+    /// approximations and the end result.
+    fn closest_bounded_fraction(denom_bound: u32, nom: u32, denom: u32) -> (u32, u32) {
+        use std::cmp::Ordering::{self, *};
+        assert!(0 < denom);
+        assert!(0 < denom_bound);
+        assert!(nom < denom);
+
+        // Avoid a few type troubles. All intermediate results are bounded by `denom_bound` which
+        // is in turn bounded by u32::MAX. Representing with u64 allows multiplication of any two
+        // values without fears of overflow.
+
+        // Compare two fractions whose parts fit into a u32.
+        fn compare_fraction((an, ad): (u64, u64), (bn, bd): (u64, u64)) -> Ordering {
+            (an * bd).cmp(&(bn * ad))
+        }
+
+        // Computes the nominator of the absolute difference between two such fractions.
+        fn abs_diff_nom((an, ad): (u64, u64), (bn, bd): (u64, u64)) -> u64 {
+            let c0 = an * bd;
+            let c1 = ad * bn;
+
+            let d0 = c0.max(c1);
+            let d1 = c0.min(c1);
+            d0 - d1
+        }
+
+        let exact = (u64::from(nom), u64::from(denom));
+        // The lower bound fraction, numerator and denominator.
+        let mut lower = (0u64, 1u64);
+        // The upper bound fraction, numerator and denominator.
+        let mut upper = (1u64, 1u64);
+        // The closest approximation for now.
+        let mut guess = (u64::from(nom * 2 > denom), 1u64);
+
+        // loop invariant: ad, bd <= denom_bound
+        // iterates the Farey sequence.
+        loop {
+            // Break if we are done.
+            if compare_fraction(guess, exact) == Equal {
+                break;
+            }
+
+            // Break if next Farey number is out-of-range.
+            if u64::from(denom_bound) - lower.1 < upper.1 {
+                break;
+            }
+
+            // Next Farey approximation n between a and b
+            let next = (lower.0 + upper.0, lower.1 + upper.1);
+            // if F < n then replace the upper bound, else replace lower.
+            if compare_fraction(exact, next) == Less {
+                upper = next;
+            } else {
+                lower = next;
+            }
+
+            // Now correct the closest guess.
+            // In other words, if |c - f| > |n - f| then replace it with the new guess.
+            // This favors the guess with smaller denominator on equality.
+
+            // |g - f| = |g_diff_nom|/(gd*fd);
+            let g_diff_nom = abs_diff_nom(guess, exact);
+            // |n - f| = |n_diff_nom|/(nd*fd);
+            let n_diff_nom = abs_diff_nom(next, exact);
+
+            // The difference |n - f| is smaller than |g - f| if either the integral part of the
+            // fraction |n_diff_nom|/nd is smaller than the one of |g_diff_nom|/gd or if they are
+            // the same but the fractional part is larger.
+            if match (n_diff_nom / next.1).cmp(&(g_diff_nom / guess.1)) {
+                Less => true,
+                Greater => false,
+                // Note that the nominator for the fractional part is smaller than its denominator
+                // which is smaller than u32 and can't overflow the multiplication with the other
+                // denominator, that is we can compare these fractions by multiplication with the
+                // respective other denominator.
+                Equal => {
+                    compare_fraction(
+                        (n_diff_nom % next.1, next.1),
+                        (g_diff_nom % guess.1, guess.1),
+                    ) == Less
+                }
+            } {
+                guess = next;
+            }
+        }
+
+        (guess.0 as u32, guess.1 as u32)
+    }
+}
+
+impl From<Delay> for Duration {
+    fn from(delay: Delay) -> Self {
+        let ratio = delay.into_ratio();
+        let ms = ratio.to_integer();
+        let rest = ratio.numer() % ratio.denom();
+        let nanos = (u64::from(rest) * 1_000_000) / u64::from(*ratio.denom());
+        Duration::from_millis(ms.into()) + Duration::from_nanos(nanos)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{Delay, Duration, Ratio};
+
+    #[test]
+    fn simple() {
+        let second = Delay::from_numer_denom_ms(1000, 1);
+        assert_eq!(Duration::from(second), Duration::from_secs(1));
+    }
+
+    #[test]
+    fn fps_30() {
+        let thirtieth = Delay::from_numer_denom_ms(1000, 30);
+        let duration = Duration::from(thirtieth);
+        assert_eq!(duration.as_secs(), 0);
+        assert_eq!(duration.subsec_millis(), 33);
+        assert_eq!(duration.subsec_nanos(), 33_333_333);
+    }
+
+    #[test]
+    fn duration_outlier() {
+        let oob = Duration::from_secs(0xFFFF_FFFF);
+        let delay = Delay::from_saturating_duration(oob);
+        assert_eq!(delay.numer_denom_ms(), (0xFFFF_FFFF, 1));
+    }
+
+    #[test]
+    fn duration_approx() {
+        let oob = Duration::from_millis(0xFFFF_FFFF) + Duration::from_micros(1);
+        let delay = Delay::from_saturating_duration(oob);
+        assert_eq!(delay.numer_denom_ms(), (0xFFFF_FFFF, 1));
+
+        let inbounds = Duration::from_millis(0xFFFF_FFFF) - Duration::from_micros(1);
+        let delay = Delay::from_saturating_duration(inbounds);
+        assert_eq!(delay.numer_denom_ms(), (0xFFFF_FFFF, 1));
+
+        let fine =
+            Duration::from_millis(0xFFFF_FFFF / 1000) + Duration::from_micros(0xFFFF_FFFF % 1000);
+        let delay = Delay::from_saturating_duration(fine);
+        // Funnily, 0xFFFF_FFFF is divisble by 5, thus we compare with a `Ratio`.
+        assert_eq!(delay.into_ratio(), Ratio::new(0xFFFF_FFFF, 1000));
+    }
+
+    #[test]
+    fn precise() {
+        // The ratio has only 32 bits in the numerator, too imprecise to get more than 11 digits
+        // correct. But it may be expressed as 1_000_000/3 instead.
+        let exceed = Duration::from_secs(333) + Duration::from_nanos(333_333_333);
+        let delay = Delay::from_saturating_duration(exceed);
+        assert_eq!(Duration::from(delay), exceed);
+    }
+
+    #[test]
+    fn small() {
+        // Not quite a delay of `1 ms`.
+        let delay = Delay::from_numer_denom_ms(1 << 16, (1 << 16) + 1);
+        let duration = Duration::from(delay);
+        assert_eq!(duration.as_millis(), 0);
+        // Not precisely the original but should be smaller than 0.
+        let delay = Delay::from_saturating_duration(duration);
+        assert_eq!(delay.into_ratio().to_integer(), 0);
+    }
+}

vendor/image/src/codecs/avif/decoder.rs

@@ -0,0 +1,177 @@
+//! Decoding of AVIF images.
+///
+/// The [AVIF] specification defines an image derivative of the AV1 bitstream, an open video codec.
+///
+/// [AVIF]: https://aomediacodec.github.io/av1-avif/
+use std::convert::TryFrom;
+use std::error::Error;
+use std::io::{self, Cursor, Read};
+use std::marker::PhantomData;
+use std::mem;
+
+use crate::error::DecodingError;
+use crate::{ColorType, ImageDecoder, ImageError, ImageFormat, ImageResult};
+
+use dav1d::{PixelLayout, PlanarImageComponent};
+use dcv_color_primitives as dcp;
+use mp4parse::{read_avif, ParseStrictness};
+
+fn error_map<E: Into<Box<dyn Error + Send + Sync>>>(err: E) -> ImageError {
+    ImageError::Decoding(DecodingError::new(ImageFormat::Avif.into(), err))
+}
+
+/// AVIF Decoder.
+///
+/// Reads one image into the chosen input.
+pub struct AvifDecoder<R> {
+    inner: PhantomData<R>,
+    picture: dav1d::Picture,
+    alpha_picture: Option<dav1d::Picture>,
+    icc_profile: Option<Vec<u8>>,
+}
+
+impl<R: Read> AvifDecoder<R> {
+    /// Create a new decoder that reads its input from `r`.
+    pub fn new(mut r: R) -> ImageResult<Self> {
+        let ctx = read_avif(&mut r, ParseStrictness::Normal).map_err(error_map)?;
+        let coded = ctx.primary_item_coded_data().unwrap_or_default();
+
+        let mut primary_decoder = dav1d::Decoder::new();
+        primary_decoder
+            .send_data(coded, None, None, None)
+            .map_err(error_map)?;
+        let picture = primary_decoder.get_picture().map_err(error_map)?;
+        let alpha_item = ctx.alpha_item_coded_data().unwrap_or_default();
+        let alpha_picture = if !alpha_item.is_empty() {
+            let mut alpha_decoder = dav1d::Decoder::new();
+            alpha_decoder
+                .send_data(alpha_item, None, None, None)
+                .map_err(error_map)?;
+            Some(alpha_decoder.get_picture().map_err(error_map)?)
+        } else {
+            None
+        };
+        let icc_profile = ctx
+            .icc_colour_information()
+            .map(|x| x.ok().unwrap_or_default())
+            .map(|x| x.to_vec());
+
+        assert_eq!(picture.bit_depth(), 8);
+        Ok(AvifDecoder {
+            inner: PhantomData,
+            picture,
+            alpha_picture,
+            icc_profile,
+        })
+    }
+}
+
+/// Wrapper struct around a `Cursor<Vec<u8>>`
+pub struct AvifReader<R>(Cursor<Vec<u8>>, PhantomData<R>);
+impl<R> Read for AvifReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        self.0.read(buf)
+    }
+    fn read_to_end(&mut self, buf: &mut Vec<u8>) -> io::Result<usize> {
+        if self.0.position() == 0 && buf.is_empty() {
+            mem::swap(buf, self.0.get_mut());
+            Ok(buf.len())
+        } else {
+            self.0.read_to_end(buf)
+        }
+    }
+}
+
+impl<'a, R: 'a + Read> ImageDecoder<'a> for AvifDecoder<R> {
+    type Reader = AvifReader<R>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        (self.picture.width(), self.picture.height())
+    }
+
+    fn color_type(&self) -> ColorType {
+        ColorType::Rgba8
+    }
+
+    fn icc_profile(&mut self) -> Option<Vec<u8>> {
+        self.icc_profile.clone()
+    }
+
+    fn into_reader(self) -> ImageResult<Self::Reader> {
+        let plane = self.picture.plane(PlanarImageComponent::Y);
+        Ok(AvifReader(
+            Cursor::new(plane.as_ref().to_vec()),
+            PhantomData,
+        ))
+    }
+
+    fn read_image(self, buf: &mut [u8]) -> ImageResult<()> {
+        assert_eq!(u64::try_from(buf.len()), Ok(self.total_bytes()));
+
+        dcp::initialize();
+
+        if self.picture.pixel_layout() != PixelLayout::I400 {
+            let pixel_format = match self.picture.pixel_layout() {
+                PixelLayout::I400 => todo!(),
+                PixelLayout::I420 => dcp::PixelFormat::I420,
+                PixelLayout::I422 => dcp::PixelFormat::I422,
+                PixelLayout::I444 => dcp::PixelFormat::I444,
+                PixelLayout::Unknown => panic!("Unknown pixel layout"),
+            };
+            let src_format = dcp::ImageFormat {
+                pixel_format,
+                color_space: dcp::ColorSpace::Bt601,
+                num_planes: 3,
+            };
+            let dst_format = dcp::ImageFormat {
+                pixel_format: dcp::PixelFormat::Rgba,
+                color_space: dcp::ColorSpace::Lrgb,
+                num_planes: 1,
+            };
+            let (width, height) = self.dimensions();
+            let planes = &[
+                self.picture.plane(PlanarImageComponent::Y),
+                self.picture.plane(PlanarImageComponent::U),
+                self.picture.plane(PlanarImageComponent::V),
+            ];
+            let src_buffers = planes.iter().map(AsRef::as_ref).collect::<Vec<_>>();
+            let strides = &[
+                self.picture.stride(PlanarImageComponent::Y) as usize,
+                self.picture.stride(PlanarImageComponent::U) as usize,
+                self.picture.stride(PlanarImageComponent::V) as usize,
+            ];
+            let dst_buffers = &mut [&mut buf[..]];
+            dcp::convert_image(
+                width,
+                height,
+                &src_format,
+                Some(strides),
+                &src_buffers,
+                &dst_format,
+                None,
+                dst_buffers,
+            )
+            .map_err(error_map)?;
+        } else {
+            let plane = self.picture.plane(PlanarImageComponent::Y);
+            buf.copy_from_slice(plane.as_ref());
+        }
+
+        if let Some(picture) = self.alpha_picture {
+            assert_eq!(picture.pixel_layout(), PixelLayout::I400);
+            let stride = picture.stride(PlanarImageComponent::Y) as usize;
+            let plane = picture.plane(PlanarImageComponent::Y);
+            let width = picture.width();
+            for (buf, slice) in Iterator::zip(
+                buf.chunks_exact_mut(width as usize * 4),
+                plane.as_ref().chunks_exact(stride),
+            ) {
+                for i in 0..width as usize {
+                    buf[3 + i * 4] = slice[i];
+                }
+            }
+        }
+
+        Ok(())
+    }
+}

vendor/image/src/codecs/avif/encoder.rs

@@ -0,0 +1,274 @@
+//! Encoding of AVIF images.
+///
+/// The [AVIF] specification defines an image derivative of the AV1 bitstream, an open video codec.
+///
+/// [AVIF]: https://aomediacodec.github.io/av1-avif/
+use std::borrow::Cow;
+use std::cmp::min;
+use std::io::Write;
+
+use crate::buffer::ConvertBuffer;
+use crate::color::{FromColor, Luma, LumaA, Rgb, Rgba};
+use crate::error::{
+    EncodingError, ParameterError, ParameterErrorKind, UnsupportedError, UnsupportedErrorKind,
+};
+use crate::{ColorType, ImageBuffer, ImageEncoder, ImageFormat, Pixel};
+use crate::{ImageError, ImageResult};
+
+use bytemuck::{try_cast_slice, try_cast_slice_mut, Pod, PodCastError};
+use num_traits::Zero;
+use ravif::{Encoder, Img, RGB8, RGBA8};
+use rgb::AsPixels;
+
+/// AVIF Encoder.
+///
+/// Writes one image into the chosen output.
+pub struct AvifEncoder<W> {
+    inner: W,
+    encoder: Encoder,
+}
+
+/// An enumeration over supported AVIF color spaces
+#[derive(Debug, Copy, Clone, PartialEq, Eq)]
+#[non_exhaustive]
+pub enum ColorSpace {
+    /// sRGB colorspace
+    Srgb,
+    /// BT.709 colorspace
+    Bt709,
+}
+
+impl ColorSpace {
+    fn to_ravif(self) -> ravif::ColorSpace {
+        match self {
+            Self::Srgb => ravif::ColorSpace::RGB,
+            Self::Bt709 => ravif::ColorSpace::YCbCr,
+        }
+    }
+}
+
+enum RgbColor<'buf> {
+    Rgb8(Img<&'buf [RGB8]>),
+    Rgba8(Img<&'buf [RGBA8]>),
+}
+
+impl<W: Write> AvifEncoder<W> {
+    /// Create a new encoder that writes its output to `w`.
+    pub fn new(w: W) -> Self {
+        AvifEncoder::new_with_speed_quality(w, 4, 80) // `cavif` uses these defaults
+    }
+
+    /// Create a new encoder with specified speed and quality, that writes its output to `w`.
+    /// `speed` accepts a value in the range 0-10, where 0 is the slowest and 10 is the fastest.
+    /// `quality` accepts a value in the range 0-100, where 0 is the worst and 100 is the best.
+    pub fn new_with_speed_quality(w: W, speed: u8, quality: u8) -> Self {
+        // Clamp quality and speed to range
+        let quality = min(quality, 100);
+        let speed = min(speed, 10);
+
+        let encoder = Encoder::new()
+            .with_quality(f32::from(quality))
+            .with_alpha_quality(f32::from(quality))
+            .with_speed(speed);
+
+        AvifEncoder { inner: w, encoder }
+    }
+
+    /// Encode with the specified `color_space`.
+    pub fn with_colorspace(mut self, color_space: ColorSpace) -> Self {
+        self.encoder = self
+            .encoder
+            .with_internal_color_space(color_space.to_ravif());
+        self
+    }
+
+    /// Configures `rayon` thread pool size.
+    /// The default `None` is to use all threads in the default `rayon` thread pool.
+    pub fn with_num_threads(mut self, num_threads: Option<usize>) -> Self {
+        self.encoder = self.encoder.with_num_threads(num_threads);
+        self
+    }
+}
+
+impl<W: Write> ImageEncoder for AvifEncoder<W> {
+    /// Encode image data with the indicated color type.
+    ///
+    /// The encoder currently requires all data to be RGBA8, it will be converted internally if
+    /// necessary. When data is suitably aligned, i.e. u16 channels to two bytes, then the
+    /// conversion may be more efficient.
+    fn write_image(
+        mut self,
+        data: &[u8],
+        width: u32,
+        height: u32,
+        color: ColorType,
+    ) -> ImageResult<()> {
+        self.set_color(color);
+        // `ravif` needs strongly typed data so let's convert. We can either use a temporarily
+        // owned version in our own buffer or zero-copy if possible by using the input buffer.
+        // This requires going through `rgb`.
+        let mut fallback = vec![]; // This vector is used if we need to do a color conversion.
+        let result = match Self::encode_as_img(&mut fallback, data, width, height, color)? {
+            RgbColor::Rgb8(buffer) => self.encoder.encode_rgb(buffer),
+            RgbColor::Rgba8(buffer) => self.encoder.encode_rgba(buffer),
+        };
+        let data = result.map_err(|err| {
+            ImageError::Encoding(EncodingError::new(ImageFormat::Avif.into(), err))
+        })?;
+        self.inner.write_all(&data.avif_file)?;
+        Ok(())
+    }
+}
+
+impl<W: Write> AvifEncoder<W> {
+    // Does not currently do anything. Mirrors behaviour of old config function.
+    fn set_color(&mut self, _color: ColorType) {
+        // self.config.color_space = ColorSpace::RGB;
+    }
+
+    fn encode_as_img<'buf>(
+        fallback: &'buf mut Vec<u8>,
+        data: &'buf [u8],
+        width: u32,
+        height: u32,
+        color: ColorType,
+    ) -> ImageResult<RgbColor<'buf>> {
+        // Error wrapping utility for color dependent buffer dimensions.
+        fn try_from_raw<P: Pixel + 'static>(
+            data: &[P::Subpixel],
+            width: u32,
+            height: u32,
+        ) -> ImageResult<ImageBuffer<P, &[P::Subpixel]>> {
+            ImageBuffer::from_raw(width, height, data).ok_or_else(|| {
+                ImageError::Parameter(ParameterError::from_kind(
+                    ParameterErrorKind::DimensionMismatch,
+                ))
+            })
+        }
+
+        // Convert to target color type using few buffer allocations.
+        fn convert_into<'buf, P>(
+            buf: &'buf mut Vec<u8>,
+            image: ImageBuffer<P, &[P::Subpixel]>,
+        ) -> Img<&'buf [RGBA8]>
+        where
+            P: Pixel + 'static,
+            Rgba<u8>: FromColor<P>,
+        {
+            let (width, height) = image.dimensions();
+            // TODO: conversion re-using the target buffer?
+            let image: ImageBuffer<Rgba<u8>, _> = image.convert();
+            *buf = image.into_raw();
+            Img::new(buf.as_pixels(), width as usize, height as usize)
+        }
+
+        // Cast the input slice using few buffer allocations if possible.
+        // In particular try not to allocate if the caller did the infallible reverse.
+        fn cast_buffer<Channel>(buf: &[u8]) -> ImageResult<Cow<[Channel]>>
+        where
+            Channel: Pod + Zero,
+        {
+            match try_cast_slice(buf) {
+                Ok(slice) => Ok(Cow::Borrowed(slice)),
+                Err(PodCastError::OutputSliceWouldHaveSlop) => Err(ImageError::Parameter(
+                    ParameterError::from_kind(ParameterErrorKind::DimensionMismatch),
+                )),
+                Err(PodCastError::TargetAlignmentGreaterAndInputNotAligned) => {
+                    // Sad, but let's allocate.
+                    // bytemuck checks alignment _before_ slop but size mismatch before this..
+                    if buf.len() % std::mem::size_of::<Channel>() != 0 {
+                        Err(ImageError::Parameter(ParameterError::from_kind(
+                            ParameterErrorKind::DimensionMismatch,
+                        )))
+                    } else {
+                        let len = buf.len() / std::mem::size_of::<Channel>();
+                        let mut data = vec![Channel::zero(); len];
+                        let view = try_cast_slice_mut::<_, u8>(data.as_mut_slice()).unwrap();
+                        view.copy_from_slice(buf);
+                        Ok(Cow::Owned(data))
+                    }
+                }
+                Err(err) => {
+                    // Are you trying to encode a ZST??
+                    Err(ImageError::Parameter(ParameterError::from_kind(
+                        ParameterErrorKind::Generic(format!("{:?}", err)),
+                    )))
+                }
+            }
+        }
+
+        match color {
+            ColorType::Rgb8 => {
+                // ravif doesn't do any checks but has some asserts, so we do the checks.
+                let img = try_from_raw::<Rgb<u8>>(data, width, height)?;
+                // Now, internally ravif uses u32 but it takes usize. We could do some checked
+                // conversion but instead we use that a non-empty image must be addressable.
+                if img.pixels().len() == 0 {
+                    return Err(ImageError::Parameter(ParameterError::from_kind(
+                        ParameterErrorKind::DimensionMismatch,
+                    )));
+                }
+
+                Ok(RgbColor::Rgb8(Img::new(
+                    rgb::AsPixels::as_pixels(data),
+                    width as usize,
+                    height as usize,
+                )))
+            }
+            ColorType::Rgba8 => {
+                // ravif doesn't do any checks but has some asserts, so we do the checks.
+                let img = try_from_raw::<Rgba<u8>>(data, width, height)?;
+                // Now, internally ravif uses u32 but it takes usize. We could do some checked
+                // conversion but instead we use that a non-empty image must be addressable.
+                if img.pixels().len() == 0 {
+                    return Err(ImageError::Parameter(ParameterError::from_kind(
+                        ParameterErrorKind::DimensionMismatch,
+                    )));
+                }
+
+                Ok(RgbColor::Rgba8(Img::new(
+                    rgb::AsPixels::as_pixels(data),
+                    width as usize,
+                    height as usize,
+                )))
+            }
+            // we need a separate buffer..
+            ColorType::L8 => {
+                let image = try_from_raw::<Luma<u8>>(data, width, height)?;
+                Ok(RgbColor::Rgba8(convert_into(fallback, image)))
+            }
+            ColorType::La8 => {
+                let image = try_from_raw::<LumaA<u8>>(data, width, height)?;
+                Ok(RgbColor::Rgba8(convert_into(fallback, image)))
+            }
+            // we need to really convert data..
+            ColorType::L16 => {
+                let buffer = cast_buffer(data)?;
+                let image = try_from_raw::<Luma<u16>>(&buffer, width, height)?;
+                Ok(RgbColor::Rgba8(convert_into(fallback, image)))
+            }
+            ColorType::La16 => {
+                let buffer = cast_buffer(data)?;
+                let image = try_from_raw::<LumaA<u16>>(&buffer, width, height)?;
+                Ok(RgbColor::Rgba8(convert_into(fallback, image)))
+            }
+            ColorType::Rgb16 => {
+                let buffer = cast_buffer(data)?;
+                let image = try_from_raw::<Rgb<u16>>(&buffer, width, height)?;
+                Ok(RgbColor::Rgba8(convert_into(fallback, image)))
+            }
+            ColorType::Rgba16 => {
+                let buffer = cast_buffer(data)?;
+                let image = try_from_raw::<Rgba<u16>>(&buffer, width, height)?;
+                Ok(RgbColor::Rgba8(convert_into(fallback, image)))
+            }
+            // for cases we do not support at all?
+            _ => Err(ImageError::Unsupported(
+                UnsupportedError::from_format_and_kind(
+                    ImageFormat::Avif.into(),
+                    UnsupportedErrorKind::Color(color.into()),
+                ),
+            )),
+        }
+    }
+}

vendor/image/src/codecs/avif/mod.rs

@@ -0,0 +1,14 @@
+//! Encoding of AVIF images.
+///
+/// The [AVIF] specification defines an image derivative of the AV1 bitstream, an open video codec.
+///
+/// [AVIF]: https://aomediacodec.github.io/av1-avif/
+#[cfg(feature = "avif-decoder")]
+pub use self::decoder::AvifDecoder;
+#[cfg(feature = "avif-encoder")]
+pub use self::encoder::{AvifEncoder, ColorSpace};
+
+#[cfg(feature = "avif-decoder")]
+mod decoder;
+#[cfg(feature = "avif-encoder")]
+mod encoder;

vendor/image/src/codecs/bmp/encoder.rs

@@ -0,0 +1,388 @@
+use byteorder::{LittleEndian, WriteBytesExt};
+use std::io::{self, Write};
+
+use crate::error::{
+    EncodingError, ImageError, ImageFormatHint, ImageResult, ParameterError, ParameterErrorKind,
+};
+use crate::image::ImageEncoder;
+use crate::{color, ImageFormat};
+
+const BITMAPFILEHEADER_SIZE: u32 = 14;
+const BITMAPINFOHEADER_SIZE: u32 = 40;
+const BITMAPV4HEADER_SIZE: u32 = 108;
+
+/// The representation of a BMP encoder.
+pub struct BmpEncoder<'a, W: 'a> {
+    writer: &'a mut W,
+}
+
+impl<'a, W: Write + 'a> BmpEncoder<'a, W> {
+    /// Create a new encoder that writes its output to ```w```.
+    pub fn new(w: &'a mut W) -> Self {
+        BmpEncoder { writer: w }
+    }
+
+    /// Encodes the image ```image```
+    /// that has dimensions ```width``` and ```height```
+    /// and ```ColorType``` ```c```.
+    pub fn encode(
+        &mut self,
+        image: &[u8],
+        width: u32,
+        height: u32,
+        c: color::ColorType,
+    ) -> ImageResult<()> {
+        self.encode_with_palette(image, width, height, c, None)
+    }
+
+    /// Same as ```encode```, but allow a palette to be passed in.
+    /// The ```palette``` is ignored for color types other than Luma/Luma-with-alpha.
+    pub fn encode_with_palette(
+        &mut self,
+        image: &[u8],
+        width: u32,
+        height: u32,
+        c: color::ColorType,
+        palette: Option<&[[u8; 3]]>,
+    ) -> ImageResult<()> {
+        if palette.is_some() && c != color::ColorType::L8 && c != color::ColorType::La8 {
+            return Err(ImageError::IoError(io::Error::new(
+                io::ErrorKind::InvalidInput,
+                format!(
+                    "Unsupported color type {:?} when using a non-empty palette. Supported types: Gray(8), GrayA(8).",
+                    c
+                ),
+            )));
+        }
+
+        let bmp_header_size = BITMAPFILEHEADER_SIZE;
+
+        let (dib_header_size, written_pixel_size, palette_color_count) =
+            get_pixel_info(c, palette)?;
+        let row_pad_size = (4 - (width * written_pixel_size) % 4) % 4; // each row must be padded to a multiple of 4 bytes
+        let image_size = width
+            .checked_mul(height)
+            .and_then(|v| v.checked_mul(written_pixel_size))
+            .and_then(|v| v.checked_add(height * row_pad_size))
+            .ok_or_else(|| {
+                ImageError::Parameter(ParameterError::from_kind(
+                    ParameterErrorKind::DimensionMismatch,
+                ))
+            })?;
+        let palette_size = palette_color_count * 4; // all palette colors are BGRA
+        let file_size = bmp_header_size
+            .checked_add(dib_header_size)
+            .and_then(|v| v.checked_add(palette_size))
+            .and_then(|v| v.checked_add(image_size))
+            .ok_or_else(|| {
+                ImageError::Encoding(EncodingError::new(
+                    ImageFormatHint::Exact(ImageFormat::Bmp),
+                    "calculated BMP header size larger than 2^32",
+                ))
+            })?;
+
+        // write BMP header
+        self.writer.write_u8(b'B')?;
+        self.writer.write_u8(b'M')?;
+        self.writer.write_u32::<LittleEndian>(file_size)?; // file size
+        self.writer.write_u16::<LittleEndian>(0)?; // reserved 1
+        self.writer.write_u16::<LittleEndian>(0)?; // reserved 2
+        self.writer
+            .write_u32::<LittleEndian>(bmp_header_size + dib_header_size + palette_size)?; // image data offset
+
+        // write DIB header
+        self.writer.write_u32::<LittleEndian>(dib_header_size)?;
+        self.writer.write_i32::<LittleEndian>(width as i32)?;
+        self.writer.write_i32::<LittleEndian>(height as i32)?;
+        self.writer.write_u16::<LittleEndian>(1)?; // color planes
+        self.writer
+            .write_u16::<LittleEndian>((written_pixel_size * 8) as u16)?; // bits per pixel
+        if dib_header_size >= BITMAPV4HEADER_SIZE {
+            // Assume BGRA32
+            self.writer.write_u32::<LittleEndian>(3)?; // compression method - bitfields
+        } else {
+            self.writer.write_u32::<LittleEndian>(0)?; // compression method - no compression
+        }
+        self.writer.write_u32::<LittleEndian>(image_size)?;
+        self.writer.write_i32::<LittleEndian>(0)?; // horizontal ppm
+        self.writer.write_i32::<LittleEndian>(0)?; // vertical ppm
+        self.writer.write_u32::<LittleEndian>(palette_color_count)?;
+        self.writer.write_u32::<LittleEndian>(0)?; // all colors are important
+        if dib_header_size >= BITMAPV4HEADER_SIZE {
+            // Assume BGRA32
+            self.writer.write_u32::<LittleEndian>(0xff << 16)?; // red mask
+            self.writer.write_u32::<LittleEndian>(0xff << 8)?; // green mask
+            self.writer.write_u32::<LittleEndian>(0xff)?; // blue mask
+            self.writer.write_u32::<LittleEndian>(0xff << 24)?; // alpha mask
+            self.writer.write_u32::<LittleEndian>(0x73524742)?; // colorspace - sRGB
+
+            // endpoints (3x3) and gamma (3)
+            for _ in 0..12 {
+                self.writer.write_u32::<LittleEndian>(0)?;
+            }
+        }
+
+        // write image data
+        match c {
+            color::ColorType::Rgb8 => self.encode_rgb(image, width, height, row_pad_size, 3)?,
+            color::ColorType::Rgba8 => self.encode_rgba(image, width, height, row_pad_size, 4)?,
+            color::ColorType::L8 => {
+                self.encode_gray(image, width, height, row_pad_size, 1, palette)?
+            }
+            color::ColorType::La8 => {
+                self.encode_gray(image, width, height, row_pad_size, 2, palette)?
+            }
+            _ => {
+                return Err(ImageError::IoError(io::Error::new(
+                    io::ErrorKind::InvalidInput,
+                    &get_unsupported_error_message(c)[..],
+                )))
+            }
+        }
+
+        Ok(())
+    }
+
+    fn encode_rgb(
+        &mut self,
+        image: &[u8],
+        width: u32,
+        height: u32,
+        row_pad_size: u32,
+        bytes_per_pixel: u32,
+    ) -> io::Result<()> {
+        let width = width as usize;
+        let height = height as usize;
+        let x_stride = bytes_per_pixel as usize;
+        let y_stride = width * x_stride;
+        for row in (0..height).rev() {
+            // from the bottom up
+            let row_start = row * y_stride;
+            for px in image[row_start..][..y_stride].chunks_exact(x_stride) {
+                let r = px[0];
+                let g = px[1];
+                let b = px[2];
+                // written as BGR
+                self.writer.write_all(&[b, g, r])?;
+            }
+            self.write_row_pad(row_pad_size)?;
+        }
+
+        Ok(())
+    }
+
+    fn encode_rgba(
+        &mut self,
+        image: &[u8],
+        width: u32,
+        height: u32,
+        row_pad_size: u32,
+        bytes_per_pixel: u32,
+    ) -> io::Result<()> {
+        let width = width as usize;
+        let height = height as usize;
+        let x_stride = bytes_per_pixel as usize;
+        let y_stride = width * x_stride;
+        for row in (0..height).rev() {
+            // from the bottom up
+            let row_start = row * y_stride;
+            for px in image[row_start..][..y_stride].chunks_exact(x_stride) {
+                let r = px[0];
+                let g = px[1];
+                let b = px[2];
+                let a = px[3];
+                // written as BGRA
+                self.writer.write_all(&[b, g, r, a])?;
+            }
+            self.write_row_pad(row_pad_size)?;
+        }
+
+        Ok(())
+    }
+
+    fn encode_gray(
+        &mut self,
+        image: &[u8],
+        width: u32,
+        height: u32,
+        row_pad_size: u32,
+        bytes_per_pixel: u32,
+        palette: Option<&[[u8; 3]]>,
+    ) -> io::Result<()> {
+        // write grayscale palette
+        if let Some(palette) = palette {
+            for item in palette {
+                // each color is written as BGRA, where A is always 0
+                self.writer.write_all(&[item[2], item[1], item[0], 0])?;
+            }
+        } else {
+            for val in 0u8..=255 {
+                // each color is written as BGRA, where A is always 0 and since only grayscale is being written, B = G = R = index
+                self.writer.write_all(&[val, val, val, 0])?;
+            }
+        }
+
+        // write image data
+        let x_stride = bytes_per_pixel;
+        let y_stride = width * x_stride;
+        for row in (0..height).rev() {
+            // from the bottom up
+            let row_start = row * y_stride;
+            for col in 0..width {
+                let pixel_start = (row_start + (col * x_stride)) as usize;
+                // color value is equal to the palette index
+                self.writer.write_u8(image[pixel_start])?;
+                // alpha is never written as it's not widely supported
+            }
+
+            self.write_row_pad(row_pad_size)?;
+        }
+
+        Ok(())
+    }
+
+    fn write_row_pad(&mut self, row_pad_size: u32) -> io::Result<()> {
+        for _ in 0..row_pad_size {
+            self.writer.write_u8(0)?;
+        }
+
+        Ok(())
+    }
+}
+
+impl<'a, W: Write> ImageEncoder for BmpEncoder<'a, W> {
+    fn write_image(
+        mut self,
+        buf: &[u8],
+        width: u32,
+        height: u32,
+        color_type: color::ColorType,
+    ) -> ImageResult<()> {
+        self.encode(buf, width, height, color_type)
+    }
+}
+
+fn get_unsupported_error_message(c: color::ColorType) -> String {
+    format!(
+        "Unsupported color type {:?}.  Supported types: RGB(8), RGBA(8), Gray(8), GrayA(8).",
+        c
+    )
+}
+
+/// Returns a tuple representing: (dib header size, written pixel size, palette color count).
+fn get_pixel_info(c: color::ColorType, palette: Option<&[[u8; 3]]>) -> io::Result<(u32, u32, u32)> {
+    let sizes = match c {
+        color::ColorType::Rgb8 => (BITMAPINFOHEADER_SIZE, 3, 0),
+        color::ColorType::Rgba8 => (BITMAPV4HEADER_SIZE, 4, 0),
+        color::ColorType::L8 => (
+            BITMAPINFOHEADER_SIZE,
+            1,
+            palette.map(|p| p.len()).unwrap_or(256) as u32,
+        ),
+        color::ColorType::La8 => (
+            BITMAPINFOHEADER_SIZE,
+            1,
+            palette.map(|p| p.len()).unwrap_or(256) as u32,
+        ),
+        _ => {
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidInput,
+                &get_unsupported_error_message(c)[..],
+            ))
+        }
+    };
+
+    Ok(sizes)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::super::BmpDecoder;
+    use super::BmpEncoder;
+    use crate::color::ColorType;
+    use crate::image::ImageDecoder;
+    use std::io::Cursor;
+
+    fn round_trip_image(image: &[u8], width: u32, height: u32, c: ColorType) -> Vec<u8> {
+        let mut encoded_data = Vec::new();
+        {
+            let mut encoder = BmpEncoder::new(&mut encoded_data);
+            encoder
+                .encode(&image, width, height, c)
+                .expect("could not encode image");
+        }
+
+        let decoder = BmpDecoder::new(Cursor::new(&encoded_data)).expect("failed to decode");
+
+        let mut buf = vec![0; decoder.total_bytes() as usize];
+        decoder.read_image(&mut buf).expect("failed to decode");
+        buf
+    }
+
+    #[test]
+    fn round_trip_single_pixel_rgb() {
+        let image = [255u8, 0, 0]; // single red pixel
+        let decoded = round_trip_image(&image, 1, 1, ColorType::Rgb8);
+        assert_eq!(3, decoded.len());
+        assert_eq!(255, decoded[0]);
+        assert_eq!(0, decoded[1]);
+        assert_eq!(0, decoded[2]);
+    }
+
+    #[test]
+    #[cfg(target_pointer_width = "64")]
+    fn huge_files_return_error() {
+        let mut encoded_data = Vec::new();
+        let image = vec![0u8; 3 * 40_000 * 40_000]; // 40_000x40_000 pixels, 3 bytes per pixel, allocated on the heap
+        let mut encoder = BmpEncoder::new(&mut encoded_data);
+        let result = encoder.encode(&image, 40_000, 40_000, ColorType::Rgb8);
+        assert!(result.is_err());
+    }
+
+    #[test]
+    fn round_trip_single_pixel_rgba() {
+        let image = [1, 2, 3, 4];
+        let decoded = round_trip_image(&image, 1, 1, ColorType::Rgba8);
+        assert_eq!(&decoded[..], &image[..]);
+    }
+
+    #[test]
+    fn round_trip_3px_rgb() {
+        let image = [0u8; 3 * 3 * 3]; // 3x3 pixels, 3 bytes per pixel
+        let _decoded = round_trip_image(&image, 3, 3, ColorType::Rgb8);
+    }
+
+    #[test]
+    fn round_trip_gray() {
+        let image = [0u8, 1, 2]; // 3 pixels
+        let decoded = round_trip_image(&image, 3, 1, ColorType::L8);
+        // should be read back as 3 RGB pixels
+        assert_eq!(9, decoded.len());
+        assert_eq!(0, decoded[0]);
+        assert_eq!(0, decoded[1]);
+        assert_eq!(0, decoded[2]);
+        assert_eq!(1, decoded[3]);
+        assert_eq!(1, decoded[4]);
+        assert_eq!(1, decoded[5]);
+        assert_eq!(2, decoded[6]);
+        assert_eq!(2, decoded[7]);
+        assert_eq!(2, decoded[8]);
+    }
+
+    #[test]
+    fn round_trip_graya() {
+        let image = [0u8, 0, 1, 0, 2, 0]; // 3 pixels, each with an alpha channel
+        let decoded = round_trip_image(&image, 1, 3, ColorType::La8);
+        // should be read back as 3 RGB pixels
+        assert_eq!(9, decoded.len());
+        assert_eq!(0, decoded[0]);
+        assert_eq!(0, decoded[1]);
+        assert_eq!(0, decoded[2]);
+        assert_eq!(1, decoded[3]);
+        assert_eq!(1, decoded[4]);
+        assert_eq!(1, decoded[5]);
+        assert_eq!(2, decoded[6]);
+        assert_eq!(2, decoded[7]);
+        assert_eq!(2, decoded[8]);
+    }
+}

vendor/image/src/codecs/bmp/mod.rs

@@ -0,0 +1,14 @@
+//!  Decoding and Encoding of BMP Images
+//!
+//!  A decoder and encoder for BMP (Windows Bitmap) images
+//!
+//!  # Related Links
+//!  * <https://msdn.microsoft.com/en-us/library/windows/desktop/dd183375%28v=vs.85%29.aspx>
+//!  * <https://en.wikipedia.org/wiki/BMP_file_format>
+//!
+
+pub use self::decoder::BmpDecoder;
+pub use self::encoder::BmpEncoder;
+
+mod decoder;
+mod encoder;

vendor/image/src/codecs/dds.rs

@@ -0,0 +1,375 @@
+//!  Decoding of DDS images
+//!
+//!  DDS (DirectDraw Surface) is a container format for storing DXT (S3TC) compressed images.
+//!
+//!  # Related Links
+//!  * <https://docs.microsoft.com/en-us/windows/win32/direct3ddds/dx-graphics-dds-pguide> - Description of the DDS format.
+
+use std::io::Read;
+use std::{error, fmt};
+
+use byteorder::{LittleEndian, ReadBytesExt};
+
+#[allow(deprecated)]
+use crate::codecs::dxt::{DxtDecoder, DxtReader, DxtVariant};
+use crate::color::ColorType;
+use crate::error::{
+    DecodingError, ImageError, ImageFormatHint, ImageResult, UnsupportedError, UnsupportedErrorKind,
+};
+use crate::image::{ImageDecoder, ImageFormat};
+
+/// Errors that can occur during decoding and parsing a DDS image
+#[derive(Debug, Copy, Clone, Hash, PartialEq, Eq, PartialOrd, Ord)]
+enum DecoderError {
+    /// Wrong DDS channel width
+    PixelFormatSizeInvalid(u32),
+    /// Wrong DDS header size
+    HeaderSizeInvalid(u32),
+    /// Wrong DDS header flags
+    HeaderFlagsInvalid(u32),
+
+    /// Invalid DXGI format in DX10 header
+    DxgiFormatInvalid(u32),
+    /// Invalid resource dimension
+    ResourceDimensionInvalid(u32),
+    /// Invalid flags in DX10 header
+    Dx10FlagsInvalid(u32),
+    /// Invalid array size in DX10 header
+    Dx10ArraySizeInvalid(u32),
+
+    /// DDS "DDS " signature invalid or missing
+    DdsSignatureInvalid,
+}
+
+impl fmt::Display for DecoderError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            DecoderError::PixelFormatSizeInvalid(s) => {
+                f.write_fmt(format_args!("Invalid DDS PixelFormat size: {}", s))
+            }
+            DecoderError::HeaderSizeInvalid(s) => {
+                f.write_fmt(format_args!("Invalid DDS header size: {}", s))
+            }
+            DecoderError::HeaderFlagsInvalid(fs) => {
+                f.write_fmt(format_args!("Invalid DDS header flags: {:#010X}", fs))
+            }
+            DecoderError::DxgiFormatInvalid(df) => {
+                f.write_fmt(format_args!("Invalid DDS DXGI format: {}", df))
+            }
+            DecoderError::ResourceDimensionInvalid(d) => {
+                f.write_fmt(format_args!("Invalid DDS resource dimension: {}", d))
+            }
+            DecoderError::Dx10FlagsInvalid(fs) => {
+                f.write_fmt(format_args!("Invalid DDS DX10 header flags: {:#010X}", fs))
+            }
+            DecoderError::Dx10ArraySizeInvalid(s) => {
+                f.write_fmt(format_args!("Invalid DDS DX10 array size: {}", s))
+            }
+            DecoderError::DdsSignatureInvalid => f.write_str("DDS signature not found"),
+        }
+    }
+}
+
+impl From<DecoderError> for ImageError {
+    fn from(e: DecoderError) -> ImageError {
+        ImageError::Decoding(DecodingError::new(ImageFormat::Dds.into(), e))
+    }
+}
+
+impl error::Error for DecoderError {}
+
+/// Header used by DDS image files
+#[derive(Debug)]
+struct Header {
+    _flags: u32,
+    height: u32,
+    width: u32,
+    _pitch_or_linear_size: u32,
+    _depth: u32,
+    _mipmap_count: u32,
+    pixel_format: PixelFormat,
+    _caps: u32,
+    _caps2: u32,
+}
+
+/// Extended DX10 header used by some DDS image files
+#[derive(Debug)]
+struct DX10Header {
+    dxgi_format: u32,
+    resource_dimension: u32,
+    misc_flag: u32,
+    array_size: u32,
+    misc_flags_2: u32,
+}
+
+/// DDS pixel format
+#[derive(Debug)]
+struct PixelFormat {
+    flags: u32,
+    fourcc: [u8; 4],
+    _rgb_bit_count: u32,
+    _r_bit_mask: u32,
+    _g_bit_mask: u32,
+    _b_bit_mask: u32,
+    _a_bit_mask: u32,
+}
+
+impl PixelFormat {
+    fn from_reader(r: &mut dyn Read) -> ImageResult<Self> {
+        let size = r.read_u32::<LittleEndian>()?;
+        if size != 32 {
+            return Err(DecoderError::PixelFormatSizeInvalid(size).into());
+        }
+
+        Ok(Self {
+            flags: r.read_u32::<LittleEndian>()?,
+            fourcc: {
+                let mut v = [0; 4];
+                r.read_exact(&mut v)?;
+                v
+            },
+            _rgb_bit_count: r.read_u32::<LittleEndian>()?,
+            _r_bit_mask: r.read_u32::<LittleEndian>()?,
+            _g_bit_mask: r.read_u32::<LittleEndian>()?,
+            _b_bit_mask: r.read_u32::<LittleEndian>()?,
+            _a_bit_mask: r.read_u32::<LittleEndian>()?,
+        })
+    }
+}
+
+impl Header {
+    fn from_reader(r: &mut dyn Read) -> ImageResult<Self> {
+        let size = r.read_u32::<LittleEndian>()?;
+        if size != 124 {
+            return Err(DecoderError::HeaderSizeInvalid(size).into());
+        }
+
+        const REQUIRED_FLAGS: u32 = 0x1 | 0x2 | 0x4 | 0x1000;
+        const VALID_FLAGS: u32 = 0x1 | 0x2 | 0x4 | 0x8 | 0x1000 | 0x20000 | 0x80000 | 0x800000;
+        let flags = r.read_u32::<LittleEndian>()?;
+        if flags & (REQUIRED_FLAGS | !VALID_FLAGS) != REQUIRED_FLAGS {
+            return Err(DecoderError::HeaderFlagsInvalid(flags).into());
+        }
+
+        let height = r.read_u32::<LittleEndian>()?;
+        let width = r.read_u32::<LittleEndian>()?;
+        let pitch_or_linear_size = r.read_u32::<LittleEndian>()?;
+        let depth = r.read_u32::<LittleEndian>()?;
+        let mipmap_count = r.read_u32::<LittleEndian>()?;
+        // Skip `dwReserved1`
+        {
+            let mut skipped = [0; 4 * 11];
+            r.read_exact(&mut skipped)?;
+        }
+        let pixel_format = PixelFormat::from_reader(r)?;
+        let caps = r.read_u32::<LittleEndian>()?;
+        let caps2 = r.read_u32::<LittleEndian>()?;
+        // Skip `dwCaps3`, `dwCaps4`, `dwReserved2` (unused)
+        {
+            let mut skipped = [0; 4 + 4 + 4];
+            r.read_exact(&mut skipped)?;
+        }
+
+        Ok(Self {
+            _flags: flags,
+            height,
+            width,
+            _pitch_or_linear_size: pitch_or_linear_size,
+            _depth: depth,
+            _mipmap_count: mipmap_count,
+            pixel_format,
+            _caps: caps,
+            _caps2: caps2,
+        })
+    }
+}
+
+impl DX10Header {
+    fn from_reader(r: &mut dyn Read) -> ImageResult<Self> {
+        let dxgi_format = r.read_u32::<LittleEndian>()?;
+        let resource_dimension = r.read_u32::<LittleEndian>()?;
+        let misc_flag = r.read_u32::<LittleEndian>()?;
+        let array_size = r.read_u32::<LittleEndian>()?;
+        let misc_flags_2 = r.read_u32::<LittleEndian>()?;
+
+        let dx10_header = Self {
+            dxgi_format,
+            resource_dimension,
+            misc_flag,
+            array_size,
+            misc_flags_2,
+        };
+        dx10_header.validate()?;
+
+        Ok(dx10_header)
+    }
+
+    fn validate(&self) -> Result<(), ImageError> {
+        // Note: see https://docs.microsoft.com/en-us/windows/win32/direct3ddds/dds-header-dxt10 for info on valid values
+        if self.dxgi_format > 132 {
+            // Invalid format
+            return Err(DecoderError::DxgiFormatInvalid(self.dxgi_format).into());
+        }
+
+        if self.resource_dimension < 2 || self.resource_dimension > 4 {
+            // Invalid dimension
+            // Only 1D (2), 2D (3) and 3D (4) resource dimensions are allowed
+            return Err(DecoderError::ResourceDimensionInvalid(self.resource_dimension).into());
+        }
+
+        if self.misc_flag != 0x0 && self.misc_flag != 0x4 {
+            // Invalid flag
+            // Only no (0x0) and DDS_RESOURCE_MISC_TEXTURECUBE (0x4) flags are allowed
+            return Err(DecoderError::Dx10FlagsInvalid(self.misc_flag).into());
+        }
+
+        if self.resource_dimension == 4 && self.array_size != 1 {
+            // Invalid array size
+            // 3D textures (resource dimension == 4) must have an array size of 1
+            return Err(DecoderError::Dx10ArraySizeInvalid(self.array_size).into());
+        }
+
+        if self.misc_flags_2 > 0x4 {
+            // Invalid alpha flags
+            return Err(DecoderError::Dx10FlagsInvalid(self.misc_flags_2).into());
+        }
+
+        Ok(())
+    }
+}
+
+/// The representation of a DDS decoder
+pub struct DdsDecoder<R: Read> {
+    #[allow(deprecated)]
+    inner: DxtDecoder<R>,
+}
+
+impl<R: Read> DdsDecoder<R> {
+    /// Create a new decoder that decodes from the stream `r`
+    pub fn new(mut r: R) -> ImageResult<Self> {
+        let mut magic = [0; 4];
+        r.read_exact(&mut magic)?;
+        if magic != b"DDS "[..] {
+            return Err(DecoderError::DdsSignatureInvalid.into());
+        }
+
+        let header = Header::from_reader(&mut r)?;
+
+        if header.pixel_format.flags & 0x4 != 0 {
+            #[allow(deprecated)]
+            let variant = match &header.pixel_format.fourcc {
+                b"DXT1" => DxtVariant::DXT1,
+                b"DXT3" => DxtVariant::DXT3,
+                b"DXT5" => DxtVariant::DXT5,
+                b"DX10" => {
+                    let dx10_header = DX10Header::from_reader(&mut r)?;
+                    // Format equivalents were taken from https://docs.microsoft.com/en-us/windows/win32/direct3d11/texture-block-compression-in-direct3d-11
+                    // The enum integer values were taken from https://docs.microsoft.com/en-us/windows/win32/api/dxgiformat/ne-dxgiformat-dxgi_format
+                    // DXT1 represents the different BC1 variants, DTX3 represents the different BC2 variants and DTX5 represents the different BC3 variants
+                    match dx10_header.dxgi_format {
+                        70 | 71 | 72 => DxtVariant::DXT1, // DXGI_FORMAT_BC1_TYPELESS, DXGI_FORMAT_BC1_UNORM or DXGI_FORMAT_BC1_UNORM_SRGB
+                        73 | 74 | 75 => DxtVariant::DXT3, // DXGI_FORMAT_BC2_TYPELESS, DXGI_FORMAT_BC2_UNORM or DXGI_FORMAT_BC2_UNORM_SRGB
+                        76 | 77 | 78 => DxtVariant::DXT5, // DXGI_FORMAT_BC3_TYPELESS, DXGI_FORMAT_BC3_UNORM or DXGI_FORMAT_BC3_UNORM_SRGB
+                        _ => {
+                            return Err(ImageError::Unsupported(
+                                UnsupportedError::from_format_and_kind(
+                                    ImageFormat::Dds.into(),
+                                    UnsupportedErrorKind::GenericFeature(format!(
+                                        "DDS DXGI Format {}",
+                                        dx10_header.dxgi_format
+                                    )),
+                                ),
+                            ))
+                        }
+                    }
+                }
+                fourcc => {
+                    return Err(ImageError::Unsupported(
+                        UnsupportedError::from_format_and_kind(
+                            ImageFormat::Dds.into(),
+                            UnsupportedErrorKind::GenericFeature(format!(
+                                "DDS FourCC {:?}",
+                                fourcc
+                            )),
+                        ),
+                    ))
+                }
+            };
+
+            #[allow(deprecated)]
+            let bytes_per_pixel = variant.color_type().bytes_per_pixel();
+
+            if crate::utils::check_dimension_overflow(header.width, header.height, bytes_per_pixel)
+            {
+                return Err(ImageError::Unsupported(
+                    UnsupportedError::from_format_and_kind(
+                        ImageFormat::Dds.into(),
+                        UnsupportedErrorKind::GenericFeature(format!(
+                            "Image dimensions ({}x{}) are too large",
+                            header.width, header.height
+                        )),
+                    ),
+                ));
+            }
+
+            #[allow(deprecated)]
+            let inner = DxtDecoder::new(r, header.width, header.height, variant)?;
+            Ok(Self { inner })
+        } else {
+            // For now, supports only DXT variants
+            Err(ImageError::Unsupported(
+                UnsupportedError::from_format_and_kind(
+                    ImageFormat::Dds.into(),
+                    UnsupportedErrorKind::Format(ImageFormatHint::Name("DDS".to_string())),
+                ),
+            ))
+        }
+    }
+}
+
+impl<'a, R: 'a + Read> ImageDecoder<'a> for DdsDecoder<R> {
+    #[allow(deprecated)]
+    type Reader = DxtReader<R>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        self.inner.dimensions()
+    }
+
+    fn color_type(&self) -> ColorType {
+        self.inner.color_type()
+    }
+
+    fn scanline_bytes(&self) -> u64 {
+        self.inner.scanline_bytes()
+    }
+
+    fn into_reader(self) -> ImageResult<Self::Reader> {
+        self.inner.into_reader()
+    }
+
+    fn read_image(self, buf: &mut [u8]) -> ImageResult<()> {
+        self.inner.read_image(buf)
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn dimension_overflow() {
+        // A DXT1 header set to 0xFFFF_FFFC width and height (the highest u32%4 == 0)
+        let header = vec![
+            0x44, 0x44, 0x53, 0x20, 0x7C, 0x0, 0x0, 0x0, 0x7, 0x10, 0x8, 0x0, 0xFC, 0xFF, 0xFF,
+            0xFF, 0xFC, 0xFF, 0xFF, 0xFF, 0x0, 0xC0, 0x12, 0x0, 0x0, 0x0, 0x0, 0x0, 0x1, 0x0, 0x0,
+            0x0, 0x49, 0x4D, 0x41, 0x47, 0x45, 0x4D, 0x41, 0x47, 0x49, 0x43, 0x4B, 0x0, 0x0, 0x0,
+            0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
+            0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x20, 0x0, 0x0, 0x0,
+            0x4, 0x0, 0x0, 0x0, 0x44, 0x58, 0x54, 0x31, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
+            0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x10, 0x0, 0x0, 0x0,
+            0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
+        ];
+
+        assert!(DdsDecoder::new(&header[..]).is_err());
+    }
+}

vendor/image/src/codecs/dxt.rs

@@ -0,0 +1,869 @@
+//!  Decoding of DXT (S3TC) compression
+//!
+//!  DXT is an image format that supports lossy compression
+//!
+//!  # Related Links
+//!  * <https://www.khronos.org/registry/OpenGL/extensions/EXT/EXT_texture_compression_s3tc.txt> - Description of the DXT compression OpenGL extensions.
+//!
+//!  Note: this module only implements bare DXT encoding/decoding, it does not parse formats that can contain DXT files like .dds
+
+use std::convert::TryFrom;
+use std::io::{self, Read, Seek, SeekFrom, Write};
+
+use crate::color::ColorType;
+use crate::error::{ImageError, ImageResult, ParameterError, ParameterErrorKind};
+use crate::image::{self, ImageDecoder, ImageDecoderRect, ImageReadBuffer, Progress};
+
+/// What version of DXT compression are we using?
+/// Note that DXT2 and DXT4 are left away as they're
+/// just DXT3 and DXT5 with premultiplied alpha
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum DxtVariant {
+    /// The DXT1 format. 48 bytes of RGB data in a 4x4 pixel square is
+    /// compressed into an 8 byte block of DXT1 data
+    DXT1,
+    /// The DXT3 format. 64 bytes of RGBA data in a 4x4 pixel square is
+    /// compressed into a 16 byte block of DXT3 data
+    DXT3,
+    /// The DXT5 format. 64 bytes of RGBA data in a 4x4 pixel square is
+    /// compressed into a 16 byte block of DXT5 data
+    DXT5,
+}
+
+impl DxtVariant {
+    /// Returns the amount of bytes of raw image data
+    /// that is encoded in a single DXTn block
+    fn decoded_bytes_per_block(self) -> usize {
+        match self {
+            DxtVariant::DXT1 => 48,
+            DxtVariant::DXT3 | DxtVariant::DXT5 => 64,
+        }
+    }
+
+    /// Returns the amount of bytes per block of encoded DXTn data
+    fn encoded_bytes_per_block(self) -> usize {
+        match self {
+            DxtVariant::DXT1 => 8,
+            DxtVariant::DXT3 | DxtVariant::DXT5 => 16,
+        }
+    }
+
+    /// Returns the color type that is stored in this DXT variant
+    pub fn color_type(self) -> ColorType {
+        match self {
+            DxtVariant::DXT1 => ColorType::Rgb8,
+            DxtVariant::DXT3 | DxtVariant::DXT5 => ColorType::Rgba8,
+        }
+    }
+}
+
+/// DXT decoder
+pub struct DxtDecoder<R: Read> {
+    inner: R,
+    width_blocks: u32,
+    height_blocks: u32,
+    variant: DxtVariant,
+    row: u32,
+}
+
+impl<R: Read> DxtDecoder<R> {
+    /// Create a new DXT decoder that decodes from the stream ```r```.
+    /// As DXT is often stored as raw buffers with the width/height
+    /// somewhere else the width and height of the image need
+    /// to be passed in ```width``` and ```height```, as well as the
+    /// DXT variant in ```variant```.
+    /// width and height are required to be powers of 2 and at least 4.
+    /// otherwise an error will be returned
+    pub fn new(
+        r: R,
+        width: u32,
+        height: u32,
+        variant: DxtVariant,
+    ) -> Result<DxtDecoder<R>, ImageError> {
+        if width % 4 != 0 || height % 4 != 0 {
+            // TODO: this is actually a bit of a weird case. We could return `DecodingError` but
+            // it's not really the format that is wrong However, the encoder should surely return
+            // `EncodingError` so it would be the logical choice for symmetry.
+            return Err(ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::DimensionMismatch,
+            )));
+        }
+        let width_blocks = width / 4;
+        let height_blocks = height / 4;
+        Ok(DxtDecoder {
+            inner: r,
+            width_blocks,
+            height_blocks,
+            variant,
+            row: 0,
+        })
+    }
+
+    fn read_scanline(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        assert_eq!(u64::try_from(buf.len()), Ok(self.scanline_bytes()));
+
+        let mut src =
+            vec![0u8; self.variant.encoded_bytes_per_block() * self.width_blocks as usize];
+        self.inner.read_exact(&mut src)?;
+        match self.variant {
+            DxtVariant::DXT1 => decode_dxt1_row(&src, buf),
+            DxtVariant::DXT3 => decode_dxt3_row(&src, buf),
+            DxtVariant::DXT5 => decode_dxt5_row(&src, buf),
+        }
+        self.row += 1;
+        Ok(buf.len())
+    }
+}
+
+// Note that, due to the way that DXT compression works, a scanline is considered to consist out of
+// 4 lines of pixels.
+impl<'a, R: 'a + Read> ImageDecoder<'a> for DxtDecoder<R> {
+    type Reader = DxtReader<R>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        (self.width_blocks * 4, self.height_blocks * 4)
+    }
+
+    fn color_type(&self) -> ColorType {
+        self.variant.color_type()
+    }
+
+    fn scanline_bytes(&self) -> u64 {
+        self.variant.decoded_bytes_per_block() as u64 * u64::from(self.width_blocks)
+    }
+
+    fn into_reader(self) -> ImageResult<Self::Reader> {
+        Ok(DxtReader {
+            buffer: ImageReadBuffer::new(self.scanline_bytes(), self.total_bytes()),
+            decoder: self,
+        })
+    }
+
+    fn read_image(mut self, buf: &mut [u8]) -> ImageResult<()> {
+        assert_eq!(u64::try_from(buf.len()), Ok(self.total_bytes()));
+
+        for chunk in buf.chunks_mut(self.scanline_bytes().max(1) as usize) {
+            self.read_scanline(chunk)?;
+        }
+        Ok(())
+    }
+}
+
+impl<'a, R: 'a + Read + Seek> ImageDecoderRect<'a> for DxtDecoder<R> {
+    fn read_rect_with_progress<F: Fn(Progress)>(
+        &mut self,
+        x: u32,
+        y: u32,
+        width: u32,
+        height: u32,
+        buf: &mut [u8],
+        progress_callback: F,
+    ) -> ImageResult<()> {
+        let encoded_scanline_bytes =
+            self.variant.encoded_bytes_per_block() as u64 * u64::from(self.width_blocks);
+
+        let start = self.inner.stream_position()?;
+        image::load_rect(
+            x,
+            y,
+            width,
+            height,
+            buf,
+            progress_callback,
+            self,
+            |s, scanline| {
+                s.inner
+                    .seek(SeekFrom::Start(start + scanline * encoded_scanline_bytes))?;
+                Ok(())
+            },
+            |s, buf| s.read_scanline(buf).map(|_| ()),
+        )?;
+        self.inner.seek(SeekFrom::Start(start))?;
+        Ok(())
+    }
+}
+
+/// DXT reader
+pub struct DxtReader<R: Read> {
+    buffer: ImageReadBuffer,
+    decoder: DxtDecoder<R>,
+}
+
+impl<R: Read> Read for DxtReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        let decoder = &mut self.decoder;
+        self.buffer.read(buf, |buf| decoder.read_scanline(buf))
+    }
+}
+
+/// DXT encoder
+pub struct DxtEncoder<W: Write> {
+    w: W,
+}
+
+impl<W: Write> DxtEncoder<W> {
+    /// Create a new encoder that writes its output to ```w```
+    pub fn new(w: W) -> DxtEncoder<W> {
+        DxtEncoder { w }
+    }
+
+    /// Encodes the image data ```data```
+    /// that has dimensions ```width``` and ```height```
+    /// in ```DxtVariant``` ```variant```
+    /// data is assumed to be in variant.color_type()
+    pub fn encode(
+        mut self,
+        data: &[u8],
+        width: u32,
+        height: u32,
+        variant: DxtVariant,
+    ) -> ImageResult<()> {
+        if width % 4 != 0 || height % 4 != 0 {
+            // TODO: this is not very idiomatic yet. Should return an EncodingError.
+            return Err(ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::DimensionMismatch,
+            )));
+        }
+        let width_blocks = width / 4;
+        let height_blocks = height / 4;
+
+        let stride = variant.decoded_bytes_per_block();
+
+        assert!(data.len() >= width_blocks as usize * height_blocks as usize * stride);
+
+        for chunk in data.chunks(width_blocks as usize * stride) {
+            let data = match variant {
+                DxtVariant::DXT1 => encode_dxt1_row(chunk),
+                DxtVariant::DXT3 => encode_dxt3_row(chunk),
+                DxtVariant::DXT5 => encode_dxt5_row(chunk),
+            };
+            self.w.write_all(&data)?;
+        }
+        Ok(())
+    }
+}
+
+/**
+ * Actual encoding/decoding logic below.
+ */
+use std::mem::swap;
+
+type Rgb = [u8; 3];
+
+/// decodes a 5-bit R, 6-bit G, 5-bit B 16-bit packed color value into 8-bit RGB
+/// mapping is done so min/max range values are preserved. So for 5-bit
+/// values 0x00 -> 0x00 and 0x1F -> 0xFF
+fn enc565_decode(value: u16) -> Rgb {
+    let red = (value >> 11) & 0x1F;
+    let green = (value >> 5) & 0x3F;
+    let blue = (value) & 0x1F;
+    [
+        (red * 0xFF / 0x1F) as u8,
+        (green * 0xFF / 0x3F) as u8,
+        (blue * 0xFF / 0x1F) as u8,
+    ]
+}
+
+/// encodes an 8-bit RGB value into a 5-bit R, 6-bit G, 5-bit B 16-bit packed color value
+/// mapping preserves min/max values. It is guaranteed that i == encode(decode(i)) for all i
+fn enc565_encode(rgb: Rgb) -> u16 {
+    let red = (u16::from(rgb[0]) * 0x1F + 0x7E) / 0xFF;
+    let green = (u16::from(rgb[1]) * 0x3F + 0x7E) / 0xFF;
+    let blue = (u16::from(rgb[2]) * 0x1F + 0x7E) / 0xFF;
+    (red << 11) | (green << 5) | blue
+}
+
+/// utility function: squares a value
+fn square(a: i32) -> i32 {
+    a * a
+}
+
+/// returns the squared error between two RGB values
+fn diff(a: Rgb, b: Rgb) -> i32 {
+    square(i32::from(a[0]) - i32::from(b[0]))
+        + square(i32::from(a[1]) - i32::from(b[1]))
+        + square(i32::from(a[2]) - i32::from(b[2]))
+}
+
+/*
+ * Functions for decoding DXT compression
+ */
+
+/// Constructs the DXT5 alpha lookup table from the two alpha entries
+/// if alpha0 > alpha1, constructs a table of [a0, a1, 6 linearly interpolated values from a0 to a1]
+/// if alpha0 <= alpha1, constructs a table of [a0, a1, 4 linearly interpolated values from a0 to a1, 0, 0xFF]
+fn alpha_table_dxt5(alpha0: u8, alpha1: u8) -> [u8; 8] {
+    let mut table = [alpha0, alpha1, 0, 0, 0, 0, 0, 0xFF];
+    if alpha0 > alpha1 {
+        for i in 2..8u16 {
+            table[i as usize] =
+                (((8 - i) * u16::from(alpha0) + (i - 1) * u16::from(alpha1)) / 7) as u8;
+        }
+    } else {
+        for i in 2..6u16 {
+            table[i as usize] =
+                (((6 - i) * u16::from(alpha0) + (i - 1) * u16::from(alpha1)) / 5) as u8;
+        }
+    }
+    table
+}
+
+/// decodes an 8-byte dxt color block into the RGB channels of a 16xRGB or 16xRGBA block.
+/// source should have a length of 8, dest a length of 48 (RGB) or 64 (RGBA)
+fn decode_dxt_colors(source: &[u8], dest: &mut [u8], is_dxt1: bool) {
+    // sanity checks, also enable the compiler to elide all following bound checks
+    assert!(source.len() == 8 && (dest.len() == 48 || dest.len() == 64));
+    // calculate pitch to store RGB values in dest (3 for RGB, 4 for RGBA)
+    let pitch = dest.len() / 16;
+
+    // extract color data
+    let color0 = u16::from(source[0]) | (u16::from(source[1]) << 8);
+    let color1 = u16::from(source[2]) | (u16::from(source[3]) << 8);
+    let color_table = u32::from(source[4])
+        | (u32::from(source[5]) << 8)
+        | (u32::from(source[6]) << 16)
+        | (u32::from(source[7]) << 24);
+    // let color_table = source[4..8].iter().rev().fold(0, |t, &b| (t << 8) | b as u32);
+
+    // decode the colors to rgb format
+    let mut colors = [[0; 3]; 4];
+    colors[0] = enc565_decode(color0);
+    colors[1] = enc565_decode(color1);
+
+    // determine color interpolation method
+    if color0 > color1 || !is_dxt1 {
+        // linearly interpolate the other two color table entries
+        for i in 0..3 {
+            colors[2][i] = ((u16::from(colors[0][i]) * 2 + u16::from(colors[1][i]) + 1) / 3) as u8;
+            colors[3][i] = ((u16::from(colors[0][i]) + u16::from(colors[1][i]) * 2 + 1) / 3) as u8;
+        }
+    } else {
+        // linearly interpolate one other entry, keep the other at 0
+        for i in 0..3 {
+            colors[2][i] = ((u16::from(colors[0][i]) + u16::from(colors[1][i]) + 1) / 2) as u8;
+        }
+    }
+
+    // serialize the result. Every color is determined by looking up
+    // two bits in color_table which identify which color to actually pick from the 4 possible colors
+    for i in 0..16 {
+        dest[i * pitch..i * pitch + 3]
+            .copy_from_slice(&colors[(color_table >> (i * 2)) as usize & 3]);
+    }
+}
+
+/// Decodes a 16-byte bock of dxt5 data to a 16xRGBA block
+fn decode_dxt5_block(source: &[u8], dest: &mut [u8]) {
+    assert!(source.len() == 16 && dest.len() == 64);
+
+    // extract alpha index table (stored as little endian 64-bit value)
+    let alpha_table = source[2..8]
+        .iter()
+        .rev()
+        .fold(0, |t, &b| (t << 8) | u64::from(b));
+
+    // alhpa level decode
+    let alphas = alpha_table_dxt5(source[0], source[1]);
+
+    // serialize alpha
+    for i in 0..16 {
+        dest[i * 4 + 3] = alphas[(alpha_table >> (i * 3)) as usize & 7];
+    }
+
+    // handle colors
+    decode_dxt_colors(&source[8..16], dest, false);
+}
+
+/// Decodes a 16-byte bock of dxt3 data to a 16xRGBA block
+fn decode_dxt3_block(source: &[u8], dest: &mut [u8]) {
+    assert!(source.len() == 16 && dest.len() == 64);
+
+    // extract alpha index table (stored as little endian 64-bit value)
+    let alpha_table = source[0..8]
+        .iter()
+        .rev()
+        .fold(0, |t, &b| (t << 8) | u64::from(b));
+
+    // serialize alpha (stored as 4-bit values)
+    for i in 0..16 {
+        dest[i * 4 + 3] = ((alpha_table >> (i * 4)) as u8 & 0xF) * 0x11;
+    }
+
+    // handle colors
+    decode_dxt_colors(&source[8..16], dest, false);
+}
+
+/// Decodes a 8-byte bock of dxt5 data to a 16xRGB block
+fn decode_dxt1_block(source: &[u8], dest: &mut [u8]) {
+    assert!(source.len() == 8 && dest.len() == 48);
+    decode_dxt_colors(source, dest, true);
+}
+
+/// Decode a row of DXT1 data to four rows of RGB data.
+/// source.len() should be a multiple of 8, otherwise this panics.
+fn decode_dxt1_row(source: &[u8], dest: &mut [u8]) {
+    assert!(source.len() % 8 == 0);
+    let block_count = source.len() / 8;
+    assert!(dest.len() >= block_count * 48);
+
+    // contains the 16 decoded pixels per block
+    let mut decoded_block = [0u8; 48];
+
+    for (x, encoded_block) in source.chunks(8).enumerate() {
+        decode_dxt1_block(encoded_block, &mut decoded_block);
+
+        // copy the values from the decoded block to linewise RGB layout
+        for line in 0..4 {
+            let offset = (block_count * line + x) * 12;
+            dest[offset..offset + 12].copy_from_slice(&decoded_block[line * 12..(line + 1) * 12]);
+        }
+    }
+}
+
+/// Decode a row of DXT3 data to four rows of RGBA data.
+/// source.len() should be a multiple of 16, otherwise this panics.
+fn decode_dxt3_row(source: &[u8], dest: &mut [u8]) {
+    assert!(source.len() % 16 == 0);
+    let block_count = source.len() / 16;
+    assert!(dest.len() >= block_count * 64);
+
+    // contains the 16 decoded pixels per block
+    let mut decoded_block = [0u8; 64];
+
+    for (x, encoded_block) in source.chunks(16).enumerate() {
+        decode_dxt3_block(encoded_block, &mut decoded_block);
+
+        // copy the values from the decoded block to linewise RGB layout
+        for line in 0..4 {
+            let offset = (block_count * line + x) * 16;
+            dest[offset..offset + 16].copy_from_slice(&decoded_block[line * 16..(line + 1) * 16]);
+        }
+    }
+}
+
+/// Decode a row of DXT5 data to four rows of RGBA data.
+/// source.len() should be a multiple of 16, otherwise this panics.
+fn decode_dxt5_row(source: &[u8], dest: &mut [u8]) {
+    assert!(source.len() % 16 == 0);
+    let block_count = source.len() / 16;
+    assert!(dest.len() >= block_count * 64);
+
+    // contains the 16 decoded pixels per block
+    let mut decoded_block = [0u8; 64];
+
+    for (x, encoded_block) in source.chunks(16).enumerate() {
+        decode_dxt5_block(encoded_block, &mut decoded_block);
+
+        // copy the values from the decoded block to linewise RGB layout
+        for line in 0..4 {
+            let offset = (block_count * line + x) * 16;
+            dest[offset..offset + 16].copy_from_slice(&decoded_block[line * 16..(line + 1) * 16]);
+        }
+    }
+}
+
+/*
+ * Functions for encoding DXT compression
+ */
+
+/// Tries to perform the color encoding part of dxt compression
+/// the approach taken is simple, it picks unique combinations
+/// of the colors present in the block, and attempts to encode the
+/// block with each, picking the encoding that yields the least
+/// squared error out of all of them.
+///
+/// This could probably be faster but is already reasonably fast
+/// and a good reference impl to optimize others against.
+///
+/// Another way to perform this analysis would be to perform a
+/// singular value decomposition of the different colors, and
+/// then pick 2 points on this line as the base colors. But
+/// this is still rather unwieldy math and has issues
+/// with the 3-linear-colors-and-0 case, it's also worse
+/// at conserving the original colors.
+///
+/// source: should be RGBAx16 or RGBx16 bytes of data,
+/// dest 8 bytes of resulting encoded color data
+fn encode_dxt_colors(source: &[u8], dest: &mut [u8], is_dxt1: bool) {
+    // sanity checks and determine stride when parsing the source data
+    assert!((source.len() == 64 || source.len() == 48) && dest.len() == 8);
+    let stride = source.len() / 16;
+
+    // reference colors array
+    let mut colors = [[0u8; 3]; 4];
+
+    // Put the colors we're going to be processing in an array with pure RGB layout
+    // note: we reverse the pixel order here. The reason for this is found in the inner quantization loop.
+    let mut targets = [[0u8; 3]; 16];
+    for (s, d) in source.chunks(stride).rev().zip(&mut targets) {
+        *d = [s[0], s[1], s[2]];
+    }
+
+    // roundtrip all colors through the r5g6b5 encoding
+    for rgb in &mut targets {
+        *rgb = enc565_decode(enc565_encode(*rgb));
+    }
+
+    // and deduplicate the set of colors to choose from as the algorithm is O(N^2) in this
+    let mut colorspace_ = [[0u8; 3]; 16];
+    let mut colorspace_len = 0;
+    for color in &targets {
+        if !colorspace_[..colorspace_len].contains(color) {
+            colorspace_[colorspace_len] = *color;
+            colorspace_len += 1;
+        }
+    }
+    let mut colorspace = &colorspace_[..colorspace_len];
+
+    // in case of slight gradients it can happen that there's only one entry left in the color table.
+    // as the resulting banding can be quite bad if we would just left the block at the closest
+    // encodable color, we have a special path here that tries to emulate the wanted color
+    // using the linear interpolation between gradients
+    if colorspace.len() == 1 {
+        // the base color we got from colorspace reduction
+        let ref_rgb = colorspace[0];
+        // the unreduced color in this block that's the furthest away from the actual block
+        let mut rgb = targets
+            .iter()
+            .cloned()
+            .max_by_key(|rgb| diff(*rgb, ref_rgb))
+            .unwrap();
+        // amplify differences by 2.5, which should push them to the next quantized value
+        // if possible without overshoot
+        for i in 0..3 {
+            rgb[i] =
+                ((i16::from(rgb[i]) - i16::from(ref_rgb[i])) * 5 / 2 + i16::from(ref_rgb[i])) as u8;
+        }
+
+        // roundtrip it through quantization
+        let encoded = enc565_encode(rgb);
+        let rgb = enc565_decode(encoded);
+
+        // in case this didn't land us a different color the best way to represent this field is
+        // as a single color block
+        if rgb == ref_rgb {
+            dest[0] = encoded as u8;
+            dest[1] = (encoded >> 8) as u8;
+
+            for d in dest.iter_mut().take(8).skip(2) {
+                *d = 0;
+            }
+            return;
+        }
+
+        // we did find a separate value: add it to the options so after one round of quantization
+        // we're done
+        colorspace_[1] = rgb;
+        colorspace = &colorspace_[..2];
+    }
+
+    // block quantization loop: we basically just try every possible combination, returning
+    // the combination with the least squared error
+    // stores the best candidate colors
+    let mut chosen_colors = [[0; 3]; 4];
+    // did this index table use the [0,0,0] variant
+    let mut chosen_use_0 = false;
+    // error calculated for the last entry
+    let mut chosen_error = 0xFFFF_FFFFu32;
+
+    // loop through unique permutations of the colorspace, where c1 != c2
+    'search: for (i, &c1) in colorspace.iter().enumerate() {
+        colors[0] = c1;
+
+        for &c2 in &colorspace[0..i] {
+            colors[1] = c2;
+
+            if is_dxt1 {
+                // what's inside here is ran at most 120 times.
+                for use_0 in 0..2 {
+                    // and 240 times here.
+
+                    if use_0 != 0 {
+                        // interpolate one color, set the other to 0
+                        for i in 0..3 {
+                            colors[2][i] =
+                                ((u16::from(colors[0][i]) + u16::from(colors[1][i]) + 1) / 2) as u8;
+                        }
+                        colors[3] = [0, 0, 0];
+                    } else {
+                        // interpolate to get 2 more colors
+                        for i in 0..3 {
+                            colors[2][i] =
+                                ((u16::from(colors[0][i]) * 2 + u16::from(colors[1][i]) + 1) / 3)
+                                    as u8;
+                            colors[3][i] =
+                                ((u16::from(colors[0][i]) + u16::from(colors[1][i]) * 2 + 1) / 3)
+                                    as u8;
+                        }
+                    }
+
+                    // calculate the total error if we were to quantize the block with these color combinations
+                    // both these loops have statically known iteration counts and are well vectorizable
+                    // note that the inside of this can be run about 15360 times worst case, i.e. 960 times per
+                    // pixel.
+                    let total_error = targets
+                        .iter()
+                        .map(|t| colors.iter().map(|c| diff(*c, *t) as u32).min().unwrap())
+                        .sum();
+
+                    // update the match if we found a better one
+                    if total_error < chosen_error {
+                        chosen_colors = colors;
+                        chosen_use_0 = use_0 != 0;
+                        chosen_error = total_error;
+
+                        // if we've got a perfect or at most 1 LSB off match, we're done
+                        if total_error < 4 {
+                            break 'search;
+                        }
+                    }
+                }
+            } else {
+                // what's inside here is ran at most 120 times.
+
+                // interpolate to get 2 more colors
+                for i in 0..3 {
+                    colors[2][i] =
+                        ((u16::from(colors[0][i]) * 2 + u16::from(colors[1][i]) + 1) / 3) as u8;
+                    colors[3][i] =
+                        ((u16::from(colors[0][i]) + u16::from(colors[1][i]) * 2 + 1) / 3) as u8;
+                }
+
+                // calculate the total error if we were to quantize the block with these color combinations
+                // both these loops have statically known iteration counts and are well vectorizable
+                // note that the inside of this can be run about 15360 times worst case, i.e. 960 times per
+                // pixel.
+                let total_error = targets
+                    .iter()
+                    .map(|t| colors.iter().map(|c| diff(*c, *t) as u32).min().unwrap())
+                    .sum();
+
+                // update the match if we found a better one
+                if total_error < chosen_error {
+                    chosen_colors = colors;
+                    chosen_error = total_error;
+
+                    // if we've got a perfect or at most 1 LSB off match, we're done
+                    if total_error < 4 {
+                        break 'search;
+                    }
+                }
+            }
+        }
+    }
+
+    // calculate the final indices
+    // note that targets is already in reverse pixel order, to make the index computation easy.
+    let mut chosen_indices = 0u32;
+    for t in &targets {
+        let (idx, _) = chosen_colors
+            .iter()
+            .enumerate()
+            .min_by_key(|&(_, c)| diff(*c, *t))
+            .unwrap();
+        chosen_indices = (chosen_indices << 2) | idx as u32;
+    }
+
+    // encode the colors
+    let mut color0 = enc565_encode(chosen_colors[0]);
+    let mut color1 = enc565_encode(chosen_colors[1]);
+
+    // determine encoding. Note that color0 == color1 is impossible at this point
+    if is_dxt1 {
+        if color0 > color1 {
+            if chosen_use_0 {
+                swap(&mut color0, &mut color1);
+                // Indexes are packed 2 bits wide, swap index 0/1 but preserve 2/3.
+                let filter = (chosen_indices & 0xAAAA_AAAA) >> 1;
+                chosen_indices ^= filter ^ 0x5555_5555;
+            }
+        } else if !chosen_use_0 {
+            swap(&mut color0, &mut color1);
+            // Indexes are packed 2 bits wide, swap index 0/1 and 2/3.
+            chosen_indices ^= 0x5555_5555;
+        }
+    }
+
+    // encode everything.
+    dest[0] = color0 as u8;
+    dest[1] = (color0 >> 8) as u8;
+    dest[2] = color1 as u8;
+    dest[3] = (color1 >> 8) as u8;
+    for i in 0..4 {
+        dest[i + 4] = (chosen_indices >> (i * 8)) as u8;
+    }
+}
+
+/// Encodes a buffer of 16 alpha bytes into a dxt5 alpha index table,
+/// where the alpha table they are indexed against is created by
+/// calling alpha_table_dxt5(alpha0, alpha1)
+/// returns the resulting error and alpha table
+fn encode_dxt5_alpha(alpha0: u8, alpha1: u8, alphas: &[u8; 16]) -> (i32, u64) {
+    // create a table for the given alpha ranges
+    let table = alpha_table_dxt5(alpha0, alpha1);
+    let mut indices = 0u64;
+    let mut total_error = 0i32;
+
+    // least error brute force search
+    for (i, &a) in alphas.iter().enumerate() {
+        let (index, error) = table
+            .iter()
+            .enumerate()
+            .map(|(i, &e)| (i, square(i32::from(e) - i32::from(a))))
+            .min_by_key(|&(_, e)| e)
+            .unwrap();
+        total_error += error;
+        indices |= (index as u64) << (i * 3);
+    }
+
+    (total_error, indices)
+}
+
+/// Encodes a RGBAx16 sequence of bytes to a 16 bytes DXT5 block
+fn encode_dxt5_block(source: &[u8], dest: &mut [u8]) {
+    assert!(source.len() == 64 && dest.len() == 16);
+
+    // perform dxt color encoding
+    encode_dxt_colors(source, &mut dest[8..16], false);
+
+    // copy out the alpha bytes
+    let mut alphas = [0; 16];
+    for i in 0..16 {
+        alphas[i] = source[i * 4 + 3];
+    }
+
+    // try both alpha compression methods, see which has the least error.
+    let alpha07 = alphas.iter().cloned().min().unwrap();
+    let alpha17 = alphas.iter().cloned().max().unwrap();
+    let (error7, indices7) = encode_dxt5_alpha(alpha07, alpha17, &alphas);
+
+    // if all alphas are 0 or 255 it doesn't particularly matter what we do here.
+    let alpha05 = alphas
+        .iter()
+        .cloned()
+        .filter(|&i| i != 255)
+        .max()
+        .unwrap_or(255);
+    let alpha15 = alphas
+        .iter()
+        .cloned()
+        .filter(|&i| i != 0)
+        .min()
+        .unwrap_or(0);
+    let (error5, indices5) = encode_dxt5_alpha(alpha05, alpha15, &alphas);
+
+    // pick the best one, encode the min/max values
+    let mut alpha_table = if error5 < error7 {
+        dest[0] = alpha05;
+        dest[1] = alpha15;
+        indices5
+    } else {
+        dest[0] = alpha07;
+        dest[1] = alpha17;
+        indices7
+    };
+
+    // encode the alphas
+    for byte in dest[2..8].iter_mut() {
+        *byte = alpha_table as u8;
+        alpha_table >>= 8;
+    }
+}
+
+/// Encodes a RGBAx16 sequence of bytes into a 16 bytes DXT3 block
+fn encode_dxt3_block(source: &[u8], dest: &mut [u8]) {
+    assert!(source.len() == 64 && dest.len() == 16);
+
+    // perform dxt color encoding
+    encode_dxt_colors(source, &mut dest[8..16], false);
+
+    // DXT3 alpha compression is very simple, just round towards the nearest value
+
+    // index the alpha values into the 64bit alpha table
+    let mut alpha_table = 0u64;
+    for i in 0..16 {
+        let alpha = u64::from(source[i * 4 + 3]);
+        let alpha = (alpha + 0x8) / 0x11;
+        alpha_table |= alpha << (i * 4);
+    }
+
+    // encode the alpha values
+    for byte in &mut dest[0..8] {
+        *byte = alpha_table as u8;
+        alpha_table >>= 8;
+    }
+}
+
+/// Encodes a RGBx16 sequence of bytes into a 8 bytes DXT1 block
+fn encode_dxt1_block(source: &[u8], dest: &mut [u8]) {
+    assert!(source.len() == 48 && dest.len() == 8);
+
+    // perform dxt color encoding
+    encode_dxt_colors(source, dest, true);
+}
+
+/// Decode a row of DXT1 data to four rows of RGBA data.
+/// source.len() should be a multiple of 8, otherwise this panics.
+fn encode_dxt1_row(source: &[u8]) -> Vec<u8> {
+    assert!(source.len() % 48 == 0);
+    let block_count = source.len() / 48;
+
+    let mut dest = vec![0u8; block_count * 8];
+    // contains the 16 decoded pixels per block
+    let mut decoded_block = [0u8; 48];
+
+    for (x, encoded_block) in dest.chunks_mut(8).enumerate() {
+        // copy the values from the decoded block to linewise RGB layout
+        for line in 0..4 {
+            let offset = (block_count * line + x) * 12;
+            decoded_block[line * 12..(line + 1) * 12].copy_from_slice(&source[offset..offset + 12]);
+        }
+
+        encode_dxt1_block(&decoded_block, encoded_block);
+    }
+    dest
+}
+
+/// Decode a row of DXT3 data to four rows of RGBA data.
+/// source.len() should be a multiple of 16, otherwise this panics.
+fn encode_dxt3_row(source: &[u8]) -> Vec<u8> {
+    assert!(source.len() % 64 == 0);
+    let block_count = source.len() / 64;
+
+    let mut dest = vec![0u8; block_count * 16];
+    // contains the 16 decoded pixels per block
+    let mut decoded_block = [0u8; 64];
+
+    for (x, encoded_block) in dest.chunks_mut(16).enumerate() {
+        // copy the values from the decoded block to linewise RGB layout
+        for line in 0..4 {
+            let offset = (block_count * line + x) * 16;
+            decoded_block[line * 16..(line + 1) * 16].copy_from_slice(&source[offset..offset + 16]);
+        }
+
+        encode_dxt3_block(&decoded_block, encoded_block);
+    }
+    dest
+}
+
+/// Decode a row of DXT5 data to four rows of RGBA data.
+/// source.len() should be a multiple of 16, otherwise this panics.
+fn encode_dxt5_row(source: &[u8]) -> Vec<u8> {
+    assert!(source.len() % 64 == 0);
+    let block_count = source.len() / 64;
+
+    let mut dest = vec![0u8; block_count * 16];
+    // contains the 16 decoded pixels per block
+    let mut decoded_block = [0u8; 64];
+
+    for (x, encoded_block) in dest.chunks_mut(16).enumerate() {
+        // copy the values from the decoded block to linewise RGB layout
+        for line in 0..4 {
+            let offset = (block_count * line + x) * 16;
+            decoded_block[line * 16..(line + 1) * 16].copy_from_slice(&source[offset..offset + 16]);
+        }
+
+        encode_dxt5_block(&decoded_block, encoded_block);
+    }
+    dest
+}

vendor/image/src/codecs/farbfeld.rs

@@ -0,0 +1,400 @@
+//! Decoding of farbfeld images
+//!
+//! farbfeld is a lossless image format which is easy to parse, pipe and compress.
+//!
+//! It has the following format:
+//!
+//! | Bytes  | Description                                             |
+//! |--------|---------------------------------------------------------|
+//! | 8      | "farbfeld" magic value                                  |
+//! | 4      | 32-Bit BE unsigned integer (width)                      |
+//! | 4      | 32-Bit BE unsigned integer (height)                     |
+//! | [2222] | 4⋅16-Bit BE unsigned integers [RGBA] / pixel, row-major |
+//!
+//! The RGB-data should be sRGB for best interoperability and not alpha-premultiplied.
+//!
+//! # Related Links
+//! * <https://tools.suckless.org/farbfeld/> - the farbfeld specification
+
+use std::convert::TryFrom;
+use std::i64;
+use std::io::{self, Read, Seek, SeekFrom, Write};
+
+use byteorder::{BigEndian, ByteOrder, NativeEndian};
+
+use crate::color::ColorType;
+use crate::error::{
+    DecodingError, ImageError, ImageResult, UnsupportedError, UnsupportedErrorKind,
+};
+use crate::image::{self, ImageDecoder, ImageDecoderRect, ImageEncoder, ImageFormat, Progress};
+
+/// farbfeld Reader
+pub struct FarbfeldReader<R: Read> {
+    width: u32,
+    height: u32,
+    inner: R,
+    /// Relative to the start of the pixel data
+    current_offset: u64,
+    cached_byte: Option<u8>,
+}
+
+impl<R: Read> FarbfeldReader<R> {
+    fn new(mut buffered_read: R) -> ImageResult<FarbfeldReader<R>> {
+        fn read_dimm<R: Read>(from: &mut R) -> ImageResult<u32> {
+            let mut buf = [0u8; 4];
+            from.read_exact(&mut buf).map_err(|err| {
+                ImageError::Decoding(DecodingError::new(ImageFormat::Farbfeld.into(), err))
+            })?;
+            Ok(BigEndian::read_u32(&buf))
+        }
+
+        let mut magic = [0u8; 8];
+        buffered_read.read_exact(&mut magic).map_err(|err| {
+            ImageError::Decoding(DecodingError::new(ImageFormat::Farbfeld.into(), err))
+        })?;
+        if &magic != b"farbfeld" {
+            return Err(ImageError::Decoding(DecodingError::new(
+                ImageFormat::Farbfeld.into(),
+                format!("Invalid magic: {:02x?}", magic),
+            )));
+        }
+
+        let reader = FarbfeldReader {
+            width: read_dimm(&mut buffered_read)?,
+            height: read_dimm(&mut buffered_read)?,
+            inner: buffered_read,
+            current_offset: 0,
+            cached_byte: None,
+        };
+
+        if crate::utils::check_dimension_overflow(
+            reader.width,
+            reader.height,
+            // ColorType is always rgba16
+            ColorType::Rgba16.bytes_per_pixel(),
+        ) {
+            return Err(ImageError::Unsupported(
+                UnsupportedError::from_format_and_kind(
+                    ImageFormat::Farbfeld.into(),
+                    UnsupportedErrorKind::GenericFeature(format!(
+                        "Image dimensions ({}x{}) are too large",
+                        reader.width, reader.height
+                    )),
+                ),
+            ));
+        }
+
+        Ok(reader)
+    }
+}
+
+impl<R: Read> Read for FarbfeldReader<R> {
+    fn read(&mut self, mut buf: &mut [u8]) -> io::Result<usize> {
+        let mut bytes_written = 0;
+        if let Some(byte) = self.cached_byte.take() {
+            buf[0] = byte;
+            buf = &mut buf[1..];
+            bytes_written = 1;
+            self.current_offset += 1;
+        }
+
+        if buf.len() == 1 {
+            buf[0] = cache_byte(&mut self.inner, &mut self.cached_byte)?;
+            bytes_written += 1;
+            self.current_offset += 1;
+        } else {
+            for channel_out in buf.chunks_exact_mut(2) {
+                consume_channel(&mut self.inner, channel_out)?;
+                bytes_written += 2;
+                self.current_offset += 2;
+            }
+        }
+
+        Ok(bytes_written)
+    }
+}
+
+impl<R: Read + Seek> Seek for FarbfeldReader<R> {
+    fn seek(&mut self, pos: SeekFrom) -> io::Result<u64> {
+        fn parse_offset(original_offset: u64, end_offset: u64, pos: SeekFrom) -> Option<i64> {
+            match pos {
+                SeekFrom::Start(off) => i64::try_from(off)
+                    .ok()?
+                    .checked_sub(i64::try_from(original_offset).ok()?),
+                SeekFrom::End(off) => {
+                    if off < i64::try_from(end_offset).unwrap_or(i64::MAX) {
+                        None
+                    } else {
+                        Some(i64::try_from(end_offset.checked_sub(original_offset)?).ok()? + off)
+                    }
+                }
+                SeekFrom::Current(off) => {
+                    if off < i64::try_from(original_offset).unwrap_or(i64::MAX) {
+                        None
+                    } else {
+                        Some(off)
+                    }
+                }
+            }
+        }
+
+        let original_offset = self.current_offset;
+        let end_offset = self.width as u64 * self.height as u64 * 2;
+        let offset_from_current =
+            parse_offset(original_offset, end_offset, pos).ok_or_else(|| {
+                io::Error::new(
+                    io::ErrorKind::InvalidInput,
+                    "invalid seek to a negative or overflowing position",
+                )
+            })?;
+
+        // TODO: convert to seek_relative() once that gets stabilised
+        self.inner.seek(SeekFrom::Current(offset_from_current))?;
+        self.current_offset = if offset_from_current < 0 {
+            original_offset.checked_sub(offset_from_current.wrapping_neg() as u64)
+        } else {
+            original_offset.checked_add(offset_from_current as u64)
+        }
+        .expect("This should've been checked above");
+
+        if self.current_offset < end_offset && self.current_offset % 2 == 1 {
+            let curr = self.inner.seek(SeekFrom::Current(-1))?;
+            cache_byte(&mut self.inner, &mut self.cached_byte)?;
+            self.inner.seek(SeekFrom::Start(curr))?;
+        } else {
+            self.cached_byte = None;
+        }
+
+        Ok(original_offset)
+    }
+}
+
+fn consume_channel<R: Read>(from: &mut R, to: &mut [u8]) -> io::Result<()> {
+    let mut ibuf = [0u8; 2];
+    from.read_exact(&mut ibuf)?;
+    NativeEndian::write_u16(to, BigEndian::read_u16(&ibuf));
+    Ok(())
+}
+
+fn cache_byte<R: Read>(from: &mut R, cached_byte: &mut Option<u8>) -> io::Result<u8> {
+    let mut obuf = [0u8; 2];
+    consume_channel(from, &mut obuf)?;
+    *cached_byte = Some(obuf[1]);
+    Ok(obuf[0])
+}
+
+/// farbfeld decoder
+pub struct FarbfeldDecoder<R: Read> {
+    reader: FarbfeldReader<R>,
+}
+
+impl<R: Read> FarbfeldDecoder<R> {
+    /// Creates a new decoder that decodes from the stream ```r```
+    pub fn new(buffered_read: R) -> ImageResult<FarbfeldDecoder<R>> {
+        Ok(FarbfeldDecoder {
+            reader: FarbfeldReader::new(buffered_read)?,
+        })
+    }
+}
+
+impl<'a, R: 'a + Read> ImageDecoder<'a> for FarbfeldDecoder<R> {
+    type Reader = FarbfeldReader<R>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        (self.reader.width, self.reader.height)
+    }
+
+    fn color_type(&self) -> ColorType {
+        ColorType::Rgba16
+    }
+
+    fn into_reader(self) -> ImageResult<Self::Reader> {
+        Ok(self.reader)
+    }
+
+    fn scanline_bytes(&self) -> u64 {
+        2
+    }
+}
+
+impl<'a, R: 'a + Read + Seek> ImageDecoderRect<'a> for FarbfeldDecoder<R> {
+    fn read_rect_with_progress<F: Fn(Progress)>(
+        &mut self,
+        x: u32,
+        y: u32,
+        width: u32,
+        height: u32,
+        buf: &mut [u8],
+        progress_callback: F,
+    ) -> ImageResult<()> {
+        // A "scanline" (defined as "shortest non-caching read" in the doc) is just one channel in this case
+
+        let start = self.reader.stream_position()?;
+        image::load_rect(
+            x,
+            y,
+            width,
+            height,
+            buf,
+            progress_callback,
+            self,
+            |s, scanline| s.reader.seek(SeekFrom::Start(scanline * 2)).map(|_| ()),
+            |s, buf| s.reader.read_exact(buf),
+        )?;
+        self.reader.seek(SeekFrom::Start(start))?;
+        Ok(())
+    }
+}
+
+/// farbfeld encoder
+pub struct FarbfeldEncoder<W: Write> {
+    w: W,
+}
+
+impl<W: Write> FarbfeldEncoder<W> {
+    /// Create a new encoder that writes its output to ```w```. The writer should be buffered.
+    pub fn new(buffered_writer: W) -> FarbfeldEncoder<W> {
+        FarbfeldEncoder { w: buffered_writer }
+    }
+
+    /// Encodes the image ```data``` (native endian)
+    /// that has dimensions ```width``` and ```height```
+    pub fn encode(self, data: &[u8], width: u32, height: u32) -> ImageResult<()> {
+        self.encode_impl(data, width, height)?;
+        Ok(())
+    }
+
+    fn encode_impl(mut self, data: &[u8], width: u32, height: u32) -> io::Result<()> {
+        self.w.write_all(b"farbfeld")?;
+
+        let mut buf = [0u8; 4];
+        BigEndian::write_u32(&mut buf, width);
+        self.w.write_all(&buf)?;
+
+        BigEndian::write_u32(&mut buf, height);
+        self.w.write_all(&buf)?;
+
+        for channel in data.chunks_exact(2) {
+            BigEndian::write_u16(&mut buf, NativeEndian::read_u16(channel));
+            self.w.write_all(&buf[..2])?;
+        }
+
+        Ok(())
+    }
+}
+
+impl<W: Write> ImageEncoder for FarbfeldEncoder<W> {
+    fn write_image(
+        self,
+        buf: &[u8],
+        width: u32,
+        height: u32,
+        color_type: ColorType,
+    ) -> ImageResult<()> {
+        if color_type != ColorType::Rgba16 {
+            return Err(ImageError::Unsupported(
+                UnsupportedError::from_format_and_kind(
+                    ImageFormat::Farbfeld.into(),
+                    UnsupportedErrorKind::Color(color_type.into()),
+                ),
+            ));
+        }
+
+        self.encode(buf, width, height)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::codecs::farbfeld::FarbfeldDecoder;
+    use crate::ImageDecoderRect;
+    use byteorder::{ByteOrder, NativeEndian};
+    use std::io::{Cursor, Seek, SeekFrom};
+
+    static RECTANGLE_IN: &[u8] =     b"farbfeld\
+                                       \x00\x00\x00\x02\x00\x00\x00\x03\
+                                       \xFF\x01\xFE\x02\xFD\x03\xFC\x04\xFB\x05\xFA\x06\xF9\x07\xF8\x08\
+                                       \xF7\x09\xF6\x0A\xF5\x0B\xF4\x0C\xF3\x0D\xF2\x0E\xF1\x0F\xF0\x10\
+                                       \xEF\x11\xEE\x12\xED\x13\xEC\x14\xEB\x15\xEA\x16\xE9\x17\xE8\x18";
+
+    #[test]
+    fn read_rect_1x2() {
+        static RECTANGLE_OUT: &[u16] = &[
+            0xF30D, 0xF20E, 0xF10F, 0xF010, 0xEB15, 0xEA16, 0xE917, 0xE818,
+        ];
+
+        read_rect(1, 1, 1, 2, RECTANGLE_OUT);
+    }
+
+    #[test]
+    fn read_rect_2x2() {
+        static RECTANGLE_OUT: &[u16] = &[
+            0xFF01, 0xFE02, 0xFD03, 0xFC04, 0xFB05, 0xFA06, 0xF907, 0xF808, 0xF709, 0xF60A, 0xF50B,
+            0xF40C, 0xF30D, 0xF20E, 0xF10F, 0xF010,
+        ];
+
+        read_rect(0, 0, 2, 2, RECTANGLE_OUT);
+    }
+
+    #[test]
+    fn read_rect_2x1() {
+        static RECTANGLE_OUT: &[u16] = &[
+            0xEF11, 0xEE12, 0xED13, 0xEC14, 0xEB15, 0xEA16, 0xE917, 0xE818,
+        ];
+
+        read_rect(0, 2, 2, 1, RECTANGLE_OUT);
+    }
+
+    #[test]
+    fn read_rect_2x3() {
+        static RECTANGLE_OUT: &[u16] = &[
+            0xFF01, 0xFE02, 0xFD03, 0xFC04, 0xFB05, 0xFA06, 0xF907, 0xF808, 0xF709, 0xF60A, 0xF50B,
+            0xF40C, 0xF30D, 0xF20E, 0xF10F, 0xF010, 0xEF11, 0xEE12, 0xED13, 0xEC14, 0xEB15, 0xEA16,
+            0xE917, 0xE818,
+        ];
+
+        read_rect(0, 0, 2, 3, RECTANGLE_OUT);
+    }
+
+    #[test]
+    fn read_rect_in_stream() {
+        static RECTANGLE_OUT: &[u16] = &[0xEF11, 0xEE12, 0xED13, 0xEC14];
+
+        let mut input = vec![];
+        input.extend_from_slice(b"This is a 31-byte-long prologue");
+        input.extend_from_slice(RECTANGLE_IN);
+        let mut input_cur = Cursor::new(input);
+        input_cur.seek(SeekFrom::Start(31)).unwrap();
+
+        let mut out_buf = [0u8; 64];
+        FarbfeldDecoder::new(input_cur)
+            .unwrap()
+            .read_rect(0, 2, 1, 1, &mut out_buf)
+            .unwrap();
+        let exp = degenerate_pixels(RECTANGLE_OUT);
+        assert_eq!(&out_buf[..exp.len()], &exp[..]);
+    }
+
+    #[test]
+    fn dimension_overflow() {
+        let header = b"farbfeld\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF";
+
+        assert!(FarbfeldDecoder::new(Cursor::new(header)).is_err());
+    }
+
+    fn read_rect(x: u32, y: u32, width: u32, height: u32, exp_wide: &[u16]) {
+        let mut out_buf = [0u8; 64];
+        FarbfeldDecoder::new(Cursor::new(RECTANGLE_IN))
+            .unwrap()
+            .read_rect(x, y, width, height, &mut out_buf)
+            .unwrap();
+        let exp = degenerate_pixels(exp_wide);
+        assert_eq!(&out_buf[..exp.len()], &exp[..]);
+    }
+
+    fn degenerate_pixels(exp_wide: &[u16]) -> Vec<u8> {
+        let mut exp = vec![0u8; exp_wide.len() * 2];
+        NativeEndian::write_u16_into(exp_wide, &mut exp);
+        exp
+    }
+}

vendor/image/src/codecs/gif.rs

@@ -0,0 +1,606 @@
+//!  Decoding of GIF Images
+//!
+//!  GIF (Graphics Interchange Format) is an image format that supports lossless compression.
+//!
+//!  # Related Links
+//!  * <http://www.w3.org/Graphics/GIF/spec-gif89a.txt> - The GIF Specification
+//!
+//! # Examples
+//! ```rust,no_run
+//! use image::codecs::gif::{GifDecoder, GifEncoder};
+//! use image::{ImageDecoder, AnimationDecoder};
+//! use std::fs::File;
+//! # fn main() -> std::io::Result<()> {
+//! // Decode a gif into frames
+//! let file_in = File::open("foo.gif")?;
+//! let mut decoder = GifDecoder::new(file_in).unwrap();
+//! let frames = decoder.into_frames();
+//! let frames = frames.collect_frames().expect("error decoding gif");
+//!
+//! // Encode frames into a gif and save to a file
+//! let mut file_out = File::open("out.gif")?;
+//! let mut encoder = GifEncoder::new(file_out);
+//! encoder.encode_frames(frames.into_iter());
+//! # Ok(())
+//! # }
+//! ```
+#![allow(clippy::while_let_loop)]
+
+use std::convert::TryFrom;
+use std::convert::TryInto;
+use std::io::{self, Cursor, Read, Write};
+use std::marker::PhantomData;
+use std::mem;
+
+use gif::ColorOutput;
+use gif::{DisposalMethod, Frame};
+use num_rational::Ratio;
+
+use crate::animation;
+use crate::color::{ColorType, Rgba};
+use crate::error::{
+    DecodingError, EncodingError, ImageError, ImageResult, ParameterError, ParameterErrorKind,
+    UnsupportedError, UnsupportedErrorKind,
+};
+use crate::image::{self, AnimationDecoder, ImageDecoder, ImageFormat};
+use crate::io::Limits;
+use crate::traits::Pixel;
+use crate::ImageBuffer;
+
+/// GIF decoder
+pub struct GifDecoder<R: Read> {
+    reader: gif::Decoder<R>,
+    limits: Limits,
+}
+
+impl<R: Read> GifDecoder<R> {
+    /// Creates a new decoder that decodes the input steam `r`
+    pub fn new(r: R) -> ImageResult<GifDecoder<R>> {
+        let mut decoder = gif::DecodeOptions::new();
+        decoder.set_color_output(ColorOutput::RGBA);
+
+        Ok(GifDecoder {
+            reader: decoder.read_info(r).map_err(ImageError::from_decoding)?,
+            limits: Limits::default(),
+        })
+    }
+
+    /// Creates a new decoder that decodes the input steam `r`, using limits `limits`
+    pub fn with_limits(r: R, limits: Limits) -> ImageResult<GifDecoder<R>> {
+        let mut decoder = gif::DecodeOptions::new();
+        decoder.set_color_output(ColorOutput::RGBA);
+
+        Ok(GifDecoder {
+            reader: decoder.read_info(r).map_err(ImageError::from_decoding)?,
+            limits,
+        })
+    }
+}
+
+/// Wrapper struct around a `Cursor<Vec<u8>>`
+pub struct GifReader<R>(Cursor<Vec<u8>>, PhantomData<R>);
+impl<R> Read for GifReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        self.0.read(buf)
+    }
+    fn read_to_end(&mut self, buf: &mut Vec<u8>) -> io::Result<usize> {
+        if self.0.position() == 0 && buf.is_empty() {
+            mem::swap(buf, self.0.get_mut());
+            Ok(buf.len())
+        } else {
+            self.0.read_to_end(buf)
+        }
+    }
+}
+
+impl<'a, R: 'a + Read> ImageDecoder<'a> for GifDecoder<R> {
+    type Reader = GifReader<R>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        (
+            u32::from(self.reader.width()),
+            u32::from(self.reader.height()),
+        )
+    }
+
+    fn color_type(&self) -> ColorType {
+        ColorType::Rgba8
+    }
+
+    fn into_reader(self) -> ImageResult<Self::Reader> {
+        Ok(GifReader(
+            Cursor::new(image::decoder_to_vec(self)?),
+            PhantomData,
+        ))
+    }
+
+    fn read_image(mut self, buf: &mut [u8]) -> ImageResult<()> {
+        assert_eq!(u64::try_from(buf.len()), Ok(self.total_bytes()));
+
+        let frame = match self
+            .reader
+            .next_frame_info()
+            .map_err(ImageError::from_decoding)?
+        {
+            Some(frame) => FrameInfo::new_from_frame(frame),
+            None => {
+                return Err(ImageError::Parameter(ParameterError::from_kind(
+                    ParameterErrorKind::NoMoreData,
+                )))
+            }
+        };
+
+        let (width, height) = self.dimensions();
+
+        if frame.left == 0
+            && frame.width == width
+            && (frame.top as u64 + frame.height as u64 <= height as u64)
+        {
+            // If the frame matches the logical screen, or, as a more general case,
+            // fits into it and touches its left and right borders, then
+            // we can directly write it into the buffer without causing line wraparound.
+            let line_length = usize::try_from(width)
+                .unwrap()
+                .checked_mul(self.color_type().bytes_per_pixel() as usize)
+                .unwrap();
+
+            // isolate the portion of the buffer to read the frame data into.
+            // the chunks above and below it are going to be zeroed.
+            let (blank_top, rest) =
+                buf.split_at_mut(line_length.checked_mul(frame.top as usize).unwrap());
+            let (buf, blank_bottom) =
+                rest.split_at_mut(line_length.checked_mul(frame.height as usize).unwrap());
+
+            debug_assert_eq!(buf.len(), self.reader.buffer_size());
+
+            // this is only necessary in case the buffer is not zeroed
+            for b in blank_top {
+                *b = 0;
+            }
+            // fill the middle section with the frame data
+            self.reader
+                .read_into_buffer(buf)
+                .map_err(ImageError::from_decoding)?;
+            // this is only necessary in case the buffer is not zeroed
+            for b in blank_bottom {
+                *b = 0;
+            }
+        } else {
+            // If the frame does not match the logical screen, read into an extra buffer
+            // and 'insert' the frame from left/top to logical screen width/height.
+            let buffer_size = self.reader.buffer_size();
+
+            self.limits.reserve_usize(buffer_size)?;
+
+            let mut frame_buffer = vec![0; buffer_size];
+
+            self.limits.free_usize(buffer_size);
+
+            self.reader
+                .read_into_buffer(&mut frame_buffer[..])
+                .map_err(ImageError::from_decoding)?;
+
+            let frame_buffer = ImageBuffer::from_raw(frame.width, frame.height, frame_buffer);
+            let image_buffer = ImageBuffer::from_raw(width, height, buf);
+
+            // `buffer_size` uses wrapping arithmetic, thus might not report the
+            // correct storage requirement if the result does not fit in `usize`.
+            // `ImageBuffer::from_raw` detects overflow and reports by returning `None`.
+            if frame_buffer.is_none() || image_buffer.is_none() {
+                return Err(ImageError::Unsupported(
+                    UnsupportedError::from_format_and_kind(
+                        ImageFormat::Gif.into(),
+                        UnsupportedErrorKind::GenericFeature(format!(
+                            "Image dimensions ({}, {}) are too large",
+                            frame.width, frame.height
+                        )),
+                    ),
+                ));
+            }
+
+            let frame_buffer = frame_buffer.unwrap();
+            let mut image_buffer = image_buffer.unwrap();
+
+            for (x, y, pixel) in image_buffer.enumerate_pixels_mut() {
+                let frame_x = x.wrapping_sub(frame.left);
+                let frame_y = y.wrapping_sub(frame.top);
+
+                if frame_x < frame.width && frame_y < frame.height {
+                    *pixel = *frame_buffer.get_pixel(frame_x, frame_y);
+                } else {
+                    // this is only necessary in case the buffer is not zeroed
+                    *pixel = Rgba([0, 0, 0, 0]);
+                }
+            }
+        }
+
+        Ok(())
+    }
+}
+
+struct GifFrameIterator<R: Read> {
+    reader: gif::Decoder<R>,
+
+    width: u32,
+    height: u32,
+
+    non_disposed_frame: ImageBuffer<Rgba<u8>, Vec<u8>>,
+}
+
+impl<R: Read> GifFrameIterator<R> {
+    fn new(decoder: GifDecoder<R>) -> GifFrameIterator<R> {
+        let (width, height) = decoder.dimensions();
+
+        // intentionally ignore the background color for web compatibility
+
+        // create the first non disposed frame
+        let non_disposed_frame = ImageBuffer::from_pixel(width, height, Rgba([0, 0, 0, 0]));
+
+        GifFrameIterator {
+            reader: decoder.reader,
+            width,
+            height,
+            non_disposed_frame,
+        }
+    }
+}
+
+impl<R: Read> Iterator for GifFrameIterator<R> {
+    type Item = ImageResult<animation::Frame>;
+
+    fn next(&mut self) -> Option<ImageResult<animation::Frame>> {
+        // begin looping over each frame
+
+        let frame = match self.reader.next_frame_info() {
+            Ok(frame_info) => {
+                if let Some(frame) = frame_info {
+                    FrameInfo::new_from_frame(frame)
+                } else {
+                    // no more frames
+                    return None;
+                }
+            }
+            Err(err) => return Some(Err(ImageError::from_decoding(err))),
+        };
+
+        let mut vec = vec![0; self.reader.buffer_size()];
+        if let Err(err) = self.reader.read_into_buffer(&mut vec) {
+            return Some(Err(ImageError::from_decoding(err)));
+        }
+
+        // create the image buffer from the raw frame.
+        // `buffer_size` uses wrapping arithmetic, thus might not report the
+        // correct storage requirement if the result does not fit in `usize`.
+        // on the other hand, `ImageBuffer::from_raw` detects overflow and
+        // reports by returning `None`.
+        let mut frame_buffer = match ImageBuffer::from_raw(frame.width, frame.height, vec) {
+            Some(frame_buffer) => frame_buffer,
+            None => {
+                return Some(Err(ImageError::Unsupported(
+                    UnsupportedError::from_format_and_kind(
+                        ImageFormat::Gif.into(),
+                        UnsupportedErrorKind::GenericFeature(format!(
+                            "Image dimensions ({}, {}) are too large",
+                            frame.width, frame.height
+                        )),
+                    ),
+                )))
+            }
+        };
+
+        // blend the current frame with the non-disposed frame, then update
+        // the non-disposed frame according to the disposal method.
+        fn blend_and_dispose_pixel(
+            dispose: DisposalMethod,
+            previous: &mut Rgba<u8>,
+            current: &mut Rgba<u8>,
+        ) {
+            let pixel_alpha = current.channels()[3];
+            if pixel_alpha == 0 {
+                *current = *previous;
+            }
+
+            match dispose {
+                DisposalMethod::Any | DisposalMethod::Keep => {
+                    // do not dispose
+                    // (keep pixels from this frame)
+                    // note: the `Any` disposal method is underspecified in the GIF
+                    // spec, but most viewers treat it identically to `Keep`
+                    *previous = *current;
+                }
+                DisposalMethod::Background => {
+                    // restore to background color
+                    // (background shows through transparent pixels in the next frame)
+                    *previous = Rgba([0, 0, 0, 0]);
+                }
+                DisposalMethod::Previous => {
+                    // restore to previous
+                    // (dispose frames leaving the last none disposal frame)
+                }
+            }
+        }
+
+        // if `frame_buffer`'s frame exactly matches the entire image, then
+        // use it directly, else create a new buffer to hold the composited
+        // image.
+        let image_buffer = if (frame.left, frame.top) == (0, 0)
+            && (self.width, self.height) == frame_buffer.dimensions()
+        {
+            for (x, y, pixel) in frame_buffer.enumerate_pixels_mut() {
+                let previous_pixel = self.non_disposed_frame.get_pixel_mut(x, y);
+                blend_and_dispose_pixel(frame.disposal_method, previous_pixel, pixel);
+            }
+            frame_buffer
+        } else {
+            ImageBuffer::from_fn(self.width, self.height, |x, y| {
+                let frame_x = x.wrapping_sub(frame.left);
+                let frame_y = y.wrapping_sub(frame.top);
+                let previous_pixel = self.non_disposed_frame.get_pixel_mut(x, y);
+
+                if frame_x < frame_buffer.width() && frame_y < frame_buffer.height() {
+                    let mut pixel = *frame_buffer.get_pixel(frame_x, frame_y);
+                    blend_and_dispose_pixel(frame.disposal_method, previous_pixel, &mut pixel);
+                    pixel
+                } else {
+                    // out of bounds, return pixel from previous frame
+                    *previous_pixel
+                }
+            })
+        };
+
+        Some(Ok(animation::Frame::from_parts(
+            image_buffer,
+            0,
+            0,
+            frame.delay,
+        )))
+    }
+}
+
+impl<'a, R: Read + 'a> AnimationDecoder<'a> for GifDecoder<R> {
+    fn into_frames(self) -> animation::Frames<'a> {
+        animation::Frames::new(Box::new(GifFrameIterator::new(self)))
+    }
+}
+
+struct FrameInfo {
+    left: u32,
+    top: u32,
+    width: u32,
+    height: u32,
+    disposal_method: DisposalMethod,
+    delay: animation::Delay,
+}
+
+impl FrameInfo {
+    fn new_from_frame(frame: &Frame) -> FrameInfo {
+        FrameInfo {
+            left: u32::from(frame.left),
+            top: u32::from(frame.top),
+            width: u32::from(frame.width),
+            height: u32::from(frame.height),
+            disposal_method: frame.dispose,
+            // frame.delay is in units of 10ms so frame.delay*10 is in ms
+            delay: animation::Delay::from_ratio(Ratio::new(u32::from(frame.delay) * 10, 1)),
+        }
+    }
+}
+
+/// Number of repetitions for a GIF animation
+#[derive(Clone, Copy, Debug)]
+pub enum Repeat {
+    /// Finite number of repetitions
+    Finite(u16),
+    /// Looping GIF
+    Infinite,
+}
+
+impl Repeat {
+    pub(crate) fn to_gif_enum(&self) -> gif::Repeat {
+        match self {
+            Repeat::Finite(n) => gif::Repeat::Finite(*n),
+            Repeat::Infinite => gif::Repeat::Infinite,
+        }
+    }
+}
+
+/// GIF encoder.
+pub struct GifEncoder<W: Write> {
+    w: Option<W>,
+    gif_encoder: Option<gif::Encoder<W>>,
+    speed: i32,
+    repeat: Option<Repeat>,
+}
+
+impl<W: Write> GifEncoder<W> {
+    /// Creates a new GIF encoder with a speed of 1. This prioritizes quality over performance at any cost.
+    pub fn new(w: W) -> GifEncoder<W> {
+        Self::new_with_speed(w, 1)
+    }
+
+    /// Create a new GIF encoder, and has the speed parameter `speed`. See
+    /// [`Frame::from_rgba_speed`](https://docs.rs/gif/latest/gif/struct.Frame.html#method.from_rgba_speed)
+    /// for more information.
+    pub fn new_with_speed(w: W, speed: i32) -> GifEncoder<W> {
+        assert!(
+            (1..=30).contains(&speed),
+            "speed needs to be in the range [1, 30]"
+        );
+        GifEncoder {
+            w: Some(w),
+            gif_encoder: None,
+            speed,
+            repeat: None,
+        }
+    }
+
+    /// Set the repeat behaviour of the encoded GIF
+    pub fn set_repeat(&mut self, repeat: Repeat) -> ImageResult<()> {
+        if let Some(ref mut encoder) = self.gif_encoder {
+            encoder
+                .set_repeat(repeat.to_gif_enum())
+                .map_err(ImageError::from_encoding)?;
+        }
+        self.repeat = Some(repeat);
+        Ok(())
+    }
+
+    /// Encode a single image.
+    pub fn encode(
+        &mut self,
+        data: &[u8],
+        width: u32,
+        height: u32,
+        color: ColorType,
+    ) -> ImageResult<()> {
+        let (width, height) = self.gif_dimensions(width, height)?;
+        match color {
+            ColorType::Rgb8 => self.encode_gif(Frame::from_rgb(width, height, data)),
+            ColorType::Rgba8 => {
+                self.encode_gif(Frame::from_rgba(width, height, &mut data.to_owned()))
+            }
+            _ => Err(ImageError::Unsupported(
+                UnsupportedError::from_format_and_kind(
+                    ImageFormat::Gif.into(),
+                    UnsupportedErrorKind::Color(color.into()),
+                ),
+            )),
+        }
+    }
+
+    /// Encode one frame of animation.
+    pub fn encode_frame(&mut self, img_frame: animation::Frame) -> ImageResult<()> {
+        let frame = self.convert_frame(img_frame)?;
+        self.encode_gif(frame)
+    }
+
+    /// Encodes Frames.
+    /// Consider using `try_encode_frames` instead to encode an `animation::Frames` like iterator.
+    pub fn encode_frames<F>(&mut self, frames: F) -> ImageResult<()>
+    where
+        F: IntoIterator<Item = animation::Frame>,
+    {
+        for img_frame in frames {
+            self.encode_frame(img_frame)?;
+        }
+        Ok(())
+    }
+
+    /// Try to encode a collection of `ImageResult<animation::Frame>` objects.
+    /// Use this function to encode an `animation::Frames` like iterator.
+    /// Whenever an `Err` item is encountered, that value is returned without further actions.
+    pub fn try_encode_frames<F>(&mut self, frames: F) -> ImageResult<()>
+    where
+        F: IntoIterator<Item = ImageResult<animation::Frame>>,
+    {
+        for img_frame in frames {
+            self.encode_frame(img_frame?)?;
+        }
+        Ok(())
+    }
+
+    pub(crate) fn convert_frame(
+        &mut self,
+        img_frame: animation::Frame,
+    ) -> ImageResult<Frame<'static>> {
+        // get the delay before converting img_frame
+        let frame_delay = img_frame.delay().into_ratio().to_integer();
+        // convert img_frame into RgbaImage
+        let mut rbga_frame = img_frame.into_buffer();
+        let (width, height) = self.gif_dimensions(rbga_frame.width(), rbga_frame.height())?;
+
+        // Create the gif::Frame from the animation::Frame
+        let mut frame = Frame::from_rgba_speed(width, height, &mut rbga_frame, self.speed);
+        // Saturate the conversion to u16::MAX instead of returning an error as that
+        // would require a new special cased variant in ParameterErrorKind which most
+        // likely couldn't be reused for other cases. This isn't a bad trade-off given
+        // that the current algorithm is already lossy.
+        frame.delay = (frame_delay / 10).try_into().unwrap_or(std::u16::MAX);
+
+        Ok(frame)
+    }
+
+    fn gif_dimensions(&self, width: u32, height: u32) -> ImageResult<(u16, u16)> {
+        fn inner_dimensions(width: u32, height: u32) -> Option<(u16, u16)> {
+            let width = u16::try_from(width).ok()?;
+            let height = u16::try_from(height).ok()?;
+            Some((width, height))
+        }
+
+        // TODO: this is not very idiomatic yet. Should return an EncodingError.
+        inner_dimensions(width, height).ok_or_else(|| {
+            ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::DimensionMismatch,
+            ))
+        })
+    }
+
+    pub(crate) fn encode_gif(&mut self, mut frame: Frame) -> ImageResult<()> {
+        let gif_encoder;
+        if let Some(ref mut encoder) = self.gif_encoder {
+            gif_encoder = encoder;
+        } else {
+            let writer = self.w.take().unwrap();
+            let mut encoder = gif::Encoder::new(writer, frame.width, frame.height, &[])
+                .map_err(ImageError::from_encoding)?;
+            if let Some(ref repeat) = self.repeat {
+                encoder
+                    .set_repeat(repeat.to_gif_enum())
+                    .map_err(ImageError::from_encoding)?;
+            }
+            self.gif_encoder = Some(encoder);
+            gif_encoder = self.gif_encoder.as_mut().unwrap()
+        }
+
+        frame.dispose = gif::DisposalMethod::Background;
+
+        gif_encoder
+            .write_frame(&frame)
+            .map_err(ImageError::from_encoding)
+    }
+}
+
+impl ImageError {
+    fn from_decoding(err: gif::DecodingError) -> ImageError {
+        use gif::DecodingError::*;
+        match err {
+            err @ Format(_) => {
+                ImageError::Decoding(DecodingError::new(ImageFormat::Gif.into(), err))
+            }
+            Io(io_err) => ImageError::IoError(io_err),
+        }
+    }
+
+    fn from_encoding(err: gif::EncodingError) -> ImageError {
+        use gif::EncodingError::*;
+        match err {
+            err @ Format(_) => {
+                ImageError::Encoding(EncodingError::new(ImageFormat::Gif.into(), err))
+            }
+            Io(io_err) => ImageError::IoError(io_err),
+        }
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn frames_exceeding_logical_screen_size() {
+        // This is a gif with 10x10 logical screen, but a 16x16 frame + 6px offset inside.
+        let data = vec![
+            0x47, 0x49, 0x46, 0x38, 0x39, 0x61, 0x0A, 0x00, 0x0A, 0x00, 0xF0, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x0E, 0xFF, 0x1F, 0x21, 0xF9, 0x04, 0x09, 0x64, 0x00, 0x00, 0x00, 0x2C,
+            0x06, 0x00, 0x06, 0x00, 0x10, 0x00, 0x10, 0x00, 0x00, 0x02, 0x23, 0x84, 0x8F, 0xA9,
+            0xBB, 0xE1, 0xE8, 0x42, 0x8A, 0x0F, 0x50, 0x79, 0xAE, 0xD1, 0xF9, 0x7A, 0xE8, 0x71,
+            0x5B, 0x48, 0x81, 0x64, 0xD5, 0x91, 0xCA, 0x89, 0x4D, 0x21, 0x63, 0x89, 0x4C, 0x09,
+            0x77, 0xF5, 0x6D, 0x14, 0x00, 0x3B,
+        ];
+
+        let decoder = GifDecoder::new(Cursor::new(data)).unwrap();
+        let mut buf = vec![0u8; decoder.total_bytes() as usize];
+
+        assert!(decoder.read_image(&mut buf).is_ok());
+    }
+}

vendor/image/src/codecs/hdr/encoder.rs

@@ -0,0 +1,433 @@
+use crate::codecs::hdr::{rgbe8, Rgbe8Pixel, SIGNATURE};
+use crate::color::Rgb;
+use crate::error::ImageResult;
+use std::cmp::Ordering;
+use std::io::{Result, Write};
+
+/// Radiance HDR encoder
+pub struct HdrEncoder<W: Write> {
+    w: W,
+}
+
+impl<W: Write> HdrEncoder<W> {
+    /// Creates encoder
+    pub fn new(w: W) -> HdrEncoder<W> {
+        HdrEncoder { w }
+    }
+
+    /// Encodes the image ```data```
+    /// that has dimensions ```width``` and ```height```
+    pub fn encode(mut self, data: &[Rgb<f32>], width: usize, height: usize) -> ImageResult<()> {
+        assert!(data.len() >= width * height);
+        let w = &mut self.w;
+        w.write_all(SIGNATURE)?;
+        w.write_all(b"\n")?;
+        w.write_all(b"# Rust HDR encoder\n")?;
+        w.write_all(b"FORMAT=32-bit_rle_rgbe\n\n")?;
+        w.write_all(format!("-Y {} +X {}\n", height, width).as_bytes())?;
+
+        if !(8..=32_768).contains(&width) {
+            for &pix in data {
+                write_rgbe8(w, to_rgbe8(pix))?;
+            }
+        } else {
+            // new RLE marker contains scanline width
+            let marker = rgbe8(2, 2, (width / 256) as u8, (width % 256) as u8);
+            // buffers for encoded pixels
+            let mut bufr = vec![0; width];
+            let mut bufg = vec![0; width];
+            let mut bufb = vec![0; width];
+            let mut bufe = vec![0; width];
+            let mut rle_buf = vec![0; width];
+            for scanline in data.chunks(width) {
+                for ((((r, g), b), e), &pix) in bufr
+                    .iter_mut()
+                    .zip(bufg.iter_mut())
+                    .zip(bufb.iter_mut())
+                    .zip(bufe.iter_mut())
+                    .zip(scanline.iter())
+                {
+                    let cp = to_rgbe8(pix);
+                    *r = cp.c[0];
+                    *g = cp.c[1];
+                    *b = cp.c[2];
+                    *e = cp.e;
+                }
+                write_rgbe8(w, marker)?; // New RLE encoding marker
+                rle_buf.clear();
+                rle_compress(&bufr[..], &mut rle_buf);
+                w.write_all(&rle_buf[..])?;
+                rle_buf.clear();
+                rle_compress(&bufg[..], &mut rle_buf);
+                w.write_all(&rle_buf[..])?;
+                rle_buf.clear();
+                rle_compress(&bufb[..], &mut rle_buf);
+                w.write_all(&rle_buf[..])?;
+                rle_buf.clear();
+                rle_compress(&bufe[..], &mut rle_buf);
+                w.write_all(&rle_buf[..])?;
+            }
+        }
+        Ok(())
+    }
+}
+
+#[derive(Debug, PartialEq, Eq)]
+enum RunOrNot {
+    Run(u8, usize),
+    Norun(usize, usize),
+}
+use self::RunOrNot::{Norun, Run};
+
+const RUN_MAX_LEN: usize = 127;
+const NORUN_MAX_LEN: usize = 128;
+
+struct RunIterator<'a> {
+    data: &'a [u8],
+    curidx: usize,
+}
+
+impl<'a> RunIterator<'a> {
+    fn new(data: &'a [u8]) -> RunIterator<'a> {
+        RunIterator { data, curidx: 0 }
+    }
+}
+
+impl<'a> Iterator for RunIterator<'a> {
+    type Item = RunOrNot;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        if self.curidx == self.data.len() {
+            None
+        } else {
+            let cv = self.data[self.curidx];
+            let crun = self.data[self.curidx..]
+                .iter()
+                .take_while(|&&v| v == cv)
+                .take(RUN_MAX_LEN)
+                .count();
+            let ret = if crun > 2 {
+                Run(cv, crun)
+            } else {
+                Norun(self.curidx, crun)
+            };
+            self.curidx += crun;
+            Some(ret)
+        }
+    }
+}
+
+struct NorunCombineIterator<'a> {
+    runiter: RunIterator<'a>,
+    prev: Option<RunOrNot>,
+}
+
+impl<'a> NorunCombineIterator<'a> {
+    fn new(data: &'a [u8]) -> NorunCombineIterator<'a> {
+        NorunCombineIterator {
+            runiter: RunIterator::new(data),
+            prev: None,
+        }
+    }
+}
+
+// Combines sequential noruns produced by RunIterator
+impl<'a> Iterator for NorunCombineIterator<'a> {
+    type Item = RunOrNot;
+    fn next(&mut self) -> Option<Self::Item> {
+        loop {
+            match self.prev.take() {
+                Some(Run(c, len)) => {
+                    // Just return stored run
+                    return Some(Run(c, len));
+                }
+                Some(Norun(idx, len)) => {
+                    // Let's see if we need to continue norun
+                    match self.runiter.next() {
+                        Some(Norun(_, len1)) => {
+                            // norun continues
+                            let clen = len + len1; // combined length
+                            match clen.cmp(&NORUN_MAX_LEN) {
+                                Ordering::Equal => return Some(Norun(idx, clen)),
+                                Ordering::Greater => {
+                                    // combined norun exceeds maximum length. store extra part of norun
+                                    self.prev =
+                                        Some(Norun(idx + NORUN_MAX_LEN, clen - NORUN_MAX_LEN));
+                                    // then return maximal norun
+                                    return Some(Norun(idx, NORUN_MAX_LEN));
+                                }
+                                Ordering::Less => {
+                                    // len + len1 < NORUN_MAX_LEN
+                                    self.prev = Some(Norun(idx, len + len1));
+                                    // combine and continue loop
+                                }
+                            }
+                        }
+                        Some(Run(c, len1)) => {
+                            // Run encountered. Store it
+                            self.prev = Some(Run(c, len1));
+                            return Some(Norun(idx, len)); // and return combined norun
+                        }
+                        None => {
+                            // End of sequence
+                            return Some(Norun(idx, len)); // return combined norun
+                        }
+                    }
+                } // End match self.prev.take() == Some(NoRun())
+                None => {
+                    // No norun to combine
+                    match self.runiter.next() {
+                        Some(Norun(idx, len)) => {
+                            self.prev = Some(Norun(idx, len));
+                            // store for combine and continue the loop
+                        }
+                        Some(Run(c, len)) => {
+                            // Some run. Just return it
+                            return Some(Run(c, len));
+                        }
+                        None => {
+                            // That's all, folks
+                            return None;
+                        }
+                    }
+                } // End match self.prev.take() == None
+            } // End match
+        } // End loop
+    }
+}
+
+// Appends RLE compressed ```data``` to ```rle```
+fn rle_compress(data: &[u8], rle: &mut Vec<u8>) {
+    rle.clear();
+    if data.is_empty() {
+        rle.push(0); // Technically correct. It means read next 0 bytes.
+        return;
+    }
+    // Task: split data into chunks of repeating (max 127) and non-repeating bytes (max 128)
+    // Prepend non-repeating chunk with its length
+    // Replace repeating byte with (run length + 128) and the byte
+    for rnr in NorunCombineIterator::new(data) {
+        match rnr {
+            Run(c, len) => {
+                assert!(len <= 127);
+                rle.push(128u8 + len as u8);
+                rle.push(c);
+            }
+            Norun(idx, len) => {
+                assert!(len <= 128);
+                rle.push(len as u8);
+                rle.extend_from_slice(&data[idx..idx + len]);
+            }
+        }
+    }
+}
+
+fn write_rgbe8<W: Write>(w: &mut W, v: Rgbe8Pixel) -> Result<()> {
+    w.write_all(&[v.c[0], v.c[1], v.c[2], v.e])
+}
+
+/// Converts ```Rgb<f32>``` into ```Rgbe8Pixel```
+pub fn to_rgbe8(pix: Rgb<f32>) -> Rgbe8Pixel {
+    let pix = pix.0;
+    let mx = f32::max(pix[0], f32::max(pix[1], pix[2]));
+    if mx <= 0.0 {
+        Rgbe8Pixel { c: [0, 0, 0], e: 0 }
+    } else {
+        // let (frac, exp) = mx.frexp(); // unstable yet
+        let exp = mx.log2().floor() as i32 + 1;
+        let mul = f32::powi(2.0, exp);
+        let mut conv = [0u8; 3];
+        for (cv, &sv) in conv.iter_mut().zip(pix.iter()) {
+            *cv = f32::trunc(sv / mul * 256.0) as u8;
+        }
+        Rgbe8Pixel {
+            c: conv,
+            e: (exp + 128) as u8,
+        }
+    }
+}
+
+#[test]
+fn to_rgbe8_test() {
+    use crate::codecs::hdr::rgbe8;
+    let test_cases = vec![rgbe8(0, 0, 0, 0), rgbe8(1, 1, 128, 128)];
+    for &pix in &test_cases {
+        assert_eq!(pix, to_rgbe8(pix.to_hdr()));
+    }
+    for mc in 128..255 {
+        // TODO: use inclusive range when stable
+        let pix = rgbe8(mc, mc, mc, 100);
+        assert_eq!(pix, to_rgbe8(pix.to_hdr()));
+        let pix = rgbe8(mc, 0, mc, 130);
+        assert_eq!(pix, to_rgbe8(pix.to_hdr()));
+        let pix = rgbe8(0, 0, mc, 140);
+        assert_eq!(pix, to_rgbe8(pix.to_hdr()));
+        let pix = rgbe8(1, 0, mc, 150);
+        assert_eq!(pix, to_rgbe8(pix.to_hdr()));
+        let pix = rgbe8(1, mc, 10, 128);
+        assert_eq!(pix, to_rgbe8(pix.to_hdr()));
+        for c in 0..255 {
+            // Radiance HDR seems to be pre IEEE 754.
+            // exponent can be -128 (represented as 0u8), so some colors cannot be represented in normalized f32
+            // Let's exclude exponent value of -128 (0u8) from testing
+            let pix = rgbe8(1, mc, c, if c == 0 { 1 } else { c });
+            assert_eq!(pix, to_rgbe8(pix.to_hdr()));
+        }
+    }
+    fn relative_dist(a: Rgb<f32>, b: Rgb<f32>) -> f32 {
+        // maximal difference divided by maximal value
+        let max_diff =
+            a.0.iter()
+                .zip(b.0.iter())
+                .fold(0.0, |diff, (&a, &b)| f32::max(diff, (a - b).abs()));
+        let max_val =
+            a.0.iter()
+                .chain(b.0.iter())
+                .fold(0.0, |maxv, &a| f32::max(maxv, a));
+        if max_val == 0.0 {
+            0.0
+        } else {
+            max_diff / max_val
+        }
+    }
+    let test_values = vec![
+        0.000_001, 0.000_02, 0.000_3, 0.004, 0.05, 0.6, 7.0, 80.0, 900.0, 1_000.0, 20_000.0,
+        300_000.0,
+    ];
+    for &r in &test_values {
+        for &g in &test_values {
+            for &b in &test_values {
+                let c1 = Rgb([r, g, b]);
+                let c2 = to_rgbe8(c1).to_hdr();
+                let rel_dist = relative_dist(c1, c2);
+                // Maximal value is normalized to the range 128..256, thus we have 1/128 precision
+                assert!(
+                    rel_dist <= 1.0 / 128.0,
+                    "Relative distance ({}) exceeds 1/128 for {:?} and {:?}",
+                    rel_dist,
+                    c1,
+                    c2
+                );
+            }
+        }
+    }
+}
+
+#[test]
+fn runiterator_test() {
+    let data = [];
+    let mut run_iter = RunIterator::new(&data[..]);
+    assert_eq!(run_iter.next(), None);
+    let data = [5];
+    let mut run_iter = RunIterator::new(&data[..]);
+    assert_eq!(run_iter.next(), Some(Norun(0, 1)));
+    assert_eq!(run_iter.next(), None);
+    let data = [1, 1];
+    let mut run_iter = RunIterator::new(&data[..]);
+    assert_eq!(run_iter.next(), Some(Norun(0, 2)));
+    assert_eq!(run_iter.next(), None);
+    let data = [0, 0, 0];
+    let mut run_iter = RunIterator::new(&data[..]);
+    assert_eq!(run_iter.next(), Some(Run(0u8, 3)));
+    assert_eq!(run_iter.next(), None);
+    let data = [0, 0, 1, 1];
+    let mut run_iter = RunIterator::new(&data[..]);
+    assert_eq!(run_iter.next(), Some(Norun(0, 2)));
+    assert_eq!(run_iter.next(), Some(Norun(2, 2)));
+    assert_eq!(run_iter.next(), None);
+    let data = [0, 0, 0, 1, 1];
+    let mut run_iter = RunIterator::new(&data[..]);
+    assert_eq!(run_iter.next(), Some(Run(0u8, 3)));
+    assert_eq!(run_iter.next(), Some(Norun(3, 2)));
+    assert_eq!(run_iter.next(), None);
+    let data = [1, 2, 2, 2];
+    let mut run_iter = RunIterator::new(&data[..]);
+    assert_eq!(run_iter.next(), Some(Norun(0, 1)));
+    assert_eq!(run_iter.next(), Some(Run(2u8, 3)));
+    assert_eq!(run_iter.next(), None);
+    let data = [1, 1, 2, 2, 2];
+    let mut run_iter = RunIterator::new(&data[..]);
+    assert_eq!(run_iter.next(), Some(Norun(0, 2)));
+    assert_eq!(run_iter.next(), Some(Run(2u8, 3)));
+    assert_eq!(run_iter.next(), None);
+    let data = [2; 128];
+    let mut run_iter = RunIterator::new(&data[..]);
+    assert_eq!(run_iter.next(), Some(Run(2u8, 127)));
+    assert_eq!(run_iter.next(), Some(Norun(127, 1)));
+    assert_eq!(run_iter.next(), None);
+    let data = [2; 129];
+    let mut run_iter = RunIterator::new(&data[..]);
+    assert_eq!(run_iter.next(), Some(Run(2u8, 127)));
+    assert_eq!(run_iter.next(), Some(Norun(127, 2)));
+    assert_eq!(run_iter.next(), None);
+    let data = [2; 130];
+    let mut run_iter = RunIterator::new(&data[..]);
+    assert_eq!(run_iter.next(), Some(Run(2u8, 127)));
+    assert_eq!(run_iter.next(), Some(Run(2u8, 3)));
+    assert_eq!(run_iter.next(), None);
+}
+
+#[test]
+fn noruncombine_test() {
+    fn a<T>(mut v: Vec<T>, mut other: Vec<T>) -> Vec<T> {
+        v.append(&mut other);
+        v
+    }
+
+    let v = vec![];
+    let mut rsi = NorunCombineIterator::new(&v[..]);
+    assert_eq!(rsi.next(), None);
+
+    let v = vec![1];
+    let mut rsi = NorunCombineIterator::new(&v[..]);
+    assert_eq!(rsi.next(), Some(Norun(0, 1)));
+    assert_eq!(rsi.next(), None);
+
+    let v = vec![2, 2];
+    let mut rsi = NorunCombineIterator::new(&v[..]);
+    assert_eq!(rsi.next(), Some(Norun(0, 2)));
+    assert_eq!(rsi.next(), None);
+
+    let v = vec![3, 3, 3];
+    let mut rsi = NorunCombineIterator::new(&v[..]);
+    assert_eq!(rsi.next(), Some(Run(3, 3)));
+    assert_eq!(rsi.next(), None);
+
+    let v = vec![4, 4, 3, 3, 3];
+    let mut rsi = NorunCombineIterator::new(&v[..]);
+    assert_eq!(rsi.next(), Some(Norun(0, 2)));
+    assert_eq!(rsi.next(), Some(Run(3, 3)));
+    assert_eq!(rsi.next(), None);
+
+    let v = vec![40; 400];
+    let mut rsi = NorunCombineIterator::new(&v[..]);
+    assert_eq!(rsi.next(), Some(Run(40, 127)));
+    assert_eq!(rsi.next(), Some(Run(40, 127)));
+    assert_eq!(rsi.next(), Some(Run(40, 127)));
+    assert_eq!(rsi.next(), Some(Run(40, 19)));
+    assert_eq!(rsi.next(), None);
+
+    let v = a(a(vec![5; 3], vec![6; 129]), vec![7, 3, 7, 10, 255]);
+    let mut rsi = NorunCombineIterator::new(&v[..]);
+    assert_eq!(rsi.next(), Some(Run(5, 3)));
+    assert_eq!(rsi.next(), Some(Run(6, 127)));
+    assert_eq!(rsi.next(), Some(Norun(130, 7)));
+    assert_eq!(rsi.next(), None);
+
+    let v = a(a(vec![5; 2], vec![6; 129]), vec![7, 3, 7, 7, 255]);
+    let mut rsi = NorunCombineIterator::new(&v[..]);
+    assert_eq!(rsi.next(), Some(Norun(0, 2)));
+    assert_eq!(rsi.next(), Some(Run(6, 127)));
+    assert_eq!(rsi.next(), Some(Norun(129, 7)));
+    assert_eq!(rsi.next(), None);
+
+    let v: Vec<_> = ::std::iter::repeat(())
+        .flat_map(|_| (0..2))
+        .take(257)
+        .collect();
+    let mut rsi = NorunCombineIterator::new(&v[..]);
+    assert_eq!(rsi.next(), Some(Norun(0, 128)));
+    assert_eq!(rsi.next(), Some(Norun(128, 128)));
+    assert_eq!(rsi.next(), Some(Norun(256, 1)));
+    assert_eq!(rsi.next(), None);
+}

vendor/image/src/codecs/hdr/mod.rs

@@ -0,0 +1,15 @@
+//!  Decoding of Radiance HDR Images
+//!
+//!  A decoder for Radiance HDR images
+//!
+//!  # Related Links
+//!
+//!  * <http://radsite.lbl.gov/radiance/refer/filefmts.pdf>
+//!  * <http://www.graphics.cornell.edu/~bjw/rgbe/rgbe.c>
+//!
+
+mod decoder;
+mod encoder;
+
+pub use self::decoder::*;
+pub use self::encoder::*;

vendor/image/src/codecs/ico/decoder.rs

@@ -0,0 +1,470 @@
+use byteorder::{LittleEndian, ReadBytesExt};
+use std::convert::TryFrom;
+use std::io::{self, Cursor, Read, Seek, SeekFrom};
+use std::marker::PhantomData;
+use std::{error, fmt, mem};
+
+use crate::color::ColorType;
+use crate::error::{
+    DecodingError, ImageError, ImageResult, UnsupportedError, UnsupportedErrorKind,
+};
+use crate::image::{self, ImageDecoder, ImageFormat};
+
+use self::InnerDecoder::*;
+use crate::codecs::bmp::BmpDecoder;
+use crate::codecs::png::{PngDecoder, PNG_SIGNATURE};
+
+/// Errors that can occur during decoding and parsing an ICO image or one of its enclosed images.
+#[derive(Debug, Copy, Clone, Hash, PartialEq, Eq, PartialOrd, Ord)]
+enum DecoderError {
+    /// The ICO directory is empty
+    NoEntries,
+    /// The number of color planes (0 or 1), or the horizontal coordinate of the hotspot for CUR files too big.
+    IcoEntryTooManyPlanesOrHotspot,
+    /// The bit depth (may be 0 meaning unspecified), or the vertical coordinate of the hotspot for CUR files too big.
+    IcoEntryTooManyBitsPerPixelOrHotspot,
+
+    /// The entry is in PNG format and specified a length that is shorter than PNG header.
+    PngShorterThanHeader,
+    /// The enclosed PNG is not in RGBA, which is invalid: https://blogs.msdn.microsoft.com/oldnewthing/20101022-00/?p=12473/.
+    PngNotRgba,
+
+    /// The entry is in BMP format and specified a data size that is not correct for the image and optional mask data.
+    InvalidDataSize,
+
+    /// The dimensions specified by the entry does not match the dimensions in the header of the enclosed image.
+    ImageEntryDimensionMismatch {
+        /// The mismatched subimage's type
+        format: IcoEntryImageFormat,
+        /// The dimensions specified by the entry
+        entry: (u16, u16),
+        /// The dimensions of the image itself
+        image: (u32, u32),
+    },
+}
+
+impl fmt::Display for DecoderError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            DecoderError::NoEntries => f.write_str("ICO directory contains no image"),
+            DecoderError::IcoEntryTooManyPlanesOrHotspot => {
+                f.write_str("ICO image entry has too many color planes or too large hotspot value")
+            }
+            DecoderError::IcoEntryTooManyBitsPerPixelOrHotspot => f.write_str(
+                "ICO image entry has too many bits per pixel or too large hotspot value",
+            ),
+            DecoderError::PngShorterThanHeader => {
+                f.write_str("Entry specified a length that is shorter than PNG header!")
+            }
+            DecoderError::PngNotRgba => f.write_str("The PNG is not in RGBA format!"),
+            DecoderError::InvalidDataSize => {
+                f.write_str("ICO image data size did not match expected size")
+            }
+            DecoderError::ImageEntryDimensionMismatch {
+                format,
+                entry,
+                image,
+            } => f.write_fmt(format_args!(
+                "Entry{:?} and {}{:?} dimensions do not match!",
+                entry, format, image
+            )),
+        }
+    }
+}
+
+impl From<DecoderError> for ImageError {
+    fn from(e: DecoderError) -> ImageError {
+        ImageError::Decoding(DecodingError::new(ImageFormat::Ico.into(), e))
+    }
+}
+
+impl error::Error for DecoderError {}
+
+/// The image formats an ICO may contain
+#[derive(Debug, Copy, Clone, Hash, PartialEq, Eq, PartialOrd, Ord)]
+enum IcoEntryImageFormat {
+    /// PNG in ARGB
+    Png,
+    /// BMP with optional alpha mask
+    Bmp,
+}
+
+impl fmt::Display for IcoEntryImageFormat {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        f.write_str(match self {
+            IcoEntryImageFormat::Png => "PNG",
+            IcoEntryImageFormat::Bmp => "BMP",
+        })
+    }
+}
+
+impl From<IcoEntryImageFormat> for ImageFormat {
+    fn from(val: IcoEntryImageFormat) -> Self {
+        match val {
+            IcoEntryImageFormat::Png => ImageFormat::Png,
+            IcoEntryImageFormat::Bmp => ImageFormat::Bmp,
+        }
+    }
+}
+
+/// An ico decoder
+pub struct IcoDecoder<R: Read> {
+    selected_entry: DirEntry,
+    inner_decoder: InnerDecoder<R>,
+}
+
+enum InnerDecoder<R: Read> {
+    Bmp(BmpDecoder<R>),
+    Png(Box<PngDecoder<R>>),
+}
+
+#[derive(Clone, Copy, Default)]
+struct DirEntry {
+    width: u8,
+    height: u8,
+    // We ignore some header fields as they will be replicated in the PNG, BMP and they are not
+    // necessary for determining the best_entry.
+    #[allow(unused)]
+    color_count: u8,
+    // Wikipedia has this to say:
+    // Although Microsoft's technical documentation states that this value must be zero, the icon
+    // encoder built into .NET (System.Drawing.Icon.Save) sets this value to 255. It appears that
+    // the operating system ignores this value altogether.
+    #[allow(unused)]
+    reserved: u8,
+
+    // We ignore some header fields as they will be replicated in the PNG, BMP and they are not
+    // necessary for determining the best_entry.
+    #[allow(unused)]
+    num_color_planes: u16,
+    bits_per_pixel: u16,
+
+    image_length: u32,
+    image_offset: u32,
+}
+
+impl<R: Read + Seek> IcoDecoder<R> {
+    /// Create a new decoder that decodes from the stream ```r```
+    pub fn new(mut r: R) -> ImageResult<IcoDecoder<R>> {
+        let entries = read_entries(&mut r)?;
+        let entry = best_entry(entries)?;
+        let decoder = entry.decoder(r)?;
+
+        Ok(IcoDecoder {
+            selected_entry: entry,
+            inner_decoder: decoder,
+        })
+    }
+}
+
+fn read_entries<R: Read>(r: &mut R) -> ImageResult<Vec<DirEntry>> {
+    let _reserved = r.read_u16::<LittleEndian>()?;
+    let _type = r.read_u16::<LittleEndian>()?;
+    let count = r.read_u16::<LittleEndian>()?;
+    (0..count).map(|_| read_entry(r)).collect()
+}
+
+fn read_entry<R: Read>(r: &mut R) -> ImageResult<DirEntry> {
+    Ok(DirEntry {
+        width: r.read_u8()?,
+        height: r.read_u8()?,
+        color_count: r.read_u8()?,
+        reserved: r.read_u8()?,
+        num_color_planes: {
+            // This may be either the number of color planes (0 or 1), or the horizontal coordinate
+            // of the hotspot for CUR files.
+            let num = r.read_u16::<LittleEndian>()?;
+            if num > 256 {
+                return Err(DecoderError::IcoEntryTooManyPlanesOrHotspot.into());
+            }
+            num
+        },
+        bits_per_pixel: {
+            // This may be either the bit depth (may be 0 meaning unspecified),
+            // or the vertical coordinate of the hotspot for CUR files.
+            let num = r.read_u16::<LittleEndian>()?;
+            if num > 256 {
+                return Err(DecoderError::IcoEntryTooManyBitsPerPixelOrHotspot.into());
+            }
+            num
+        },
+        image_length: r.read_u32::<LittleEndian>()?,
+        image_offset: r.read_u32::<LittleEndian>()?,
+    })
+}
+
+/// Find the entry with the highest (color depth, size).
+fn best_entry(mut entries: Vec<DirEntry>) -> ImageResult<DirEntry> {
+    let mut best = entries.pop().ok_or(DecoderError::NoEntries)?;
+
+    let mut best_score = (
+        best.bits_per_pixel,
+        u32::from(best.real_width()) * u32::from(best.real_height()),
+    );
+
+    for entry in entries {
+        let score = (
+            entry.bits_per_pixel,
+            u32::from(entry.real_width()) * u32::from(entry.real_height()),
+        );
+        if score > best_score {
+            best = entry;
+            best_score = score;
+        }
+    }
+    Ok(best)
+}
+
+impl DirEntry {
+    fn real_width(&self) -> u16 {
+        match self.width {
+            0 => 256,
+            w => u16::from(w),
+        }
+    }
+
+    fn real_height(&self) -> u16 {
+        match self.height {
+            0 => 256,
+            h => u16::from(h),
+        }
+    }
+
+    fn matches_dimensions(&self, width: u32, height: u32) -> bool {
+        u32::from(self.real_width()) == width.min(256)
+            && u32::from(self.real_height()) == height.min(256)
+    }
+
+    fn seek_to_start<R: Read + Seek>(&self, r: &mut R) -> ImageResult<()> {
+        r.seek(SeekFrom::Start(u64::from(self.image_offset)))?;
+        Ok(())
+    }
+
+    fn is_png<R: Read + Seek>(&self, r: &mut R) -> ImageResult<bool> {
+        self.seek_to_start(r)?;
+
+        // Read the first 8 bytes to sniff the image.
+        let mut signature = [0u8; 8];
+        r.read_exact(&mut signature)?;
+
+        Ok(signature == PNG_SIGNATURE)
+    }
+
+    fn decoder<R: Read + Seek>(&self, mut r: R) -> ImageResult<InnerDecoder<R>> {
+        let is_png = self.is_png(&mut r)?;
+        self.seek_to_start(&mut r)?;
+
+        if is_png {
+            Ok(Png(Box::new(PngDecoder::new(r)?)))
+        } else {
+            Ok(Bmp(BmpDecoder::new_with_ico_format(r)?))
+        }
+    }
+}
+
+/// Wrapper struct around a `Cursor<Vec<u8>>`
+pub struct IcoReader<R>(Cursor<Vec<u8>>, PhantomData<R>);
+impl<R> Read for IcoReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        self.0.read(buf)
+    }
+    fn read_to_end(&mut self, buf: &mut Vec<u8>) -> io::Result<usize> {
+        if self.0.position() == 0 && buf.is_empty() {
+            mem::swap(buf, self.0.get_mut());
+            Ok(buf.len())
+        } else {
+            self.0.read_to_end(buf)
+        }
+    }
+}
+
+impl<'a, R: 'a + Read + Seek> ImageDecoder<'a> for IcoDecoder<R> {
+    type Reader = IcoReader<R>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        match self.inner_decoder {
+            Bmp(ref decoder) => decoder.dimensions(),
+            Png(ref decoder) => decoder.dimensions(),
+        }
+    }
+
+    fn color_type(&self) -> ColorType {
+        match self.inner_decoder {
+            Bmp(ref decoder) => decoder.color_type(),
+            Png(ref decoder) => decoder.color_type(),
+        }
+    }
+
+    fn into_reader(self) -> ImageResult<Self::Reader> {
+        Ok(IcoReader(
+            Cursor::new(image::decoder_to_vec(self)?),
+            PhantomData,
+        ))
+    }
+
+    fn read_image(self, buf: &mut [u8]) -> ImageResult<()> {
+        assert_eq!(u64::try_from(buf.len()), Ok(self.total_bytes()));
+        match self.inner_decoder {
+            Png(decoder) => {
+                if self.selected_entry.image_length < PNG_SIGNATURE.len() as u32 {
+                    return Err(DecoderError::PngShorterThanHeader.into());
+                }
+
+                // Check if the image dimensions match the ones in the image data.
+                let (width, height) = decoder.dimensions();
+                if !self.selected_entry.matches_dimensions(width, height) {
+                    return Err(DecoderError::ImageEntryDimensionMismatch {
+                        format: IcoEntryImageFormat::Png,
+                        entry: (
+                            self.selected_entry.real_width(),
+                            self.selected_entry.real_height(),
+                        ),
+                        image: (width, height),
+                    }
+                    .into());
+                }
+
+                // Embedded PNG images can only be of the 32BPP RGBA format.
+                // https://blogs.msdn.microsoft.com/oldnewthing/20101022-00/?p=12473/
+                if decoder.color_type() != ColorType::Rgba8 {
+                    return Err(DecoderError::PngNotRgba.into());
+                }
+
+                decoder.read_image(buf)
+            }
+            Bmp(mut decoder) => {
+                let (width, height) = decoder.dimensions();
+                if !self.selected_entry.matches_dimensions(width, height) {
+                    return Err(DecoderError::ImageEntryDimensionMismatch {
+                        format: IcoEntryImageFormat::Bmp,
+                        entry: (
+                            self.selected_entry.real_width(),
+                            self.selected_entry.real_height(),
+                        ),
+                        image: (width, height),
+                    }
+                    .into());
+                }
+
+                // The ICO decoder needs an alpha channel to apply the AND mask.
+                if decoder.color_type() != ColorType::Rgba8 {
+                    return Err(ImageError::Unsupported(
+                        UnsupportedError::from_format_and_kind(
+                            ImageFormat::Bmp.into(),
+                            UnsupportedErrorKind::Color(decoder.color_type().into()),
+                        ),
+                    ));
+                }
+
+                decoder.read_image_data(buf)?;
+
+                let r = decoder.reader();
+                let image_end = r.stream_position()?;
+                let data_end = u64::from(self.selected_entry.image_offset)
+                    + u64::from(self.selected_entry.image_length);
+
+                let mask_row_bytes = ((width + 31) / 32) * 4;
+                let mask_length = u64::from(mask_row_bytes) * u64::from(height);
+
+                // data_end should be image_end + the mask length (mask_row_bytes * height).
+                // According to
+                // https://devblogs.microsoft.com/oldnewthing/20101021-00/?p=12483
+                // the mask is required, but according to Wikipedia
+                // https://en.wikipedia.org/wiki/ICO_(file_format)
+                // the mask is not required. Unfortunately, Wikipedia does not have a citation
+                // for that claim, so we can't be sure which is correct.
+                if data_end >= image_end + mask_length {
+                    // If there's an AND mask following the image, read and apply it.
+                    for y in 0..height {
+                        let mut x = 0;
+                        for _ in 0..mask_row_bytes {
+                            // Apply the bits of each byte until we reach the end of the row.
+                            let mask_byte = r.read_u8()?;
+                            for bit in (0..8).rev() {
+                                if x >= width {
+                                    break;
+                                }
+                                if mask_byte & (1 << bit) != 0 {
+                                    // Set alpha channel to transparent.
+                                    buf[((height - y - 1) * width + x) as usize * 4 + 3] = 0;
+                                }
+                                x += 1;
+                            }
+                        }
+                    }
+
+                    Ok(())
+                } else if data_end == image_end {
+                    // accept images with no mask data
+                    Ok(())
+                } else {
+                    Err(DecoderError::InvalidDataSize.into())
+                }
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    // Test if BMP images without alpha channel inside ICOs don't panic.
+    // Because the test data is invalid decoding should produce an error.
+    #[test]
+    fn bmp_16_with_missing_alpha_channel() {
+        let data = vec![
+            0x00, 0x00, 0x01, 0x00, 0x01, 0x00, 0x0e, 0x04, 0xc3, 0x7e, 0x00, 0x00, 0x00, 0x00,
+            0x7c, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x00, 0x00, 0xf8, 0xff, 0xff, 0xff, 0x01, 0x00,
+            0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x12, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x8f, 0xf6, 0xff, 0xff, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x20, 0x66, 0x74, 0x83, 0x70, 0x61, 0x76, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02,
+            0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xeb, 0x00, 0x9b, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x4e, 0x47, 0x0d,
+            0x0a, 0x1a, 0x0a, 0x00, 0x00, 0x00, 0x62, 0x49, 0x48, 0x44, 0x52, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x0c,
+            0x00, 0x00, 0x00, 0xc3, 0x3f, 0x94, 0x61, 0xaa, 0x17, 0x4d, 0x8d, 0x79, 0x1d, 0x8b,
+            0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x14, 0x2e, 0x28, 0x40, 0xe5, 0x9f,
+            0x4b, 0x4d, 0xe9, 0x87, 0xd3, 0xda, 0xd6, 0x89, 0x81, 0xc5, 0xa4, 0xa1, 0x60, 0x98,
+            0x31, 0xc7, 0x1d, 0xb6, 0x8f, 0x20, 0xc8, 0x3e, 0xee, 0xd8, 0xe4, 0x8f, 0xee, 0x7b,
+            0x48, 0x9b, 0x88, 0x25, 0x13, 0xda, 0xa4, 0x13, 0xa4, 0x00, 0x00, 0x00, 0x00, 0x40,
+            0x16, 0x01, 0xff, 0xff, 0xff, 0xff, 0xe9, 0x0c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0xa3, 0x66, 0x64, 0x41, 0x54, 0xa3, 0xa3, 0x00, 0x00, 0x00, 0xb8, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xa3, 0x66, 0x64, 0x41, 0x54, 0xa3, 0xa3,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x8f, 0xf6, 0xff, 0xff,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x66, 0x74, 0x83, 0x70, 0x61, 0x76,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff,
+            0xeb, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a, 0x00, 0x00, 0x00, 0x62, 0x49,
+            0x48, 0x44, 0x52, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0xff, 0xff, 0x94, 0xc8, 0x00, 0x02, 0x0c, 0x00, 0xff, 0xff, 0xc6,
+            0x84, 0x00, 0x2a, 0x75, 0x03, 0xa3, 0x05, 0xfb, 0xe1, 0x6e, 0xe8, 0x27, 0xd6, 0xd3,
+            0x96, 0xc1, 0xe4, 0x30, 0x0c, 0x05, 0xb9, 0xa3, 0x8b, 0x29, 0xda, 0xa4, 0xf1, 0x4d,
+            0xf3, 0xb2, 0x98, 0x2b, 0xe6, 0x93, 0x07, 0xf9, 0xca, 0x2b, 0xc2, 0x39, 0x20, 0xba,
+            0x7c, 0xa0, 0xb1, 0x43, 0xe6, 0xf9, 0xdc, 0xd1, 0xc2, 0x52, 0xdc, 0x41, 0xc1, 0x2f,
+            0x29, 0xf7, 0x46, 0x32, 0xda, 0x1b, 0x72, 0x8c, 0xe6, 0x2b, 0x01, 0xe5, 0x49, 0x21,
+            0x89, 0x89, 0xe4, 0x3d, 0xa1, 0xdb, 0x3b, 0x4a, 0x0b, 0x52, 0x86, 0x52, 0x33, 0x9d,
+            0xb2, 0xcf, 0x4a, 0x86, 0x53, 0xd7, 0xa9, 0x4b, 0xaf, 0x62, 0x06, 0x49, 0x53, 0x00,
+            0xc3, 0x3f, 0x94, 0x61, 0xaa, 0x17, 0x4d, 0x8d, 0x79, 0x1d, 0x8b, 0x10, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x14, 0x2e, 0x28, 0x40, 0xe5, 0x9f, 0x4b, 0x4d, 0xe9,
+            0x87, 0xd3, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xe7, 0xc5, 0x00,
+            0x02, 0x00, 0x00, 0x00, 0x06, 0x00, 0x0b, 0x00, 0x50, 0x31, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x06, 0x76, 0x76, 0x01, 0x00, 0x00, 0x00, 0x76, 0x00,
+            0x00, 0x23, 0x3f, 0x52, 0x41, 0x44, 0x49, 0x41, 0x4e, 0x43, 0x45, 0x61, 0x50, 0x35,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x4d, 0x47, 0x49, 0x46, 0x38, 0x37, 0x61, 0x05,
+            0x50, 0x37, 0x00, 0x00, 0x00, 0x00, 0x00, 0xc7, 0x37, 0x61,
+        ];
+
+        let decoder = IcoDecoder::new(Cursor::new(&data)).unwrap();
+        let mut buf = vec![0; usize::try_from(decoder.total_bytes()).unwrap()];
+        assert!(decoder.read_image(&mut buf).is_err());
+    }
+}

vendor/image/src/codecs/ico/encoder.rs

@@ -0,0 +1,194 @@
+use byteorder::{LittleEndian, WriteBytesExt};
+use std::borrow::Cow;
+use std::io::{self, Write};
+
+use crate::color::ColorType;
+use crate::error::{ImageError, ImageResult, ParameterError, ParameterErrorKind};
+use crate::image::ImageEncoder;
+
+use crate::codecs::png::PngEncoder;
+
+// Enum value indicating an ICO image (as opposed to a CUR image):
+const ICO_IMAGE_TYPE: u16 = 1;
+// The length of an ICO file ICONDIR structure, in bytes:
+const ICO_ICONDIR_SIZE: u32 = 6;
+// The length of an ICO file DIRENTRY structure, in bytes:
+const ICO_DIRENTRY_SIZE: u32 = 16;
+
+/// ICO encoder
+pub struct IcoEncoder<W: Write> {
+    w: W,
+}
+
+/// An ICO image entry
+pub struct IcoFrame<'a> {
+    // Pre-encoded PNG or BMP
+    encoded_image: Cow<'a, [u8]>,
+    // Stored as `0 => 256, n => n`
+    width: u8,
+    // Stored as `0 => 256, n => n`
+    height: u8,
+    color_type: ColorType,
+}
+
+impl<'a> IcoFrame<'a> {
+    /// Construct a new `IcoFrame` using a pre-encoded PNG or BMP
+    ///
+    /// The `width` and `height` must be between 1 and 256 (inclusive).
+    pub fn with_encoded(
+        encoded_image: impl Into<Cow<'a, [u8]>>,
+        width: u32,
+        height: u32,
+        color_type: ColorType,
+    ) -> ImageResult<Self> {
+        let encoded_image = encoded_image.into();
+
+        if !(1..=256).contains(&width) {
+            return Err(ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::Generic(format!(
+                    "the image width must be `1..=256`, instead width {} was provided",
+                    width,
+                )),
+            )));
+        }
+
+        if !(1..=256).contains(&height) {
+            return Err(ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::Generic(format!(
+                    "the image height must be `1..=256`, instead height {} was provided",
+                    height,
+                )),
+            )));
+        }
+
+        Ok(Self {
+            encoded_image,
+            width: width as u8,
+            height: height as u8,
+            color_type,
+        })
+    }
+
+    /// Construct a new `IcoFrame` by encoding `buf` as a PNG
+    ///
+    /// The `width` and `height` must be between 1 and 256 (inclusive)
+    pub fn as_png(buf: &[u8], width: u32, height: u32, color_type: ColorType) -> ImageResult<Self> {
+        let mut image_data: Vec<u8> = Vec::new();
+        PngEncoder::new(&mut image_data).write_image(buf, width, height, color_type)?;
+
+        let frame = Self::with_encoded(image_data, width, height, color_type)?;
+        Ok(frame)
+    }
+}
+
+impl<W: Write> IcoEncoder<W> {
+    /// Create a new encoder that writes its output to ```w```.
+    pub fn new(w: W) -> IcoEncoder<W> {
+        IcoEncoder { w }
+    }
+
+    /// Encodes the image ```image``` that has dimensions ```width``` and
+    /// ```height``` and ```ColorType``` ```c```.  The dimensions of the image
+    /// must be between 1 and 256 (inclusive) or an error will be returned.
+    ///
+    /// Expects data to be big endian.
+    #[deprecated = "Use `IcoEncoder::write_image` instead. Beware that `write_image` has a different endianness convention"]
+    pub fn encode(self, data: &[u8], width: u32, height: u32, color: ColorType) -> ImageResult<()> {
+        let mut image_data: Vec<u8> = Vec::new();
+        #[allow(deprecated)]
+        PngEncoder::new(&mut image_data).encode(data, width, height, color)?;
+
+        let image = IcoFrame::with_encoded(&image_data, width, height, color)?;
+        self.encode_images(&[image])
+    }
+
+    /// Takes some [`IcoFrame`]s and encodes them into an ICO.
+    ///
+    /// `images` is a list of images, usually ordered by dimension, which
+    /// must be between 1 and 65535 (inclusive) in length.
+    pub fn encode_images(mut self, images: &[IcoFrame<'_>]) -> ImageResult<()> {
+        if !(1..=usize::from(u16::MAX)).contains(&images.len()) {
+            return Err(ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::Generic(format!(
+                    "the number of images must be `1..=u16::MAX`, instead {} images were provided",
+                    images.len(),
+                )),
+            )));
+        }
+        let num_images = images.len() as u16;
+
+        let mut offset = ICO_ICONDIR_SIZE + (ICO_DIRENTRY_SIZE * (images.len() as u32));
+        write_icondir(&mut self.w, num_images)?;
+        for image in images {
+            write_direntry(
+                &mut self.w,
+                image.width,
+                image.height,
+                image.color_type,
+                offset,
+                image.encoded_image.len() as u32,
+            )?;
+
+            offset += image.encoded_image.len() as u32;
+        }
+        for image in images {
+            self.w.write_all(&image.encoded_image)?;
+        }
+        Ok(())
+    }
+}
+
+impl<W: Write> ImageEncoder for IcoEncoder<W> {
+    /// Write an ICO image with the specified width, height, and color type.
+    ///
+    /// For color types with 16-bit per channel or larger, the contents of `buf` should be in
+    /// native endian.
+    ///
+    /// WARNING: In image 0.23.14 and earlier this method erroneously expected buf to be in big endian.
+    fn write_image(
+        self,
+        buf: &[u8],
+        width: u32,
+        height: u32,
+        color_type: ColorType,
+    ) -> ImageResult<()> {
+        let image = IcoFrame::as_png(buf, width, height, color_type)?;
+        self.encode_images(&[image])
+    }
+}
+
+fn write_icondir<W: Write>(w: &mut W, num_images: u16) -> io::Result<()> {
+    // Reserved field (must be zero):
+    w.write_u16::<LittleEndian>(0)?;
+    // Image type (ICO or CUR):
+    w.write_u16::<LittleEndian>(ICO_IMAGE_TYPE)?;
+    // Number of images in the file:
+    w.write_u16::<LittleEndian>(num_images)?;
+    Ok(())
+}
+
+fn write_direntry<W: Write>(
+    w: &mut W,
+    width: u8,
+    height: u8,
+    color: ColorType,
+    data_start: u32,
+    data_size: u32,
+) -> io::Result<()> {
+    // Image dimensions:
+    w.write_u8(width)?;
+    w.write_u8(height)?;
+    // Number of colors in palette (or zero for no palette):
+    w.write_u8(0)?;
+    // Reserved field (must be zero):
+    w.write_u8(0)?;
+    // Color planes:
+    w.write_u16::<LittleEndian>(0)?;
+    // Bits per pixel:
+    w.write_u16::<LittleEndian>(color.bits_per_pixel())?;
+    // Image data size, in bytes:
+    w.write_u32::<LittleEndian>(data_size)?;
+    // Image data offset, in bytes:
+    w.write_u32::<LittleEndian>(data_start)?;
+    Ok(())
+}

vendor/image/src/codecs/ico/mod.rs

@@ -0,0 +1,14 @@
+//!  Decoding and Encoding of ICO files
+//!
+//!  A decoder and encoder for ICO (Windows Icon) image container files.
+//!
+//!  # Related Links
+//!  * <https://msdn.microsoft.com/en-us/library/ms997538.aspx>
+//!  * <https://en.wikipedia.org/wiki/ICO_%28file_format%29>
+
+pub use self::decoder::IcoDecoder;
+#[allow(deprecated)]
+pub use self::encoder::{IcoEncoder, IcoFrame};
+
+mod decoder;
+mod encoder;

vendor/image/src/codecs/jpeg/entropy.rs

@@ -0,0 +1,63 @@
+/// Given an array containing the number of codes of each code length,
+/// this function generates the huffman codes lengths and their respective
+/// code lengths as specified by the JPEG spec.
+const fn derive_codes_and_sizes(bits: &[u8; 16]) -> ([u8; 256], [u16; 256]) {
+    let mut huffsize = [0u8; 256];
+    let mut huffcode = [0u16; 256];
+
+    let mut k = 0;
+
+    // Annex C.2
+    // Figure C.1
+    // Generate table of individual code lengths
+    let mut i = 0;
+    while i < 16 {
+        let mut j = 0;
+        while j < bits[i as usize] {
+            huffsize[k] = i + 1;
+            k += 1;
+            j += 1;
+        }
+        i += 1;
+    }
+
+    huffsize[k] = 0;
+
+    // Annex C.2
+    // Figure C.2
+    // Generate table of huffman codes
+    k = 0;
+    let mut code = 0u16;
+    let mut size = huffsize[0];
+
+    while huffsize[k] != 0 {
+        huffcode[k] = code;
+        code += 1;
+        k += 1;
+
+        if huffsize[k] == size {
+            continue;
+        }
+
+        // FIXME there is something wrong with this code
+        let diff = huffsize[k].wrapping_sub(size);
+        code = if diff < 16 { code << diff as usize } else { 0 };
+
+        size = size.wrapping_add(diff);
+    }
+
+    (huffsize, huffcode)
+}
+
+pub(crate) const fn build_huff_lut_const(bits: &[u8; 16], huffval: &[u8]) -> [(u8, u16); 256] {
+    let mut lut = [(17u8, 0u16); 256];
+    let (huffsize, huffcode) = derive_codes_and_sizes(bits);
+
+    let mut i = 0;
+    while i < huffval.len() {
+        lut[huffval[i] as usize] = (huffsize[i], huffcode[i]);
+        i += 1;
+    }
+
+    lut
+}

vendor/image/src/codecs/jpeg/mod.rs

@@ -0,0 +1,16 @@
+//! Decoding and Encoding of JPEG Images
+//!
+//! JPEG (Joint Photographic Experts Group) is an image format that supports lossy compression.
+//! This module implements the Baseline JPEG standard.
+//!
+//! # Related Links
+//! * <http://www.w3.org/Graphics/JPEG/itu-t81.pdf> - The JPEG specification
+//!
+
+pub use self::decoder::JpegDecoder;
+pub use self::encoder::{JpegEncoder, PixelDensity, PixelDensityUnit};
+
+mod decoder;
+mod encoder;
+mod entropy;
+mod transform;

vendor/image/src/codecs/jpeg/transform.rs

@@ -0,0 +1,196 @@
+/*
+fdct is a Rust translation of jfdctint.c from the
+Independent JPEG Group's libjpeg version 9a
+obtained from http://www.ijg.org/files/jpegsr9a.zip
+It comes with the following conditions of distribution and use:
+
+    In plain English:
+
+    1. We don't promise that this software works.  (But if you find any bugs,
+        please let us know!)
+    2. You can use this software for whatever you want.  You don't have to pay us.
+    3. You may not pretend that you wrote this software.  If you use it in a
+       program, you must acknowledge somewhere in your documentation that
+       you've used the IJG code.
+
+    In legalese:
+
+    The authors make NO WARRANTY or representation, either express or implied,
+    with respect to this software, its quality, accuracy, merchantability, or
+    fitness for a particular purpose.  This software is provided "AS IS", and you,
+    its user, assume the entire risk as to its quality and accuracy.
+
+    This software is copyright (C) 1991-2014, Thomas G. Lane, Guido Vollbeding.
+    All Rights Reserved except as specified below.
+
+    Permission is hereby granted to use, copy, modify, and distribute this
+    software (or portions thereof) for any purpose, without fee, subject to these
+    conditions:
+    (1) If any part of the source code for this software is distributed, then this
+    README file must be included, with this copyright and no-warranty notice
+    unaltered; and any additions, deletions, or changes to the original files
+    must be clearly indicated in accompanying documentation.
+    (2) If only executable code is distributed, then the accompanying
+    documentation must state that "this software is based in part on the work of
+    the Independent JPEG Group".
+    (3) Permission for use of this software is granted only if the user accepts
+    full responsibility for any undesirable consequences; the authors accept
+    NO LIABILITY for damages of any kind.
+
+    These conditions apply to any software derived from or based on the IJG code,
+    not just to the unmodified library.  If you use our work, you ought to
+    acknowledge us.
+
+    Permission is NOT granted for the use of any IJG author's name or company name
+    in advertising or publicity relating to this software or products derived from
+    it.  This software may be referred to only as "the Independent JPEG Group's
+    software".
+
+    We specifically permit and encourage the use of this software as the basis of
+    commercial products, provided that all warranty or liability claims are
+    assumed by the product vendor.
+*/
+
+static CONST_BITS: i32 = 13;
+static PASS1_BITS: i32 = 2;
+
+static FIX_0_298631336: i32 = 2446;
+static FIX_0_390180644: i32 = 3196;
+static FIX_0_541196100: i32 = 4433;
+static FIX_0_765366865: i32 = 6270;
+static FIX_0_899976223: i32 = 7373;
+static FIX_1_175875602: i32 = 9633;
+static FIX_1_501321110: i32 = 12_299;
+static FIX_1_847759065: i32 = 15_137;
+static FIX_1_961570560: i32 = 16_069;
+static FIX_2_053119869: i32 = 16_819;
+static FIX_2_562915447: i32 = 20_995;
+static FIX_3_072711026: i32 = 25_172;
+
+pub(crate) fn fdct(samples: &[u8; 64], coeffs: &mut [i32; 64]) {
+    // Pass 1: process rows.
+    // Results are scaled by sqrt(8) compared to a true DCT
+    // furthermore we scale the results by 2**PASS1_BITS
+    for y in 0usize..8 {
+        let y0 = y * 8;
+
+        // Even part
+        let t0 = i32::from(samples[y0]) + i32::from(samples[y0 + 7]);
+        let t1 = i32::from(samples[y0 + 1]) + i32::from(samples[y0 + 6]);
+        let t2 = i32::from(samples[y0 + 2]) + i32::from(samples[y0 + 5]);
+        let t3 = i32::from(samples[y0 + 3]) + i32::from(samples[y0 + 4]);
+
+        let t10 = t0 + t3;
+        let t12 = t0 - t3;
+        let t11 = t1 + t2;
+        let t13 = t1 - t2;
+
+        let t0 = i32::from(samples[y0]) - i32::from(samples[y0 + 7]);
+        let t1 = i32::from(samples[y0 + 1]) - i32::from(samples[y0 + 6]);
+        let t2 = i32::from(samples[y0 + 2]) - i32::from(samples[y0 + 5]);
+        let t3 = i32::from(samples[y0 + 3]) - i32::from(samples[y0 + 4]);
+
+        // Apply unsigned -> signed conversion
+        coeffs[y0] = (t10 + t11 - 8 * 128) << PASS1_BITS as usize;
+        coeffs[y0 + 4] = (t10 - t11) << PASS1_BITS as usize;
+
+        let mut z1 = (t12 + t13) * FIX_0_541196100;
+        // Add fudge factor here for final descale
+        z1 += 1 << (CONST_BITS - PASS1_BITS - 1) as usize;
+
+        coeffs[y0 + 2] = (z1 + t12 * FIX_0_765366865) >> (CONST_BITS - PASS1_BITS) as usize;
+        coeffs[y0 + 6] = (z1 - t13 * FIX_1_847759065) >> (CONST_BITS - PASS1_BITS) as usize;
+
+        // Odd part
+        let t12 = t0 + t2;
+        let t13 = t1 + t3;
+
+        let mut z1 = (t12 + t13) * FIX_1_175875602;
+        // Add fudge factor here for final descale
+        z1 += 1 << (CONST_BITS - PASS1_BITS - 1) as usize;
+
+        let mut t12 = t12 * (-FIX_0_390180644);
+        let mut t13 = t13 * (-FIX_1_961570560);
+        t12 += z1;
+        t13 += z1;
+
+        let z1 = (t0 + t3) * (-FIX_0_899976223);
+        let mut t0 = t0 * FIX_1_501321110;
+        let mut t3 = t3 * FIX_0_298631336;
+        t0 += z1 + t12;
+        t3 += z1 + t13;
+
+        let z1 = (t1 + t2) * (-FIX_2_562915447);
+        let mut t1 = t1 * FIX_3_072711026;
+        let mut t2 = t2 * FIX_2_053119869;
+        t1 += z1 + t13;
+        t2 += z1 + t12;
+
+        coeffs[y0 + 1] = t0 >> (CONST_BITS - PASS1_BITS) as usize;
+        coeffs[y0 + 3] = t1 >> (CONST_BITS - PASS1_BITS) as usize;
+        coeffs[y0 + 5] = t2 >> (CONST_BITS - PASS1_BITS) as usize;
+        coeffs[y0 + 7] = t3 >> (CONST_BITS - PASS1_BITS) as usize;
+    }
+
+    // Pass 2: process columns
+    // We remove the PASS1_BITS scaling but leave the results scaled up an
+    // overall factor of 8
+    for x in (0usize..8).rev() {
+        // Even part
+        let t0 = coeffs[x] + coeffs[x + 8 * 7];
+        let t1 = coeffs[x + 8] + coeffs[x + 8 * 6];
+        let t2 = coeffs[x + 8 * 2] + coeffs[x + 8 * 5];
+        let t3 = coeffs[x + 8 * 3] + coeffs[x + 8 * 4];
+
+        // Add fudge factor here for final descale
+        let t10 = t0 + t3 + (1 << (PASS1_BITS - 1) as usize);
+        let t12 = t0 - t3;
+        let t11 = t1 + t2;
+        let t13 = t1 - t2;
+
+        let t0 = coeffs[x] - coeffs[x + 8 * 7];
+        let t1 = coeffs[x + 8] - coeffs[x + 8 * 6];
+        let t2 = coeffs[x + 8 * 2] - coeffs[x + 8 * 5];
+        let t3 = coeffs[x + 8 * 3] - coeffs[x + 8 * 4];
+
+        coeffs[x] = (t10 + t11) >> PASS1_BITS as usize;
+        coeffs[x + 8 * 4] = (t10 - t11) >> PASS1_BITS as usize;
+
+        let mut z1 = (t12 + t13) * FIX_0_541196100;
+        // Add fudge factor here for final descale
+        z1 += 1 << (CONST_BITS + PASS1_BITS - 1) as usize;
+
+        coeffs[x + 8 * 2] = (z1 + t12 * FIX_0_765366865) >> (CONST_BITS + PASS1_BITS) as usize;
+        coeffs[x + 8 * 6] = (z1 - t13 * FIX_1_847759065) >> (CONST_BITS + PASS1_BITS) as usize;
+
+        // Odd part
+        let t12 = t0 + t2;
+        let t13 = t1 + t3;
+
+        let mut z1 = (t12 + t13) * FIX_1_175875602;
+        // Add fudge factor here for final descale
+        z1 += 1 << (CONST_BITS - PASS1_BITS - 1) as usize;
+
+        let mut t12 = t12 * (-FIX_0_390180644);
+        let mut t13 = t13 * (-FIX_1_961570560);
+        t12 += z1;
+        t13 += z1;
+
+        let z1 = (t0 + t3) * (-FIX_0_899976223);
+        let mut t0 = t0 * FIX_1_501321110;
+        let mut t3 = t3 * FIX_0_298631336;
+        t0 += z1 + t12;
+        t3 += z1 + t13;
+
+        let z1 = (t1 + t2) * (-FIX_2_562915447);
+        let mut t1 = t1 * FIX_3_072711026;
+        let mut t2 = t2 * FIX_2_053119869;
+        t1 += z1 + t13;
+        t2 += z1 + t12;
+
+        coeffs[x + 8] = t0 >> (CONST_BITS + PASS1_BITS) as usize;
+        coeffs[x + 8 * 3] = t1 >> (CONST_BITS + PASS1_BITS) as usize;
+        coeffs[x + 8 * 5] = t2 >> (CONST_BITS + PASS1_BITS) as usize;
+        coeffs[x + 8 * 7] = t3 >> (CONST_BITS + PASS1_BITS) as usize;
+    }
+}

vendor/image/src/codecs/openexr.rs

@@ -0,0 +1,592 @@
+//! Decoding of OpenEXR (.exr) Images
+//!
+//! OpenEXR is an image format that is widely used, especially in VFX,
+//! because it supports lossless and lossy compression for float data.
+//!
+//! This decoder only supports RGB and RGBA images.
+//! If an image does not contain alpha information,
+//! it is defaulted to `1.0` (no transparency).
+//!
+//! # Related Links
+//! * <https://www.openexr.com/documentation.html> - The OpenEXR reference.
+//!
+//!
+//! Current limitations (July 2021):
+//!     - only pixel type `Rgba32F` and `Rgba16F` are supported
+//!     - only non-deep rgb/rgba files supported, no conversion from/to YCbCr or similar
+//!     - only the first non-deep rgb layer is used
+//!     - only the largest mip map level is used
+//!     - pixels outside display window are lost
+//!     - meta data is lost
+//!     - dwaa/dwab compressed images not supported yet by the exr library
+//!     - (chroma) subsampling not supported yet by the exr library
+use exr::prelude::*;
+
+use crate::error::{DecodingError, EncodingError, ImageFormatHint};
+use crate::image::decoder_to_vec;
+use crate::{
+    ColorType, ExtendedColorType, ImageDecoder, ImageEncoder, ImageError, ImageFormat, ImageResult,
+    Progress,
+};
+use std::convert::TryInto;
+use std::io::{Cursor, Read, Seek, Write};
+
+/// An OpenEXR decoder. Immediately reads the meta data from the file.
+#[derive(Debug)]
+pub struct OpenExrDecoder<R> {
+    exr_reader: exr::block::reader::Reader<R>,
+
+    // select a header that is rgb and not deep
+    header_index: usize,
+
+    // decode either rgb or rgba.
+    // can be specified to include or discard alpha channels.
+    // if none, the alpha channel will only be allocated where the file contains data for it.
+    alpha_preference: Option<bool>,
+
+    alpha_present_in_file: bool,
+}
+
+impl<R: Read + Seek> OpenExrDecoder<R> {
+    /// Create a decoder. Consumes the first few bytes of the source to extract image dimensions.
+    /// Assumes the reader is buffered. In most cases,
+    /// you should wrap your reader in a `BufReader` for best performance.
+    /// Loads an alpha channel if the file has alpha samples.
+    /// Use `with_alpha_preference` if you want to load or not load alpha unconditionally.
+    pub fn new(source: R) -> ImageResult<Self> {
+        Self::with_alpha_preference(source, None)
+    }
+
+    /// Create a decoder. Consumes the first few bytes of the source to extract image dimensions.
+    /// Assumes the reader is buffered. In most cases,
+    /// you should wrap your reader in a `BufReader` for best performance.
+    /// If alpha preference is specified, an alpha channel will
+    /// always be present or always be not present in the returned image.
+    /// If alpha preference is none, the alpha channel will only be returned if it is found in the file.
+    pub fn with_alpha_preference(source: R, alpha_preference: Option<bool>) -> ImageResult<Self> {
+        // read meta data, then wait for further instructions, keeping the file open and ready
+        let exr_reader = exr::block::read(source, false).map_err(to_image_err)?;
+
+        let header_index = exr_reader
+            .headers()
+            .iter()
+            .position(|header| {
+                // check if r/g/b exists in the channels
+                let has_rgb = ["R", "G", "B"]
+                    .iter()
+                    .all(|&required|  // alpha will be optional
+                    header.channels.find_index_of_channel(&Text::from(required)).is_some());
+
+                // we currently dont support deep images, or images with other color spaces than rgb
+                !header.deep && has_rgb
+            })
+            .ok_or_else(|| {
+                ImageError::Decoding(DecodingError::new(
+                    ImageFormatHint::Exact(ImageFormat::OpenExr),
+                    "image does not contain non-deep rgb channels",
+                ))
+            })?;
+
+        let has_alpha = exr_reader.headers()[header_index]
+            .channels
+            .find_index_of_channel(&Text::from("A"))
+            .is_some();
+
+        Ok(Self {
+            alpha_preference,
+            exr_reader,
+            header_index,
+            alpha_present_in_file: has_alpha,
+        })
+    }
+
+    // does not leak exrs-specific meta data into public api, just does it for this module
+    fn selected_exr_header(&self) -> &exr::meta::header::Header {
+        &self.exr_reader.meta_data().headers[self.header_index]
+    }
+}
+
+impl<'a, R: 'a + Read + Seek> ImageDecoder<'a> for OpenExrDecoder<R> {
+    type Reader = Cursor<Vec<u8>>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        let size = self
+            .selected_exr_header()
+            .shared_attributes
+            .display_window
+            .size;
+        (size.width() as u32, size.height() as u32)
+    }
+
+    fn color_type(&self) -> ColorType {
+        let returns_alpha = self.alpha_preference.unwrap_or(self.alpha_present_in_file);
+        if returns_alpha {
+            ColorType::Rgba32F
+        } else {
+            ColorType::Rgb32F
+        }
+    }
+
+    fn original_color_type(&self) -> ExtendedColorType {
+        if self.alpha_present_in_file {
+            ExtendedColorType::Rgba32F
+        } else {
+            ExtendedColorType::Rgb32F
+        }
+    }
+
+    /// Use `read_image` instead if possible,
+    /// as this method creates a whole new buffer just to contain the entire image.
+    fn into_reader(self) -> ImageResult<Self::Reader> {
+        Ok(Cursor::new(decoder_to_vec(self)?))
+    }
+
+    fn scanline_bytes(&self) -> u64 {
+        // we cannot always read individual scan lines for every file,
+        // as the tiles or lines in the file could be in random or reversed order.
+        // therefore we currently read all lines at once
+        // Todo: optimize for specific exr.line_order?
+        self.total_bytes()
+    }
+
+    // reads with or without alpha, depending on `self.alpha_preference` and `self.alpha_present_in_file`
+    fn read_image_with_progress<F: Fn(Progress)>(
+        self,
+        unaligned_bytes: &mut [u8],
+        progress_callback: F,
+    ) -> ImageResult<()> {
+        let blocks_in_header = self.selected_exr_header().chunk_count as u64;
+        let channel_count = self.color_type().channel_count() as usize;
+
+        let display_window = self.selected_exr_header().shared_attributes.display_window;
+        let data_window_offset =
+            self.selected_exr_header().own_attributes.layer_position - display_window.position;
+
+        {
+            // check whether the buffer is large enough for the dimensions of the file
+            let (width, height) = self.dimensions();
+            let bytes_per_pixel = self.color_type().bytes_per_pixel() as usize;
+            let expected_byte_count = (width as usize)
+                .checked_mul(height as usize)
+                .and_then(|size| size.checked_mul(bytes_per_pixel));
+
+            // if the width and height does not match the length of the bytes, the arguments are invalid
+            let has_invalid_size_or_overflowed = expected_byte_count
+                .map(|expected_byte_count| unaligned_bytes.len() != expected_byte_count)
+                // otherwise, size calculation overflowed, is bigger than memory,
+                // therefore data is too small, so it is invalid.
+                .unwrap_or(true);
+
+            if has_invalid_size_or_overflowed {
+                panic!("byte buffer not large enough for the specified dimensions and f32 pixels");
+            }
+        }
+
+        let result = read()
+            .no_deep_data()
+            .largest_resolution_level()
+            .rgba_channels(
+                move |_size, _channels| vec![0_f32; display_window.size.area() * channel_count],
+                move |buffer, index_in_data_window, (r, g, b, a_or_1): (f32, f32, f32, f32)| {
+                    let index_in_display_window =
+                        index_in_data_window.to_i32() + data_window_offset;
+
+                    // only keep pixels inside the data window
+                    // TODO filter chunks based on this
+                    if index_in_display_window.x() >= 0
+                        && index_in_display_window.y() >= 0
+                        && index_in_display_window.x() < display_window.size.width() as i32
+                        && index_in_display_window.y() < display_window.size.height() as i32
+                    {
+                        let index_in_display_window =
+                            index_in_display_window.to_usize("index bug").unwrap();
+                        let first_f32_index =
+                            index_in_display_window.flat_index_for_size(display_window.size);
+
+                        buffer[first_f32_index * channel_count
+                            ..(first_f32_index + 1) * channel_count]
+                            .copy_from_slice(&[r, g, b, a_or_1][0..channel_count]);
+
+                        // TODO white point chromaticities + srgb/linear conversion?
+                    }
+                },
+            )
+            .first_valid_layer() // TODO select exact layer by self.header_index?
+            .all_attributes()
+            .on_progress(|progress| {
+                progress_callback(
+                    Progress::new(
+                        (progress * blocks_in_header as f64) as u64,
+                        blocks_in_header,
+                    ), // TODO precision errors?
+                );
+            })
+            .from_chunks(self.exr_reader)
+            .map_err(to_image_err)?;
+
+        // TODO this copy is strictly not necessary, but the exr api is a little too simple for reading into a borrowed target slice
+
+        // this cast is safe and works with any alignment, as bytes are copied, and not f32 values.
+        // note: buffer slice length is checked in the beginning of this function and will be correct at this point
+        unaligned_bytes.copy_from_slice(bytemuck::cast_slice(
+            result.layer_data.channel_data.pixels.as_slice(),
+        ));
+        Ok(())
+    }
+}
+
+/// Write a raw byte buffer of pixels,
+/// returning an Error if it has an invalid length.
+///
+/// Assumes the writer is buffered. In most cases,
+/// you should wrap your writer in a `BufWriter` for best performance.
+// private. access via `OpenExrEncoder`
+fn write_buffer(
+    mut buffered_write: impl Write + Seek,
+    unaligned_bytes: &[u8],
+    width: u32,
+    height: u32,
+    color_type: ColorType,
+) -> ImageResult<()> {
+    let width = width as usize;
+    let height = height as usize;
+
+    {
+        // check whether the buffer is large enough for the specified dimensions
+        let expected_byte_count = width
+            .checked_mul(height)
+            .and_then(|size| size.checked_mul(color_type.bytes_per_pixel() as usize));
+
+        // if the width and height does not match the length of the bytes, the arguments are invalid
+        let has_invalid_size_or_overflowed = expected_byte_count
+            .map(|expected_byte_count| unaligned_bytes.len() < expected_byte_count)
+            // otherwise, size calculation overflowed, is bigger than memory,
+            // therefore data is too small, so it is invalid.
+            .unwrap_or(true);
+
+        if has_invalid_size_or_overflowed {
+            return Err(ImageError::Encoding(EncodingError::new(
+                ImageFormatHint::Exact(ImageFormat::OpenExr),
+                "byte buffer not large enough for the specified dimensions and f32 pixels",
+            )));
+        }
+    }
+
+    // bytes might be unaligned so we cannot cast the whole thing, instead lookup each f32 individually
+    let lookup_f32 = move |f32_index: usize| {
+        let unaligned_f32_bytes_slice = &unaligned_bytes[f32_index * 4..(f32_index + 1) * 4];
+        let f32_bytes_array = unaligned_f32_bytes_slice
+            .try_into()
+            .expect("indexing error");
+        f32::from_ne_bytes(f32_bytes_array)
+    };
+
+    match color_type {
+        ColorType::Rgb32F => {
+            exr::prelude::Image // TODO compression method zip??
+                ::from_channels(
+                (width, height),
+                SpecificChannels::rgb(|pixel: Vec2<usize>| {
+                    let pixel_index = 3 * pixel.flat_index_for_size(Vec2(width, height));
+                    (
+                        lookup_f32(pixel_index),
+                        lookup_f32(pixel_index + 1),
+                        lookup_f32(pixel_index + 2),
+                    )
+                }),
+            )
+            .write()
+            // .on_progress(|progress| todo!())
+            .to_buffered(&mut buffered_write)
+            .map_err(to_image_err)?;
+        }
+
+        ColorType::Rgba32F => {
+            exr::prelude::Image // TODO compression method zip??
+                ::from_channels(
+                (width, height),
+                SpecificChannels::rgba(|pixel: Vec2<usize>| {
+                    let pixel_index = 4 * pixel.flat_index_for_size(Vec2(width, height));
+                    (
+                        lookup_f32(pixel_index),
+                        lookup_f32(pixel_index + 1),
+                        lookup_f32(pixel_index + 2),
+                        lookup_f32(pixel_index + 3),
+                    )
+                }),
+            )
+            .write()
+            // .on_progress(|progress| todo!())
+            .to_buffered(&mut buffered_write)
+            .map_err(to_image_err)?;
+        }
+
+        // TODO other color types and channel types
+        unsupported_color_type => {
+            return Err(ImageError::Encoding(EncodingError::new(
+                ImageFormatHint::Exact(ImageFormat::OpenExr),
+                format!(
+                    "writing color type {:?} not yet supported",
+                    unsupported_color_type
+                ),
+            )))
+        }
+    }
+
+    Ok(())
+}
+
+// TODO is this struct and trait actually used anywhere?
+/// A thin wrapper that implements `ImageEncoder` for OpenEXR images. Will behave like `image::codecs::openexr::write_buffer`.
+#[derive(Debug)]
+pub struct OpenExrEncoder<W>(W);
+
+impl<W> OpenExrEncoder<W> {
+    /// Create an `ImageEncoder`. Does not write anything yet. Writing later will behave like `image::codecs::openexr::write_buffer`.
+    // use constructor, not public field, for future backwards-compatibility
+    pub fn new(write: W) -> Self {
+        Self(write)
+    }
+}
+
+impl<W> ImageEncoder for OpenExrEncoder<W>
+where
+    W: Write + Seek,
+{
+    /// Writes the complete image.
+    ///
+    /// Returns an Error if it has an invalid length.
+    /// Assumes the writer is buffered. In most cases,
+    /// you should wrap your writer in a `BufWriter` for best performance.
+    fn write_image(
+        self,
+        buf: &[u8],
+        width: u32,
+        height: u32,
+        color_type: ColorType,
+    ) -> ImageResult<()> {
+        write_buffer(self.0, buf, width, height, color_type)
+    }
+}
+
+fn to_image_err(exr_error: Error) -> ImageError {
+    ImageError::Decoding(DecodingError::new(
+        ImageFormatHint::Exact(ImageFormat::OpenExr),
+        exr_error.to_string(),
+    ))
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    use std::io::BufReader;
+    use std::path::{Path, PathBuf};
+
+    use crate::buffer_::{Rgb32FImage, Rgba32FImage};
+    use crate::error::{LimitError, LimitErrorKind};
+    use crate::{ImageBuffer, Rgb, Rgba};
+
+    const BASE_PATH: &[&str] = &[".", "tests", "images", "exr"];
+
+    /// Write an `Rgb32FImage`.
+    /// Assumes the writer is buffered. In most cases,
+    /// you should wrap your writer in a `BufWriter` for best performance.
+    fn write_rgb_image(write: impl Write + Seek, image: &Rgb32FImage) -> ImageResult<()> {
+        write_buffer(
+            write,
+            bytemuck::cast_slice(image.as_raw().as_slice()),
+            image.width(),
+            image.height(),
+            ColorType::Rgb32F,
+        )
+    }
+
+    /// Write an `Rgba32FImage`.
+    /// Assumes the writer is buffered. In most cases,
+    /// you should wrap your writer in a `BufWriter` for best performance.
+    fn write_rgba_image(write: impl Write + Seek, image: &Rgba32FImage) -> ImageResult<()> {
+        write_buffer(
+            write,
+            bytemuck::cast_slice(image.as_raw().as_slice()),
+            image.width(),
+            image.height(),
+            ColorType::Rgba32F,
+        )
+    }
+
+    /// Read the file from the specified path into an `Rgba32FImage`.
+    fn read_as_rgba_image_from_file(path: impl AsRef<Path>) -> ImageResult<Rgba32FImage> {
+        read_as_rgba_image(BufReader::new(std::fs::File::open(path)?))
+    }
+
+    /// Read the file from the specified path into an `Rgb32FImage`.
+    fn read_as_rgb_image_from_file(path: impl AsRef<Path>) -> ImageResult<Rgb32FImage> {
+        read_as_rgb_image(BufReader::new(std::fs::File::open(path)?))
+    }
+
+    /// Read the file from the specified path into an `Rgb32FImage`.
+    fn read_as_rgb_image(read: impl Read + Seek) -> ImageResult<Rgb32FImage> {
+        let decoder = OpenExrDecoder::with_alpha_preference(read, Some(false))?;
+        let (width, height) = decoder.dimensions();
+        let buffer: Vec<f32> = decoder_to_vec(decoder)?;
+
+        ImageBuffer::from_raw(width, height, buffer)
+            // this should be the only reason for the "from raw" call to fail,
+            // even though such a large allocation would probably cause an error much earlier
+            .ok_or_else(|| {
+                ImageError::Limits(LimitError::from_kind(LimitErrorKind::InsufficientMemory))
+            })
+    }
+
+    /// Read the file from the specified path into an `Rgba32FImage`.
+    fn read_as_rgba_image(read: impl Read + Seek) -> ImageResult<Rgba32FImage> {
+        let decoder = OpenExrDecoder::with_alpha_preference(read, Some(true))?;
+        let (width, height) = decoder.dimensions();
+        let buffer: Vec<f32> = decoder_to_vec(decoder)?;
+
+        ImageBuffer::from_raw(width, height, buffer)
+            // this should be the only reason for the "from raw" call to fail,
+            // even though such a large allocation would probably cause an error much earlier
+            .ok_or_else(|| {
+                ImageError::Limits(LimitError::from_kind(LimitErrorKind::InsufficientMemory))
+            })
+    }
+
+    #[test]
+    fn compare_exr_hdr() {
+        if cfg!(not(feature = "hdr")) {
+            eprintln!("warning: to run all the openexr tests, activate the hdr feature flag");
+        }
+
+        #[cfg(feature = "hdr")]
+        {
+            let folder = BASE_PATH.iter().collect::<PathBuf>();
+            let reference_path = folder.clone().join("overexposed gradient.hdr");
+            let exr_path = folder
+                .clone()
+                .join("overexposed gradient - data window equals display window.exr");
+
+            let hdr: Vec<Rgb<f32>> = crate::codecs::hdr::HdrDecoder::new(std::io::BufReader::new(
+                std::fs::File::open(&reference_path).unwrap(),
+            ))
+            .unwrap()
+            .read_image_hdr()
+            .unwrap();
+
+            let exr_pixels: Rgb32FImage = read_as_rgb_image_from_file(exr_path).unwrap();
+            assert_eq!(
+                exr_pixels.dimensions().0 * exr_pixels.dimensions().1,
+                hdr.len() as u32
+            );
+
+            for (expected, found) in hdr.iter().zip(exr_pixels.pixels()) {
+                for (expected, found) in expected.0.iter().zip(found.0.iter()) {
+                    // the large tolerance seems to be caused by
+                    // the RGBE u8x4 pixel quantization of the hdr image format
+                    assert!(
+                        (expected - found).abs() < 0.1,
+                        "expected {}, found {}",
+                        expected,
+                        found
+                    );
+                }
+            }
+        }
+    }
+
+    #[test]
+    fn roundtrip_rgba() {
+        let mut next_random = vec![1.0, 0.0, -1.0, -3.14, 27.0, 11.0, 31.0]
+            .into_iter()
+            .cycle();
+        let mut next_random = move || next_random.next().unwrap();
+
+        let generated_image: Rgba32FImage = ImageBuffer::from_fn(9, 31, |_x, _y| {
+            Rgba([next_random(), next_random(), next_random(), next_random()])
+        });
+
+        let mut bytes = vec![];
+        write_rgba_image(Cursor::new(&mut bytes), &generated_image).unwrap();
+        let decoded_image = read_as_rgba_image(Cursor::new(bytes)).unwrap();
+
+        debug_assert_eq!(generated_image, decoded_image);
+    }
+
+    #[test]
+    fn roundtrip_rgb() {
+        let mut next_random = vec![1.0, 0.0, -1.0, -3.14, 27.0, 11.0, 31.0]
+            .into_iter()
+            .cycle();
+        let mut next_random = move || next_random.next().unwrap();
+
+        let generated_image: Rgb32FImage = ImageBuffer::from_fn(9, 31, |_x, _y| {
+            Rgb([next_random(), next_random(), next_random()])
+        });
+
+        let mut bytes = vec![];
+        write_rgb_image(Cursor::new(&mut bytes), &generated_image).unwrap();
+        let decoded_image = read_as_rgb_image(Cursor::new(bytes)).unwrap();
+
+        debug_assert_eq!(generated_image, decoded_image);
+    }
+
+    #[test]
+    fn compare_rgba_rgb() {
+        let exr_path = BASE_PATH
+            .iter()
+            .collect::<PathBuf>()
+            .join("overexposed gradient - data window equals display window.exr");
+
+        let rgb: Rgb32FImage = read_as_rgb_image_from_file(&exr_path).unwrap();
+        let rgba: Rgba32FImage = read_as_rgba_image_from_file(&exr_path).unwrap();
+
+        assert_eq!(rgba.dimensions(), rgb.dimensions());
+
+        for (Rgb(rgb), Rgba(rgba)) in rgb.pixels().zip(rgba.pixels()) {
+            assert_eq!(rgb, &rgba[..3]);
+        }
+    }
+
+    #[test]
+    fn compare_cropped() {
+        // like in photoshop, exr images may have layers placed anywhere in a canvas.
+        // we don't want to load the pixels from the layer, but we want to load the pixels from the canvas.
+        // a layer might be smaller than the canvas, in that case the canvas should be transparent black
+        // where no layer was covering it. a layer might also be larger than the canvas,
+        // these pixels should be discarded.
+        //
+        // in this test we want to make sure that an
+        // auto-cropped image will be reproduced to the original.
+
+        let exr_path = BASE_PATH.iter().collect::<PathBuf>();
+        let original = exr_path.clone().join("cropping - uncropped original.exr");
+        let cropped = exr_path
+            .clone()
+            .join("cropping - data window differs display window.exr");
+
+        // smoke-check that the exr files are actually not the same
+        {
+            let original_exr = read_first_flat_layer_from_file(&original).unwrap();
+            let cropped_exr = read_first_flat_layer_from_file(&cropped).unwrap();
+            assert_eq!(
+                original_exr.attributes.display_window,
+                cropped_exr.attributes.display_window
+            );
+            assert_ne!(
+                original_exr.layer_data.attributes.layer_position,
+                cropped_exr.layer_data.attributes.layer_position
+            );
+            assert_ne!(original_exr.layer_data.size, cropped_exr.layer_data.size);
+        }
+
+        // check that they result in the same image
+        let original: Rgba32FImage = read_as_rgba_image_from_file(&original).unwrap();
+        let cropped: Rgba32FImage = read_as_rgba_image_from_file(&cropped).unwrap();
+        assert_eq!(original.dimensions(), cropped.dimensions());
+
+        // the following is not a simple assert_eq, as in case of an error,
+        // the whole image would be printed to the console, which takes forever
+        assert!(original.pixels().zip(cropped.pixels()).all(|(a, b)| a == b));
+    }
+}

vendor/image/src/codecs/png.rs

@@ -0,0 +1,778 @@
+//! Decoding and Encoding of PNG Images
+//!
+//! PNG (Portable Network Graphics) is an image format that supports lossless compression.
+//!
+//! # Related Links
+//! * <http://www.w3.org/TR/PNG/> - The PNG Specification
+//!
+
+use std::convert::TryFrom;
+use std::fmt;
+use std::io::{self, Read, Write};
+
+use num_rational::Ratio;
+use png::{BlendOp, DisposeOp};
+
+use crate::animation::{Delay, Frame, Frames};
+use crate::color::{Blend, ColorType, ExtendedColorType};
+use crate::error::{
+    DecodingError, EncodingError, ImageError, ImageResult, LimitError, LimitErrorKind,
+    ParameterError, ParameterErrorKind, UnsupportedError, UnsupportedErrorKind,
+};
+use crate::image::{AnimationDecoder, ImageDecoder, ImageEncoder, ImageFormat};
+use crate::io::Limits;
+use crate::{DynamicImage, GenericImage, ImageBuffer, Luma, LumaA, Rgb, Rgba, RgbaImage};
+
+// http://www.w3.org/TR/PNG-Structure.html
+// The first eight bytes of a PNG file always contain the following (decimal) values:
+pub(crate) const PNG_SIGNATURE: [u8; 8] = [137, 80, 78, 71, 13, 10, 26, 10];
+
+/// Png Reader
+///
+/// This reader will try to read the png one row at a time,
+/// however for interlaced png files this is not possible and
+/// these are therefore read at once.
+pub struct PngReader<R: Read> {
+    reader: png::Reader<R>,
+    buffer: Vec<u8>,
+    index: usize,
+}
+
+impl<R: Read> PngReader<R> {
+    fn new(mut reader: png::Reader<R>) -> ImageResult<PngReader<R>> {
+        let len = reader.output_buffer_size();
+        // Since interlaced images do not come in
+        // scanline order it is almost impossible to
+        // read them in a streaming fashion, however
+        // this shouldn't be a too big of a problem
+        // as most interlaced images should fit in memory.
+        let buffer = if reader.info().interlaced {
+            let mut buffer = vec![0; len];
+            reader
+                .next_frame(&mut buffer)
+                .map_err(ImageError::from_png)?;
+            buffer
+        } else {
+            Vec::new()
+        };
+
+        Ok(PngReader {
+            reader,
+            buffer,
+            index: 0,
+        })
+    }
+}
+
+impl<R: Read> Read for PngReader<R> {
+    fn read(&mut self, mut buf: &mut [u8]) -> io::Result<usize> {
+        // io::Write::write for slice cannot fail
+        let readed = buf.write(&self.buffer[self.index..]).unwrap();
+
+        let mut bytes = readed;
+        self.index += readed;
+
+        while self.index >= self.buffer.len() {
+            match self.reader.next_row()? {
+                Some(row) => {
+                    // Faster to copy directly to external buffer
+                    let readed = buf.write(row.data()).unwrap();
+                    bytes += readed;
+
+                    self.buffer = row.data()[readed..].to_owned();
+                    self.index = 0;
+                }
+                None => return Ok(bytes),
+            }
+        }
+
+        Ok(bytes)
+    }
+
+    fn read_to_end(&mut self, buf: &mut Vec<u8>) -> io::Result<usize> {
+        let mut bytes = self.buffer.len();
+        if buf.is_empty() {
+            std::mem::swap(&mut self.buffer, buf);
+        } else {
+            buf.extend_from_slice(&self.buffer);
+            self.buffer.clear();
+        }
+
+        self.index = 0;
+
+        while let Some(row) = self.reader.next_row()? {
+            buf.extend_from_slice(row.data());
+            bytes += row.data().len();
+        }
+
+        Ok(bytes)
+    }
+}
+
+/// PNG decoder
+pub struct PngDecoder<R: Read> {
+    color_type: ColorType,
+    reader: png::Reader<R>,
+}
+
+impl<R: Read> PngDecoder<R> {
+    /// Creates a new decoder that decodes from the stream ```r```
+    pub fn new(r: R) -> ImageResult<PngDecoder<R>> {
+        Self::with_limits(r, Limits::default())
+    }
+
+    /// Creates a new decoder that decodes from the stream ```r``` with the given limits.
+    pub fn with_limits(r: R, limits: Limits) -> ImageResult<PngDecoder<R>> {
+        limits.check_support(&crate::io::LimitSupport::default())?;
+
+        let max_bytes = usize::try_from(limits.max_alloc.unwrap_or(u64::MAX)).unwrap_or(usize::MAX);
+        let mut decoder = png::Decoder::new_with_limits(r, png::Limits { bytes: max_bytes });
+
+        let info = decoder.read_header_info().map_err(ImageError::from_png)?;
+        limits.check_dimensions(info.width, info.height)?;
+
+        // By default the PNG decoder will scale 16 bpc to 8 bpc, so custom
+        // transformations must be set. EXPAND preserves the default behavior
+        // expanding bpc < 8 to 8 bpc.
+        decoder.set_transformations(png::Transformations::EXPAND);
+        let reader = decoder.read_info().map_err(ImageError::from_png)?;
+        let (color_type, bits) = reader.output_color_type();
+        let color_type = match (color_type, bits) {
+            (png::ColorType::Grayscale, png::BitDepth::Eight) => ColorType::L8,
+            (png::ColorType::Grayscale, png::BitDepth::Sixteen) => ColorType::L16,
+            (png::ColorType::GrayscaleAlpha, png::BitDepth::Eight) => ColorType::La8,
+            (png::ColorType::GrayscaleAlpha, png::BitDepth::Sixteen) => ColorType::La16,
+            (png::ColorType::Rgb, png::BitDepth::Eight) => ColorType::Rgb8,
+            (png::ColorType::Rgb, png::BitDepth::Sixteen) => ColorType::Rgb16,
+            (png::ColorType::Rgba, png::BitDepth::Eight) => ColorType::Rgba8,
+            (png::ColorType::Rgba, png::BitDepth::Sixteen) => ColorType::Rgba16,
+
+            (png::ColorType::Grayscale, png::BitDepth::One) => {
+                return Err(unsupported_color(ExtendedColorType::L1))
+            }
+            (png::ColorType::GrayscaleAlpha, png::BitDepth::One) => {
+                return Err(unsupported_color(ExtendedColorType::La1))
+            }
+            (png::ColorType::Rgb, png::BitDepth::One) => {
+                return Err(unsupported_color(ExtendedColorType::Rgb1))
+            }
+            (png::ColorType::Rgba, png::BitDepth::One) => {
+                return Err(unsupported_color(ExtendedColorType::Rgba1))
+            }
+
+            (png::ColorType::Grayscale, png::BitDepth::Two) => {
+                return Err(unsupported_color(ExtendedColorType::L2))
+            }
+            (png::ColorType::GrayscaleAlpha, png::BitDepth::Two) => {
+                return Err(unsupported_color(ExtendedColorType::La2))
+            }
+            (png::ColorType::Rgb, png::BitDepth::Two) => {
+                return Err(unsupported_color(ExtendedColorType::Rgb2))
+            }
+            (png::ColorType::Rgba, png::BitDepth::Two) => {
+                return Err(unsupported_color(ExtendedColorType::Rgba2))
+            }
+
+            (png::ColorType::Grayscale, png::BitDepth::Four) => {
+                return Err(unsupported_color(ExtendedColorType::L4))
+            }
+            (png::ColorType::GrayscaleAlpha, png::BitDepth::Four) => {
+                return Err(unsupported_color(ExtendedColorType::La4))
+            }
+            (png::ColorType::Rgb, png::BitDepth::Four) => {
+                return Err(unsupported_color(ExtendedColorType::Rgb4))
+            }
+            (png::ColorType::Rgba, png::BitDepth::Four) => {
+                return Err(unsupported_color(ExtendedColorType::Rgba4))
+            }
+
+            (png::ColorType::Indexed, bits) => {
+                return Err(unsupported_color(ExtendedColorType::Unknown(bits as u8)))
+            }
+        };
+
+        Ok(PngDecoder { color_type, reader })
+    }
+
+    /// Turn this into an iterator over the animation frames.
+    ///
+    /// Reading the complete animation requires more memory than reading the data from the IDAT
+    /// frame–multiple frame buffers need to be reserved at the same time. We further do not
+    /// support compositing 16-bit colors. In any case this would be lossy as the interface of
+    /// animation decoders does not support 16-bit colors.
+    ///
+    /// If something is not supported or a limit is violated then the decoding step that requires
+    /// them will fail and an error will be returned instead of the frame. No further frames will
+    /// be returned.
+    pub fn apng(self) -> ApngDecoder<R> {
+        ApngDecoder::new(self)
+    }
+
+    /// Returns if the image contains an animation.
+    ///
+    /// Note that the file itself decides if the default image is considered to be part of the
+    /// animation. When it is not the common interpretation is to use it as a thumbnail.
+    ///
+    /// If a non-animated image is converted into an `ApngDecoder` then its iterator is empty.
+    pub fn is_apng(&self) -> bool {
+        self.reader.info().animation_control.is_some()
+    }
+}
+
+fn unsupported_color(ect: ExtendedColorType) -> ImageError {
+    ImageError::Unsupported(UnsupportedError::from_format_and_kind(
+        ImageFormat::Png.into(),
+        UnsupportedErrorKind::Color(ect),
+    ))
+}
+
+impl<'a, R: 'a + Read> ImageDecoder<'a> for PngDecoder<R> {
+    type Reader = PngReader<R>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        self.reader.info().size()
+    }
+
+    fn color_type(&self) -> ColorType {
+        self.color_type
+    }
+
+    fn icc_profile(&mut self) -> Option<Vec<u8>> {
+        self.reader.info().icc_profile.as_ref().map(|x| x.to_vec())
+    }
+
+    fn into_reader(self) -> ImageResult<Self::Reader> {
+        PngReader::new(self.reader)
+    }
+
+    fn read_image(mut self, buf: &mut [u8]) -> ImageResult<()> {
+        use byteorder::{BigEndian, ByteOrder, NativeEndian};
+
+        assert_eq!(u64::try_from(buf.len()), Ok(self.total_bytes()));
+        self.reader.next_frame(buf).map_err(ImageError::from_png)?;
+        // PNG images are big endian. For 16 bit per channel and larger types,
+        // the buffer may need to be reordered to native endianness per the
+        // contract of `read_image`.
+        // TODO: assumes equal channel bit depth.
+        let bpc = self.color_type().bytes_per_pixel() / self.color_type().channel_count();
+
+        match bpc {
+            1 => (), // No reodering necessary for u8
+            2 => buf.chunks_mut(2).for_each(|c| {
+                let v = BigEndian::read_u16(c);
+                NativeEndian::write_u16(c, v)
+            }),
+            _ => unreachable!(),
+        }
+        Ok(())
+    }
+
+    fn scanline_bytes(&self) -> u64 {
+        let width = self.reader.info().width;
+        self.reader.output_line_size(width) as u64
+    }
+}
+
+/// An [`AnimationDecoder`] adapter of [`PngDecoder`].
+///
+/// See [`PngDecoder::apng`] for more information.
+///
+/// [`AnimationDecoder`]: ../trait.AnimationDecoder.html
+/// [`PngDecoder`]: struct.PngDecoder.html
+/// [`PngDecoder::apng`]: struct.PngDecoder.html#method.apng
+pub struct ApngDecoder<R: Read> {
+    inner: PngDecoder<R>,
+    /// The current output buffer.
+    current: RgbaImage,
+    /// The previous output buffer, used for dispose op previous.
+    previous: RgbaImage,
+    /// The dispose op of the current frame.
+    dispose: DisposeOp,
+    /// The number of image still expected to be able to load.
+    remaining: u32,
+    /// The next (first) image is the thumbnail.
+    has_thumbnail: bool,
+}
+
+impl<R: Read> ApngDecoder<R> {
+    fn new(inner: PngDecoder<R>) -> Self {
+        let (width, height) = inner.dimensions();
+        let info = inner.reader.info();
+        let remaining = match info.animation_control() {
+            // The expected number of fcTL in the remaining image.
+            Some(actl) => actl.num_frames,
+            None => 0,
+        };
+        // If the IDAT has no fcTL then it is not part of the animation counted by
+        // num_frames. All following fdAT chunks must be preceded by an fcTL
+        let has_thumbnail = info.frame_control.is_none();
+        ApngDecoder {
+            inner,
+            // TODO: should we delay this allocation? At least if we support limits we should.
+            current: RgbaImage::new(width, height),
+            previous: RgbaImage::new(width, height),
+            dispose: DisposeOp::Background,
+            remaining,
+            has_thumbnail,
+        }
+    }
+
+    // TODO: thumbnail(&mut self) -> Option<impl ImageDecoder<'_>>
+
+    /// Decode one subframe and overlay it on the canvas.
+    fn mix_next_frame(&mut self) -> Result<Option<&RgbaImage>, ImageError> {
+        // Remove this image from remaining.
+        self.remaining = match self.remaining.checked_sub(1) {
+            None => return Ok(None),
+            Some(next) => next,
+        };
+
+        // Shorten ourselves to 0 in case of error.
+        let remaining = self.remaining;
+        self.remaining = 0;
+
+        // Skip the thumbnail that is not part of the animation.
+        if self.has_thumbnail {
+            self.has_thumbnail = false;
+            let mut buffer = vec![0; self.inner.reader.output_buffer_size()];
+            self.inner
+                .reader
+                .next_frame(&mut buffer)
+                .map_err(ImageError::from_png)?;
+        }
+
+        self.animatable_color_type()?;
+
+        // Dispose of the previous frame.
+        match self.dispose {
+            DisposeOp::None => {
+                self.previous.clone_from(&self.current);
+            }
+            DisposeOp::Background => {
+                self.previous.clone_from(&self.current);
+                self.current
+                    .pixels_mut()
+                    .for_each(|pixel| *pixel = Rgba([0, 0, 0, 0]));
+            }
+            DisposeOp::Previous => {
+                self.current.clone_from(&self.previous);
+            }
+        }
+
+        // Read next frame data.
+        let mut buffer = vec![0; self.inner.reader.output_buffer_size()];
+        self.inner
+            .reader
+            .next_frame(&mut buffer)
+            .map_err(ImageError::from_png)?;
+        let info = self.inner.reader.info();
+
+        // Find out how to interpret the decoded frame.
+        let (width, height, px, py, blend);
+        match info.frame_control() {
+            None => {
+                width = info.width;
+                height = info.height;
+                px = 0;
+                py = 0;
+                blend = BlendOp::Source;
+            }
+            Some(fc) => {
+                width = fc.width;
+                height = fc.height;
+                px = fc.x_offset;
+                py = fc.y_offset;
+                blend = fc.blend_op;
+                self.dispose = fc.dispose_op;
+            }
+        };
+
+        // Turn the data into an rgba image proper.
+        let source = match self.inner.color_type {
+            ColorType::L8 => {
+                let image = ImageBuffer::<Luma<_>, _>::from_raw(width, height, buffer).unwrap();
+                DynamicImage::ImageLuma8(image).into_rgba8()
+            }
+            ColorType::La8 => {
+                let image = ImageBuffer::<LumaA<_>, _>::from_raw(width, height, buffer).unwrap();
+                DynamicImage::ImageLumaA8(image).into_rgba8()
+            }
+            ColorType::Rgb8 => {
+                let image = ImageBuffer::<Rgb<_>, _>::from_raw(width, height, buffer).unwrap();
+                DynamicImage::ImageRgb8(image).into_rgba8()
+            }
+            ColorType::Rgba8 => ImageBuffer::<Rgba<_>, _>::from_raw(width, height, buffer).unwrap(),
+            ColorType::L16 | ColorType::Rgb16 | ColorType::La16 | ColorType::Rgba16 => {
+                // TODO: to enable remove restriction in `animatable_color_type` method.
+                unreachable!("16-bit apng not yet support")
+            }
+            _ => unreachable!("Invalid png color"),
+        };
+
+        match blend {
+            BlendOp::Source => {
+                self.current
+                    .copy_from(&source, px, py)
+                    .expect("Invalid png image not detected in png");
+            }
+            BlendOp::Over => {
+                // TODO: investigate speed, speed-ups, and bounds-checks.
+                for (x, y, p) in source.enumerate_pixels() {
+                    self.current.get_pixel_mut(x + px, y + py).blend(p);
+                }
+            }
+        }
+
+        // Ok, we can proceed with actually remaining images.
+        self.remaining = remaining;
+        // Return composited output buffer.
+        Ok(Some(&self.current))
+    }
+
+    fn animatable_color_type(&self) -> Result<(), ImageError> {
+        match self.inner.color_type {
+            ColorType::L8 | ColorType::Rgb8 | ColorType::La8 | ColorType::Rgba8 => Ok(()),
+            // TODO: do not handle multi-byte colors. Remember to implement it in `mix_next_frame`.
+            ColorType::L16 | ColorType::Rgb16 | ColorType::La16 | ColorType::Rgba16 => {
+                Err(unsupported_color(self.inner.color_type.into()))
+            }
+            _ => unreachable!("{:?} not a valid png color", self.inner.color_type),
+        }
+    }
+}
+
+impl<'a, R: Read + 'a> AnimationDecoder<'a> for ApngDecoder<R> {
+    fn into_frames(self) -> Frames<'a> {
+        struct FrameIterator<R: Read>(ApngDecoder<R>);
+
+        impl<R: Read> Iterator for FrameIterator<R> {
+            type Item = ImageResult<Frame>;
+
+            fn next(&mut self) -> Option<Self::Item> {
+                let image = match self.0.mix_next_frame() {
+                    Ok(Some(image)) => image.clone(),
+                    Ok(None) => return None,
+                    Err(err) => return Some(Err(err)),
+                };
+
+                let info = self.0.inner.reader.info();
+                let fc = info.frame_control().unwrap();
+                // PNG delays are rations in seconds.
+                let num = u32::from(fc.delay_num) * 1_000u32;
+                let denom = match fc.delay_den {
+                    // The standard dictates to replace by 100 when the denominator is 0.
+                    0 => 100,
+                    d => u32::from(d),
+                };
+                let delay = Delay::from_ratio(Ratio::new(num, denom));
+                Some(Ok(Frame::from_parts(image, 0, 0, delay)))
+            }
+        }
+
+        Frames::new(Box::new(FrameIterator(self)))
+    }
+}
+
+/// PNG encoder
+pub struct PngEncoder<W: Write> {
+    w: W,
+    compression: CompressionType,
+    filter: FilterType,
+}
+
+/// Compression level of a PNG encoder. The default setting is `Fast`.
+#[derive(Clone, Copy, Debug, Eq, PartialEq)]
+#[non_exhaustive]
+pub enum CompressionType {
+    /// Default compression level
+    Default,
+    /// Fast, minimal compression
+    Fast,
+    /// High compression level
+    Best,
+    /// Huffman coding compression
+    #[deprecated(note = "use one of the other compression levels instead, such as 'Fast'")]
+    Huffman,
+    /// Run-length encoding compression
+    #[deprecated(note = "use one of the other compression levels instead, such as 'Fast'")]
+    Rle,
+}
+
+/// Filter algorithms used to process image data to improve compression.
+///
+/// The default filter is `Adaptive`.
+#[derive(Clone, Copy, Debug, Eq, PartialEq)]
+#[non_exhaustive]
+pub enum FilterType {
+    /// No processing done, best used for low bit depth grayscale or data with a
+    /// low color count
+    NoFilter,
+    /// Filters based on previous pixel in the same scanline
+    Sub,
+    /// Filters based on the scanline above
+    Up,
+    /// Filters based on the average of left and right neighbor pixels
+    Avg,
+    /// Algorithm that takes into account the left, upper left, and above pixels
+    Paeth,
+    /// Uses a heuristic to select one of the preceding filters for each
+    /// scanline rather than one filter for the entire image
+    Adaptive,
+}
+
+#[derive(Clone, Copy, Debug, Eq, PartialEq)]
+#[non_exhaustive]
+enum BadPngRepresentation {
+    ColorType(ColorType),
+}
+
+impl<W: Write> PngEncoder<W> {
+    /// Create a new encoder that writes its output to ```w```
+    pub fn new(w: W) -> PngEncoder<W> {
+        PngEncoder {
+            w,
+            compression: CompressionType::default(),
+            filter: FilterType::default(),
+        }
+    }
+
+    /// Create a new encoder that writes its output to `w` with `CompressionType` `compression` and
+    /// `FilterType` `filter`.
+    ///
+    /// It is best to view the options as a _hint_ to the implementation on the smallest or fastest
+    /// option for encoding a particular image. That is, using options that map directly to a PNG
+    /// image parameter will use this parameter where possible. But variants that have no direct
+    /// mapping may be interpreted differently in minor versions. The exact output is expressly
+    /// __not__ part the SemVer stability guarantee.
+    ///
+    /// Note that it is not optimal to use a single filter type, so an adaptive
+    /// filter type is selected as the default. The filter which best minimizes
+    /// file size may change with the type of compression used.
+    pub fn new_with_quality(
+        w: W,
+        compression: CompressionType,
+        filter: FilterType,
+    ) -> PngEncoder<W> {
+        PngEncoder {
+            w,
+            compression,
+            filter,
+        }
+    }
+
+    /// Encodes the image `data` that has dimensions `width` and `height` and `ColorType` `c`.
+    ///
+    /// Expects data in big endian.
+    #[deprecated = "Use `PngEncoder::write_image` instead. Beware that `write_image` has a different endianness convention"]
+    pub fn encode(self, data: &[u8], width: u32, height: u32, color: ColorType) -> ImageResult<()> {
+        self.encode_inner(data, width, height, color)
+    }
+
+    fn encode_inner(
+        self,
+        data: &[u8],
+        width: u32,
+        height: u32,
+        color: ColorType,
+    ) -> ImageResult<()> {
+        let (ct, bits) = match color {
+            ColorType::L8 => (png::ColorType::Grayscale, png::BitDepth::Eight),
+            ColorType::L16 => (png::ColorType::Grayscale, png::BitDepth::Sixteen),
+            ColorType::La8 => (png::ColorType::GrayscaleAlpha, png::BitDepth::Eight),
+            ColorType::La16 => (png::ColorType::GrayscaleAlpha, png::BitDepth::Sixteen),
+            ColorType::Rgb8 => (png::ColorType::Rgb, png::BitDepth::Eight),
+            ColorType::Rgb16 => (png::ColorType::Rgb, png::BitDepth::Sixteen),
+            ColorType::Rgba8 => (png::ColorType::Rgba, png::BitDepth::Eight),
+            ColorType::Rgba16 => (png::ColorType::Rgba, png::BitDepth::Sixteen),
+            _ => {
+                return Err(ImageError::Unsupported(
+                    UnsupportedError::from_format_and_kind(
+                        ImageFormat::Png.into(),
+                        UnsupportedErrorKind::Color(color.into()),
+                    ),
+                ))
+            }
+        };
+        let comp = match self.compression {
+            CompressionType::Default => png::Compression::Default,
+            CompressionType::Best => png::Compression::Best,
+            _ => png::Compression::Fast,
+        };
+        let (filter, adaptive_filter) = match self.filter {
+            FilterType::NoFilter => (
+                png::FilterType::NoFilter,
+                png::AdaptiveFilterType::NonAdaptive,
+            ),
+            FilterType::Sub => (png::FilterType::Sub, png::AdaptiveFilterType::NonAdaptive),
+            FilterType::Up => (png::FilterType::Up, png::AdaptiveFilterType::NonAdaptive),
+            FilterType::Avg => (png::FilterType::Avg, png::AdaptiveFilterType::NonAdaptive),
+            FilterType::Paeth => (png::FilterType::Paeth, png::AdaptiveFilterType::NonAdaptive),
+            FilterType::Adaptive => (png::FilterType::Sub, png::AdaptiveFilterType::Adaptive),
+        };
+
+        let mut encoder = png::Encoder::new(self.w, width, height);
+        encoder.set_color(ct);
+        encoder.set_depth(bits);
+        encoder.set_compression(comp);
+        encoder.set_filter(filter);
+        encoder.set_adaptive_filter(adaptive_filter);
+        let mut writer = encoder
+            .write_header()
+            .map_err(|e| ImageError::IoError(e.into()))?;
+        writer
+            .write_image_data(data)
+            .map_err(|e| ImageError::IoError(e.into()))
+    }
+}
+
+impl<W: Write> ImageEncoder for PngEncoder<W> {
+    /// Write a PNG image with the specified width, height, and color type.
+    ///
+    /// For color types with 16-bit per channel or larger, the contents of `buf` should be in
+    /// native endian. PngEncoder will automatically convert to big endian as required by the
+    /// underlying PNG format.
+    fn write_image(
+        self,
+        buf: &[u8],
+        width: u32,
+        height: u32,
+        color_type: ColorType,
+    ) -> ImageResult<()> {
+        use byteorder::{BigEndian, ByteOrder, NativeEndian};
+        use ColorType::*;
+
+        // PNG images are big endian. For 16 bit per channel and larger types,
+        // the buffer may need to be reordered to big endian per the
+        // contract of `write_image`.
+        // TODO: assumes equal channel bit depth.
+        match color_type {
+            L8 | La8 | Rgb8 | Rgba8 => {
+                // No reodering necessary for u8
+                self.encode_inner(buf, width, height, color_type)
+            }
+            L16 | La16 | Rgb16 | Rgba16 => {
+                // Because the buffer is immutable and the PNG encoder does not
+                // yet take Write/Read traits, create a temporary buffer for
+                // big endian reordering.
+                let mut reordered = vec![0; buf.len()];
+                buf.chunks(2)
+                    .zip(reordered.chunks_mut(2))
+                    .for_each(|(b, r)| BigEndian::write_u16(r, NativeEndian::read_u16(b)));
+                self.encode_inner(&reordered, width, height, color_type)
+            }
+            _ => Err(ImageError::Encoding(EncodingError::new(
+                ImageFormat::Png.into(),
+                BadPngRepresentation::ColorType(color_type),
+            ))),
+        }
+    }
+}
+
+impl ImageError {
+    fn from_png(err: png::DecodingError) -> ImageError {
+        use png::DecodingError::*;
+        match err {
+            IoError(err) => ImageError::IoError(err),
+            // The input image was not a valid PNG.
+            err @ Format(_) => {
+                ImageError::Decoding(DecodingError::new(ImageFormat::Png.into(), err))
+            }
+            // Other is used when:
+            // - The decoder is polled for more animation frames despite being done (or not being animated
+            //   in the first place).
+            // - The output buffer does not have the required size.
+            err @ Parameter(_) => ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::Generic(err.to_string()),
+            )),
+            LimitsExceeded => {
+                ImageError::Limits(LimitError::from_kind(LimitErrorKind::InsufficientMemory))
+            }
+        }
+    }
+}
+
+impl Default for CompressionType {
+    fn default() -> Self {
+        CompressionType::Fast
+    }
+}
+
+impl Default for FilterType {
+    fn default() -> Self {
+        FilterType::Adaptive
+    }
+}
+
+impl fmt::Display for BadPngRepresentation {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        match self {
+            Self::ColorType(color_type) => write!(
+                f,
+                "The color {:?} can not be represented in PNG.",
+                color_type
+            ),
+        }
+    }
+}
+
+impl std::error::Error for BadPngRepresentation {}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::image::ImageDecoder;
+    use crate::ImageOutputFormat;
+
+    use std::io::{Cursor, Read};
+
+    #[test]
+    fn ensure_no_decoder_off_by_one() {
+        let dec = PngDecoder::new(
+            std::fs::File::open("tests/images/png/bugfixes/debug_triangle_corners_widescreen.png")
+                .unwrap(),
+        )
+        .expect("Unable to read PNG file (does it exist?)");
+
+        assert_eq![(2000, 1000), dec.dimensions()];
+
+        assert_eq![
+            ColorType::Rgb8,
+            dec.color_type(),
+            "Image MUST have the Rgb8 format"
+        ];
+
+        let correct_bytes = dec
+            .into_reader()
+            .expect("Unable to read file")
+            .bytes()
+            .map(|x| x.expect("Unable to read byte"))
+            .collect::<Vec<u8>>();
+
+        assert_eq![6_000_000, correct_bytes.len()];
+    }
+
+    #[test]
+    fn underlying_error() {
+        use std::error::Error;
+
+        let mut not_png =
+            std::fs::read("tests/images/png/bugfixes/debug_triangle_corners_widescreen.png")
+                .unwrap();
+        not_png[0] = 0;
+
+        let error = PngDecoder::new(&not_png[..]).err().unwrap();
+        let _ = error
+            .source()
+            .unwrap()
+            .downcast_ref::<png::DecodingError>()
+            .expect("Caused by a png error");
+    }
+
+    #[test]
+    fn encode_bad_color_type() {
+        // regression test for issue #1663
+        let image = DynamicImage::new_rgb32f(1, 1);
+        let mut target = Cursor::new(vec![]);
+        let _ = image.write_to(&mut target, ImageOutputFormat::Png);
+    }
+}

vendor/image/src/codecs/pnm/autobreak.rs

@@ -0,0 +1,124 @@
+//! Insert line breaks between written buffers when they would overflow the line length.
+use std::io;
+
+// The pnm standard says to insert line breaks after 70 characters. Assumes that no line breaks
+// are actually written. We have to be careful to fully commit buffers or not commit them at all,
+// otherwise we might insert a newline in the middle of a token.
+pub(crate) struct AutoBreak<W: io::Write> {
+    wrapped: W,
+    line_capacity: usize,
+    line: Vec<u8>,
+    has_newline: bool,
+    panicked: bool, // see https://github.com/rust-lang/rust/issues/30888
+}
+
+impl<W: io::Write> AutoBreak<W> {
+    pub(crate) fn new(writer: W, line_capacity: usize) -> Self {
+        AutoBreak {
+            wrapped: writer,
+            line_capacity,
+            line: Vec::with_capacity(line_capacity + 1),
+            has_newline: false,
+            panicked: false,
+        }
+    }
+
+    fn flush_buf(&mut self) -> io::Result<()> {
+        // from BufWriter
+        let mut written = 0;
+        let len = self.line.len();
+        let mut ret = Ok(());
+        while written < len {
+            self.panicked = true;
+            let r = self.wrapped.write(&self.line[written..]);
+            self.panicked = false;
+            match r {
+                Ok(0) => {
+                    ret = Err(io::Error::new(
+                        io::ErrorKind::WriteZero,
+                        "failed to write the buffered data",
+                    ));
+                    break;
+                }
+                Ok(n) => written += n,
+                Err(ref e) if e.kind() == io::ErrorKind::Interrupted => {}
+                Err(e) => {
+                    ret = Err(e);
+                    break;
+                }
+            }
+        }
+        if written > 0 {
+            self.line.drain(..written);
+        }
+        ret
+    }
+}
+
+impl<W: io::Write> io::Write for AutoBreak<W> {
+    fn write(&mut self, buffer: &[u8]) -> io::Result<usize> {
+        if self.has_newline {
+            self.flush()?;
+            self.has_newline = false;
+        }
+
+        if !self.line.is_empty() && self.line.len() + buffer.len() > self.line_capacity {
+            self.line.push(b'\n');
+            self.has_newline = true;
+            self.flush()?;
+            self.has_newline = false;
+        }
+
+        self.line.extend_from_slice(buffer);
+        Ok(buffer.len())
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        self.flush_buf()?;
+        self.wrapped.flush()
+    }
+}
+
+impl<W: io::Write> Drop for AutoBreak<W> {
+    fn drop(&mut self) {
+        if !self.panicked {
+            let _r = self.flush_buf();
+            // internal writer flushed automatically by Drop
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::io::Write;
+
+    #[test]
+    fn test_aligned_writes() {
+        let mut output = Vec::new();
+
+        {
+            let mut writer = AutoBreak::new(&mut output, 10);
+            writer.write_all(b"0123456789").unwrap();
+            writer.write_all(b"0123456789").unwrap();
+        }
+
+        assert_eq!(output.as_slice(), b"0123456789\n0123456789");
+    }
+
+    #[test]
+    fn test_greater_writes() {
+        let mut output = Vec::new();
+
+        {
+            let mut writer = AutoBreak::new(&mut output, 10);
+            writer.write_all(b"012").unwrap();
+            writer.write_all(b"345").unwrap();
+            writer.write_all(b"0123456789").unwrap();
+            writer.write_all(b"012345678910").unwrap();
+            writer.write_all(b"_").unwrap();
+        }
+
+        assert_eq!(output.as_slice(), b"012345\n0123456789\n012345678910\n_");
+    }
+}

vendor/image/src/codecs/pnm/encoder.rs

@@ -0,0 +1,673 @@
+//! Encoding of PNM Images
+use std::fmt;
+use std::io;
+
+use std::io::Write;
+
+use super::AutoBreak;
+use super::{ArbitraryHeader, ArbitraryTuplType, BitmapHeader, GraymapHeader, PixmapHeader};
+use super::{HeaderRecord, PnmHeader, PnmSubtype, SampleEncoding};
+use crate::color::{ColorType, ExtendedColorType};
+use crate::error::{
+    ImageError, ImageResult, ParameterError, ParameterErrorKind, UnsupportedError,
+    UnsupportedErrorKind,
+};
+use crate::image::{ImageEncoder, ImageFormat};
+
+use byteorder::{BigEndian, WriteBytesExt};
+
+enum HeaderStrategy {
+    Dynamic,
+    Subtype(PnmSubtype),
+    Chosen(PnmHeader),
+}
+
+#[derive(Clone, Copy)]
+pub enum FlatSamples<'a> {
+    U8(&'a [u8]),
+    U16(&'a [u16]),
+}
+
+/// Encodes images to any of the `pnm` image formats.
+pub struct PnmEncoder<W: Write> {
+    writer: W,
+    header: HeaderStrategy,
+}
+
+/// Encapsulate the checking system in the type system. Non of the fields are actually accessed
+/// but requiring them forces us to validly construct the struct anyways.
+struct CheckedImageBuffer<'a> {
+    _image: FlatSamples<'a>,
+    _width: u32,
+    _height: u32,
+    _color: ExtendedColorType,
+}
+
+// Check the header against the buffer. Each struct produces the next after a check.
+struct UncheckedHeader<'a> {
+    header: &'a PnmHeader,
+}
+
+struct CheckedDimensions<'a> {
+    unchecked: UncheckedHeader<'a>,
+    width: u32,
+    height: u32,
+}
+
+struct CheckedHeaderColor<'a> {
+    dimensions: CheckedDimensions<'a>,
+    color: ExtendedColorType,
+}
+
+struct CheckedHeader<'a> {
+    color: CheckedHeaderColor<'a>,
+    encoding: TupleEncoding<'a>,
+    _image: CheckedImageBuffer<'a>,
+}
+
+enum TupleEncoding<'a> {
+    PbmBits {
+        samples: FlatSamples<'a>,
+        width: u32,
+    },
+    Ascii {
+        samples: FlatSamples<'a>,
+    },
+    Bytes {
+        samples: FlatSamples<'a>,
+    },
+}
+
+impl<W: Write> PnmEncoder<W> {
+    /// Create new PnmEncoder from the `writer`.
+    ///
+    /// The encoded images will have some `pnm` format. If more control over the image type is
+    /// required, use either one of `with_subtype` or `with_header`. For more information on the
+    /// behaviour, see `with_dynamic_header`.
+    pub fn new(writer: W) -> Self {
+        PnmEncoder {
+            writer,
+            header: HeaderStrategy::Dynamic,
+        }
+    }
+
+    /// Encode a specific pnm subtype image.
+    ///
+    /// The magic number and encoding type will be chosen as provided while the rest of the header
+    /// data will be generated dynamically. Trying to encode incompatible images (e.g. encoding an
+    /// RGB image as Graymap) will result in an error.
+    ///
+    /// This will overwrite the effect of earlier calls to `with_header` and `with_dynamic_header`.
+    pub fn with_subtype(self, subtype: PnmSubtype) -> Self {
+        PnmEncoder {
+            writer: self.writer,
+            header: HeaderStrategy::Subtype(subtype),
+        }
+    }
+
+    /// Enforce the use of a chosen header.
+    ///
+    /// While this option gives the most control over the actual written data, the encoding process
+    /// will error in case the header data and image parameters do not agree. It is the users
+    /// obligation to ensure that the width and height are set accordingly, for example.
+    ///
+    /// Choose this option if you want a lossless decoding/encoding round trip.
+    ///
+    /// This will overwrite the effect of earlier calls to `with_subtype` and `with_dynamic_header`.
+    pub fn with_header(self, header: PnmHeader) -> Self {
+        PnmEncoder {
+            writer: self.writer,
+            header: HeaderStrategy::Chosen(header),
+        }
+    }
+
+    /// Create the header dynamically for each image.
+    ///
+    /// This is the default option upon creation of the encoder. With this, most images should be
+    /// encodable but the specific format chosen is out of the users control. The pnm subtype is
+    /// chosen arbitrarily by the library.
+    ///
+    /// This will overwrite the effect of earlier calls to `with_subtype` and `with_header`.
+    pub fn with_dynamic_header(self) -> Self {
+        PnmEncoder {
+            writer: self.writer,
+            header: HeaderStrategy::Dynamic,
+        }
+    }
+
+    /// Encode an image whose samples are represented as `u8`.
+    ///
+    /// Some `pnm` subtypes are incompatible with some color options, a chosen header most
+    /// certainly with any deviation from the original decoded image.
+    pub fn encode<'s, S>(
+        &mut self,
+        image: S,
+        width: u32,
+        height: u32,
+        color: ColorType,
+    ) -> ImageResult<()>
+    where
+        S: Into<FlatSamples<'s>>,
+    {
+        let image = image.into();
+        match self.header {
+            HeaderStrategy::Dynamic => {
+                self.write_dynamic_header(image, width, height, color.into())
+            }
+            HeaderStrategy::Subtype(subtype) => {
+                self.write_subtyped_header(subtype, image, width, height, color.into())
+            }
+            HeaderStrategy::Chosen(ref header) => Self::write_with_header(
+                &mut self.writer,
+                header,
+                image,
+                width,
+                height,
+                color.into(),
+            ),
+        }
+    }
+
+    /// Choose any valid pnm format that the image can be expressed in and write its header.
+    ///
+    /// Returns how the body should be written if successful.
+    fn write_dynamic_header(
+        &mut self,
+        image: FlatSamples,
+        width: u32,
+        height: u32,
+        color: ExtendedColorType,
+    ) -> ImageResult<()> {
+        let depth = u32::from(color.channel_count());
+        let (maxval, tupltype) = match color {
+            ExtendedColorType::L1 => (1, ArbitraryTuplType::BlackAndWhite),
+            ExtendedColorType::L8 => (0xff, ArbitraryTuplType::Grayscale),
+            ExtendedColorType::L16 => (0xffff, ArbitraryTuplType::Grayscale),
+            ExtendedColorType::La1 => (1, ArbitraryTuplType::BlackAndWhiteAlpha),
+            ExtendedColorType::La8 => (0xff, ArbitraryTuplType::GrayscaleAlpha),
+            ExtendedColorType::La16 => (0xffff, ArbitraryTuplType::GrayscaleAlpha),
+            ExtendedColorType::Rgb8 => (0xff, ArbitraryTuplType::RGB),
+            ExtendedColorType::Rgb16 => (0xffff, ArbitraryTuplType::RGB),
+            ExtendedColorType::Rgba8 => (0xff, ArbitraryTuplType::RGBAlpha),
+            ExtendedColorType::Rgba16 => (0xffff, ArbitraryTuplType::RGBAlpha),
+            _ => {
+                return Err(ImageError::Unsupported(
+                    UnsupportedError::from_format_and_kind(
+                        ImageFormat::Pnm.into(),
+                        UnsupportedErrorKind::Color(color),
+                    ),
+                ))
+            }
+        };
+
+        let header = PnmHeader {
+            decoded: HeaderRecord::Arbitrary(ArbitraryHeader {
+                width,
+                height,
+                depth,
+                maxval,
+                tupltype: Some(tupltype),
+            }),
+            encoded: None,
+        };
+
+        Self::write_with_header(&mut self.writer, &header, image, width, height, color)
+    }
+
+    /// Try to encode the image with the chosen format, give its corresponding pixel encoding type.
+    fn write_subtyped_header(
+        &mut self,
+        subtype: PnmSubtype,
+        image: FlatSamples,
+        width: u32,
+        height: u32,
+        color: ExtendedColorType,
+    ) -> ImageResult<()> {
+        let header = match (subtype, color) {
+            (PnmSubtype::ArbitraryMap, color) => {
+                return self.write_dynamic_header(image, width, height, color)
+            }
+            (PnmSubtype::Pixmap(encoding), ExtendedColorType::Rgb8) => PnmHeader {
+                decoded: HeaderRecord::Pixmap(PixmapHeader {
+                    encoding,
+                    width,
+                    height,
+                    maxval: 255,
+                }),
+                encoded: None,
+            },
+            (PnmSubtype::Graymap(encoding), ExtendedColorType::L8) => PnmHeader {
+                decoded: HeaderRecord::Graymap(GraymapHeader {
+                    encoding,
+                    width,
+                    height,
+                    maxwhite: 255,
+                }),
+                encoded: None,
+            },
+            (PnmSubtype::Bitmap(encoding), ExtendedColorType::L8)
+            | (PnmSubtype::Bitmap(encoding), ExtendedColorType::L1) => PnmHeader {
+                decoded: HeaderRecord::Bitmap(BitmapHeader {
+                    encoding,
+                    width,
+                    height,
+                }),
+                encoded: None,
+            },
+            (_, _) => {
+                return Err(ImageError::Parameter(ParameterError::from_kind(
+                    ParameterErrorKind::Generic(
+                        "Color type can not be represented in the chosen format".to_owned(),
+                    ),
+                )));
+            }
+        };
+
+        Self::write_with_header(&mut self.writer, &header, image, width, height, color)
+    }
+
+    /// Try to encode the image with the chosen header, checking if values are correct.
+    ///
+    /// Returns how the body should be written if successful.
+    fn write_with_header(
+        writer: &mut dyn Write,
+        header: &PnmHeader,
+        image: FlatSamples,
+        width: u32,
+        height: u32,
+        color: ExtendedColorType,
+    ) -> ImageResult<()> {
+        let unchecked = UncheckedHeader { header };
+
+        unchecked
+            .check_header_dimensions(width, height)?
+            .check_header_color(color)?
+            .check_sample_values(image)?
+            .write_header(writer)?
+            .write_image(writer)
+    }
+}
+
+impl<W: Write> ImageEncoder for PnmEncoder<W> {
+    fn write_image(
+        mut self,
+        buf: &[u8],
+        width: u32,
+        height: u32,
+        color_type: ColorType,
+    ) -> ImageResult<()> {
+        self.encode(buf, width, height, color_type)
+    }
+}
+
+impl<'a> CheckedImageBuffer<'a> {
+    fn check(
+        image: FlatSamples<'a>,
+        width: u32,
+        height: u32,
+        color: ExtendedColorType,
+    ) -> ImageResult<CheckedImageBuffer<'a>> {
+        let components = color.channel_count() as usize;
+        let uwidth = width as usize;
+        let uheight = height as usize;
+        let expected_len = components
+            .checked_mul(uwidth)
+            .and_then(|v| v.checked_mul(uheight));
+        if Some(image.len()) != expected_len {
+            // Image buffer does not correspond to size and colour.
+            return Err(ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::DimensionMismatch,
+            )));
+        }
+        Ok(CheckedImageBuffer {
+            _image: image,
+            _width: width,
+            _height: height,
+            _color: color,
+        })
+    }
+}
+
+impl<'a> UncheckedHeader<'a> {
+    fn check_header_dimensions(
+        self,
+        width: u32,
+        height: u32,
+    ) -> ImageResult<CheckedDimensions<'a>> {
+        if self.header.width() != width || self.header.height() != height {
+            // Chosen header does not match Image dimensions.
+            return Err(ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::DimensionMismatch,
+            )));
+        }
+
+        Ok(CheckedDimensions {
+            unchecked: self,
+            width,
+            height,
+        })
+    }
+}
+
+impl<'a> CheckedDimensions<'a> {
+    // Check color compatibility with the header. This will only error when we are certain that
+    // the combination is bogus (e.g. combining Pixmap and Palette) but allows uncertain
+    // combinations (basically a ArbitraryTuplType::Custom with any color of fitting depth).
+    fn check_header_color(self, color: ExtendedColorType) -> ImageResult<CheckedHeaderColor<'a>> {
+        let components = u32::from(color.channel_count());
+
+        match *self.unchecked.header {
+            PnmHeader {
+                decoded: HeaderRecord::Bitmap(_),
+                ..
+            } => match color {
+                ExtendedColorType::L1 | ExtendedColorType::L8 | ExtendedColorType::L16 => (),
+                _ => {
+                    return Err(ImageError::Parameter(ParameterError::from_kind(
+                        ParameterErrorKind::Generic(
+                            "PBM format only support luma color types".to_owned(),
+                        ),
+                    )))
+                }
+            },
+            PnmHeader {
+                decoded: HeaderRecord::Graymap(_),
+                ..
+            } => match color {
+                ExtendedColorType::L1 | ExtendedColorType::L8 | ExtendedColorType::L16 => (),
+                _ => {
+                    return Err(ImageError::Parameter(ParameterError::from_kind(
+                        ParameterErrorKind::Generic(
+                            "PGM format only support luma color types".to_owned(),
+                        ),
+                    )))
+                }
+            },
+            PnmHeader {
+                decoded: HeaderRecord::Pixmap(_),
+                ..
+            } => match color {
+                ExtendedColorType::Rgb8 => (),
+                _ => {
+                    return Err(ImageError::Parameter(ParameterError::from_kind(
+                        ParameterErrorKind::Generic(
+                            "PPM format only support ExtendedColorType::Rgb8".to_owned(),
+                        ),
+                    )))
+                }
+            },
+            PnmHeader {
+                decoded:
+                    HeaderRecord::Arbitrary(ArbitraryHeader {
+                        depth,
+                        ref tupltype,
+                        ..
+                    }),
+                ..
+            } => match (tupltype, color) {
+                (&Some(ArbitraryTuplType::BlackAndWhite), ExtendedColorType::L1) => (),
+                (&Some(ArbitraryTuplType::BlackAndWhiteAlpha), ExtendedColorType::La8) => (),
+
+                (&Some(ArbitraryTuplType::Grayscale), ExtendedColorType::L1) => (),
+                (&Some(ArbitraryTuplType::Grayscale), ExtendedColorType::L8) => (),
+                (&Some(ArbitraryTuplType::Grayscale), ExtendedColorType::L16) => (),
+                (&Some(ArbitraryTuplType::GrayscaleAlpha), ExtendedColorType::La8) => (),
+
+                (&Some(ArbitraryTuplType::RGB), ExtendedColorType::Rgb8) => (),
+                (&Some(ArbitraryTuplType::RGBAlpha), ExtendedColorType::Rgba8) => (),
+
+                (&None, _) if depth == components => (),
+                (&Some(ArbitraryTuplType::Custom(_)), _) if depth == components => (),
+                _ if depth != components => {
+                    return Err(ImageError::Parameter(ParameterError::from_kind(
+                        ParameterErrorKind::Generic(format!(
+                            "Depth mismatch: header {} vs. color {}",
+                            depth, components
+                        )),
+                    )))
+                }
+                _ => {
+                    return Err(ImageError::Parameter(ParameterError::from_kind(
+                        ParameterErrorKind::Generic(
+                            "Invalid color type for selected PAM color type".to_owned(),
+                        ),
+                    )))
+                }
+            },
+        }
+
+        Ok(CheckedHeaderColor {
+            dimensions: self,
+            color,
+        })
+    }
+}
+
+impl<'a> CheckedHeaderColor<'a> {
+    fn check_sample_values(self, image: FlatSamples<'a>) -> ImageResult<CheckedHeader<'a>> {
+        let header_maxval = match self.dimensions.unchecked.header.decoded {
+            HeaderRecord::Bitmap(_) => 1,
+            HeaderRecord::Graymap(GraymapHeader { maxwhite, .. }) => maxwhite,
+            HeaderRecord::Pixmap(PixmapHeader { maxval, .. }) => maxval,
+            HeaderRecord::Arbitrary(ArbitraryHeader { maxval, .. }) => maxval,
+        };
+
+        // We trust the image color bit count to be correct at least.
+        let max_sample = match self.color {
+            ExtendedColorType::Unknown(n) if n <= 16 => (1 << n) - 1,
+            ExtendedColorType::L1 => 1,
+            ExtendedColorType::L8
+            | ExtendedColorType::La8
+            | ExtendedColorType::Rgb8
+            | ExtendedColorType::Rgba8
+            | ExtendedColorType::Bgr8
+            | ExtendedColorType::Bgra8 => 0xff,
+            ExtendedColorType::L16
+            | ExtendedColorType::La16
+            | ExtendedColorType::Rgb16
+            | ExtendedColorType::Rgba16 => 0xffff,
+            _ => {
+                // Unsupported target color type.
+                return Err(ImageError::Unsupported(
+                    UnsupportedError::from_format_and_kind(
+                        ImageFormat::Pnm.into(),
+                        UnsupportedErrorKind::Color(self.color),
+                    ),
+                ));
+            }
+        };
+
+        // Avoid the performance heavy check if possible, e.g. if the header has been chosen by us.
+        if header_maxval < max_sample && !image.all_smaller(header_maxval) {
+            // Sample value greater than allowed for chosen header.
+            return Err(ImageError::Unsupported(
+                UnsupportedError::from_format_and_kind(
+                    ImageFormat::Pnm.into(),
+                    UnsupportedErrorKind::GenericFeature(
+                        "Sample value greater than allowed for chosen header".to_owned(),
+                    ),
+                ),
+            ));
+        }
+
+        let encoding = image.encoding_for(&self.dimensions.unchecked.header.decoded);
+
+        let image = CheckedImageBuffer::check(
+            image,
+            self.dimensions.width,
+            self.dimensions.height,
+            self.color,
+        )?;
+
+        Ok(CheckedHeader {
+            color: self,
+            encoding,
+            _image: image,
+        })
+    }
+}
+
+impl<'a> CheckedHeader<'a> {
+    fn write_header(self, writer: &mut dyn Write) -> ImageResult<TupleEncoding<'a>> {
+        self.header().write(writer)?;
+        Ok(self.encoding)
+    }
+
+    fn header(&self) -> &PnmHeader {
+        self.color.dimensions.unchecked.header
+    }
+}
+
+struct SampleWriter<'a>(&'a mut dyn Write);
+
+impl<'a> SampleWriter<'a> {
+    fn write_samples_ascii<V>(self, samples: V) -> io::Result<()>
+    where
+        V: Iterator,
+        V::Item: fmt::Display,
+    {
+        let mut auto_break_writer = AutoBreak::new(self.0, 70);
+        for value in samples {
+            write!(auto_break_writer, "{} ", value)?;
+        }
+        auto_break_writer.flush()
+    }
+
+    fn write_pbm_bits<V>(self, samples: &[V], width: u32) -> io::Result<()>
+    /* Default gives 0 for all primitives. TODO: replace this with `Zeroable` once it hits stable */
+    where
+        V: Default + Eq + Copy,
+    {
+        // The length of an encoded scanline
+        let line_width = (width - 1) / 8 + 1;
+
+        // We'll be writing single bytes, so buffer
+        let mut line_buffer = Vec::with_capacity(line_width as usize);
+
+        for line in samples.chunks(width as usize) {
+            for byte_bits in line.chunks(8) {
+                let mut byte = 0u8;
+                for i in 0..8 {
+                    // Black pixels are encoded as 1s
+                    if let Some(&v) = byte_bits.get(i) {
+                        if v == V::default() {
+                            byte |= 1u8 << (7 - i)
+                        }
+                    }
+                }
+                line_buffer.push(byte)
+            }
+            self.0.write_all(line_buffer.as_slice())?;
+            line_buffer.clear();
+        }
+
+        self.0.flush()
+    }
+}
+
+impl<'a> FlatSamples<'a> {
+    fn len(&self) -> usize {
+        match *self {
+            FlatSamples::U8(arr) => arr.len(),
+            FlatSamples::U16(arr) => arr.len(),
+        }
+    }
+
+    fn all_smaller(&self, max_val: u32) -> bool {
+        match *self {
+            FlatSamples::U8(arr) => arr.iter().any(|&val| u32::from(val) > max_val),
+            FlatSamples::U16(arr) => arr.iter().any(|&val| u32::from(val) > max_val),
+        }
+    }
+
+    fn encoding_for(&self, header: &HeaderRecord) -> TupleEncoding<'a> {
+        match *header {
+            HeaderRecord::Bitmap(BitmapHeader {
+                encoding: SampleEncoding::Binary,
+                width,
+                ..
+            }) => TupleEncoding::PbmBits {
+                samples: *self,
+                width,
+            },
+
+            HeaderRecord::Bitmap(BitmapHeader {
+                encoding: SampleEncoding::Ascii,
+                ..
+            }) => TupleEncoding::Ascii { samples: *self },
+
+            HeaderRecord::Arbitrary(_) => TupleEncoding::Bytes { samples: *self },
+
+            HeaderRecord::Graymap(GraymapHeader {
+                encoding: SampleEncoding::Ascii,
+                ..
+            })
+            | HeaderRecord::Pixmap(PixmapHeader {
+                encoding: SampleEncoding::Ascii,
+                ..
+            }) => TupleEncoding::Ascii { samples: *self },
+
+            HeaderRecord::Graymap(GraymapHeader {
+                encoding: SampleEncoding::Binary,
+                ..
+            })
+            | HeaderRecord::Pixmap(PixmapHeader {
+                encoding: SampleEncoding::Binary,
+                ..
+            }) => TupleEncoding::Bytes { samples: *self },
+        }
+    }
+}
+
+impl<'a> From<&'a [u8]> for FlatSamples<'a> {
+    fn from(samples: &'a [u8]) -> Self {
+        FlatSamples::U8(samples)
+    }
+}
+
+impl<'a> From<&'a [u16]> for FlatSamples<'a> {
+    fn from(samples: &'a [u16]) -> Self {
+        FlatSamples::U16(samples)
+    }
+}
+
+impl<'a> TupleEncoding<'a> {
+    fn write_image(&self, writer: &mut dyn Write) -> ImageResult<()> {
+        match *self {
+            TupleEncoding::PbmBits {
+                samples: FlatSamples::U8(samples),
+                width,
+            } => SampleWriter(writer)
+                .write_pbm_bits(samples, width)
+                .map_err(ImageError::IoError),
+            TupleEncoding::PbmBits {
+                samples: FlatSamples::U16(samples),
+                width,
+            } => SampleWriter(writer)
+                .write_pbm_bits(samples, width)
+                .map_err(ImageError::IoError),
+
+            TupleEncoding::Bytes {
+                samples: FlatSamples::U8(samples),
+            } => writer.write_all(samples).map_err(ImageError::IoError),
+            TupleEncoding::Bytes {
+                samples: FlatSamples::U16(samples),
+            } => samples.iter().try_for_each(|&sample| {
+                writer
+                    .write_u16::<BigEndian>(sample)
+                    .map_err(ImageError::IoError)
+            }),
+
+            TupleEncoding::Ascii {
+                samples: FlatSamples::U8(samples),
+            } => SampleWriter(writer)
+                .write_samples_ascii(samples.iter())
+                .map_err(ImageError::IoError),
+            TupleEncoding::Ascii {
+                samples: FlatSamples::U16(samples),
+            } => SampleWriter(writer)
+                .write_samples_ascii(samples.iter())
+                .map_err(ImageError::IoError),
+        }
+    }
+}

vendor/image/src/codecs/pnm/header.rs

@@ -0,0 +1,354 @@
+use std::{fmt, io};
+
+/// The kind of encoding used to store sample values
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+pub enum SampleEncoding {
+    /// Samples are unsigned binary integers in big endian
+    Binary,
+
+    /// Samples are encoded as decimal ascii strings separated by whitespace
+    Ascii,
+}
+
+/// Denotes the category of the magic number
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+pub enum PnmSubtype {
+    /// Magic numbers P1 and P4
+    Bitmap(SampleEncoding),
+
+    /// Magic numbers P2 and P5
+    Graymap(SampleEncoding),
+
+    /// Magic numbers P3 and P6
+    Pixmap(SampleEncoding),
+
+    /// Magic number P7
+    ArbitraryMap,
+}
+
+/// Stores the complete header data of a file.
+///
+/// Internally, provides mechanisms for lossless reencoding. After reading a file with the decoder
+/// it is possible to recover the header and construct an encoder. Using the encoder on the just
+/// loaded image should result in a byte copy of the original file (for single image pnms without
+/// additional trailing data).
+pub struct PnmHeader {
+    pub(crate) decoded: HeaderRecord,
+    pub(crate) encoded: Option<Vec<u8>>,
+}
+
+pub(crate) enum HeaderRecord {
+    Bitmap(BitmapHeader),
+    Graymap(GraymapHeader),
+    Pixmap(PixmapHeader),
+    Arbitrary(ArbitraryHeader),
+}
+
+/// Header produced by a `pbm` file ("Portable Bit Map")
+#[derive(Clone, Copy, Debug)]
+pub struct BitmapHeader {
+    /// Binary or Ascii encoded file
+    pub encoding: SampleEncoding,
+
+    /// Height of the image file
+    pub height: u32,
+
+    /// Width of the image file
+    pub width: u32,
+}
+
+/// Header produced by a `pgm` file ("Portable Gray Map")
+#[derive(Clone, Copy, Debug)]
+pub struct GraymapHeader {
+    /// Binary or Ascii encoded file
+    pub encoding: SampleEncoding,
+
+    /// Height of the image file
+    pub height: u32,
+
+    /// Width of the image file
+    pub width: u32,
+
+    /// Maximum sample value within the image
+    pub maxwhite: u32,
+}
+
+/// Header produced by a `ppm` file ("Portable Pixel Map")
+#[derive(Clone, Copy, Debug)]
+pub struct PixmapHeader {
+    /// Binary or Ascii encoded file
+    pub encoding: SampleEncoding,
+
+    /// Height of the image file
+    pub height: u32,
+
+    /// Width of the image file
+    pub width: u32,
+
+    /// Maximum sample value within the image
+    pub maxval: u32,
+}
+
+/// Header produced by a `pam` file ("Portable Arbitrary Map")
+#[derive(Clone, Debug)]
+pub struct ArbitraryHeader {
+    /// Height of the image file
+    pub height: u32,
+
+    /// Width of the image file
+    pub width: u32,
+
+    /// Number of color channels
+    pub depth: u32,
+
+    /// Maximum sample value within the image
+    pub maxval: u32,
+
+    /// Color interpretation of image pixels
+    pub tupltype: Option<ArbitraryTuplType>,
+}
+
+/// Standardized tuple type specifiers in the header of a `pam`.
+#[derive(Clone, Debug)]
+pub enum ArbitraryTuplType {
+    /// Pixels are either black (0) or white (1)
+    BlackAndWhite,
+
+    /// Pixels are either black (0) or white (1) and a second alpha channel
+    BlackAndWhiteAlpha,
+
+    /// Pixels represent the amount of white
+    Grayscale,
+
+    /// Grayscale with an additional alpha channel
+    GrayscaleAlpha,
+
+    /// Three channels: Red, Green, Blue
+    RGB,
+
+    /// Four channels: Red, Green, Blue, Alpha
+    RGBAlpha,
+
+    /// An image format which is not standardized
+    Custom(String),
+}
+
+impl ArbitraryTuplType {
+    pub(crate) fn name(&self) -> &str {
+        match self {
+            ArbitraryTuplType::BlackAndWhite => "BLACKANDWHITE",
+            ArbitraryTuplType::BlackAndWhiteAlpha => "BLACKANDWHITE_ALPHA",
+            ArbitraryTuplType::Grayscale => "GRAYSCALE",
+            ArbitraryTuplType::GrayscaleAlpha => "GRAYSCALE_ALPHA",
+            ArbitraryTuplType::RGB => "RGB",
+            ArbitraryTuplType::RGBAlpha => "RGB_ALPHA",
+            ArbitraryTuplType::Custom(custom) => custom,
+        }
+    }
+}
+
+impl PnmSubtype {
+    /// Get the two magic constant bytes corresponding to this format subtype.
+    pub fn magic_constant(self) -> &'static [u8; 2] {
+        match self {
+            PnmSubtype::Bitmap(SampleEncoding::Ascii) => b"P1",
+            PnmSubtype::Graymap(SampleEncoding::Ascii) => b"P2",
+            PnmSubtype::Pixmap(SampleEncoding::Ascii) => b"P3",
+            PnmSubtype::Bitmap(SampleEncoding::Binary) => b"P4",
+            PnmSubtype::Graymap(SampleEncoding::Binary) => b"P5",
+            PnmSubtype::Pixmap(SampleEncoding::Binary) => b"P6",
+            PnmSubtype::ArbitraryMap => b"P7",
+        }
+    }
+
+    /// Whether samples are stored as binary or as decimal ascii
+    pub fn sample_encoding(self) -> SampleEncoding {
+        match self {
+            PnmSubtype::ArbitraryMap => SampleEncoding::Binary,
+            PnmSubtype::Bitmap(enc) => enc,
+            PnmSubtype::Graymap(enc) => enc,
+            PnmSubtype::Pixmap(enc) => enc,
+        }
+    }
+}
+
+impl PnmHeader {
+    /// Retrieve the format subtype from which the header was created.
+    pub fn subtype(&self) -> PnmSubtype {
+        match self.decoded {
+            HeaderRecord::Bitmap(BitmapHeader { encoding, .. }) => PnmSubtype::Bitmap(encoding),
+            HeaderRecord::Graymap(GraymapHeader { encoding, .. }) => PnmSubtype::Graymap(encoding),
+            HeaderRecord::Pixmap(PixmapHeader { encoding, .. }) => PnmSubtype::Pixmap(encoding),
+            HeaderRecord::Arbitrary(ArbitraryHeader { .. }) => PnmSubtype::ArbitraryMap,
+        }
+    }
+
+    /// The width of the image this header is for.
+    pub fn width(&self) -> u32 {
+        match self.decoded {
+            HeaderRecord::Bitmap(BitmapHeader { width, .. }) => width,
+            HeaderRecord::Graymap(GraymapHeader { width, .. }) => width,
+            HeaderRecord::Pixmap(PixmapHeader { width, .. }) => width,
+            HeaderRecord::Arbitrary(ArbitraryHeader { width, .. }) => width,
+        }
+    }
+
+    /// The height of the image this header is for.
+    pub fn height(&self) -> u32 {
+        match self.decoded {
+            HeaderRecord::Bitmap(BitmapHeader { height, .. }) => height,
+            HeaderRecord::Graymap(GraymapHeader { height, .. }) => height,
+            HeaderRecord::Pixmap(PixmapHeader { height, .. }) => height,
+            HeaderRecord::Arbitrary(ArbitraryHeader { height, .. }) => height,
+        }
+    }
+
+    /// The biggest value a sample can have. In other words, the colour resolution.
+    pub fn maximal_sample(&self) -> u32 {
+        match self.decoded {
+            HeaderRecord::Bitmap(BitmapHeader { .. }) => 1,
+            HeaderRecord::Graymap(GraymapHeader { maxwhite, .. }) => maxwhite,
+            HeaderRecord::Pixmap(PixmapHeader { maxval, .. }) => maxval,
+            HeaderRecord::Arbitrary(ArbitraryHeader { maxval, .. }) => maxval,
+        }
+    }
+
+    /// Retrieve the underlying bitmap header if any
+    pub fn as_bitmap(&self) -> Option<&BitmapHeader> {
+        match self.decoded {
+            HeaderRecord::Bitmap(ref bitmap) => Some(bitmap),
+            _ => None,
+        }
+    }
+
+    /// Retrieve the underlying graymap header if any
+    pub fn as_graymap(&self) -> Option<&GraymapHeader> {
+        match self.decoded {
+            HeaderRecord::Graymap(ref graymap) => Some(graymap),
+            _ => None,
+        }
+    }
+
+    /// Retrieve the underlying pixmap header if any
+    pub fn as_pixmap(&self) -> Option<&PixmapHeader> {
+        match self.decoded {
+            HeaderRecord::Pixmap(ref pixmap) => Some(pixmap),
+            _ => None,
+        }
+    }
+
+    /// Retrieve the underlying arbitrary header if any
+    pub fn as_arbitrary(&self) -> Option<&ArbitraryHeader> {
+        match self.decoded {
+            HeaderRecord::Arbitrary(ref arbitrary) => Some(arbitrary),
+            _ => None,
+        }
+    }
+
+    /// Write the header back into a binary stream
+    pub fn write(&self, writer: &mut dyn io::Write) -> io::Result<()> {
+        writer.write_all(self.subtype().magic_constant())?;
+        match *self {
+            PnmHeader {
+                encoded: Some(ref content),
+                ..
+            } => writer.write_all(content),
+            PnmHeader {
+                decoded:
+                    HeaderRecord::Bitmap(BitmapHeader {
+                        encoding: _encoding,
+                        width,
+                        height,
+                    }),
+                ..
+            } => writeln!(writer, "\n{} {}", width, height),
+            PnmHeader {
+                decoded:
+                    HeaderRecord::Graymap(GraymapHeader {
+                        encoding: _encoding,
+                        width,
+                        height,
+                        maxwhite,
+                    }),
+                ..
+            } => writeln!(writer, "\n{} {} {}", width, height, maxwhite),
+            PnmHeader {
+                decoded:
+                    HeaderRecord::Pixmap(PixmapHeader {
+                        encoding: _encoding,
+                        width,
+                        height,
+                        maxval,
+                    }),
+                ..
+            } => writeln!(writer, "\n{} {} {}", width, height, maxval),
+            PnmHeader {
+                decoded:
+                    HeaderRecord::Arbitrary(ArbitraryHeader {
+                        width,
+                        height,
+                        depth,
+                        maxval,
+                        ref tupltype,
+                    }),
+                ..
+            } => {
+                struct TupltypeWriter<'a>(&'a Option<ArbitraryTuplType>);
+                impl<'a> fmt::Display for TupltypeWriter<'a> {
+                    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+                        match self.0 {
+                            Some(tt) => writeln!(f, "TUPLTYPE {}", tt.name()),
+                            None => Ok(()),
+                        }
+                    }
+                }
+
+                writeln!(
+                    writer,
+                    "\nWIDTH {}\nHEIGHT {}\nDEPTH {}\nMAXVAL {}\n{}ENDHDR",
+                    width,
+                    height,
+                    depth,
+                    maxval,
+                    TupltypeWriter(tupltype)
+                )
+            }
+        }
+    }
+}
+
+impl From<BitmapHeader> for PnmHeader {
+    fn from(header: BitmapHeader) -> Self {
+        PnmHeader {
+            decoded: HeaderRecord::Bitmap(header),
+            encoded: None,
+        }
+    }
+}
+
+impl From<GraymapHeader> for PnmHeader {
+    fn from(header: GraymapHeader) -> Self {
+        PnmHeader {
+            decoded: HeaderRecord::Graymap(header),
+            encoded: None,
+        }
+    }
+}
+
+impl From<PixmapHeader> for PnmHeader {
+    fn from(header: PixmapHeader) -> Self {
+        PnmHeader {
+            decoded: HeaderRecord::Pixmap(header),
+            encoded: None,
+        }
+    }
+}
+
+impl From<ArbitraryHeader> for PnmHeader {
+    fn from(header: ArbitraryHeader) -> Self {
+        PnmHeader {
+            decoded: HeaderRecord::Arbitrary(header),
+            encoded: None,
+        }
+    }
+}

vendor/image/src/codecs/pnm/mod.rs

@@ -0,0 +1,184 @@
+//! Decoding of netpbm image formats (pbm, pgm, ppm and pam).
+//!
+//! The formats pbm, pgm and ppm are fully supported. The pam decoder recognizes the tuple types
+//! `BLACKANDWHITE`, `GRAYSCALE` and `RGB` and explicitly recognizes but rejects their `_ALPHA`
+//! variants for now as alpha color types are unsupported.
+use self::autobreak::AutoBreak;
+pub use self::decoder::PnmDecoder;
+pub use self::encoder::PnmEncoder;
+use self::header::HeaderRecord;
+pub use self::header::{
+    ArbitraryHeader, ArbitraryTuplType, BitmapHeader, GraymapHeader, PixmapHeader,
+};
+pub use self::header::{PnmHeader, PnmSubtype, SampleEncoding};
+
+mod autobreak;
+mod decoder;
+mod encoder;
+mod header;
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::color::ColorType;
+    use crate::image::ImageDecoder;
+    use byteorder::{ByteOrder, NativeEndian};
+
+    fn execute_roundtrip_default(buffer: &[u8], width: u32, height: u32, color: ColorType) {
+        let mut encoded_buffer = Vec::new();
+
+        {
+            let mut encoder = PnmEncoder::new(&mut encoded_buffer);
+            encoder
+                .encode(buffer, width, height, color)
+                .expect("Failed to encode the image buffer");
+        }
+
+        let (header, loaded_color, loaded_image) = {
+            let decoder = PnmDecoder::new(&encoded_buffer[..]).unwrap();
+            let color_type = decoder.color_type();
+            let mut image = vec![0; decoder.total_bytes() as usize];
+            decoder
+                .read_image(&mut image)
+                .expect("Failed to decode the image");
+            let (_, header) = PnmDecoder::new(&encoded_buffer[..]).unwrap().into_inner();
+            (header, color_type, image)
+        };
+
+        assert_eq!(header.width(), width);
+        assert_eq!(header.height(), height);
+        assert_eq!(loaded_color, color);
+        assert_eq!(loaded_image.as_slice(), buffer);
+    }
+
+    fn execute_roundtrip_with_subtype(
+        buffer: &[u8],
+        width: u32,
+        height: u32,
+        color: ColorType,
+        subtype: PnmSubtype,
+    ) {
+        let mut encoded_buffer = Vec::new();
+
+        {
+            let mut encoder = PnmEncoder::new(&mut encoded_buffer).with_subtype(subtype);
+            encoder
+                .encode(buffer, width, height, color)
+                .expect("Failed to encode the image buffer");
+        }
+
+        let (header, loaded_color, loaded_image) = {
+            let decoder = PnmDecoder::new(&encoded_buffer[..]).unwrap();
+            let color_type = decoder.color_type();
+            let mut image = vec![0; decoder.total_bytes() as usize];
+            decoder
+                .read_image(&mut image)
+                .expect("Failed to decode the image");
+            let (_, header) = PnmDecoder::new(&encoded_buffer[..]).unwrap().into_inner();
+            (header, color_type, image)
+        };
+
+        assert_eq!(header.width(), width);
+        assert_eq!(header.height(), height);
+        assert_eq!(header.subtype(), subtype);
+        assert_eq!(loaded_color, color);
+        assert_eq!(loaded_image.as_slice(), buffer);
+    }
+
+    fn execute_roundtrip_u16(buffer: &[u16], width: u32, height: u32, color: ColorType) {
+        let mut encoded_buffer = Vec::new();
+
+        {
+            let mut encoder = PnmEncoder::new(&mut encoded_buffer);
+            encoder
+                .encode(buffer, width, height, color)
+                .expect("Failed to encode the image buffer");
+        }
+
+        let (header, loaded_color, loaded_image) = {
+            let decoder = PnmDecoder::new(&encoded_buffer[..]).unwrap();
+            let color_type = decoder.color_type();
+            let mut image = vec![0; decoder.total_bytes() as usize];
+            decoder
+                .read_image(&mut image)
+                .expect("Failed to decode the image");
+            let (_, header) = PnmDecoder::new(&encoded_buffer[..]).unwrap().into_inner();
+            (header, color_type, image)
+        };
+
+        let mut buffer_u8 = vec![0; buffer.len() * 2];
+        NativeEndian::write_u16_into(buffer, &mut buffer_u8[..]);
+
+        assert_eq!(header.width(), width);
+        assert_eq!(header.height(), height);
+        assert_eq!(loaded_color, color);
+        assert_eq!(loaded_image, buffer_u8);
+    }
+
+    #[test]
+    fn roundtrip_gray() {
+        #[rustfmt::skip]
+        let buf: [u8; 16] = [
+            0, 0, 0, 255,
+            255, 255, 255, 255,
+            255, 0, 255, 0,
+            255, 0, 0, 0,
+        ];
+
+        execute_roundtrip_default(&buf, 4, 4, ColorType::L8);
+        execute_roundtrip_with_subtype(&buf, 4, 4, ColorType::L8, PnmSubtype::ArbitraryMap);
+        execute_roundtrip_with_subtype(
+            &buf,
+            4,
+            4,
+            ColorType::L8,
+            PnmSubtype::Graymap(SampleEncoding::Ascii),
+        );
+        execute_roundtrip_with_subtype(
+            &buf,
+            4,
+            4,
+            ColorType::L8,
+            PnmSubtype::Graymap(SampleEncoding::Binary),
+        );
+    }
+
+    #[test]
+    fn roundtrip_rgb() {
+        #[rustfmt::skip]
+        let buf: [u8; 27] = [
+              0,   0,   0,
+              0,   0, 255,
+              0, 255,   0,
+              0, 255, 255,
+            255,   0,   0,
+            255,   0, 255,
+            255, 255,   0,
+            255, 255, 255,
+            255, 255, 255,
+        ];
+        execute_roundtrip_default(&buf, 3, 3, ColorType::Rgb8);
+        execute_roundtrip_with_subtype(&buf, 3, 3, ColorType::Rgb8, PnmSubtype::ArbitraryMap);
+        execute_roundtrip_with_subtype(
+            &buf,
+            3,
+            3,
+            ColorType::Rgb8,
+            PnmSubtype::Pixmap(SampleEncoding::Binary),
+        );
+        execute_roundtrip_with_subtype(
+            &buf,
+            3,
+            3,
+            ColorType::Rgb8,
+            PnmSubtype::Pixmap(SampleEncoding::Ascii),
+        );
+    }
+
+    #[test]
+    fn roundtrip_u16() {
+        let buf: [u16; 6] = [0, 1, 0xFFFF, 0x1234, 0x3412, 0xBEAF];
+
+        execute_roundtrip_u16(&buf, 6, 1, ColorType::L16);
+    }
+}

vendor/image/src/codecs/qoi.rs

@@ -0,0 +1,104 @@
+//! Decoding and encoding of QOI images
+
+use crate::{
+    error::{DecodingError, EncodingError},
+    ColorType, ImageDecoder, ImageEncoder, ImageError, ImageFormat, ImageResult,
+};
+use std::io::{Cursor, Read, Write};
+
+/// QOI decoder
+pub struct QoiDecoder<R> {
+    decoder: qoi::Decoder<R>,
+}
+
+impl<R> QoiDecoder<R>
+where
+    R: Read,
+{
+    /// Creates a new decoder that decodes from the stream ```reader```
+    pub fn new(reader: R) -> ImageResult<Self> {
+        let decoder = qoi::Decoder::from_stream(reader).map_err(decoding_error)?;
+        Ok(Self { decoder })
+    }
+}
+
+impl<'a, R: Read + 'a> ImageDecoder<'a> for QoiDecoder<R> {
+    type Reader = Cursor<Vec<u8>>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        (self.decoder.header().width, self.decoder.header().height)
+    }
+
+    fn color_type(&self) -> ColorType {
+        match self.decoder.header().channels {
+            qoi::Channels::Rgb => ColorType::Rgb8,
+            qoi::Channels::Rgba => ColorType::Rgba8,
+        }
+    }
+
+    fn into_reader(mut self) -> ImageResult<Self::Reader> {
+        let buffer = self.decoder.decode_to_vec().map_err(decoding_error)?;
+        Ok(Cursor::new(buffer))
+    }
+}
+
+fn decoding_error(error: qoi::Error) -> ImageError {
+    ImageError::Decoding(DecodingError::new(ImageFormat::Qoi.into(), error))
+}
+
+fn encoding_error(error: qoi::Error) -> ImageError {
+    ImageError::Encoding(EncodingError::new(ImageFormat::Qoi.into(), error))
+}
+
+/// QOI encoder
+pub struct QoiEncoder<W> {
+    writer: W,
+}
+
+impl<W: Write> QoiEncoder<W> {
+    /// Creates a new encoder that writes its output to ```writer```
+    pub fn new(writer: W) -> Self {
+        Self { writer }
+    }
+}
+
+impl<W: Write> ImageEncoder for QoiEncoder<W> {
+    fn write_image(
+        mut self,
+        buf: &[u8],
+        width: u32,
+        height: u32,
+        color_type: ColorType,
+    ) -> ImageResult<()> {
+        if !matches!(color_type, ColorType::Rgba8 | ColorType::Rgb8) {
+            return Err(ImageError::Encoding(EncodingError::new(
+                ImageFormat::Qoi.into(),
+                format!("unsupported color type {color_type:?}. Supported are Rgba8 and Rgb8."),
+            )));
+        }
+
+        // Encode data in QOI
+        let data = qoi::encode_to_vec(buf, width, height).map_err(encoding_error)?;
+
+        // Write data to buffer
+        self.writer.write_all(&data[..])?;
+        self.writer.flush()?;
+
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::fs::File;
+
+    #[test]
+    fn decode_test_image() {
+        let decoder = QoiDecoder::new(File::open("tests/images/qoi/basic-test.qoi").unwrap())
+            .expect("Unable to read QOI file");
+
+        assert_eq!((5, 5), decoder.dimensions());
+        assert_eq!(ColorType::Rgba8, decoder.color_type());
+    }
+}

vendor/image/src/codecs/tga/decoder.rs

@@ -0,0 +1,502 @@
+use super::header::{Header, ImageType, ALPHA_BIT_MASK, SCREEN_ORIGIN_BIT_MASK};
+use crate::{
+    color::{ColorType, ExtendedColorType},
+    error::{
+        ImageError, ImageResult, LimitError, LimitErrorKind, UnsupportedError, UnsupportedErrorKind,
+    },
+    image::{ImageDecoder, ImageFormat, ImageReadBuffer},
+};
+use byteorder::ReadBytesExt;
+use std::{
+    convert::TryFrom,
+    io::{self, Read, Seek},
+    mem,
+};
+
+struct ColorMap {
+    /// sizes in bytes
+    start_offset: usize,
+    entry_size: usize,
+    bytes: Vec<u8>,
+}
+
+impl ColorMap {
+    pub(crate) fn from_reader(
+        r: &mut dyn Read,
+        start_offset: u16,
+        num_entries: u16,
+        bits_per_entry: u8,
+    ) -> ImageResult<ColorMap> {
+        let bytes_per_entry = (bits_per_entry as usize + 7) / 8;
+
+        let mut bytes = vec![0; bytes_per_entry * num_entries as usize];
+        r.read_exact(&mut bytes)?;
+
+        Ok(ColorMap {
+            entry_size: bytes_per_entry,
+            start_offset: start_offset as usize,
+            bytes,
+        })
+    }
+
+    /// Get one entry from the color map
+    pub(crate) fn get(&self, index: usize) -> Option<&[u8]> {
+        let entry = self.start_offset + self.entry_size * index;
+        self.bytes.get(entry..entry + self.entry_size)
+    }
+}
+
+/// The representation of a TGA decoder
+pub struct TgaDecoder<R> {
+    r: R,
+
+    width: usize,
+    height: usize,
+    bytes_per_pixel: usize,
+    has_loaded_metadata: bool,
+
+    image_type: ImageType,
+    color_type: ColorType,
+    original_color_type: Option<ExtendedColorType>,
+
+    header: Header,
+    color_map: Option<ColorMap>,
+
+    // Used in read_scanline
+    line_read: Option<usize>,
+    line_remain_buff: Vec<u8>,
+}
+
+impl<R: Read + Seek> TgaDecoder<R> {
+    /// Create a new decoder that decodes from the stream `r`
+    pub fn new(r: R) -> ImageResult<TgaDecoder<R>> {
+        let mut decoder = TgaDecoder {
+            r,
+
+            width: 0,
+            height: 0,
+            bytes_per_pixel: 0,
+            has_loaded_metadata: false,
+
+            image_type: ImageType::Unknown,
+            color_type: ColorType::L8,
+            original_color_type: None,
+
+            header: Header::default(),
+            color_map: None,
+
+            line_read: None,
+            line_remain_buff: Vec::new(),
+        };
+        decoder.read_metadata()?;
+        Ok(decoder)
+    }
+
+    fn read_header(&mut self) -> ImageResult<()> {
+        self.header = Header::from_reader(&mut self.r)?;
+        self.image_type = ImageType::new(self.header.image_type);
+        self.width = self.header.image_width as usize;
+        self.height = self.header.image_height as usize;
+        self.bytes_per_pixel = (self.header.pixel_depth as usize + 7) / 8;
+        Ok(())
+    }
+
+    fn read_metadata(&mut self) -> ImageResult<()> {
+        if !self.has_loaded_metadata {
+            self.read_header()?;
+            self.read_image_id()?;
+            self.read_color_map()?;
+            self.read_color_information()?;
+            self.has_loaded_metadata = true;
+        }
+        Ok(())
+    }
+
+    /// Loads the color information for the decoder
+    ///
+    /// To keep things simple, we won't handle bit depths that aren't divisible
+    /// by 8 and are larger than 32.
+    fn read_color_information(&mut self) -> ImageResult<()> {
+        if self.header.pixel_depth % 8 != 0 || self.header.pixel_depth > 32 {
+            // Bit depth must be divisible by 8, and must be less than or equal
+            // to 32.
+            return Err(ImageError::Unsupported(
+                UnsupportedError::from_format_and_kind(
+                    ImageFormat::Tga.into(),
+                    UnsupportedErrorKind::Color(ExtendedColorType::Unknown(
+                        self.header.pixel_depth,
+                    )),
+                ),
+            ));
+        }
+
+        let num_alpha_bits = self.header.image_desc & ALPHA_BIT_MASK;
+
+        let other_channel_bits = if self.header.map_type != 0 {
+            self.header.map_entry_size
+        } else {
+            if num_alpha_bits > self.header.pixel_depth {
+                return Err(ImageError::Unsupported(
+                    UnsupportedError::from_format_and_kind(
+                        ImageFormat::Tga.into(),
+                        UnsupportedErrorKind::Color(ExtendedColorType::Unknown(
+                            self.header.pixel_depth,
+                        )),
+                    ),
+                ));
+            }
+
+            self.header.pixel_depth - num_alpha_bits
+        };
+        let color = self.image_type.is_color();
+
+        match (num_alpha_bits, other_channel_bits, color) {
+            // really, the encoding is BGR and BGRA, this is fixed
+            // up with `TgaDecoder::reverse_encoding`.
+            (0, 32, true) => self.color_type = ColorType::Rgba8,
+            (8, 24, true) => self.color_type = ColorType::Rgba8,
+            (0, 24, true) => self.color_type = ColorType::Rgb8,
+            (8, 8, false) => self.color_type = ColorType::La8,
+            (0, 8, false) => self.color_type = ColorType::L8,
+            (8, 0, false) => {
+                // alpha-only image is treated as L8
+                self.color_type = ColorType::L8;
+                self.original_color_type = Some(ExtendedColorType::A8);
+            }
+            _ => {
+                return Err(ImageError::Unsupported(
+                    UnsupportedError::from_format_and_kind(
+                        ImageFormat::Tga.into(),
+                        UnsupportedErrorKind::Color(ExtendedColorType::Unknown(
+                            self.header.pixel_depth,
+                        )),
+                    ),
+                ))
+            }
+        }
+        Ok(())
+    }
+
+    /// Read the image id field
+    ///
+    /// We're not interested in this field, so this function skips it if it
+    /// is present
+    fn read_image_id(&mut self) -> ImageResult<()> {
+        self.r
+            .seek(io::SeekFrom::Current(i64::from(self.header.id_length)))?;
+        Ok(())
+    }
+
+    fn read_color_map(&mut self) -> ImageResult<()> {
+        if self.header.map_type == 1 {
+            // FIXME: we could reverse the map entries, which avoids having to reverse all pixels
+            // in the final output individually.
+            self.color_map = Some(ColorMap::from_reader(
+                &mut self.r,
+                self.header.map_origin,
+                self.header.map_length,
+                self.header.map_entry_size,
+            )?);
+        }
+        Ok(())
+    }
+
+    /// Expands indices into its mapped color
+    fn expand_color_map(&self, pixel_data: &[u8]) -> io::Result<Vec<u8>> {
+        #[inline]
+        fn bytes_to_index(bytes: &[u8]) -> usize {
+            let mut result = 0usize;
+            for byte in bytes.iter() {
+                result = result << 8 | *byte as usize;
+            }
+            result
+        }
+
+        let bytes_per_entry = (self.header.map_entry_size as usize + 7) / 8;
+        let mut result = Vec::with_capacity(self.width * self.height * bytes_per_entry);
+
+        if self.bytes_per_pixel == 0 {
+            return Err(io::ErrorKind::Other.into());
+        }
+
+        let color_map = self
+            .color_map
+            .as_ref()
+            .ok_or_else(|| io::Error::from(io::ErrorKind::Other))?;
+
+        for chunk in pixel_data.chunks(self.bytes_per_pixel) {
+            let index = bytes_to_index(chunk);
+            if let Some(color) = color_map.get(index) {
+                result.extend_from_slice(color);
+            } else {
+                return Err(io::ErrorKind::Other.into());
+            }
+        }
+
+        Ok(result)
+    }
+
+    /// Reads a run length encoded data for given number of bytes
+    fn read_encoded_data(&mut self, num_bytes: usize) -> io::Result<Vec<u8>> {
+        let mut pixel_data = Vec::with_capacity(num_bytes);
+        let mut repeat_buf = Vec::with_capacity(self.bytes_per_pixel);
+
+        while pixel_data.len() < num_bytes {
+            let run_packet = self.r.read_u8()?;
+            // If the highest bit in `run_packet` is set, then we repeat pixels
+            //
+            // Note: the TGA format adds 1 to both counts because having a count
+            // of 0 would be pointless.
+            if (run_packet & 0x80) != 0 {
+                // high bit set, so we will repeat the data
+                let repeat_count = ((run_packet & !0x80) + 1) as usize;
+                self.r
+                    .by_ref()
+                    .take(self.bytes_per_pixel as u64)
+                    .read_to_end(&mut repeat_buf)?;
+
+                // get the repeating pixels from the bytes of the pixel stored in `repeat_buf`
+                let data = repeat_buf
+                    .iter()
+                    .cycle()
+                    .take(repeat_count * self.bytes_per_pixel);
+                pixel_data.extend(data);
+                repeat_buf.clear();
+            } else {
+                // not set, so `run_packet+1` is the number of non-encoded pixels
+                let num_raw_bytes = (run_packet + 1) as usize * self.bytes_per_pixel;
+                self.r
+                    .by_ref()
+                    .take(num_raw_bytes as u64)
+                    .read_to_end(&mut pixel_data)?;
+            }
+        }
+
+        if pixel_data.len() > num_bytes {
+            // FIXME: the last packet contained more data than we asked for!
+            // This is at least a warning. We truncate the data since some methods rely on the
+            // length to be accurate in the success case.
+            pixel_data.truncate(num_bytes);
+        }
+
+        Ok(pixel_data)
+    }
+
+    /// Reads a run length encoded packet
+    fn read_all_encoded_data(&mut self) -> ImageResult<Vec<u8>> {
+        let num_bytes = self.width * self.height * self.bytes_per_pixel;
+
+        Ok(self.read_encoded_data(num_bytes)?)
+    }
+
+    /// Reads a run length encoded line
+    fn read_encoded_line(&mut self) -> io::Result<Vec<u8>> {
+        let line_num_bytes = self.width * self.bytes_per_pixel;
+        let remain_len = self.line_remain_buff.len();
+
+        if remain_len >= line_num_bytes {
+            // `Vec::split_to` if std had it
+            let bytes = {
+                let bytes_after = self.line_remain_buff.split_off(line_num_bytes);
+                mem::replace(&mut self.line_remain_buff, bytes_after)
+            };
+
+            return Ok(bytes);
+        }
+
+        let num_bytes = line_num_bytes - remain_len;
+
+        let line_data = self.read_encoded_data(num_bytes)?;
+
+        let mut pixel_data = Vec::with_capacity(line_num_bytes);
+        pixel_data.append(&mut self.line_remain_buff);
+        pixel_data.extend_from_slice(&line_data[..num_bytes]);
+
+        // put the remain data to line_remain_buff.
+        // expects `self.line_remain_buff` to be empty from
+        // the above `pixel_data.append` call
+        debug_assert!(self.line_remain_buff.is_empty());
+        self.line_remain_buff
+            .extend_from_slice(&line_data[num_bytes..]);
+
+        Ok(pixel_data)
+    }
+
+    /// Reverse from BGR encoding to RGB encoding
+    ///
+    /// TGA files are stored in the BGRA encoding. This function swaps
+    /// the blue and red bytes in the `pixels` array.
+    fn reverse_encoding_in_output(&mut self, pixels: &mut [u8]) {
+        // We only need to reverse the encoding of color images
+        match self.color_type {
+            ColorType::Rgb8 | ColorType::Rgba8 => {
+                for chunk in pixels.chunks_mut(self.color_type.bytes_per_pixel().into()) {
+                    chunk.swap(0, 2);
+                }
+            }
+            _ => {}
+        }
+    }
+
+    /// Flip the image vertically depending on the screen origin bit
+    ///
+    /// The bit in position 5 of the image descriptor byte is the screen origin bit.
+    /// If it's 1, the origin is in the top left corner.
+    /// If it's 0, the origin is in the bottom left corner.
+    /// This function checks the bit, and if it's 0, flips the image vertically.
+    fn flip_vertically(&mut self, pixels: &mut [u8]) {
+        if self.is_flipped_vertically() {
+            if self.height == 0 {
+                return;
+            }
+
+            let num_bytes = pixels.len();
+
+            let width_bytes = num_bytes / self.height;
+
+            // Flip the image vertically.
+            for vertical_index in 0..(self.height / 2) {
+                let vertical_target = (self.height - vertical_index) * width_bytes - width_bytes;
+
+                for horizontal_index in 0..width_bytes {
+                    let source = vertical_index * width_bytes + horizontal_index;
+                    let target = vertical_target + horizontal_index;
+
+                    pixels.swap(target, source);
+                }
+            }
+        }
+    }
+
+    /// Check whether the image is vertically flipped
+    ///
+    /// The bit in position 5 of the image descriptor byte is the screen origin bit.
+    /// If it's 1, the origin is in the top left corner.
+    /// If it's 0, the origin is in the bottom left corner.
+    /// This function checks the bit, and if it's 0, flips the image vertically.
+    fn is_flipped_vertically(&self) -> bool {
+        let screen_origin_bit = SCREEN_ORIGIN_BIT_MASK & self.header.image_desc != 0;
+        !screen_origin_bit
+    }
+
+    fn read_scanline(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        if let Some(line_read) = self.line_read {
+            if line_read == self.height {
+                return Ok(0);
+            }
+        }
+
+        // read the pixels from the data region
+        let mut pixel_data = if self.image_type.is_encoded() {
+            self.read_encoded_line()?
+        } else {
+            let num_raw_bytes = self.width * self.bytes_per_pixel;
+            let mut buf = vec![0; num_raw_bytes];
+            self.r.by_ref().read_exact(&mut buf)?;
+            buf
+        };
+
+        // expand the indices using the color map if necessary
+        if self.image_type.is_color_mapped() {
+            pixel_data = self.expand_color_map(&pixel_data)?;
+        }
+        self.reverse_encoding_in_output(&mut pixel_data);
+
+        // copy to the output buffer
+        buf[..pixel_data.len()].copy_from_slice(&pixel_data);
+
+        self.line_read = Some(self.line_read.unwrap_or(0) + 1);
+
+        Ok(pixel_data.len())
+    }
+}
+
+impl<'a, R: 'a + Read + Seek> ImageDecoder<'a> for TgaDecoder<R> {
+    type Reader = TGAReader<R>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        (self.width as u32, self.height as u32)
+    }
+
+    fn color_type(&self) -> ColorType {
+        self.color_type
+    }
+
+    fn original_color_type(&self) -> ExtendedColorType {
+        self.original_color_type
+            .unwrap_or_else(|| self.color_type().into())
+    }
+
+    fn scanline_bytes(&self) -> u64 {
+        // This cannot overflow because TGA has a maximum width of u16::MAX_VALUE and
+        // `bytes_per_pixel` is a u8.
+        u64::from(self.color_type.bytes_per_pixel()) * self.width as u64
+    }
+
+    fn into_reader(self) -> ImageResult<Self::Reader> {
+        Ok(TGAReader {
+            buffer: ImageReadBuffer::new(self.scanline_bytes(), self.total_bytes()),
+            decoder: self,
+        })
+    }
+
+    fn read_image(mut self, buf: &mut [u8]) -> ImageResult<()> {
+        assert_eq!(u64::try_from(buf.len()), Ok(self.total_bytes()));
+
+        // In indexed images, we might need more bytes than pixels to read them. That's nonsensical
+        // to encode but we'll not want to crash.
+        let mut fallback_buf = vec![];
+        // read the pixels from the data region
+        let rawbuf = if self.image_type.is_encoded() {
+            let pixel_data = self.read_all_encoded_data()?;
+            if self.bytes_per_pixel <= usize::from(self.color_type.bytes_per_pixel()) {
+                buf[..pixel_data.len()].copy_from_slice(&pixel_data);
+                &buf[..pixel_data.len()]
+            } else {
+                fallback_buf = pixel_data;
+                &fallback_buf[..]
+            }
+        } else {
+            let num_raw_bytes = self.width * self.height * self.bytes_per_pixel;
+            if self.bytes_per_pixel <= usize::from(self.color_type.bytes_per_pixel()) {
+                self.r.by_ref().read_exact(&mut buf[..num_raw_bytes])?;
+                &buf[..num_raw_bytes]
+            } else {
+                fallback_buf.resize(num_raw_bytes, 0u8);
+                self.r
+                    .by_ref()
+                    .read_exact(&mut fallback_buf[..num_raw_bytes])?;
+                &fallback_buf[..num_raw_bytes]
+            }
+        };
+
+        // expand the indices using the color map if necessary
+        if self.image_type.is_color_mapped() {
+            let pixel_data = self.expand_color_map(rawbuf)?;
+            // not enough data to fill the buffer, or would overflow the buffer
+            if pixel_data.len() != buf.len() {
+                return Err(ImageError::Limits(LimitError::from_kind(
+                    LimitErrorKind::DimensionError,
+                )));
+            }
+            buf.copy_from_slice(&pixel_data);
+        }
+
+        self.reverse_encoding_in_output(buf);
+
+        self.flip_vertically(buf);
+
+        Ok(())
+    }
+}
+
+pub struct TGAReader<R> {
+    buffer: ImageReadBuffer,
+    decoder: TgaDecoder<R>,
+}
+impl<R: Read + Seek> Read for TGAReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        let decoder = &mut self.decoder;
+        self.buffer.read(buf, |buf| decoder.read_scanline(buf))
+    }
+}

vendor/image/src/codecs/tga/encoder.rs

@@ -0,0 +1,215 @@
+use super::header::Header;
+use crate::{error::EncodingError, ColorType, ImageEncoder, ImageError, ImageFormat, ImageResult};
+use std::{convert::TryFrom, error, fmt, io::Write};
+
+/// Errors that can occur during encoding and saving of a TGA image.
+#[derive(Debug, Copy, Clone, Hash, PartialEq, Eq, PartialOrd, Ord)]
+enum EncoderError {
+    /// Invalid TGA width.
+    WidthInvalid(u32),
+
+    /// Invalid TGA height.
+    HeightInvalid(u32),
+}
+
+impl fmt::Display for EncoderError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            EncoderError::WidthInvalid(s) => f.write_fmt(format_args!("Invalid TGA width: {}", s)),
+            EncoderError::HeightInvalid(s) => {
+                f.write_fmt(format_args!("Invalid TGA height: {}", s))
+            }
+        }
+    }
+}
+
+impl From<EncoderError> for ImageError {
+    fn from(e: EncoderError) -> ImageError {
+        ImageError::Encoding(EncodingError::new(ImageFormat::Tga.into(), e))
+    }
+}
+
+impl error::Error for EncoderError {}
+
+/// TGA encoder.
+pub struct TgaEncoder<W: Write> {
+    writer: W,
+}
+
+impl<W: Write> TgaEncoder<W> {
+    /// Create a new encoder that writes its output to ```w```.
+    pub fn new(w: W) -> TgaEncoder<W> {
+        TgaEncoder { writer: w }
+    }
+
+    /// Encodes the image ```buf``` that has dimensions ```width```
+    /// and ```height``` and ```ColorType``` ```color_type```.
+    ///
+    /// The dimensions of the image must be between 0 and 65535 (inclusive) or
+    /// an error will be returned.
+    pub fn encode(
+        mut self,
+        buf: &[u8],
+        width: u32,
+        height: u32,
+        color_type: ColorType,
+    ) -> ImageResult<()> {
+        // Validate dimensions.
+        let width = u16::try_from(width)
+            .map_err(|_| ImageError::from(EncoderError::WidthInvalid(width)))?;
+
+        let height = u16::try_from(height)
+            .map_err(|_| ImageError::from(EncoderError::HeightInvalid(height)))?;
+
+        // Write out TGA header.
+        let header = Header::from_pixel_info(color_type, width, height)?;
+        header.write_to(&mut self.writer)?;
+
+        // Write out Bgr(a)8 or L(a)8 image data.
+        match color_type {
+            ColorType::Rgb8 | ColorType::Rgba8 => {
+                let mut image = Vec::from(buf);
+
+                for chunk in image.chunks_mut(usize::from(color_type.bytes_per_pixel())) {
+                    chunk.swap(0, 2);
+                }
+
+                self.writer.write_all(&image)?;
+            }
+            _ => {
+                self.writer.write_all(buf)?;
+            }
+        }
+
+        Ok(())
+    }
+}
+
+impl<W: Write> ImageEncoder for TgaEncoder<W> {
+    fn write_image(
+        self,
+        buf: &[u8],
+        width: u32,
+        height: u32,
+        color_type: ColorType,
+    ) -> ImageResult<()> {
+        self.encode(buf, width, height, color_type)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{EncoderError, TgaEncoder};
+    use crate::{codecs::tga::TgaDecoder, ColorType, ImageDecoder, ImageError};
+    use std::{error::Error, io::Cursor};
+
+    fn round_trip_image(image: &[u8], width: u32, height: u32, c: ColorType) -> Vec<u8> {
+        let mut encoded_data = Vec::new();
+        {
+            let encoder = TgaEncoder::new(&mut encoded_data);
+            encoder
+                .encode(&image, width, height, c)
+                .expect("could not encode image");
+        }
+
+        let decoder = TgaDecoder::new(Cursor::new(&encoded_data)).expect("failed to decode");
+
+        let mut buf = vec![0; decoder.total_bytes() as usize];
+        decoder.read_image(&mut buf).expect("failed to decode");
+        buf
+    }
+
+    #[test]
+    fn test_image_width_too_large() {
+        // TGA cannot encode images larger than 65,535×65,535
+        // create a 65,536×1 8-bit black image buffer
+        let size = usize::from(u16::MAX) + 1;
+        let dimension = size as u32;
+        let img = vec![0u8; size];
+        // Try to encode an image that is too large
+        let mut encoded = Vec::new();
+        let encoder = TgaEncoder::new(&mut encoded);
+        let result = encoder.encode(&img, dimension, 1, ColorType::L8);
+        match result {
+            Err(ImageError::Encoding(err)) => {
+                let err = err
+                    .source()
+                    .unwrap()
+                    .downcast_ref::<EncoderError>()
+                    .unwrap();
+                assert_eq!(*err, EncoderError::WidthInvalid(dimension));
+            }
+            other => panic!(
+                "Encoding an image that is too wide should return a InvalidWidth \
+                it returned {:?} instead",
+                other
+            ),
+        }
+    }
+
+    #[test]
+    fn test_image_height_too_large() {
+        // TGA cannot encode images larger than 65,535×65,535
+        // create a 65,536×1 8-bit black image buffer
+        let size = usize::from(u16::MAX) + 1;
+        let dimension = size as u32;
+        let img = vec![0u8; size];
+        // Try to encode an image that is too large
+        let mut encoded = Vec::new();
+        let encoder = TgaEncoder::new(&mut encoded);
+        let result = encoder.encode(&img, 1, dimension, ColorType::L8);
+        match result {
+            Err(ImageError::Encoding(err)) => {
+                let err = err
+                    .source()
+                    .unwrap()
+                    .downcast_ref::<EncoderError>()
+                    .unwrap();
+                assert_eq!(*err, EncoderError::HeightInvalid(dimension));
+            }
+            other => panic!(
+                "Encoding an image that is too tall should return a InvalidHeight \
+                it returned {:?} instead",
+                other
+            ),
+        }
+    }
+
+    #[test]
+    fn round_trip_single_pixel_rgb() {
+        let image = [0, 1, 2];
+        let decoded = round_trip_image(&image, 1, 1, ColorType::Rgb8);
+        assert_eq!(decoded.len(), image.len());
+        assert_eq!(decoded.as_slice(), image);
+    }
+
+    #[test]
+    fn round_trip_single_pixel_rgba() {
+        let image = [0, 1, 2, 3];
+        let decoded = round_trip_image(&image, 1, 1, ColorType::Rgba8);
+        assert_eq!(decoded.len(), image.len());
+        assert_eq!(decoded.as_slice(), image);
+    }
+
+    #[test]
+    fn round_trip_gray() {
+        let image = [0, 1, 2];
+        let decoded = round_trip_image(&image, 3, 1, ColorType::L8);
+        assert_eq!(decoded.len(), image.len());
+        assert_eq!(decoded.as_slice(), image);
+    }
+
+    #[test]
+    fn round_trip_graya() {
+        let image = [0, 1, 2, 3, 4, 5];
+        let decoded = round_trip_image(&image, 1, 3, ColorType::La8);
+        assert_eq!(decoded.len(), image.len());
+        assert_eq!(decoded.as_slice(), image);
+    }
+
+    #[test]
+    fn round_trip_3px_rgb() {
+        let image = [0; 3 * 3 * 3]; // 3x3 pixels, 3 bytes per pixel
+        let _decoded = round_trip_image(&image, 3, 3, ColorType::Rgb8);
+    }
+}

vendor/image/src/codecs/tga/header.rs

@@ -0,0 +1,150 @@
+use crate::{
+    error::{UnsupportedError, UnsupportedErrorKind},
+    ColorType, ImageError, ImageFormat, ImageResult,
+};
+use byteorder::{LittleEndian, ReadBytesExt, WriteBytesExt};
+use std::io::{Read, Write};
+
+pub(crate) const ALPHA_BIT_MASK: u8 = 0b1111;
+pub(crate) const SCREEN_ORIGIN_BIT_MASK: u8 = 0b10_0000;
+
+pub(crate) enum ImageType {
+    NoImageData = 0,
+    /// Uncompressed images.
+    RawColorMap = 1,
+    RawTrueColor = 2,
+    RawGrayScale = 3,
+    /// Run length encoded images.
+    RunColorMap = 9,
+    RunTrueColor = 10,
+    RunGrayScale = 11,
+    Unknown,
+}
+
+impl ImageType {
+    /// Create a new image type from a u8.
+    pub(crate) fn new(img_type: u8) -> ImageType {
+        match img_type {
+            0 => ImageType::NoImageData,
+
+            1 => ImageType::RawColorMap,
+            2 => ImageType::RawTrueColor,
+            3 => ImageType::RawGrayScale,
+
+            9 => ImageType::RunColorMap,
+            10 => ImageType::RunTrueColor,
+            11 => ImageType::RunGrayScale,
+
+            _ => ImageType::Unknown,
+        }
+    }
+
+    /// Check if the image format uses colors as opposed to gray scale.
+    pub(crate) fn is_color(&self) -> bool {
+        matches! { *self,
+            ImageType::RawColorMap
+            | ImageType::RawTrueColor
+            | ImageType::RunTrueColor
+            | ImageType::RunColorMap
+        }
+    }
+
+    /// Does the image use a color map.
+    pub(crate) fn is_color_mapped(&self) -> bool {
+        matches! { *self, ImageType::RawColorMap | ImageType::RunColorMap }
+    }
+
+    /// Is the image run length encoded.
+    pub(crate) fn is_encoded(&self) -> bool {
+        matches! {*self, ImageType::RunColorMap | ImageType::RunTrueColor | ImageType::RunGrayScale }
+    }
+}
+
+/// Header used by TGA image files.
+#[derive(Debug, Default)]
+pub(crate) struct Header {
+    pub(crate) id_length: u8,      // length of ID string
+    pub(crate) map_type: u8,       // color map type
+    pub(crate) image_type: u8,     // image type code
+    pub(crate) map_origin: u16,    // starting index of map
+    pub(crate) map_length: u16,    // length of map
+    pub(crate) map_entry_size: u8, // size of map entries in bits
+    pub(crate) x_origin: u16,      // x-origin of image
+    pub(crate) y_origin: u16,      // y-origin of image
+    pub(crate) image_width: u16,   // width of image
+    pub(crate) image_height: u16,  // height of image
+    pub(crate) pixel_depth: u8,    // bits per pixel
+    pub(crate) image_desc: u8,     // image descriptor
+}
+
+impl Header {
+    /// Load the header with values from pixel information.
+    pub(crate) fn from_pixel_info(
+        color_type: ColorType,
+        width: u16,
+        height: u16,
+    ) -> ImageResult<Self> {
+        let mut header = Self::default();
+
+        if width > 0 && height > 0 {
+            let (num_alpha_bits, other_channel_bits, image_type) = match color_type {
+                ColorType::Rgba8 => (8, 24, ImageType::RawTrueColor),
+                ColorType::Rgb8 => (0, 24, ImageType::RawTrueColor),
+                ColorType::La8 => (8, 8, ImageType::RawGrayScale),
+                ColorType::L8 => (0, 8, ImageType::RawGrayScale),
+                _ => {
+                    return Err(ImageError::Unsupported(
+                        UnsupportedError::from_format_and_kind(
+                            ImageFormat::Tga.into(),
+                            UnsupportedErrorKind::Color(color_type.into()),
+                        ),
+                    ))
+                }
+            };
+
+            header.image_type = image_type as u8;
+            header.image_width = width;
+            header.image_height = height;
+            header.pixel_depth = num_alpha_bits + other_channel_bits;
+            header.image_desc = num_alpha_bits & ALPHA_BIT_MASK;
+            header.image_desc |= SCREEN_ORIGIN_BIT_MASK; // Upper left origin.
+        }
+
+        Ok(header)
+    }
+
+    /// Load the header with values from the reader.
+    pub(crate) fn from_reader(r: &mut dyn Read) -> ImageResult<Self> {
+        Ok(Self {
+            id_length: r.read_u8()?,
+            map_type: r.read_u8()?,
+            image_type: r.read_u8()?,
+            map_origin: r.read_u16::<LittleEndian>()?,
+            map_length: r.read_u16::<LittleEndian>()?,
+            map_entry_size: r.read_u8()?,
+            x_origin: r.read_u16::<LittleEndian>()?,
+            y_origin: r.read_u16::<LittleEndian>()?,
+            image_width: r.read_u16::<LittleEndian>()?,
+            image_height: r.read_u16::<LittleEndian>()?,
+            pixel_depth: r.read_u8()?,
+            image_desc: r.read_u8()?,
+        })
+    }
+
+    /// Write out the header values.
+    pub(crate) fn write_to(&self, w: &mut dyn Write) -> ImageResult<()> {
+        w.write_u8(self.id_length)?;
+        w.write_u8(self.map_type)?;
+        w.write_u8(self.image_type)?;
+        w.write_u16::<LittleEndian>(self.map_origin)?;
+        w.write_u16::<LittleEndian>(self.map_length)?;
+        w.write_u8(self.map_entry_size)?;
+        w.write_u16::<LittleEndian>(self.x_origin)?;
+        w.write_u16::<LittleEndian>(self.y_origin)?;
+        w.write_u16::<LittleEndian>(self.image_width)?;
+        w.write_u16::<LittleEndian>(self.image_height)?;
+        w.write_u8(self.pixel_depth)?;
+        w.write_u8(self.image_desc)?;
+        Ok(())
+    }
+}

vendor/image/src/codecs/tga/mod.rs

@@ -0,0 +1,17 @@
+//! Decoding of TGA Images
+//!
+//! # Related Links
+//! <http://googlesites.inequation.org/tgautilities>
+
+/// A decoder for TGA images
+///
+/// Currently this decoder does not support 8, 15 and 16 bit color images.
+pub use self::decoder::TgaDecoder;
+
+//TODO add 8, 15, 16 bit color support
+
+pub use self::encoder::TgaEncoder;
+
+mod decoder;
+mod encoder;
+mod header;

vendor/image/src/codecs/tiff.rs

@@ -0,0 +1,353 @@
+//! Decoding and Encoding of TIFF Images
+//!
+//! TIFF (Tagged Image File Format) is a versatile image format that supports
+//! lossless and lossy compression.
+//!
+//! # Related Links
+//! * <http://partners.adobe.com/public/developer/tiff/index.html> - The TIFF specification
+
+extern crate tiff;
+
+use std::convert::TryFrom;
+use std::io::{self, Cursor, Read, Seek, Write};
+use std::marker::PhantomData;
+use std::mem;
+
+use crate::color::{ColorType, ExtendedColorType};
+use crate::error::{
+    DecodingError, EncodingError, ImageError, ImageResult, LimitError, LimitErrorKind,
+    ParameterError, ParameterErrorKind, UnsupportedError, UnsupportedErrorKind,
+};
+use crate::image::{ImageDecoder, ImageEncoder, ImageFormat};
+use crate::utils;
+
+/// Decoder for TIFF images.
+pub struct TiffDecoder<R>
+where
+    R: Read + Seek,
+{
+    dimensions: (u32, u32),
+    color_type: ColorType,
+
+    // We only use an Option here so we can call with_limits on the decoder without moving.
+    inner: Option<tiff::decoder::Decoder<R>>,
+}
+
+impl<R> TiffDecoder<R>
+where
+    R: Read + Seek,
+{
+    /// Create a new TiffDecoder.
+    pub fn new(r: R) -> Result<TiffDecoder<R>, ImageError> {
+        let mut inner = tiff::decoder::Decoder::new(r).map_err(ImageError::from_tiff_decode)?;
+
+        let dimensions = inner.dimensions().map_err(ImageError::from_tiff_decode)?;
+        let color_type = inner.colortype().map_err(ImageError::from_tiff_decode)?;
+        match inner.find_tag_unsigned_vec::<u16>(tiff::tags::Tag::SampleFormat) {
+            Ok(Some(sample_formats)) => {
+                for format in sample_formats {
+                    check_sample_format(format)?;
+                }
+            }
+            Ok(None) => { /* assume UInt format */ }
+            Err(other) => return Err(ImageError::from_tiff_decode(other)),
+        };
+
+        let color_type = match color_type {
+            tiff::ColorType::Gray(8) => ColorType::L8,
+            tiff::ColorType::Gray(16) => ColorType::L16,
+            tiff::ColorType::GrayA(8) => ColorType::La8,
+            tiff::ColorType::GrayA(16) => ColorType::La16,
+            tiff::ColorType::RGB(8) => ColorType::Rgb8,
+            tiff::ColorType::RGB(16) => ColorType::Rgb16,
+            tiff::ColorType::RGBA(8) => ColorType::Rgba8,
+            tiff::ColorType::RGBA(16) => ColorType::Rgba16,
+
+            tiff::ColorType::Palette(n) | tiff::ColorType::Gray(n) => {
+                return Err(err_unknown_color_type(n))
+            }
+            tiff::ColorType::GrayA(n) => return Err(err_unknown_color_type(n.saturating_mul(2))),
+            tiff::ColorType::RGB(n) => return Err(err_unknown_color_type(n.saturating_mul(3))),
+            tiff::ColorType::YCbCr(n) => return Err(err_unknown_color_type(n.saturating_mul(3))),
+            tiff::ColorType::RGBA(n) | tiff::ColorType::CMYK(n) => {
+                return Err(err_unknown_color_type(n.saturating_mul(4)))
+            }
+        };
+
+        Ok(TiffDecoder {
+            dimensions,
+            color_type,
+            inner: Some(inner),
+        })
+    }
+}
+
+fn check_sample_format(sample_format: u16) -> Result<(), ImageError> {
+    match tiff::tags::SampleFormat::from_u16(sample_format) {
+        Some(tiff::tags::SampleFormat::Uint) => Ok(()),
+        Some(other) => Err(ImageError::Unsupported(
+            UnsupportedError::from_format_and_kind(
+                ImageFormat::Tiff.into(),
+                UnsupportedErrorKind::GenericFeature(format!(
+                    "Unhandled TIFF sample format {:?}",
+                    other
+                )),
+            ),
+        )),
+        None => Err(ImageError::Decoding(DecodingError::from_format_hint(
+            ImageFormat::Tiff.into(),
+        ))),
+    }
+}
+
+fn err_unknown_color_type(value: u8) -> ImageError {
+    ImageError::Unsupported(UnsupportedError::from_format_and_kind(
+        ImageFormat::Tiff.into(),
+        UnsupportedErrorKind::Color(ExtendedColorType::Unknown(value)),
+    ))
+}
+
+impl ImageError {
+    fn from_tiff_decode(err: tiff::TiffError) -> ImageError {
+        match err {
+            tiff::TiffError::IoError(err) => ImageError::IoError(err),
+            err @ tiff::TiffError::FormatError(_)
+            | err @ tiff::TiffError::IntSizeError
+            | err @ tiff::TiffError::UsageError(_) => {
+                ImageError::Decoding(DecodingError::new(ImageFormat::Tiff.into(), err))
+            }
+            tiff::TiffError::UnsupportedError(desc) => {
+                ImageError::Unsupported(UnsupportedError::from_format_and_kind(
+                    ImageFormat::Tiff.into(),
+                    UnsupportedErrorKind::GenericFeature(desc.to_string()),
+                ))
+            }
+            tiff::TiffError::LimitsExceeded => {
+                ImageError::Limits(LimitError::from_kind(LimitErrorKind::InsufficientMemory))
+            }
+        }
+    }
+
+    fn from_tiff_encode(err: tiff::TiffError) -> ImageError {
+        match err {
+            tiff::TiffError::IoError(err) => ImageError::IoError(err),
+            err @ tiff::TiffError::FormatError(_)
+            | err @ tiff::TiffError::IntSizeError
+            | err @ tiff::TiffError::UsageError(_) => {
+                ImageError::Encoding(EncodingError::new(ImageFormat::Tiff.into(), err))
+            }
+            tiff::TiffError::UnsupportedError(desc) => {
+                ImageError::Unsupported(UnsupportedError::from_format_and_kind(
+                    ImageFormat::Tiff.into(),
+                    UnsupportedErrorKind::GenericFeature(desc.to_string()),
+                ))
+            }
+            tiff::TiffError::LimitsExceeded => {
+                ImageError::Limits(LimitError::from_kind(LimitErrorKind::InsufficientMemory))
+            }
+        }
+    }
+}
+
+/// Wrapper struct around a `Cursor<Vec<u8>>`
+pub struct TiffReader<R>(Cursor<Vec<u8>>, PhantomData<R>);
+impl<R> Read for TiffReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        self.0.read(buf)
+    }
+    fn read_to_end(&mut self, buf: &mut Vec<u8>) -> io::Result<usize> {
+        if self.0.position() == 0 && buf.is_empty() {
+            mem::swap(buf, self.0.get_mut());
+            Ok(buf.len())
+        } else {
+            self.0.read_to_end(buf)
+        }
+    }
+}
+
+impl<'a, R: 'a + Read + Seek> ImageDecoder<'a> for TiffDecoder<R> {
+    type Reader = TiffReader<R>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        self.dimensions
+    }
+
+    fn color_type(&self) -> ColorType {
+        self.color_type
+    }
+
+    fn icc_profile(&mut self) -> Option<Vec<u8>> {
+        if let Some(decoder) = &mut self.inner {
+            decoder.get_tag_u8_vec(tiff::tags::Tag::Unknown(34675)).ok()
+        } else {
+            None
+        }
+    }
+
+    fn set_limits(&mut self, limits: crate::io::Limits) -> ImageResult<()> {
+        limits.check_support(&crate::io::LimitSupport::default())?;
+
+        let (width, height) = self.dimensions();
+        limits.check_dimensions(width, height)?;
+
+        let max_alloc = limits.max_alloc.unwrap_or(u64::MAX);
+        let max_intermediate_alloc = max_alloc.saturating_sub(self.total_bytes());
+
+        let mut tiff_limits: tiff::decoder::Limits = Default::default();
+        tiff_limits.decoding_buffer_size =
+            usize::try_from(max_alloc - max_intermediate_alloc).unwrap_or(usize::MAX);
+        tiff_limits.intermediate_buffer_size =
+            usize::try_from(max_intermediate_alloc).unwrap_or(usize::MAX);
+        tiff_limits.ifd_value_size = tiff_limits.intermediate_buffer_size;
+        self.inner = Some(self.inner.take().unwrap().with_limits(tiff_limits));
+
+        Ok(())
+    }
+
+    fn into_reader(self) -> ImageResult<Self::Reader> {
+        let buf = match self
+            .inner
+            .unwrap()
+            .read_image()
+            .map_err(ImageError::from_tiff_decode)?
+        {
+            tiff::decoder::DecodingResult::U8(v) => v,
+            tiff::decoder::DecodingResult::U16(v) => utils::vec_copy_to_u8(&v),
+            tiff::decoder::DecodingResult::U32(v) => utils::vec_copy_to_u8(&v),
+            tiff::decoder::DecodingResult::U64(v) => utils::vec_copy_to_u8(&v),
+            tiff::decoder::DecodingResult::I8(v) => utils::vec_copy_to_u8(&v),
+            tiff::decoder::DecodingResult::I16(v) => utils::vec_copy_to_u8(&v),
+            tiff::decoder::DecodingResult::I32(v) => utils::vec_copy_to_u8(&v),
+            tiff::decoder::DecodingResult::I64(v) => utils::vec_copy_to_u8(&v),
+            tiff::decoder::DecodingResult::F32(v) => utils::vec_copy_to_u8(&v),
+            tiff::decoder::DecodingResult::F64(v) => utils::vec_copy_to_u8(&v),
+        };
+
+        Ok(TiffReader(Cursor::new(buf), PhantomData))
+    }
+
+    fn read_image(self, buf: &mut [u8]) -> ImageResult<()> {
+        assert_eq!(u64::try_from(buf.len()), Ok(self.total_bytes()));
+        match self
+            .inner
+            .unwrap()
+            .read_image()
+            .map_err(ImageError::from_tiff_decode)?
+        {
+            tiff::decoder::DecodingResult::U8(v) => {
+                buf.copy_from_slice(&v);
+            }
+            tiff::decoder::DecodingResult::U16(v) => {
+                buf.copy_from_slice(bytemuck::cast_slice(&v));
+            }
+            tiff::decoder::DecodingResult::U32(v) => {
+                buf.copy_from_slice(bytemuck::cast_slice(&v));
+            }
+            tiff::decoder::DecodingResult::U64(v) => {
+                buf.copy_from_slice(bytemuck::cast_slice(&v));
+            }
+            tiff::decoder::DecodingResult::I8(v) => {
+                buf.copy_from_slice(bytemuck::cast_slice(&v));
+            }
+            tiff::decoder::DecodingResult::I16(v) => {
+                buf.copy_from_slice(bytemuck::cast_slice(&v));
+            }
+            tiff::decoder::DecodingResult::I32(v) => {
+                buf.copy_from_slice(bytemuck::cast_slice(&v));
+            }
+            tiff::decoder::DecodingResult::I64(v) => {
+                buf.copy_from_slice(bytemuck::cast_slice(&v));
+            }
+            tiff::decoder::DecodingResult::F32(v) => {
+                buf.copy_from_slice(bytemuck::cast_slice(&v));
+            }
+            tiff::decoder::DecodingResult::F64(v) => {
+                buf.copy_from_slice(bytemuck::cast_slice(&v));
+            }
+        }
+        Ok(())
+    }
+}
+
+/// Encoder for tiff images
+pub struct TiffEncoder<W> {
+    w: W,
+}
+
+// Utility to simplify and deduplicate error handling during 16-bit encoding.
+fn u8_slice_as_u16(buf: &[u8]) -> ImageResult<&[u16]> {
+    bytemuck::try_cast_slice(buf).map_err(|err| {
+        // If the buffer is not aligned or the correct length for a u16 slice, err.
+        //
+        // `bytemuck::PodCastError` of bytemuck-1.2.0 does not implement
+        // `Error` and `Display` trait.
+        // See <https://github.com/Lokathor/bytemuck/issues/22>.
+        ImageError::Parameter(ParameterError::from_kind(ParameterErrorKind::Generic(
+            format!("{:?}", err),
+        )))
+    })
+}
+
+impl<W: Write + Seek> TiffEncoder<W> {
+    /// Create a new encoder that writes its output to `w`
+    pub fn new(w: W) -> TiffEncoder<W> {
+        TiffEncoder { w }
+    }
+
+    /// Encodes the image `image` that has dimensions `width` and `height` and `ColorType` `c`.
+    ///
+    /// 16-bit types assume the buffer is native endian.
+    pub fn encode(self, data: &[u8], width: u32, height: u32, color: ColorType) -> ImageResult<()> {
+        let mut encoder =
+            tiff::encoder::TiffEncoder::new(self.w).map_err(ImageError::from_tiff_encode)?;
+        match color {
+            ColorType::L8 => {
+                encoder.write_image::<tiff::encoder::colortype::Gray8>(width, height, data)
+            }
+            ColorType::Rgb8 => {
+                encoder.write_image::<tiff::encoder::colortype::RGB8>(width, height, data)
+            }
+            ColorType::Rgba8 => {
+                encoder.write_image::<tiff::encoder::colortype::RGBA8>(width, height, data)
+            }
+            ColorType::L16 => encoder.write_image::<tiff::encoder::colortype::Gray16>(
+                width,
+                height,
+                u8_slice_as_u16(data)?,
+            ),
+            ColorType::Rgb16 => encoder.write_image::<tiff::encoder::colortype::RGB16>(
+                width,
+                height,
+                u8_slice_as_u16(data)?,
+            ),
+            ColorType::Rgba16 => encoder.write_image::<tiff::encoder::colortype::RGBA16>(
+                width,
+                height,
+                u8_slice_as_u16(data)?,
+            ),
+            _ => {
+                return Err(ImageError::Unsupported(
+                    UnsupportedError::from_format_and_kind(
+                        ImageFormat::Tiff.into(),
+                        UnsupportedErrorKind::Color(color.into()),
+                    ),
+                ))
+            }
+        }
+        .map_err(ImageError::from_tiff_encode)?;
+
+        Ok(())
+    }
+}
+
+impl<W: Write + Seek> ImageEncoder for TiffEncoder<W> {
+    fn write_image(
+        self,
+        buf: &[u8],
+        width: u32,
+        height: u32,
+        color_type: ColorType,
+    ) -> ImageResult<()> {
+        self.encode(buf, width, height, color_type)
+    }
+}

vendor/image/src/codecs/webp/decoder.rs

@@ -0,0 +1,399 @@
+use byteorder::{LittleEndian, ReadBytesExt};
+use std::convert::TryFrom;
+use std::io::{self, Cursor, Error, Read};
+use std::marker::PhantomData;
+use std::{error, fmt, mem};
+
+use crate::error::{DecodingError, ImageError, ImageResult, ParameterError, ParameterErrorKind};
+use crate::image::{ImageDecoder, ImageFormat};
+use crate::{color, AnimationDecoder, Frames, Rgba};
+
+use super::lossless::{LosslessDecoder, LosslessFrame};
+use super::vp8::{Frame as VP8Frame, Vp8Decoder};
+
+use super::extended::{read_extended_header, ExtendedImage};
+
+/// All errors that can occur when attempting to parse a WEBP container
+#[derive(Debug, Clone, Copy)]
+pub(crate) enum DecoderError {
+    /// RIFF's "RIFF" signature not found or invalid
+    RiffSignatureInvalid([u8; 4]),
+    /// WebP's "WEBP" signature not found or invalid
+    WebpSignatureInvalid([u8; 4]),
+    /// Chunk Header was incorrect or invalid in its usage
+    ChunkHeaderInvalid([u8; 4]),
+}
+
+impl fmt::Display for DecoderError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        struct SignatureWriter([u8; 4]);
+        impl fmt::Display for SignatureWriter {
+            fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+                write!(
+                    f,
+                    "[{:#04X?}, {:#04X?}, {:#04X?}, {:#04X?}]",
+                    self.0[0], self.0[1], self.0[2], self.0[3]
+                )
+            }
+        }
+
+        match self {
+            DecoderError::RiffSignatureInvalid(riff) => f.write_fmt(format_args!(
+                "Invalid RIFF signature: {}",
+                SignatureWriter(*riff)
+            )),
+            DecoderError::WebpSignatureInvalid(webp) => f.write_fmt(format_args!(
+                "Invalid WebP signature: {}",
+                SignatureWriter(*webp)
+            )),
+            DecoderError::ChunkHeaderInvalid(header) => f.write_fmt(format_args!(
+                "Invalid Chunk header: {}",
+                SignatureWriter(*header)
+            )),
+        }
+    }
+}
+
+impl From<DecoderError> for ImageError {
+    fn from(e: DecoderError) -> ImageError {
+        ImageError::Decoding(DecodingError::new(ImageFormat::WebP.into(), e))
+    }
+}
+
+impl error::Error for DecoderError {}
+
+/// All possible RIFF chunks in a WebP image file
+#[allow(clippy::upper_case_acronyms)]
+#[derive(Debug, Clone, Copy, PartialEq)]
+pub(crate) enum WebPRiffChunk {
+    RIFF,
+    WEBP,
+    VP8,
+    VP8L,
+    VP8X,
+    ANIM,
+    ANMF,
+    ALPH,
+    ICCP,
+    EXIF,
+    XMP,
+}
+
+impl WebPRiffChunk {
+    pub(crate) fn from_fourcc(chunk_fourcc: [u8; 4]) -> ImageResult<Self> {
+        match &chunk_fourcc {
+            b"RIFF" => Ok(Self::RIFF),
+            b"WEBP" => Ok(Self::WEBP),
+            b"VP8 " => Ok(Self::VP8),
+            b"VP8L" => Ok(Self::VP8L),
+            b"VP8X" => Ok(Self::VP8X),
+            b"ANIM" => Ok(Self::ANIM),
+            b"ANMF" => Ok(Self::ANMF),
+            b"ALPH" => Ok(Self::ALPH),
+            b"ICCP" => Ok(Self::ICCP),
+            b"EXIF" => Ok(Self::EXIF),
+            b"XMP " => Ok(Self::XMP),
+            _ => Err(DecoderError::ChunkHeaderInvalid(chunk_fourcc).into()),
+        }
+    }
+
+    pub(crate) fn to_fourcc(&self) -> [u8; 4] {
+        match self {
+            Self::RIFF => *b"RIFF",
+            Self::WEBP => *b"WEBP",
+            Self::VP8 => *b"VP8 ",
+            Self::VP8L => *b"VP8L",
+            Self::VP8X => *b"VP8X",
+            Self::ANIM => *b"ANIM",
+            Self::ANMF => *b"ANMF",
+            Self::ALPH => *b"ALPH",
+            Self::ICCP => *b"ICCP",
+            Self::EXIF => *b"EXIF",
+            Self::XMP => *b"XMP ",
+        }
+    }
+}
+
+enum WebPImage {
+    Lossy(VP8Frame),
+    Lossless(LosslessFrame),
+    Extended(ExtendedImage),
+}
+
+/// WebP Image format decoder. Currently only supports lossy RGB images or lossless RGBA images.
+pub struct WebPDecoder<R> {
+    r: R,
+    image: WebPImage,
+}
+
+impl<R: Read> WebPDecoder<R> {
+    /// Create a new WebPDecoder from the Reader ```r```.
+    /// This function takes ownership of the Reader.
+    pub fn new(r: R) -> ImageResult<WebPDecoder<R>> {
+        let image = WebPImage::Lossy(Default::default());
+
+        let mut decoder = WebPDecoder { r, image };
+        decoder.read_data()?;
+        Ok(decoder)
+    }
+
+    //reads the 12 bytes of the WebP file header
+    fn read_riff_header(&mut self) -> ImageResult<u32> {
+        let mut riff = [0; 4];
+        self.r.read_exact(&mut riff)?;
+        if &riff != b"RIFF" {
+            return Err(DecoderError::RiffSignatureInvalid(riff).into());
+        }
+
+        let size = self.r.read_u32::<LittleEndian>()?;
+
+        let mut webp = [0; 4];
+        self.r.read_exact(&mut webp)?;
+        if &webp != b"WEBP" {
+            return Err(DecoderError::WebpSignatureInvalid(webp).into());
+        }
+
+        Ok(size)
+    }
+
+    //reads the chunk header, decodes the frame and returns the inner decoder
+    fn read_frame(&mut self) -> ImageResult<WebPImage> {
+        let chunk = read_chunk(&mut self.r)?;
+
+        match chunk {
+            Some((cursor, WebPRiffChunk::VP8)) => {
+                let mut vp8_decoder = Vp8Decoder::new(cursor);
+                let frame = vp8_decoder.decode_frame()?;
+
+                Ok(WebPImage::Lossy(frame.clone()))
+            }
+            Some((cursor, WebPRiffChunk::VP8L)) => {
+                let mut lossless_decoder = LosslessDecoder::new(cursor);
+                let frame = lossless_decoder.decode_frame()?;
+
+                Ok(WebPImage::Lossless(frame.clone()))
+            }
+            Some((mut cursor, WebPRiffChunk::VP8X)) => {
+                let info = read_extended_header(&mut cursor)?;
+
+                let image = ExtendedImage::read_extended_chunks(&mut self.r, info)?;
+
+                Ok(WebPImage::Extended(image))
+            }
+            None => Err(ImageError::IoError(Error::from(
+                io::ErrorKind::UnexpectedEof,
+            ))),
+            Some((_, chunk)) => Err(DecoderError::ChunkHeaderInvalid(chunk.to_fourcc()).into()),
+        }
+    }
+
+    fn read_data(&mut self) -> ImageResult<()> {
+        let _size = self.read_riff_header()?;
+
+        let image = self.read_frame()?;
+
+        self.image = image;
+
+        Ok(())
+    }
+
+    /// Returns true if the image as described by the bitstream is animated.
+    pub fn has_animation(&self) -> bool {
+        match &self.image {
+            WebPImage::Lossy(_) => false,
+            WebPImage::Lossless(_) => false,
+            WebPImage::Extended(extended) => extended.has_animation(),
+        }
+    }
+
+    /// Sets the background color if the image is an extended and animated webp.
+    pub fn set_background_color(&mut self, color: Rgba<u8>) -> ImageResult<()> {
+        match &mut self.image {
+            WebPImage::Extended(image) => image.set_background_color(color),
+            _ => Err(ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::Generic(
+                    "Background color can only be set on animated webp".to_owned(),
+                ),
+            ))),
+        }
+    }
+}
+
+pub(crate) fn read_len_cursor<R>(r: &mut R) -> ImageResult<Cursor<Vec<u8>>>
+where
+    R: Read,
+{
+    let unpadded_len = u64::from(r.read_u32::<LittleEndian>()?);
+
+    // RIFF chunks containing an uneven number of bytes append
+    // an extra 0x00 at the end of the chunk
+    //
+    // The addition cannot overflow since we have a u64 that was created from a u32
+    let len = unpadded_len + (unpadded_len % 2);
+
+    let mut framedata = Vec::new();
+    r.by_ref().take(len).read_to_end(&mut framedata)?;
+
+    //remove padding byte
+    if unpadded_len % 2 == 1 {
+        framedata.pop();
+    }
+
+    Ok(io::Cursor::new(framedata))
+}
+
+/// Reads a chunk header FourCC
+/// Returns None if and only if we hit end of file reading the four character code of the chunk
+/// The inner error is `Err` if and only if the chunk header FourCC is present but unknown
+pub(crate) fn read_fourcc<R: Read>(r: &mut R) -> ImageResult<Option<ImageResult<WebPRiffChunk>>> {
+    let mut chunk_fourcc = [0; 4];
+    let result = r.read_exact(&mut chunk_fourcc);
+
+    match result {
+        Ok(()) => {}
+        Err(err) => {
+            if err.kind() == io::ErrorKind::UnexpectedEof {
+                return Ok(None);
+            } else {
+                return Err(err.into());
+            }
+        }
+    }
+
+    let chunk = WebPRiffChunk::from_fourcc(chunk_fourcc);
+    Ok(Some(chunk))
+}
+
+/// Reads a chunk
+/// Returns an error if the chunk header is not a valid webp header or some other reading error
+/// Returns None if and only if we hit end of file reading the four character code of the chunk
+pub(crate) fn read_chunk<R>(r: &mut R) -> ImageResult<Option<(Cursor<Vec<u8>>, WebPRiffChunk)>>
+where
+    R: Read,
+{
+    if let Some(chunk) = read_fourcc(r)? {
+        let chunk = chunk?;
+        let cursor = read_len_cursor(r)?;
+        Ok(Some((cursor, chunk)))
+    } else {
+        Ok(None)
+    }
+}
+
+/// Wrapper struct around a `Cursor<Vec<u8>>`
+pub struct WebpReader<R>(Cursor<Vec<u8>>, PhantomData<R>);
+impl<R> Read for WebpReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        self.0.read(buf)
+    }
+    fn read_to_end(&mut self, buf: &mut Vec<u8>) -> io::Result<usize> {
+        if self.0.position() == 0 && buf.is_empty() {
+            mem::swap(buf, self.0.get_mut());
+            Ok(buf.len())
+        } else {
+            self.0.read_to_end(buf)
+        }
+    }
+}
+
+impl<'a, R: 'a + Read> ImageDecoder<'a> for WebPDecoder<R> {
+    type Reader = WebpReader<R>;
+
+    fn dimensions(&self) -> (u32, u32) {
+        match &self.image {
+            WebPImage::Lossy(vp8_frame) => {
+                (u32::from(vp8_frame.width), u32::from(vp8_frame.height))
+            }
+            WebPImage::Lossless(lossless_frame) => (
+                u32::from(lossless_frame.width),
+                u32::from(lossless_frame.height),
+            ),
+            WebPImage::Extended(extended) => extended.dimensions(),
+        }
+    }
+
+    fn color_type(&self) -> color::ColorType {
+        match &self.image {
+            WebPImage::Lossy(_) => color::ColorType::Rgb8,
+            WebPImage::Lossless(_) => color::ColorType::Rgba8,
+            WebPImage::Extended(extended) => extended.color_type(),
+        }
+    }
+
+    fn into_reader(self) -> ImageResult<Self::Reader> {
+        match &self.image {
+            WebPImage::Lossy(vp8_frame) => {
+                let mut data = vec![0; vp8_frame.get_buf_size()];
+                vp8_frame.fill_rgb(data.as_mut_slice());
+                Ok(WebpReader(Cursor::new(data), PhantomData))
+            }
+            WebPImage::Lossless(lossless_frame) => {
+                let mut data = vec![0; lossless_frame.get_buf_size()];
+                lossless_frame.fill_rgba(data.as_mut_slice());
+                Ok(WebpReader(Cursor::new(data), PhantomData))
+            }
+            WebPImage::Extended(extended) => {
+                let mut data = vec![0; extended.get_buf_size()];
+                extended.fill_buf(data.as_mut_slice());
+                Ok(WebpReader(Cursor::new(data), PhantomData))
+            }
+        }
+    }
+
+    fn read_image(self, buf: &mut [u8]) -> ImageResult<()> {
+        assert_eq!(u64::try_from(buf.len()), Ok(self.total_bytes()));
+
+        match &self.image {
+            WebPImage::Lossy(vp8_frame) => {
+                vp8_frame.fill_rgb(buf);
+            }
+            WebPImage::Lossless(lossless_frame) => {
+                lossless_frame.fill_rgba(buf);
+            }
+            WebPImage::Extended(extended) => {
+                extended.fill_buf(buf);
+            }
+        }
+        Ok(())
+    }
+
+    fn icc_profile(&mut self) -> Option<Vec<u8>> {
+        if let WebPImage::Extended(extended) = &self.image {
+            extended.icc_profile()
+        } else {
+            None
+        }
+    }
+}
+
+impl<'a, R: 'a + Read> AnimationDecoder<'a> for WebPDecoder<R> {
+    fn into_frames(self) -> Frames<'a> {
+        match self.image {
+            WebPImage::Lossy(_) | WebPImage::Lossless(_) => {
+                Frames::new(Box::new(std::iter::empty()))
+            }
+            WebPImage::Extended(extended_image) => extended_image.into_frames(),
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn add_with_overflow_size() {
+        let bytes = vec![
+            0x52, 0x49, 0x46, 0x46, 0xaf, 0x37, 0x80, 0x47, 0x57, 0x45, 0x42, 0x50, 0x6c, 0x64,
+            0x00, 0x00, 0xff, 0xff, 0xff, 0xff, 0xfb, 0x7e, 0x73, 0x00, 0x06, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x65, 0x65, 0x65, 0x65, 0x65, 0x65,
+            0x40, 0xfb, 0xff, 0xff, 0x65, 0x65, 0x65, 0x65, 0x65, 0x65, 0x65, 0x65, 0x65, 0x65,
+            0x00, 0x00, 0x00, 0x00, 0x62, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x49,
+            0x49, 0x54, 0x55, 0x50, 0x4c, 0x54, 0x59, 0x50, 0x45, 0x33, 0x37, 0x44, 0x4d, 0x46,
+        ];
+
+        let data = std::io::Cursor::new(bytes);
+
+        let _ = WebPDecoder::new(data);
+    }
+}

vendor/image/src/codecs/webp/encoder.rs

@@ -0,0 +1,242 @@
+//! Encoding of WebP images.
+///
+/// Uses the simple encoding API from the [libwebp] library.
+///
+/// [libwebp]: https://developers.google.com/speed/webp/docs/api#simple_encoding_api
+use std::io::Write;
+
+use libwebp::{Encoder, PixelLayout, WebPMemory};
+
+use crate::error::{
+    EncodingError, ParameterError, ParameterErrorKind, UnsupportedError, UnsupportedErrorKind,
+};
+use crate::flat::SampleLayout;
+use crate::{ColorType, ImageEncoder, ImageError, ImageFormat, ImageResult};
+
+/// WebP Encoder.
+pub struct WebPEncoder<W> {
+    inner: W,
+    quality: WebPQuality,
+}
+
+/// WebP encoder quality.
+#[derive(Debug, Copy, Clone)]
+pub struct WebPQuality(Quality);
+
+#[derive(Debug, Copy, Clone)]
+enum Quality {
+    Lossless,
+    Lossy(u8),
+}
+
+impl WebPQuality {
+    /// Minimum lossy quality value (0).
+    pub const MIN: u8 = 0;
+    /// Maximum lossy quality value (100).
+    pub const MAX: u8 = 100;
+    /// Default lossy quality (80), providing a balance of quality and file size.
+    pub const DEFAULT: u8 = 80;
+
+    /// Lossless encoding.
+    pub fn lossless() -> Self {
+        Self(Quality::Lossless)
+    }
+
+    /// Lossy encoding. 0 = low quality, small size; 100 = high quality, large size.
+    ///
+    /// Values are clamped from 0 to 100.
+    pub fn lossy(quality: u8) -> Self {
+        Self(Quality::Lossy(quality.clamp(Self::MIN, Self::MAX)))
+    }
+}
+
+impl Default for WebPQuality {
+    fn default() -> Self {
+        Self::lossy(WebPQuality::DEFAULT)
+    }
+}
+
+impl<W: Write> WebPEncoder<W> {
+    /// Create a new encoder that writes its output to `w`.
+    ///
+    /// Defaults to lossy encoding, see [`WebPQuality::DEFAULT`].
+    pub fn new(w: W) -> Self {
+        WebPEncoder::new_with_quality(w, WebPQuality::default())
+    }
+
+    /// Create a new encoder with the specified quality, that writes its output to `w`.
+    pub fn new_with_quality(w: W, quality: WebPQuality) -> Self {
+        Self { inner: w, quality }
+    }
+
+    /// Encode image data with the indicated color type.
+    ///
+    /// The encoder requires image data be Rgb8 or Rgba8.
+    pub fn encode(
+        mut self,
+        data: &[u8],
+        width: u32,
+        height: u32,
+        color: ColorType,
+    ) -> ImageResult<()> {
+        // TODO: convert color types internally?
+        let layout = match color {
+            ColorType::Rgb8 => PixelLayout::Rgb,
+            ColorType::Rgba8 => PixelLayout::Rgba,
+            _ => {
+                return Err(ImageError::Unsupported(
+                    UnsupportedError::from_format_and_kind(
+                        ImageFormat::WebP.into(),
+                        UnsupportedErrorKind::Color(color.into()),
+                    ),
+                ))
+            }
+        };
+
+        // Validate dimensions upfront to avoid panics.
+        if width == 0
+            || height == 0
+            || !SampleLayout::row_major_packed(color.channel_count(), width, height)
+                .fits(data.len())
+        {
+            return Err(ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::DimensionMismatch,
+            )));
+        }
+
+        // Call the native libwebp library to encode the image.
+        let encoder = Encoder::new(data, layout, width, height);
+        let encoded: WebPMemory = match self.quality.0 {
+            Quality::Lossless => encoder.encode_lossless(),
+            Quality::Lossy(quality) => encoder.encode(quality as f32),
+        };
+
+        // The simple encoding API in libwebp does not return errors.
+        if encoded.is_empty() {
+            return Err(ImageError::Encoding(EncodingError::new(
+                ImageFormat::WebP.into(),
+                "encoding failed, output empty",
+            )));
+        }
+
+        self.inner.write_all(&encoded)?;
+        Ok(())
+    }
+}
+
+impl<W: Write> ImageEncoder for WebPEncoder<W> {
+    fn write_image(
+        self,
+        buf: &[u8],
+        width: u32,
+        height: u32,
+        color_type: ColorType,
+    ) -> ImageResult<()> {
+        self.encode(buf, width, height, color_type)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::codecs::webp::{WebPEncoder, WebPQuality};
+    use crate::{ColorType, ImageEncoder};
+
+    #[test]
+    fn webp_lossless_deterministic() {
+        // 1x1 8-bit image buffer containing a single red pixel.
+        let rgb: &[u8] = &[255, 0, 0];
+        let rgba: &[u8] = &[255, 0, 0, 128];
+        for (color, img, expected) in [
+            (
+                ColorType::Rgb8,
+                rgb,
+                [
+                    82, 73, 70, 70, 28, 0, 0, 0, 87, 69, 66, 80, 86, 80, 56, 76, 15, 0, 0, 0, 47,
+                    0, 0, 0, 0, 7, 16, 253, 143, 254, 7, 34, 162, 255, 1, 0,
+                ],
+            ),
+            (
+                ColorType::Rgba8,
+                rgba,
+                [
+                    82, 73, 70, 70, 28, 0, 0, 0, 87, 69, 66, 80, 86, 80, 56, 76, 15, 0, 0, 0, 47,
+                    0, 0, 0, 16, 7, 16, 253, 143, 2, 6, 34, 162, 255, 1, 0,
+                ],
+            ),
+        ] {
+            // Encode it into a memory buffer.
+            let mut encoded_img = Vec::new();
+            {
+                let encoder =
+                    WebPEncoder::new_with_quality(&mut encoded_img, WebPQuality::lossless());
+                encoder
+                    .write_image(&img, 1, 1, color)
+                    .expect("image encoding failed");
+            }
+
+            // WebP encoding should be deterministic.
+            assert_eq!(encoded_img, expected);
+        }
+    }
+
+    #[derive(Debug, Clone)]
+    struct MockImage {
+        width: u32,
+        height: u32,
+        color: ColorType,
+        data: Vec<u8>,
+    }
+
+    impl quickcheck::Arbitrary for MockImage {
+        fn arbitrary(g: &mut quickcheck::Gen) -> Self {
+            // Limit to small, non-empty images <= 512x512.
+            let width = u32::arbitrary(g) % 512 + 1;
+            let height = u32::arbitrary(g) % 512 + 1;
+            let (color, stride) = if bool::arbitrary(g) {
+                (ColorType::Rgb8, 3)
+            } else {
+                (ColorType::Rgba8, 4)
+            };
+            let size = width * height * stride;
+            let data: Vec<u8> = (0..size).map(|_| u8::arbitrary(g)).collect();
+            MockImage {
+                width,
+                height,
+                color,
+                data,
+            }
+        }
+    }
+
+    quickcheck! {
+        fn fuzz_webp_valid_image(image: MockImage, quality: u8) -> bool {
+            // Check valid images do not panic.
+            let mut buffer = Vec::<u8>::new();
+            for webp_quality in [WebPQuality::lossless(), WebPQuality::lossy(quality)] {
+                buffer.clear();
+                let encoder = WebPEncoder::new_with_quality(&mut buffer, webp_quality);
+                if !encoder
+                    .write_image(&image.data, image.width, image.height, image.color)
+                    .is_ok() {
+                    return false;
+                }
+            }
+            true
+        }
+
+        fn fuzz_webp_no_panic(data: Vec<u8>, width: u8, height: u8, quality: u8) -> bool {
+            // Check random (usually invalid) parameters do not panic.
+            let mut buffer = Vec::<u8>::new();
+            for color in [ColorType::Rgb8, ColorType::Rgba8] {
+                for webp_quality in [WebPQuality::lossless(), WebPQuality::lossy(quality)] {
+                    buffer.clear();
+                    let encoder = WebPEncoder::new_with_quality(&mut buffer, webp_quality);
+                    // Ignore errors.
+                    let _ = encoder
+                        .write_image(&data, width as u32, height as u32, color);
+                }
+            }
+            true
+        }
+    }
+}

vendor/image/src/codecs/webp/extended.rs

@@ -0,0 +1,839 @@
+use std::convert::TryInto;
+use std::io::{self, Cursor, Error, Read};
+use std::{error, fmt};
+
+use super::decoder::{
+    read_chunk, read_fourcc, read_len_cursor, DecoderError::ChunkHeaderInvalid, WebPRiffChunk,
+};
+use super::lossless::{LosslessDecoder, LosslessFrame};
+use super::vp8::{Frame as VP8Frame, Vp8Decoder};
+use crate::error::{DecodingError, ParameterError, ParameterErrorKind};
+use crate::image::ImageFormat;
+use crate::{
+    ColorType, Delay, Frame, Frames, ImageError, ImageResult, Rgb, RgbImage, Rgba, RgbaImage,
+};
+use byteorder::{LittleEndian, ReadBytesExt};
+
+//all errors that can occur while parsing extended chunks in a WebP file
+#[derive(Debug, Clone, Copy)]
+enum DecoderError {
+    // Some bits were invalid
+    InfoBitsInvalid { name: &'static str, value: u32 },
+    // Alpha chunk doesn't match the frame's size
+    AlphaChunkSizeMismatch,
+    // Image is too large, either for the platform's pointer size or generally
+    ImageTooLarge,
+    // Frame would go out of the canvas
+    FrameOutsideImage,
+}
+
+impl fmt::Display for DecoderError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            DecoderError::InfoBitsInvalid { name, value } => f.write_fmt(format_args!(
+                "Info bits `{}` invalid, received value: {}",
+                name, value
+            )),
+            DecoderError::AlphaChunkSizeMismatch => {
+                f.write_str("Alpha chunk doesn't match the size of the frame")
+            }
+            DecoderError::ImageTooLarge => f.write_str("Image is too large to be decoded"),
+            DecoderError::FrameOutsideImage => {
+                f.write_str("Frame is too large and would go outside the image")
+            }
+        }
+    }
+}
+
+impl From<DecoderError> for ImageError {
+    fn from(e: DecoderError) -> ImageError {
+        ImageError::Decoding(DecodingError::new(ImageFormat::WebP.into(), e))
+    }
+}
+
+impl error::Error for DecoderError {}
+
+#[derive(Debug, Clone)]
+pub(crate) struct WebPExtendedInfo {
+    _icc_profile: bool,
+    _alpha: bool,
+    _exif_metadata: bool,
+    _xmp_metadata: bool,
+    _animation: bool,
+    canvas_width: u32,
+    canvas_height: u32,
+    icc_profile: Option<Vec<u8>>,
+}
+
+#[derive(Debug)]
+enum ExtendedImageData {
+    Animation {
+        frames: Vec<AnimatedFrame>,
+        anim_info: WebPAnimatedInfo,
+    },
+    Static(WebPStatic),
+}
+
+#[derive(Debug)]
+pub(crate) struct ExtendedImage {
+    info: WebPExtendedInfo,
+    image: ExtendedImageData,
+}
+
+impl ExtendedImage {
+    pub(crate) fn dimensions(&self) -> (u32, u32) {
+        (self.info.canvas_width, self.info.canvas_height)
+    }
+
+    pub(crate) fn has_animation(&self) -> bool {
+        self.info._animation
+    }
+
+    pub(crate) fn icc_profile(&self) -> Option<Vec<u8>> {
+        self.info.icc_profile.clone()
+    }
+
+    pub(crate) fn color_type(&self) -> ColorType {
+        match &self.image {
+            ExtendedImageData::Animation { frames, .. } => &frames[0].image,
+            ExtendedImageData::Static(image) => image,
+        }
+        .color_type()
+    }
+
+    pub(crate) fn into_frames<'a>(self) -> Frames<'a> {
+        struct FrameIterator {
+            image: ExtendedImage,
+            index: usize,
+            canvas: RgbaImage,
+        }
+
+        impl Iterator for FrameIterator {
+            type Item = ImageResult<Frame>;
+
+            fn next(&mut self) -> Option<Self::Item> {
+                if let ExtendedImageData::Animation { frames, anim_info } = &self.image.image {
+                    let frame = frames.get(self.index);
+                    match frame {
+                        Some(anim_image) => {
+                            self.index += 1;
+                            ExtendedImage::draw_subimage(
+                                &mut self.canvas,
+                                anim_image,
+                                anim_info.background_color,
+                            )
+                        }
+                        None => None,
+                    }
+                } else {
+                    None
+                }
+            }
+        }
+
+        let width = self.info.canvas_width;
+        let height = self.info.canvas_height;
+        let background_color =
+            if let ExtendedImageData::Animation { ref anim_info, .. } = self.image {
+                anim_info.background_color
+            } else {
+                Rgba([0, 0, 0, 0])
+            };
+
+        let frame_iter = FrameIterator {
+            image: self,
+            index: 0,
+            canvas: RgbaImage::from_pixel(width, height, background_color),
+        };
+
+        Frames::new(Box::new(frame_iter))
+    }
+
+    pub(crate) fn read_extended_chunks<R: Read>(
+        reader: &mut R,
+        mut info: WebPExtendedInfo,
+    ) -> ImageResult<ExtendedImage> {
+        let mut anim_info: Option<WebPAnimatedInfo> = None;
+        let mut anim_frames: Vec<AnimatedFrame> = Vec::new();
+        let mut static_frame: Option<WebPStatic> = None;
+        //go until end of file and while chunk headers are valid
+        while let Some((mut cursor, chunk)) = read_extended_chunk(reader)? {
+            match chunk {
+                WebPRiffChunk::EXIF | WebPRiffChunk::XMP => {
+                    //ignore these chunks
+                }
+                WebPRiffChunk::ANIM => {
+                    if anim_info.is_none() {
+                        anim_info = Some(Self::read_anim_info(&mut cursor)?);
+                    }
+                }
+                WebPRiffChunk::ANMF => {
+                    let frame = read_anim_frame(cursor, info.canvas_width, info.canvas_height)?;
+                    anim_frames.push(frame);
+                }
+                WebPRiffChunk::ALPH => {
+                    if static_frame.is_none() {
+                        let alpha_chunk =
+                            read_alpha_chunk(&mut cursor, info.canvas_width, info.canvas_height)?;
+
+                        let vp8_frame = read_lossy_with_chunk(reader)?;
+
+                        let img = WebPStatic::from_alpha_lossy(alpha_chunk, vp8_frame)?;
+
+                        static_frame = Some(img);
+                    }
+                }
+                WebPRiffChunk::ICCP => {
+                    let mut icc_profile = Vec::new();
+                    cursor.read_to_end(&mut icc_profile)?;
+                    info.icc_profile = Some(icc_profile);
+                }
+                WebPRiffChunk::VP8 => {
+                    if static_frame.is_none() {
+                        let vp8_frame = read_lossy(cursor)?;
+
+                        let img = WebPStatic::from_lossy(vp8_frame)?;
+
+                        static_frame = Some(img);
+                    }
+                }
+                WebPRiffChunk::VP8L => {
+                    if static_frame.is_none() {
+                        let mut lossless_decoder = LosslessDecoder::new(cursor);
+                        let frame = lossless_decoder.decode_frame()?;
+                        let image = WebPStatic::Lossless(frame.clone());
+
+                        static_frame = Some(image);
+                    }
+                }
+                _ => return Err(ChunkHeaderInvalid(chunk.to_fourcc()).into()),
+            }
+        }
+
+        let image = if let Some(info) = anim_info {
+            if anim_frames.is_empty() {
+                return Err(ImageError::IoError(Error::from(
+                    io::ErrorKind::UnexpectedEof,
+                )));
+            }
+            ExtendedImageData::Animation {
+                frames: anim_frames,
+                anim_info: info,
+            }
+        } else if let Some(frame) = static_frame {
+            ExtendedImageData::Static(frame)
+        } else {
+            //reached end of file too early before image data was reached
+            return Err(ImageError::IoError(Error::from(
+                io::ErrorKind::UnexpectedEof,
+            )));
+        };
+
+        let image = ExtendedImage { image, info };
+
+        Ok(image)
+    }
+
+    fn read_anim_info<R: Read>(reader: &mut R) -> ImageResult<WebPAnimatedInfo> {
+        let mut colors: [u8; 4] = [0; 4];
+        reader.read_exact(&mut colors)?;
+
+        //background color is [blue, green, red, alpha]
+        let background_color = Rgba([colors[2], colors[1], colors[0], colors[3]]);
+
+        let loop_count = reader.read_u16::<LittleEndian>()?;
+
+        let info = WebPAnimatedInfo {
+            background_color,
+            _loop_count: loop_count,
+        };
+
+        Ok(info)
+    }
+
+    fn draw_subimage(
+        canvas: &mut RgbaImage,
+        anim_image: &AnimatedFrame,
+        background_color: Rgba<u8>,
+    ) -> Option<ImageResult<Frame>> {
+        let mut buffer = vec![0; anim_image.image.get_buf_size()];
+        anim_image.image.fill_buf(&mut buffer);
+        let has_alpha = anim_image.image.has_alpha();
+        let pixel_len: u32 = anim_image.image.color_type().bytes_per_pixel().into();
+
+        'x: for x in 0..anim_image.width {
+            for y in 0..anim_image.height {
+                let canvas_index: (u32, u32) = (x + anim_image.offset_x, y + anim_image.offset_y);
+                // Negative offsets are not possible due to unsigned ints
+                // If we go out of bounds by height, still continue by x
+                if canvas_index.1 >= canvas.height() {
+                    continue 'x;
+                }
+                // If we go out of bounds by width, it doesn't make sense to continue at all
+                if canvas_index.0 >= canvas.width() {
+                    break 'x;
+                }
+                let index: usize = ((y * anim_image.width + x) * pixel_len).try_into().unwrap();
+                canvas[canvas_index] = if anim_image.use_alpha_blending && has_alpha {
+                    let buffer: [u8; 4] = buffer[index..][..4].try_into().unwrap();
+                    ExtendedImage::do_alpha_blending(buffer, canvas[canvas_index])
+                } else {
+                    Rgba([
+                        buffer[index],
+                        buffer[index + 1],
+                        buffer[index + 2],
+                        if has_alpha { buffer[index + 3] } else { 255 },
+                    ])
+                };
+            }
+        }
+
+        let delay = Delay::from_numer_denom_ms(anim_image.duration, 1);
+        let img = canvas.clone();
+        let frame = Frame::from_parts(img, 0, 0, delay);
+
+        if anim_image.dispose {
+            for x in 0..anim_image.width {
+                for y in 0..anim_image.height {
+                    let canvas_index = (x + anim_image.offset_x, y + anim_image.offset_y);
+                    canvas[canvas_index] = background_color;
+                }
+            }
+        }
+
+        Some(Ok(frame))
+    }
+
+    fn do_alpha_blending(buffer: [u8; 4], canvas: Rgba<u8>) -> Rgba<u8> {
+        let canvas_alpha = f64::from(canvas[3]);
+        let buffer_alpha = f64::from(buffer[3]);
+        let blend_alpha_f64 = buffer_alpha + canvas_alpha * (1.0 - buffer_alpha / 255.0);
+        //value should be between 0 and 255, this truncates the fractional part
+        let blend_alpha: u8 = blend_alpha_f64 as u8;
+
+        let blend_rgb: [u8; 3] = if blend_alpha == 0 {
+            [0, 0, 0]
+        } else {
+            let mut rgb = [0u8; 3];
+            for i in 0..3 {
+                let canvas_f64 = f64::from(canvas[i]);
+                let buffer_f64 = f64::from(buffer[i]);
+
+                let val = (buffer_f64 * buffer_alpha
+                    + canvas_f64 * canvas_alpha * (1.0 - buffer_alpha / 255.0))
+                    / blend_alpha_f64;
+                //value should be between 0 and 255, this truncates the fractional part
+                rgb[i] = val as u8;
+            }
+
+            rgb
+        };
+
+        Rgba([blend_rgb[0], blend_rgb[1], blend_rgb[2], blend_alpha])
+    }
+
+    pub(crate) fn fill_buf(&self, buf: &mut [u8]) {
+        match &self.image {
+            // will always have at least one frame
+            ExtendedImageData::Animation { frames, anim_info } => {
+                let first_frame = &frames[0];
+                let (canvas_width, canvas_height) = self.dimensions();
+                if canvas_width == first_frame.width && canvas_height == first_frame.height {
+                    first_frame.image.fill_buf(buf);
+                } else {
+                    let bg_color = match &self.info._alpha {
+                        true => Rgba::from([0, 0, 0, 0]),
+                        false => anim_info.background_color,
+                    };
+                    let mut canvas = RgbaImage::from_pixel(canvas_width, canvas_height, bg_color);
+                    let _ = ExtendedImage::draw_subimage(&mut canvas, first_frame, bg_color)
+                        .unwrap()
+                        .unwrap();
+                    buf.copy_from_slice(canvas.into_raw().as_slice());
+                }
+            }
+            ExtendedImageData::Static(image) => {
+                image.fill_buf(buf);
+            }
+        }
+    }
+
+    pub(crate) fn get_buf_size(&self) -> usize {
+        match &self.image {
+            // will always have at least one frame
+            ExtendedImageData::Animation { frames, .. } => &frames[0].image,
+            ExtendedImageData::Static(image) => image,
+        }
+        .get_buf_size()
+    }
+
+    pub(crate) fn set_background_color(&mut self, color: Rgba<u8>) -> ImageResult<()> {
+        match &mut self.image {
+            ExtendedImageData::Animation { anim_info, .. } => {
+                anim_info.background_color = color;
+                Ok(())
+            }
+            _ => Err(ImageError::Parameter(ParameterError::from_kind(
+                ParameterErrorKind::Generic(
+                    "Background color can only be set on animated webp".to_owned(),
+                ),
+            ))),
+        }
+    }
+}
+
+#[derive(Debug)]
+enum WebPStatic {
+    LossyWithAlpha(RgbaImage),
+    LossyWithoutAlpha(RgbImage),
+    Lossless(LosslessFrame),
+}
+
+impl WebPStatic {
+    pub(crate) fn from_alpha_lossy(
+        alpha: AlphaChunk,
+        vp8_frame: VP8Frame,
+    ) -> ImageResult<WebPStatic> {
+        if alpha.data.len() != usize::from(vp8_frame.width) * usize::from(vp8_frame.height) {
+            return Err(DecoderError::AlphaChunkSizeMismatch.into());
+        }
+
+        let size = usize::from(vp8_frame.width).checked_mul(usize::from(vp8_frame.height) * 4);
+        let mut image_vec = match size {
+            Some(size) => vec![0u8; size],
+            None => return Err(DecoderError::ImageTooLarge.into()),
+        };
+
+        vp8_frame.fill_rgba(&mut image_vec);
+
+        for y in 0..vp8_frame.height {
+            for x in 0..vp8_frame.width {
+                let predictor: u8 = WebPStatic::get_predictor(
+                    x.into(),
+                    y.into(),
+                    vp8_frame.width.into(),
+                    alpha.filtering_method,
+                    &image_vec,
+                );
+                let predictor = u16::from(predictor);
+
+                let alpha_index = usize::from(y) * usize::from(vp8_frame.width) + usize::from(x);
+                let alpha_val = alpha.data[alpha_index];
+                let alpha: u8 = ((predictor + u16::from(alpha_val)) % 256)
+                    .try_into()
+                    .unwrap();
+
+                let alpha_index = alpha_index * 4 + 3;
+                image_vec[alpha_index] = alpha;
+            }
+        }
+
+        let image = RgbaImage::from_vec(vp8_frame.width.into(), vp8_frame.height.into(), image_vec)
+            .unwrap();
+
+        Ok(WebPStatic::LossyWithAlpha(image))
+    }
+
+    fn get_predictor(
+        x: usize,
+        y: usize,
+        width: usize,
+        filtering_method: FilteringMethod,
+        image_slice: &[u8],
+    ) -> u8 {
+        match filtering_method {
+            FilteringMethod::None => 0,
+            FilteringMethod::Horizontal => {
+                if x == 0 && y == 0 {
+                    0
+                } else if x == 0 {
+                    let index = (y - 1) * width + x;
+                    image_slice[index * 4 + 3]
+                } else {
+                    let index = y * width + x - 1;
+                    image_slice[index * 4 + 3]
+                }
+            }
+            FilteringMethod::Vertical => {
+                if x == 0 && y == 0 {
+                    0
+                } else if y == 0 {
+                    let index = y * width + x - 1;
+                    image_slice[index * 4 + 3]
+                } else {
+                    let index = (y - 1) * width + x;
+                    image_slice[index * 4 + 3]
+                }
+            }
+            FilteringMethod::Gradient => {
+                let (left, top, top_left) = match (x, y) {
+                    (0, 0) => (0, 0, 0),
+                    (0, y) => {
+                        let above_index = (y - 1) * width + x;
+                        let val = image_slice[above_index * 4 + 3];
+                        (val, val, val)
+                    }
+                    (x, 0) => {
+                        let before_index = y * width + x - 1;
+                        let val = image_slice[before_index * 4 + 3];
+                        (val, val, val)
+                    }
+                    (x, y) => {
+                        let left_index = y * width + x - 1;
+                        let left = image_slice[left_index * 4 + 3];
+                        let top_index = (y - 1) * width + x;
+                        let top = image_slice[top_index * 4 + 3];
+                        let top_left_index = (y - 1) * width + x - 1;
+                        let top_left = image_slice[top_left_index * 4 + 3];
+
+                        (left, top, top_left)
+                    }
+                };
+
+                let combination = i16::from(left) + i16::from(top) - i16::from(top_left);
+                i16::clamp(combination, 0, 255).try_into().unwrap()
+            }
+        }
+    }
+
+    pub(crate) fn from_lossy(vp8_frame: VP8Frame) -> ImageResult<WebPStatic> {
+        let mut image = RgbImage::from_pixel(
+            vp8_frame.width.into(),
+            vp8_frame.height.into(),
+            Rgb([0, 0, 0]),
+        );
+
+        vp8_frame.fill_rgb(&mut image);
+
+        Ok(WebPStatic::LossyWithoutAlpha(image))
+    }
+
+    pub(crate) fn fill_buf(&self, buf: &mut [u8]) {
+        match self {
+            WebPStatic::LossyWithAlpha(image) => {
+                buf.copy_from_slice(image);
+            }
+            WebPStatic::LossyWithoutAlpha(image) => {
+                buf.copy_from_slice(image);
+            }
+            WebPStatic::Lossless(lossless) => {
+                lossless.fill_rgba(buf);
+            }
+        }
+    }
+
+    pub(crate) fn get_buf_size(&self) -> usize {
+        match self {
+            WebPStatic::LossyWithAlpha(rgb_image) => rgb_image.len(),
+            WebPStatic::LossyWithoutAlpha(rgba_image) => rgba_image.len(),
+            WebPStatic::Lossless(lossless) => lossless.get_buf_size(),
+        }
+    }
+
+    pub(crate) fn color_type(&self) -> ColorType {
+        if self.has_alpha() {
+            ColorType::Rgba8
+        } else {
+            ColorType::Rgb8
+        }
+    }
+
+    pub(crate) fn has_alpha(&self) -> bool {
+        match self {
+            Self::LossyWithAlpha(..) | Self::Lossless(..) => true,
+            Self::LossyWithoutAlpha(..) => false,
+        }
+    }
+}
+
+#[derive(Debug)]
+struct WebPAnimatedInfo {
+    background_color: Rgba<u8>,
+    _loop_count: u16,
+}
+
+#[derive(Debug)]
+struct AnimatedFrame {
+    offset_x: u32,
+    offset_y: u32,
+    width: u32,
+    height: u32,
+    duration: u32,
+    use_alpha_blending: bool,
+    dispose: bool,
+    image: WebPStatic,
+}
+
+/// Reads a chunk, but silently ignores unknown chunks at the end of a file
+fn read_extended_chunk<R>(r: &mut R) -> ImageResult<Option<(Cursor<Vec<u8>>, WebPRiffChunk)>>
+where
+    R: Read,
+{
+    let mut unknown_chunk = Ok(());
+
+    while let Some(chunk) = read_fourcc(r)? {
+        let cursor = read_len_cursor(r)?;
+        match chunk {
+            Ok(chunk) => return unknown_chunk.and(Ok(Some((cursor, chunk)))),
+            Err(err) => unknown_chunk = unknown_chunk.and(Err(err)),
+        }
+    }
+
+    Ok(None)
+}
+
+pub(crate) fn read_extended_header<R: Read>(reader: &mut R) -> ImageResult<WebPExtendedInfo> {
+    let chunk_flags = reader.read_u8()?;
+
+    let reserved_first = chunk_flags & 0b11000000;
+    let icc_profile = chunk_flags & 0b00100000 != 0;
+    let alpha = chunk_flags & 0b00010000 != 0;
+    let exif_metadata = chunk_flags & 0b00001000 != 0;
+    let xmp_metadata = chunk_flags & 0b00000100 != 0;
+    let animation = chunk_flags & 0b00000010 != 0;
+    let reserved_second = chunk_flags & 0b00000001;
+
+    let reserved_third = read_3_bytes(reader)?;
+
+    if reserved_first != 0 || reserved_second != 0 || reserved_third != 0 {
+        let value: u32 = if reserved_first != 0 {
+            reserved_first.into()
+        } else if reserved_second != 0 {
+            reserved_second.into()
+        } else {
+            reserved_third
+        };
+        return Err(DecoderError::InfoBitsInvalid {
+            name: "reserved",
+            value,
+        }
+        .into());
+    }
+
+    let canvas_width = read_3_bytes(reader)? + 1;
+    let canvas_height = read_3_bytes(reader)? + 1;
+
+    //product of canvas dimensions cannot be larger than u32 max
+    if u32::checked_mul(canvas_width, canvas_height).is_none() {
+        return Err(DecoderError::ImageTooLarge.into());
+    }
+
+    let info = WebPExtendedInfo {
+        _icc_profile: icc_profile,
+        _alpha: alpha,
+        _exif_metadata: exif_metadata,
+        _xmp_metadata: xmp_metadata,
+        _animation: animation,
+        canvas_width,
+        canvas_height,
+        icc_profile: None,
+    };
+
+    Ok(info)
+}
+
+fn read_anim_frame<R: Read>(
+    mut reader: R,
+    canvas_width: u32,
+    canvas_height: u32,
+) -> ImageResult<AnimatedFrame> {
+    //offsets for the frames are twice the values
+    let frame_x = read_3_bytes(&mut reader)? * 2;
+    let frame_y = read_3_bytes(&mut reader)? * 2;
+
+    let frame_width = read_3_bytes(&mut reader)? + 1;
+    let frame_height = read_3_bytes(&mut reader)? + 1;
+
+    if frame_x + frame_width > canvas_width || frame_y + frame_height > canvas_height {
+        return Err(DecoderError::FrameOutsideImage.into());
+    }
+
+    let duration = read_3_bytes(&mut reader)?;
+
+    let frame_info = reader.read_u8()?;
+    let reserved = frame_info & 0b11111100;
+    if reserved != 0 {
+        return Err(DecoderError::InfoBitsInvalid {
+            name: "reserved",
+            value: reserved.into(),
+        }
+        .into());
+    }
+    let use_alpha_blending = frame_info & 0b00000010 == 0;
+    let dispose = frame_info & 0b00000001 != 0;
+
+    //read normal bitstream now
+    let static_image = read_image(&mut reader, frame_width, frame_height)?;
+
+    let frame = AnimatedFrame {
+        offset_x: frame_x,
+        offset_y: frame_y,
+        width: frame_width,
+        height: frame_height,
+        duration,
+        use_alpha_blending,
+        dispose,
+        image: static_image,
+    };
+
+    Ok(frame)
+}
+
+fn read_3_bytes<R: Read>(reader: &mut R) -> ImageResult<u32> {
+    let mut buffer: [u8; 3] = [0; 3];
+    reader.read_exact(&mut buffer)?;
+    let value: u32 =
+        (u32::from(buffer[2]) << 16) | (u32::from(buffer[1]) << 8) | u32::from(buffer[0]);
+    Ok(value)
+}
+
+fn read_lossy_with_chunk<R: Read>(reader: &mut R) -> ImageResult<VP8Frame> {
+    let (cursor, chunk) =
+        read_chunk(reader)?.ok_or_else(|| Error::from(io::ErrorKind::UnexpectedEof))?;
+
+    if chunk != WebPRiffChunk::VP8 {
+        return Err(ChunkHeaderInvalid(chunk.to_fourcc()).into());
+    }
+
+    read_lossy(cursor)
+}
+
+fn read_lossy(cursor: Cursor<Vec<u8>>) -> ImageResult<VP8Frame> {
+    let mut vp8_decoder = Vp8Decoder::new(cursor);
+    let frame = vp8_decoder.decode_frame()?;
+
+    Ok(frame.clone())
+}
+
+fn read_image<R: Read>(reader: &mut R, width: u32, height: u32) -> ImageResult<WebPStatic> {
+    let chunk = read_chunk(reader)?;
+
+    match chunk {
+        Some((cursor, WebPRiffChunk::VP8)) => {
+            let mut vp8_decoder = Vp8Decoder::new(cursor);
+            let frame = vp8_decoder.decode_frame()?;
+
+            let img = WebPStatic::from_lossy(frame.clone())?;
+
+            Ok(img)
+        }
+        Some((cursor, WebPRiffChunk::VP8L)) => {
+            let mut lossless_decoder = LosslessDecoder::new(cursor);
+            let frame = lossless_decoder.decode_frame()?;
+
+            let img = WebPStatic::Lossless(frame.clone());
+
+            Ok(img)
+        }
+        Some((mut cursor, WebPRiffChunk::ALPH)) => {
+            let alpha_chunk = read_alpha_chunk(&mut cursor, width, height)?;
+
+            let vp8_frame = read_lossy_with_chunk(reader)?;
+
+            let img = WebPStatic::from_alpha_lossy(alpha_chunk, vp8_frame)?;
+
+            Ok(img)
+        }
+        None => Err(ImageError::IoError(Error::from(
+            io::ErrorKind::UnexpectedEof,
+        ))),
+        Some((_, chunk)) => Err(ChunkHeaderInvalid(chunk.to_fourcc()).into()),
+    }
+}
+
+#[derive(Debug)]
+struct AlphaChunk {
+    _preprocessing: bool,
+    filtering_method: FilteringMethod,
+    data: Vec<u8>,
+}
+
+#[derive(Debug, Copy, Clone)]
+enum FilteringMethod {
+    None,
+    Horizontal,
+    Vertical,
+    Gradient,
+}
+
+fn read_alpha_chunk<R: Read>(reader: &mut R, width: u32, height: u32) -> ImageResult<AlphaChunk> {
+    let info_byte = reader.read_u8()?;
+
+    let reserved = info_byte & 0b11000000;
+    let preprocessing = (info_byte & 0b00110000) >> 4;
+    let filtering = (info_byte & 0b00001100) >> 2;
+    let compression = info_byte & 0b00000011;
+
+    if reserved != 0 {
+        return Err(DecoderError::InfoBitsInvalid {
+            name: "reserved",
+            value: reserved.into(),
+        }
+        .into());
+    }
+
+    let preprocessing = match preprocessing {
+        0 => false,
+        1 => true,
+        _ => {
+            return Err(DecoderError::InfoBitsInvalid {
+                name: "reserved",
+                value: preprocessing.into(),
+            }
+            .into())
+        }
+    };
+
+    let filtering_method = match filtering {
+        0 => FilteringMethod::None,
+        1 => FilteringMethod::Horizontal,
+        2 => FilteringMethod::Vertical,
+        3 => FilteringMethod::Gradient,
+        _ => unreachable!(),
+    };
+
+    let lossless_compression = match compression {
+        0 => false,
+        1 => true,
+        _ => {
+            return Err(DecoderError::InfoBitsInvalid {
+                name: "lossless compression",
+                value: compression.into(),
+            }
+            .into())
+        }
+    };
+
+    let mut framedata = Vec::new();
+    reader.read_to_end(&mut framedata)?;
+
+    let data = if lossless_compression {
+        let cursor = io::Cursor::new(framedata);
+
+        let mut decoder = LosslessDecoder::new(cursor);
+        //this is a potential problem for large images; would require rewriting lossless decoder to use u32 for width and height
+        let width: u16 = width
+            .try_into()
+            .map_err(|_| ImageError::from(DecoderError::ImageTooLarge))?;
+        let height: u16 = height
+            .try_into()
+            .map_err(|_| ImageError::from(DecoderError::ImageTooLarge))?;
+        let frame = decoder.decode_frame_implicit_dims(width, height)?;
+
+        let mut data = vec![0u8; usize::from(width) * usize::from(height)];
+
+        frame.fill_green(&mut data);
+
+        data
+    } else {
+        framedata
+    };
+
+    let chunk = AlphaChunk {
+        _preprocessing: preprocessing,
+        filtering_method,
+        data,
+    };
+
+    Ok(chunk)
+}

vendor/image/src/codecs/webp/huffman.rs

@@ -0,0 +1,202 @@
+use std::convert::TryInto;
+
+use super::lossless::BitReader;
+use super::lossless::DecoderError;
+use crate::ImageResult;
+
+/// Rudimentary utility for reading Canonical Huffman Codes.
+/// Based off https://github.com/webmproject/libwebp/blob/7f8472a610b61ec780ef0a8873cd954ac512a505/src/utils/huffman.c
+///
+
+const MAX_ALLOWED_CODE_LENGTH: usize = 15;
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+enum HuffmanTreeNode {
+    Branch(usize), //offset in vector to children
+    Leaf(u16),     //symbol stored in leaf
+    Empty,
+}
+
+/// Huffman tree
+#[derive(Clone, Debug, Default)]
+pub(crate) struct HuffmanTree {
+    tree: Vec<HuffmanTreeNode>,
+    max_nodes: usize,
+    num_nodes: usize,
+}
+
+impl HuffmanTree {
+    fn is_full(&self) -> bool {
+        self.num_nodes == self.max_nodes
+    }
+
+    /// Turns a node from empty into a branch and assigns its children
+    fn assign_children(&mut self, node_index: usize) -> usize {
+        let offset_index = self.num_nodes - node_index;
+        self.tree[node_index] = HuffmanTreeNode::Branch(offset_index);
+        self.num_nodes += 2;
+
+        offset_index
+    }
+
+    /// Init a huffman tree
+    fn init(num_leaves: usize) -> ImageResult<HuffmanTree> {
+        if num_leaves == 0 {
+            return Err(DecoderError::HuffmanError.into());
+        }
+
+        let max_nodes = 2 * num_leaves - 1;
+        let tree = vec![HuffmanTreeNode::Empty; max_nodes];
+        let num_nodes = 1;
+
+        let tree = HuffmanTree {
+            tree,
+            max_nodes,
+            num_nodes,
+        };
+
+        Ok(tree)
+    }
+
+    /// Converts code lengths to codes
+    fn code_lengths_to_codes(code_lengths: &[u16]) -> ImageResult<Vec<Option<u16>>> {
+        let max_code_length = *code_lengths
+            .iter()
+            .reduce(|a, b| if a >= b { a } else { b })
+            .unwrap();
+
+        if max_code_length > MAX_ALLOWED_CODE_LENGTH.try_into().unwrap() {
+            return Err(DecoderError::HuffmanError.into());
+        }
+
+        let mut code_length_hist = vec![0; MAX_ALLOWED_CODE_LENGTH + 1];
+
+        for &length in code_lengths.iter() {
+            code_length_hist[usize::from(length)] += 1;
+        }
+
+        code_length_hist[0] = 0;
+
+        let mut curr_code = 0;
+        let mut next_codes = vec![None; MAX_ALLOWED_CODE_LENGTH + 1];
+
+        for code_len in 1..=usize::from(max_code_length) {
+            curr_code = (curr_code + code_length_hist[code_len - 1]) << 1;
+            next_codes[code_len] = Some(curr_code);
+        }
+
+        let mut huff_codes = vec![None; code_lengths.len()];
+
+        for (symbol, &length) in code_lengths.iter().enumerate() {
+            let length = usize::from(length);
+            if length > 0 {
+                huff_codes[symbol] = next_codes[length];
+                if let Some(value) = next_codes[length].as_mut() {
+                    *value += 1;
+                }
+            } else {
+                huff_codes[symbol] = None;
+            }
+        }
+
+        Ok(huff_codes)
+    }
+
+    /// Adds a symbol to a huffman tree
+    fn add_symbol(&mut self, symbol: u16, code: u16, code_length: u16) -> ImageResult<()> {
+        let mut node_index = 0;
+        let code = usize::from(code);
+
+        for length in (0..code_length).rev() {
+            if node_index >= self.max_nodes {
+                return Err(DecoderError::HuffmanError.into());
+            }
+
+            let node = self.tree[node_index];
+
+            let offset = match node {
+                HuffmanTreeNode::Empty => {
+                    if self.is_full() {
+                        return Err(DecoderError::HuffmanError.into());
+                    }
+                    self.assign_children(node_index)
+                }
+                HuffmanTreeNode::Leaf(_) => return Err(DecoderError::HuffmanError.into()),
+                HuffmanTreeNode::Branch(offset) => offset,
+            };
+
+            node_index += offset + ((code >> length) & 1);
+        }
+
+        match self.tree[node_index] {
+            HuffmanTreeNode::Empty => self.tree[node_index] = HuffmanTreeNode::Leaf(symbol),
+            HuffmanTreeNode::Leaf(_) => return Err(DecoderError::HuffmanError.into()),
+            HuffmanTreeNode::Branch(_offset) => return Err(DecoderError::HuffmanError.into()),
+        }
+
+        Ok(())
+    }
+
+    /// Builds a tree implicitly, just from code lengths
+    pub(crate) fn build_implicit(code_lengths: Vec<u16>) -> ImageResult<HuffmanTree> {
+        let mut num_symbols = 0;
+        let mut root_symbol = 0;
+
+        for (symbol, length) in code_lengths.iter().enumerate() {
+            if *length > 0 {
+                num_symbols += 1;
+                root_symbol = symbol.try_into().unwrap();
+            }
+        }
+
+        let mut tree = HuffmanTree::init(num_symbols)?;
+
+        if num_symbols == 1 {
+            tree.add_symbol(root_symbol, 0, 0)?;
+        } else {
+            let codes = HuffmanTree::code_lengths_to_codes(&code_lengths)?;
+
+            for (symbol, &length) in code_lengths.iter().enumerate() {
+                if length > 0 && codes[symbol].is_some() {
+                    tree.add_symbol(symbol.try_into().unwrap(), codes[symbol].unwrap(), length)?;
+                }
+            }
+        }
+
+        Ok(tree)
+    }
+
+    /// Builds a tree explicitly from lengths, codes and symbols
+    pub(crate) fn build_explicit(
+        code_lengths: Vec<u16>,
+        codes: Vec<u16>,
+        symbols: Vec<u16>,
+    ) -> ImageResult<HuffmanTree> {
+        let mut tree = HuffmanTree::init(symbols.len())?;
+
+        for i in 0..symbols.len() {
+            tree.add_symbol(symbols[i], codes[i], code_lengths[i])?;
+        }
+
+        Ok(tree)
+    }
+
+    /// Reads a symbol using the bitstream
+    pub(crate) fn read_symbol(&self, bit_reader: &mut BitReader) -> ImageResult<u16> {
+        let mut index = 0;
+        let mut node = self.tree[index];
+
+        while let HuffmanTreeNode::Branch(children_offset) = node {
+            index += children_offset + bit_reader.read_bits::<usize>(1)?;
+            node = self.tree[index];
+        }
+
+        let symbol = match node {
+            HuffmanTreeNode::Branch(_) => unreachable!(),
+            HuffmanTreeNode::Empty => return Err(DecoderError::HuffmanError.into()),
+            HuffmanTreeNode::Leaf(symbol) => symbol,
+        };
+
+        Ok(symbol)
+    }
+}

vendor/image/src/codecs/webp/loop_filter.rs

@@ -0,0 +1,147 @@
+//! Does loop filtering on webp lossy images
+
+use crate::utils::clamp;
+
+#[inline]
+fn c(val: i32) -> i32 {
+    clamp(val, -128, 127)
+}
+
+//unsigned to signed
+#[inline]
+fn u2s(val: u8) -> i32 {
+    i32::from(val) - 128
+}
+
+//signed to unsigned
+#[inline]
+fn s2u(val: i32) -> u8 {
+    (c(val) + 128) as u8
+}
+
+#[inline]
+fn diff(val1: u8, val2: u8) -> u8 {
+    if val1 > val2 {
+        val1 - val2
+    } else {
+        val2 - val1
+    }
+}
+
+//15.2
+fn common_adjust(use_outer_taps: bool, pixels: &mut [u8], point: usize, stride: usize) -> i32 {
+    let p1 = u2s(pixels[point - 2 * stride]);
+    let p0 = u2s(pixels[point - stride]);
+    let q0 = u2s(pixels[point]);
+    let q1 = u2s(pixels[point + stride]);
+
+    //value for the outer 2 pixels
+    let outer = if use_outer_taps { c(p1 - q1) } else { 0 };
+
+    let mut a = c(outer + 3 * (q0 - p0));
+
+    let b = (c(a + 3)) >> 3;
+
+    a = (c(a + 4)) >> 3;
+
+    pixels[point] = s2u(q0 - a);
+    pixels[point - stride] = s2u(p0 + b);
+
+    a
+}
+
+fn simple_threshold(filter_limit: i32, pixels: &[u8], point: usize, stride: usize) -> bool {
+    i32::from(diff(pixels[point - stride], pixels[point])) * 2
+        + i32::from(diff(pixels[point - 2 * stride], pixels[point + stride])) / 2
+        <= filter_limit
+}
+
+fn should_filter(
+    interior_limit: u8,
+    edge_limit: u8,
+    pixels: &[u8],
+    point: usize,
+    stride: usize,
+) -> bool {
+    simple_threshold(i32::from(edge_limit), pixels, point, stride)
+        && diff(pixels[point - 4 * stride], pixels[point - 3 * stride]) <= interior_limit
+        && diff(pixels[point - 3 * stride], pixels[point - 2 * stride]) <= interior_limit
+        && diff(pixels[point - 2 * stride], pixels[point - stride]) <= interior_limit
+        && diff(pixels[point + 3 * stride], pixels[point + 2 * stride]) <= interior_limit
+        && diff(pixels[point + 2 * stride], pixels[point + stride]) <= interior_limit
+        && diff(pixels[point + stride], pixels[point]) <= interior_limit
+}
+
+fn high_edge_variance(threshold: u8, pixels: &[u8], point: usize, stride: usize) -> bool {
+    diff(pixels[point - 2 * stride], pixels[point - stride]) > threshold
+        || diff(pixels[point + stride], pixels[point]) > threshold
+}
+
+//simple filter
+//effects 4 pixels on an edge(2 each side)
+pub(crate) fn simple_segment(edge_limit: u8, pixels: &mut [u8], point: usize, stride: usize) {
+    if simple_threshold(i32::from(edge_limit), pixels, point, stride) {
+        common_adjust(true, pixels, point, stride);
+    }
+}
+
+//normal filter
+//works on the 8 pixels on the edges between subblocks inside a macroblock
+pub(crate) fn subblock_filter(
+    hev_threshold: u8,
+    interior_limit: u8,
+    edge_limit: u8,
+    pixels: &mut [u8],
+    point: usize,
+    stride: usize,
+) {
+    if should_filter(interior_limit, edge_limit, pixels, point, stride) {
+        let hv = high_edge_variance(hev_threshold, pixels, point, stride);
+
+        let a = (common_adjust(hv, pixels, point, stride) + 1) >> 1;
+
+        if !hv {
+            pixels[point + stride] = s2u(u2s(pixels[point + stride]) - a);
+            pixels[point - 2 * stride] = s2u(u2s(pixels[point - 2 * stride]) - a);
+        }
+    }
+}
+
+//normal filter
+//works on the 8 pixels on the edges between macroblocks
+pub(crate) fn macroblock_filter(
+    hev_threshold: u8,
+    interior_limit: u8,
+    edge_limit: u8,
+    pixels: &mut [u8],
+    point: usize,
+    stride: usize,
+) {
+    let mut spixels = [0i32; 8];
+    for i in 0..8 {
+        spixels[i] = u2s(pixels[point + i * stride - 4 * stride]);
+    }
+
+    if should_filter(interior_limit, edge_limit, pixels, point, stride) {
+        if !high_edge_variance(hev_threshold, pixels, point, stride) {
+            let w = c(c(spixels[2] - spixels[5]) + 3 * (spixels[4] - spixels[3]));
+
+            let mut a = c((27 * w + 63) >> 7);
+
+            pixels[point] = s2u(spixels[4] - a);
+            pixels[point - stride] = s2u(spixels[3] + a);
+
+            a = c((18 * w + 63) >> 7);
+
+            pixels[point + stride] = s2u(spixels[5] - a);
+            pixels[point - 2 * stride] = s2u(spixels[2] + a);
+
+            a = c((9 * w + 63) >> 7);
+
+            pixels[point + 2 * stride] = s2u(spixels[6] - a);
+            pixels[point - 3 * stride] = s2u(spixels[1] + a);
+        } else {
+            common_adjust(true, pixels, point, stride);
+        }
+    }
+}

vendor/image/src/codecs/webp/lossless.rs

@@ -0,0 +1,783 @@
+//! Decoding of lossless WebP images
+//!
+//! [Lossless spec](https://developers.google.com/speed/webp/docs/webp_lossless_bitstream_specification)
+//!
+
+use std::{
+    convert::TryFrom,
+    convert::TryInto,
+    error, fmt,
+    io::Read,
+    ops::{AddAssign, Shl},
+};
+
+use byteorder::ReadBytesExt;
+
+use crate::{error::DecodingError, ImageError, ImageFormat, ImageResult};
+
+use super::huffman::HuffmanTree;
+use super::lossless_transform::{add_pixels, TransformType};
+
+const CODE_LENGTH_CODES: usize = 19;
+const CODE_LENGTH_CODE_ORDER: [usize; CODE_LENGTH_CODES] = [
+    17, 18, 0, 1, 2, 3, 4, 5, 16, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
+];
+
+#[rustfmt::skip]
+const DISTANCE_MAP: [(i8, i8); 120] = [
+    (0, 1),  (1, 0),  (1, 1),  (-1, 1), (0, 2),  (2, 0),  (1, 2),  (-1, 2),
+    (2, 1),  (-2, 1), (2, 2),  (-2, 2), (0, 3),  (3, 0),  (1, 3),  (-1, 3),
+    (3, 1),  (-3, 1), (2, 3),  (-2, 3), (3, 2),  (-3, 2), (0, 4),  (4, 0),
+    (1, 4),  (-1, 4), (4, 1),  (-4, 1), (3, 3),  (-3, 3), (2, 4),  (-2, 4),
+    (4, 2),  (-4, 2), (0, 5),  (3, 4),  (-3, 4), (4, 3),  (-4, 3), (5, 0),
+    (1, 5),  (-1, 5), (5, 1),  (-5, 1), (2, 5),  (-2, 5), (5, 2),  (-5, 2),
+    (4, 4),  (-4, 4), (3, 5),  (-3, 5), (5, 3),  (-5, 3), (0, 6),  (6, 0),
+    (1, 6),  (-1, 6), (6, 1),  (-6, 1), (2, 6),  (-2, 6), (6, 2),  (-6, 2),
+    (4, 5),  (-4, 5), (5, 4),  (-5, 4), (3, 6),  (-3, 6), (6, 3),  (-6, 3),
+    (0, 7),  (7, 0),  (1, 7),  (-1, 7), (5, 5),  (-5, 5), (7, 1),  (-7, 1),
+    (4, 6),  (-4, 6), (6, 4),  (-6, 4), (2, 7),  (-2, 7), (7, 2),  (-7, 2),
+    (3, 7),  (-3, 7), (7, 3),  (-7, 3), (5, 6),  (-5, 6), (6, 5),  (-6, 5),
+    (8, 0),  (4, 7),  (-4, 7), (7, 4),  (-7, 4), (8, 1),  (8, 2),  (6, 6),
+    (-6, 6), (8, 3),  (5, 7),  (-5, 7), (7, 5),  (-7, 5), (8, 4),  (6, 7),
+    (-6, 7), (7, 6),  (-7, 6), (8, 5),  (7, 7),  (-7, 7), (8, 6),  (8, 7)
+];
+
+const GREEN: usize = 0;
+const RED: usize = 1;
+const BLUE: usize = 2;
+const ALPHA: usize = 3;
+const DIST: usize = 4;
+
+const HUFFMAN_CODES_PER_META_CODE: usize = 5;
+
+type HuffmanCodeGroup = [HuffmanTree; HUFFMAN_CODES_PER_META_CODE];
+
+const ALPHABET_SIZE: [u16; HUFFMAN_CODES_PER_META_CODE] = [256 + 24, 256, 256, 256, 40];
+
+#[inline]
+pub(crate) fn subsample_size(size: u16, bits: u8) -> u16 {
+    ((u32::from(size) + (1u32 << bits) - 1) >> bits)
+        .try_into()
+        .unwrap()
+}
+
+#[derive(Debug, Clone, Copy)]
+pub(crate) enum DecoderError {
+    /// Signature of 0x2f not found
+    LosslessSignatureInvalid(u8),
+    /// Version Number must be 0
+    VersionNumberInvalid(u8),
+
+    ///
+    InvalidColorCacheBits(u8),
+
+    HuffmanError,
+
+    BitStreamError,
+
+    TransformError,
+}
+
+impl fmt::Display for DecoderError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            DecoderError::LosslessSignatureInvalid(sig) => {
+                f.write_fmt(format_args!("Invalid lossless signature: {}", sig))
+            }
+            DecoderError::VersionNumberInvalid(num) => {
+                f.write_fmt(format_args!("Invalid version number: {}", num))
+            }
+            DecoderError::InvalidColorCacheBits(num) => f.write_fmt(format_args!(
+                "Invalid color cache(must be between 1-11): {}",
+                num
+            )),
+            DecoderError::HuffmanError => f.write_fmt(format_args!("Error building Huffman Tree")),
+            DecoderError::BitStreamError => {
+                f.write_fmt(format_args!("Error while reading bitstream"))
+            }
+            DecoderError::TransformError => {
+                f.write_fmt(format_args!("Error while reading or writing transforms"))
+            }
+        }
+    }
+}
+
+impl From<DecoderError> for ImageError {
+    fn from(e: DecoderError) -> ImageError {
+        ImageError::Decoding(DecodingError::new(ImageFormat::WebP.into(), e))
+    }
+}
+
+impl error::Error for DecoderError {}
+
+const NUM_TRANSFORM_TYPES: usize = 4;
+
+//Decodes lossless WebP images
+#[derive(Debug)]
+pub(crate) struct LosslessDecoder<R> {
+    r: R,
+    bit_reader: BitReader,
+    frame: LosslessFrame,
+    transforms: [Option<TransformType>; NUM_TRANSFORM_TYPES],
+    transform_order: Vec<u8>,
+}
+
+impl<R: Read> LosslessDecoder<R> {
+    /// Create a new decoder
+    pub(crate) fn new(r: R) -> LosslessDecoder<R> {
+        LosslessDecoder {
+            r,
+            bit_reader: BitReader::new(),
+            frame: Default::default(),
+            transforms: [None, None, None, None],
+            transform_order: Vec::new(),
+        }
+    }
+
+    /// Reads the frame
+    pub(crate) fn decode_frame(&mut self) -> ImageResult<&LosslessFrame> {
+        let signature = self.r.read_u8()?;
+
+        if signature != 0x2f {
+            return Err(DecoderError::LosslessSignatureInvalid(signature).into());
+        }
+
+        let mut buf = Vec::new();
+        self.r.read_to_end(&mut buf)?;
+        self.bit_reader.init(buf);
+
+        self.frame.width = self.bit_reader.read_bits::<u16>(14)? + 1;
+        self.frame.height = self.bit_reader.read_bits::<u16>(14)? + 1;
+
+        let _alpha_used = self.bit_reader.read_bits::<u8>(1)?;
+
+        let version_num = self.bit_reader.read_bits::<u8>(3)?;
+
+        if version_num != 0 {
+            return Err(DecoderError::VersionNumberInvalid(version_num).into());
+        }
+
+        let mut data = self.decode_image_stream(self.frame.width, self.frame.height, true)?;
+
+        for &trans_index in self.transform_order.iter().rev() {
+            let trans = self.transforms[usize::from(trans_index)].as_ref().unwrap();
+            trans.apply_transform(&mut data, self.frame.width, self.frame.height)?;
+        }
+
+        self.frame.buf = data;
+        Ok(&self.frame)
+    }
+
+    //used for alpha data in extended decoding
+    pub(crate) fn decode_frame_implicit_dims(
+        &mut self,
+        width: u16,
+        height: u16,
+    ) -> ImageResult<&LosslessFrame> {
+        let mut buf = Vec::new();
+        self.r.read_to_end(&mut buf)?;
+        self.bit_reader.init(buf);
+
+        self.frame.width = width;
+        self.frame.height = height;
+
+        let mut data = self.decode_image_stream(self.frame.width, self.frame.height, true)?;
+
+        //transform_order is vector of indices(0-3) into transforms in order decoded
+        for &trans_index in self.transform_order.iter().rev() {
+            let trans = self.transforms[usize::from(trans_index)].as_ref().unwrap();
+            trans.apply_transform(&mut data, self.frame.width, self.frame.height)?;
+        }
+
+        self.frame.buf = data;
+        Ok(&self.frame)
+    }
+
+    /// Reads Image data from the bitstream
+    /// Can be in any of the 5 roles described in the Specification
+    /// ARGB Image role has different behaviour to the other 4
+    /// xsize and ysize describe the size of the blocks where each block has its own entropy code
+    fn decode_image_stream(
+        &mut self,
+        xsize: u16,
+        ysize: u16,
+        is_argb_img: bool,
+    ) -> ImageResult<Vec<u32>> {
+        let trans_xsize = if is_argb_img {
+            self.read_transforms()?
+        } else {
+            xsize
+        };
+
+        let color_cache_bits = self.read_color_cache()?;
+
+        let color_cache = color_cache_bits.map(|bits| {
+            let size = 1 << bits;
+            let cache = vec![0u32; size];
+            ColorCache {
+                color_cache_bits: bits,
+                color_cache: cache,
+            }
+        });
+
+        let huffman_info = self.read_huffman_codes(is_argb_img, trans_xsize, ysize, color_cache)?;
+
+        //decode data
+        let data = self.decode_image_data(trans_xsize, ysize, huffman_info)?;
+
+        Ok(data)
+    }
+
+    /// Reads transforms and their data from the bitstream
+    fn read_transforms(&mut self) -> ImageResult<u16> {
+        let mut xsize = self.frame.width;
+
+        while self.bit_reader.read_bits::<u8>(1)? == 1 {
+            let transform_type_val = self.bit_reader.read_bits::<u8>(2)?;
+
+            if self.transforms[usize::from(transform_type_val)].is_some() {
+                //can only have one of each transform, error
+                return Err(DecoderError::TransformError.into());
+            }
+
+            self.transform_order.push(transform_type_val);
+
+            let transform_type = match transform_type_val {
+                0 => {
+                    //predictor
+
+                    let size_bits = self.bit_reader.read_bits::<u8>(3)? + 2;
+
+                    let block_xsize = subsample_size(xsize, size_bits);
+                    let block_ysize = subsample_size(self.frame.height, size_bits);
+
+                    let data = self.decode_image_stream(block_xsize, block_ysize, false)?;
+
+                    TransformType::PredictorTransform {
+                        size_bits,
+                        predictor_data: data,
+                    }
+                }
+                1 => {
+                    //color transform
+
+                    let size_bits = self.bit_reader.read_bits::<u8>(3)? + 2;
+
+                    let block_xsize = subsample_size(xsize, size_bits);
+                    let block_ysize = subsample_size(self.frame.height, size_bits);
+
+                    let data = self.decode_image_stream(block_xsize, block_ysize, false)?;
+
+                    TransformType::ColorTransform {
+                        size_bits,
+                        transform_data: data,
+                    }
+                }
+                2 => {
+                    //subtract green
+
+                    TransformType::SubtractGreen
+                }
+                3 => {
+                    let color_table_size = self.bit_reader.read_bits::<u16>(8)? + 1;
+
+                    let mut color_map = self.decode_image_stream(color_table_size, 1, false)?;
+
+                    let bits = if color_table_size <= 2 {
+                        3
+                    } else if color_table_size <= 4 {
+                        2
+                    } else if color_table_size <= 16 {
+                        1
+                    } else {
+                        0
+                    };
+                    xsize = subsample_size(xsize, bits);
+
+                    Self::adjust_color_map(&mut color_map);
+
+                    TransformType::ColorIndexingTransform {
+                        table_size: color_table_size,
+                        table_data: color_map,
+                    }
+                }
+                _ => unreachable!(),
+            };
+
+            self.transforms[usize::from(transform_type_val)] = Some(transform_type);
+        }
+
+        Ok(xsize)
+    }
+
+    /// Adjusts the color map since it's subtraction coded
+    fn adjust_color_map(color_map: &mut Vec<u32>) {
+        for i in 1..color_map.len() {
+            color_map[i] = add_pixels(color_map[i], color_map[i - 1]);
+        }
+    }
+
+    /// Reads huffman codes associated with an image
+    fn read_huffman_codes(
+        &mut self,
+        read_meta: bool,
+        xsize: u16,
+        ysize: u16,
+        color_cache: Option<ColorCache>,
+    ) -> ImageResult<HuffmanInfo> {
+        let mut num_huff_groups = 1;
+
+        let mut huffman_bits = 0;
+        let mut huffman_xsize = 1;
+        let mut huffman_ysize = 1;
+        let mut entropy_image = Vec::new();
+
+        if read_meta && self.bit_reader.read_bits::<u8>(1)? == 1 {
+            //meta huffman codes
+            huffman_bits = self.bit_reader.read_bits::<u8>(3)? + 2;
+            huffman_xsize = subsample_size(xsize, huffman_bits);
+            huffman_ysize = subsample_size(ysize, huffman_bits);
+
+            entropy_image = self.decode_image_stream(huffman_xsize, huffman_ysize, false)?;
+
+            for pixel in entropy_image.iter_mut() {
+                let meta_huff_code = (*pixel >> 8) & 0xffff;
+
+                *pixel = meta_huff_code;
+
+                if meta_huff_code >= num_huff_groups {
+                    num_huff_groups = meta_huff_code + 1;
+                }
+            }
+        }
+
+        let mut hufftree_groups = Vec::new();
+
+        for _i in 0..num_huff_groups {
+            let mut group: HuffmanCodeGroup = Default::default();
+            for j in 0..HUFFMAN_CODES_PER_META_CODE {
+                let mut alphabet_size = ALPHABET_SIZE[j];
+                if j == 0 {
+                    if let Some(color_cache) = color_cache.as_ref() {
+                        alphabet_size += 1 << color_cache.color_cache_bits;
+                    }
+                }
+
+                let tree = self.read_huffman_code(alphabet_size)?;
+                group[j] = tree;
+            }
+            hufftree_groups.push(group);
+        }
+
+        let huffman_mask = if huffman_bits == 0 {
+            !0
+        } else {
+            (1 << huffman_bits) - 1
+        };
+
+        let info = HuffmanInfo {
+            xsize: huffman_xsize,
+            _ysize: huffman_ysize,
+            color_cache,
+            image: entropy_image,
+            bits: huffman_bits,
+            mask: huffman_mask,
+            huffman_code_groups: hufftree_groups,
+        };
+
+        Ok(info)
+    }
+
+    /// Decodes and returns a single huffman tree
+    fn read_huffman_code(&mut self, alphabet_size: u16) -> ImageResult<HuffmanTree> {
+        let simple = self.bit_reader.read_bits::<u8>(1)? == 1;
+
+        if simple {
+            let num_symbols = self.bit_reader.read_bits::<u8>(1)? + 1;
+
+            let mut code_lengths = vec![u16::from(num_symbols - 1)];
+            let mut codes = vec![0];
+            let mut symbols = Vec::new();
+
+            let is_first_8bits = self.bit_reader.read_bits::<u8>(1)?;
+            symbols.push(self.bit_reader.read_bits::<u16>(1 + 7 * is_first_8bits)?);
+
+            if num_symbols == 2 {
+                symbols.push(self.bit_reader.read_bits::<u16>(8)?);
+                code_lengths.push(1);
+                codes.push(1);
+            }
+
+            HuffmanTree::build_explicit(code_lengths, codes, symbols)
+        } else {
+            let mut code_length_code_lengths = vec![0; CODE_LENGTH_CODES];
+
+            let num_code_lengths = 4 + self.bit_reader.read_bits::<usize>(4)?;
+            for i in 0..num_code_lengths {
+                code_length_code_lengths[CODE_LENGTH_CODE_ORDER[i]] =
+                    self.bit_reader.read_bits(3)?;
+            }
+
+            let new_code_lengths =
+                self.read_huffman_code_lengths(code_length_code_lengths, alphabet_size)?;
+
+            HuffmanTree::build_implicit(new_code_lengths)
+        }
+    }
+
+    /// Reads huffman code lengths
+    fn read_huffman_code_lengths(
+        &mut self,
+        code_length_code_lengths: Vec<u16>,
+        num_symbols: u16,
+    ) -> ImageResult<Vec<u16>> {
+        let table = HuffmanTree::build_implicit(code_length_code_lengths)?;
+
+        let mut max_symbol = if self.bit_reader.read_bits::<u8>(1)? == 1 {
+            let length_nbits = 2 + 2 * self.bit_reader.read_bits::<u8>(3)?;
+            2 + self.bit_reader.read_bits::<u16>(length_nbits)?
+        } else {
+            num_symbols
+        };
+
+        let mut code_lengths = vec![0; usize::from(num_symbols)];
+        let mut prev_code_len = 8; //default code length
+
+        let mut symbol = 0;
+        while symbol < num_symbols {
+            if max_symbol == 0 {
+                break;
+            }
+            max_symbol -= 1;
+
+            let code_len = table.read_symbol(&mut self.bit_reader)?;
+
+            if code_len < 16 {
+                code_lengths[usize::from(symbol)] = code_len;
+                symbol += 1;
+                if code_len != 0 {
+                    prev_code_len = code_len;
+                }
+            } else {
+                let use_prev = code_len == 16;
+                let slot = code_len - 16;
+                let extra_bits = match slot {
+                    0 => 2,
+                    1 => 3,
+                    2 => 7,
+                    _ => return Err(DecoderError::BitStreamError.into()),
+                };
+                let repeat_offset = match slot {
+                    0 | 1 => 3,
+                    2 => 11,
+                    _ => return Err(DecoderError::BitStreamError.into()),
+                };
+
+                let mut repeat = self.bit_reader.read_bits::<u16>(extra_bits)? + repeat_offset;
+
+                if symbol + repeat > num_symbols {
+                    return Err(DecoderError::BitStreamError.into());
+                } else {
+                    let length = if use_prev { prev_code_len } else { 0 };
+                    while repeat > 0 {
+                        repeat -= 1;
+                        code_lengths[usize::from(symbol)] = length;
+                        symbol += 1;
+                    }
+                }
+            }
+        }
+
+        Ok(code_lengths)
+    }
+
+    /// Decodes the image data using the huffman trees and either of the 3 methods of decoding
+    fn decode_image_data(
+        &mut self,
+        width: u16,
+        height: u16,
+        mut huffman_info: HuffmanInfo,
+    ) -> ImageResult<Vec<u32>> {
+        let num_values = usize::from(width) * usize::from(height);
+        let mut data = vec![0; num_values];
+
+        let huff_index = huffman_info.get_huff_index(0, 0);
+        let mut tree = &huffman_info.huffman_code_groups[huff_index];
+        let mut last_cached = 0;
+        let mut index = 0;
+        let mut x = 0;
+        let mut y = 0;
+        while index < num_values {
+            if (x & huffman_info.mask) == 0 {
+                let index = huffman_info.get_huff_index(x, y);
+                tree = &huffman_info.huffman_code_groups[index];
+            }
+
+            let code = tree[GREEN].read_symbol(&mut self.bit_reader)?;
+
+            //check code
+            if code < 256 {
+                //literal, so just use huffman codes and read as argb
+                let red = tree[RED].read_symbol(&mut self.bit_reader)?;
+                let blue = tree[BLUE].read_symbol(&mut self.bit_reader)?;
+                let alpha = tree[ALPHA].read_symbol(&mut self.bit_reader)?;
+
+                data[index] = (u32::from(alpha) << 24)
+                    + (u32::from(red) << 16)
+                    + (u32::from(code) << 8)
+                    + u32::from(blue);
+
+                index += 1;
+                x += 1;
+                if x >= width {
+                    x = 0;
+                    y += 1;
+                }
+            } else if code < 256 + 24 {
+                //backward reference, so go back and use that to add image data
+                let length_symbol = code - 256;
+                let length = Self::get_copy_distance(&mut self.bit_reader, length_symbol)?;
+
+                let dist_symbol = tree[DIST].read_symbol(&mut self.bit_reader)?;
+                let dist_code = Self::get_copy_distance(&mut self.bit_reader, dist_symbol)?;
+                let dist = Self::plane_code_to_distance(width, dist_code);
+
+                if index < dist || num_values - index < length {
+                    return Err(DecoderError::BitStreamError.into());
+                }
+
+                for i in 0..length {
+                    data[index + i] = data[index + i - dist];
+                }
+                index += length;
+                x += u16::try_from(length).unwrap();
+                while x >= width {
+                    x -= width;
+                    y += 1;
+                }
+                if index < num_values {
+                    let index = huffman_info.get_huff_index(x, y);
+                    tree = &huffman_info.huffman_code_groups[index];
+                }
+            } else {
+                //color cache, so use previously stored pixels to get this pixel
+                let key = code - 256 - 24;
+
+                if let Some(color_cache) = huffman_info.color_cache.as_mut() {
+                    //cache old colors
+                    while last_cached < index {
+                        color_cache.insert(data[last_cached]);
+                        last_cached += 1;
+                    }
+                    data[index] = color_cache.lookup(key.into())?;
+                } else {
+                    return Err(DecoderError::BitStreamError.into());
+                }
+                index += 1;
+                x += 1;
+                if x >= width {
+                    x = 0;
+                    y += 1;
+                }
+            }
+        }
+
+        Ok(data)
+    }
+
+    /// Reads color cache data from the bitstream
+    fn read_color_cache(&mut self) -> ImageResult<Option<u8>> {
+        if self.bit_reader.read_bits::<u8>(1)? == 1 {
+            let code_bits = self.bit_reader.read_bits::<u8>(4)?;
+
+            if !(1..=11).contains(&code_bits) {
+                return Err(DecoderError::InvalidColorCacheBits(code_bits).into());
+            }
+
+            Ok(Some(code_bits))
+        } else {
+            Ok(None)
+        }
+    }
+
+    /// Gets the copy distance from the prefix code and bitstream
+    fn get_copy_distance(bit_reader: &mut BitReader, prefix_code: u16) -> ImageResult<usize> {
+        if prefix_code < 4 {
+            return Ok(usize::from(prefix_code + 1));
+        }
+        let extra_bits: u8 = ((prefix_code - 2) >> 1).try_into().unwrap();
+        let offset = (2 + (usize::from(prefix_code) & 1)) << extra_bits;
+
+        Ok(offset + bit_reader.read_bits::<usize>(extra_bits)? + 1)
+    }
+
+    /// Gets distance to pixel
+    fn plane_code_to_distance(xsize: u16, plane_code: usize) -> usize {
+        if plane_code > 120 {
+            plane_code - 120
+        } else {
+            let (xoffset, yoffset) = DISTANCE_MAP[plane_code - 1];
+
+            let dist = i32::from(xoffset) + i32::from(yoffset) * i32::from(xsize);
+            if dist < 1 {
+                return 1;
+            }
+            dist.try_into().unwrap()
+        }
+    }
+}
+
+#[derive(Debug, Clone)]
+struct HuffmanInfo {
+    xsize: u16,
+    _ysize: u16,
+    color_cache: Option<ColorCache>,
+    image: Vec<u32>,
+    bits: u8,
+    mask: u16,
+    huffman_code_groups: Vec<HuffmanCodeGroup>,
+}
+
+impl HuffmanInfo {
+    fn get_huff_index(&self, x: u16, y: u16) -> usize {
+        if self.bits == 0 {
+            return 0;
+        }
+        let position = usize::from((y >> self.bits) * self.xsize + (x >> self.bits));
+        let meta_huff_code: usize = self.image[position].try_into().unwrap();
+        meta_huff_code
+    }
+}
+
+#[derive(Debug, Clone)]
+struct ColorCache {
+    color_cache_bits: u8,
+    color_cache: Vec<u32>,
+}
+
+impl ColorCache {
+    fn insert(&mut self, color: u32) {
+        let index = (0x1e35a7bdu32.overflowing_mul(color).0) >> (32 - self.color_cache_bits);
+        self.color_cache[index as usize] = color;
+    }
+
+    fn lookup(&self, index: usize) -> ImageResult<u32> {
+        match self.color_cache.get(index) {
+            Some(&value) => Ok(value),
+            None => Err(DecoderError::BitStreamError.into()),
+        }
+    }
+}
+
+#[derive(Debug, Clone)]
+pub(crate) struct BitReader {
+    buf: Vec<u8>,
+    index: usize,
+    bit_count: u8,
+}
+
+impl BitReader {
+    fn new() -> BitReader {
+        BitReader {
+            buf: Vec::new(),
+            index: 0,
+            bit_count: 0,
+        }
+    }
+
+    fn init(&mut self, buf: Vec<u8>) {
+        self.buf = buf;
+    }
+
+    pub(crate) fn read_bits<T>(&mut self, num: u8) -> ImageResult<T>
+    where
+        T: num_traits::Unsigned + Shl<u8, Output = T> + AddAssign<T> + From<bool>,
+    {
+        let mut value: T = T::zero();
+
+        for i in 0..num {
+            if self.buf.len() <= self.index {
+                return Err(DecoderError::BitStreamError.into());
+            }
+            let bit_true = self.buf[self.index] & (1 << self.bit_count) != 0;
+            value += T::from(bit_true) << i;
+            self.bit_count = if self.bit_count == 7 {
+                self.index += 1;
+                0
+            } else {
+                self.bit_count + 1
+            };
+        }
+
+        Ok(value)
+    }
+}
+
+#[derive(Debug, Clone, Default)]
+pub(crate) struct LosslessFrame {
+    pub(crate) width: u16,
+    pub(crate) height: u16,
+
+    pub(crate) buf: Vec<u32>,
+}
+
+impl LosslessFrame {
+    /// Fills a buffer by converting from argb to rgba
+    pub(crate) fn fill_rgba(&self, buf: &mut [u8]) {
+        for (&argb_val, chunk) in self.buf.iter().zip(buf.chunks_exact_mut(4)) {
+            chunk[0] = ((argb_val >> 16) & 0xff).try_into().unwrap();
+            chunk[1] = ((argb_val >> 8) & 0xff).try_into().unwrap();
+            chunk[2] = (argb_val & 0xff).try_into().unwrap();
+            chunk[3] = ((argb_val >> 24) & 0xff).try_into().unwrap();
+        }
+    }
+
+    /// Get buffer size from the image
+    pub(crate) fn get_buf_size(&self) -> usize {
+        usize::from(self.width) * usize::from(self.height) * 4
+    }
+
+    /// Fills a buffer with just the green values from the lossless decoding
+    /// Used in extended alpha decoding
+    pub(crate) fn fill_green(&self, buf: &mut [u8]) {
+        for (&argb_val, buf_value) in self.buf.iter().zip(buf.iter_mut()) {
+            *buf_value = ((argb_val >> 8) & 0xff).try_into().unwrap();
+        }
+    }
+}
+
+#[cfg(test)]
+mod test {
+
+    use super::BitReader;
+
+    #[test]
+    fn bit_read_test() {
+        let mut bit_reader = BitReader::new();
+
+        //10011100 01000001 11100001
+        let buf = vec![0x9C, 0x41, 0xE1];
+
+        bit_reader.init(buf);
+
+        assert_eq!(bit_reader.read_bits::<u8>(3).unwrap(), 4); //100
+        assert_eq!(bit_reader.read_bits::<u8>(2).unwrap(), 3); //11
+        assert_eq!(bit_reader.read_bits::<u8>(6).unwrap(), 12); //001100
+        assert_eq!(bit_reader.read_bits::<u16>(10).unwrap(), 40); //0000101000
+        assert_eq!(bit_reader.read_bits::<u8>(3).unwrap(), 7); //111
+    }
+
+    #[test]
+    fn bit_read_error_test() {
+        let mut bit_reader = BitReader::new();
+
+        //01101010
+        let buf = vec![0x6A];
+
+        bit_reader.init(buf);
+
+        assert_eq!(bit_reader.read_bits::<u8>(3).unwrap(), 2); //010
+        assert_eq!(bit_reader.read_bits::<u8>(5).unwrap(), 13); //01101
+        assert!(bit_reader.read_bits::<u8>(4).is_err()); //error
+    }
+}

vendor/image/src/codecs/webp/lossless_transform.rs

@@ -0,0 +1,464 @@
+use std::convert::TryFrom;
+use std::convert::TryInto;
+
+use super::lossless::subsample_size;
+use super::lossless::DecoderError;
+
+#[derive(Debug, Clone)]
+pub(crate) enum TransformType {
+    PredictorTransform {
+        size_bits: u8,
+        predictor_data: Vec<u32>,
+    },
+    ColorTransform {
+        size_bits: u8,
+        transform_data: Vec<u32>,
+    },
+    SubtractGreen,
+    ColorIndexingTransform {
+        table_size: u16,
+        table_data: Vec<u32>,
+    },
+}
+
+impl TransformType {
+    /// Applies a transform to the image data
+    pub(crate) fn apply_transform(
+        &self,
+        image_data: &mut Vec<u32>,
+        width: u16,
+        height: u16,
+    ) -> Result<(), DecoderError> {
+        match self {
+            TransformType::PredictorTransform {
+                size_bits,
+                predictor_data,
+            } => {
+                let block_xsize = usize::from(subsample_size(width, *size_bits));
+                let width = usize::from(width);
+                let height = usize::from(height);
+
+                if image_data.len() < width * height {
+                    return Err(DecoderError::TransformError);
+                }
+
+                //handle top and left borders specially
+                //this involves ignoring mode and just setting prediction values like this
+                image_data[0] = add_pixels(image_data[0], 0xff000000);
+
+                for x in 1..width {
+                    image_data[x] = add_pixels(image_data[x], get_left(image_data, x, 0, width));
+                }
+
+                for y in 1..height {
+                    image_data[y * width] =
+                        add_pixels(image_data[y * width], get_top(image_data, 0, y, width));
+                }
+
+                for y in 1..height {
+                    for x in 1..width {
+                        let block_index = (y >> size_bits) * block_xsize + (x >> size_bits);
+
+                        let index = y * width + x;
+
+                        let green = (predictor_data[block_index] >> 8) & 0xff;
+
+                        match green {
+                            0 => image_data[index] = add_pixels(image_data[index], 0xff000000),
+                            1 => {
+                                image_data[index] =
+                                    add_pixels(image_data[index], get_left(image_data, x, y, width))
+                            }
+                            2 => {
+                                image_data[index] =
+                                    add_pixels(image_data[index], get_top(image_data, x, y, width))
+                            }
+                            3 => {
+                                image_data[index] = add_pixels(
+                                    image_data[index],
+                                    get_top_right(image_data, x, y, width),
+                                )
+                            }
+                            4 => {
+                                image_data[index] = add_pixels(
+                                    image_data[index],
+                                    get_top_left(image_data, x, y, width),
+                                )
+                            }
+                            5 => {
+                                image_data[index] = add_pixels(image_data[index], {
+                                    let first = average2(
+                                        get_left(image_data, x, y, width),
+                                        get_top_right(image_data, x, y, width),
+                                    );
+                                    average2(first, get_top(image_data, x, y, width))
+                                })
+                            }
+                            6 => {
+                                image_data[index] = add_pixels(
+                                    image_data[index],
+                                    average2(
+                                        get_left(image_data, x, y, width),
+                                        get_top_left(image_data, x, y, width),
+                                    ),
+                                )
+                            }
+                            7 => {
+                                image_data[index] = add_pixels(
+                                    image_data[index],
+                                    average2(
+                                        get_left(image_data, x, y, width),
+                                        get_top(image_data, x, y, width),
+                                    ),
+                                )
+                            }
+                            8 => {
+                                image_data[index] = add_pixels(
+                                    image_data[index],
+                                    average2(
+                                        get_top_left(image_data, x, y, width),
+                                        get_top(image_data, x, y, width),
+                                    ),
+                                )
+                            }
+                            9 => {
+                                image_data[index] = add_pixels(
+                                    image_data[index],
+                                    average2(
+                                        get_top(image_data, x, y, width),
+                                        get_top_right(image_data, x, y, width),
+                                    ),
+                                )
+                            }
+                            10 => {
+                                image_data[index] = add_pixels(image_data[index], {
+                                    let first = average2(
+                                        get_left(image_data, x, y, width),
+                                        get_top_left(image_data, x, y, width),
+                                    );
+                                    let second = average2(
+                                        get_top(image_data, x, y, width),
+                                        get_top_right(image_data, x, y, width),
+                                    );
+                                    average2(first, second)
+                                })
+                            }
+                            11 => {
+                                image_data[index] = add_pixels(
+                                    image_data[index],
+                                    select(
+                                        get_left(image_data, x, y, width),
+                                        get_top(image_data, x, y, width),
+                                        get_top_left(image_data, x, y, width),
+                                    ),
+                                )
+                            }
+                            12 => {
+                                image_data[index] = add_pixels(
+                                    image_data[index],
+                                    clamp_add_subtract_full(
+                                        get_left(image_data, x, y, width),
+                                        get_top(image_data, x, y, width),
+                                        get_top_left(image_data, x, y, width),
+                                    ),
+                                )
+                            }
+                            13 => {
+                                image_data[index] = add_pixels(image_data[index], {
+                                    let first = average2(
+                                        get_left(image_data, x, y, width),
+                                        get_top(image_data, x, y, width),
+                                    );
+                                    clamp_add_subtract_half(
+                                        first,
+                                        get_top_left(image_data, x, y, width),
+                                    )
+                                })
+                            }
+                            _ => {}
+                        }
+                    }
+                }
+            }
+            TransformType::ColorTransform {
+                size_bits,
+                transform_data,
+            } => {
+                let block_xsize = usize::from(subsample_size(width, *size_bits));
+                let width = usize::from(width);
+                let height = usize::from(height);
+
+                for y in 0..height {
+                    for x in 0..width {
+                        let block_index = (y >> size_bits) * block_xsize + (x >> size_bits);
+
+                        let index = y * width + x;
+
+                        let multiplier =
+                            ColorTransformElement::from_color_code(transform_data[block_index]);
+
+                        image_data[index] = transform_color(&multiplier, image_data[index]);
+                    }
+                }
+            }
+            TransformType::SubtractGreen => {
+                let width = usize::from(width);
+                for y in 0..usize::from(height) {
+                    for x in 0..width {
+                        image_data[y * width + x] = add_green(image_data[y * width + x]);
+                    }
+                }
+            }
+            TransformType::ColorIndexingTransform {
+                table_size,
+                table_data,
+            } => {
+                let mut new_image_data =
+                    Vec::with_capacity(usize::from(width) * usize::from(height));
+
+                let table_size = *table_size;
+                let width_bits: u8 = if table_size <= 2 {
+                    3
+                } else if table_size <= 4 {
+                    2
+                } else if table_size <= 16 {
+                    1
+                } else {
+                    0
+                };
+
+                let bits_per_pixel = 8 >> width_bits;
+                let mask = (1 << bits_per_pixel) - 1;
+
+                let mut src = 0;
+                let width = usize::from(width);
+
+                let pixels_per_byte = 1 << width_bits;
+                let count_mask = pixels_per_byte - 1;
+                let mut packed_pixels = 0;
+
+                for _y in 0..usize::from(height) {
+                    for x in 0..width {
+                        if (x & count_mask) == 0 {
+                            packed_pixels = (image_data[src] >> 8) & 0xff;
+                            src += 1;
+                        }
+
+                        let pixels: usize = (packed_pixels & mask).try_into().unwrap();
+                        let new_val = if pixels >= table_size.into() {
+                            0x00000000
+                        } else {
+                            table_data[pixels]
+                        };
+
+                        new_image_data.push(new_val);
+
+                        packed_pixels >>= bits_per_pixel;
+                    }
+                }
+
+                *image_data = new_image_data;
+            }
+        }
+
+        Ok(())
+    }
+}
+
+//predictor functions
+
+/// Adds 2 pixels mod 256 for each pixel
+pub(crate) fn add_pixels(a: u32, b: u32) -> u32 {
+    let new_alpha = ((a >> 24) + (b >> 24)) & 0xff;
+    let new_red = (((a >> 16) & 0xff) + ((b >> 16) & 0xff)) & 0xff;
+    let new_green = (((a >> 8) & 0xff) + ((b >> 8) & 0xff)) & 0xff;
+    let new_blue = ((a & 0xff) + (b & 0xff)) & 0xff;
+
+    (new_alpha << 24) + (new_red << 16) + (new_green << 8) + new_blue
+}
+
+/// Get left pixel
+fn get_left(data: &[u32], x: usize, y: usize, width: usize) -> u32 {
+    data[y * width + x - 1]
+}
+
+/// Get top pixel
+fn get_top(data: &[u32], x: usize, y: usize, width: usize) -> u32 {
+    data[(y - 1) * width + x]
+}
+
+/// Get pixel to top right
+fn get_top_right(data: &[u32], x: usize, y: usize, width: usize) -> u32 {
+    // if x == width - 1 this gets the left most pixel of the current row
+    // as described in the specification
+    data[(y - 1) * width + x + 1]
+}
+
+/// Get pixel to top left
+fn get_top_left(data: &[u32], x: usize, y: usize, width: usize) -> u32 {
+    data[(y - 1) * width + x - 1]
+}
+
+/// Get average of 2 pixels
+fn average2(a: u32, b: u32) -> u32 {
+    let mut avg = 0u32;
+    for i in 0..4 {
+        let sub_a: u8 = ((a >> (i * 8)) & 0xff).try_into().unwrap();
+        let sub_b: u8 = ((b >> (i * 8)) & 0xff).try_into().unwrap();
+        avg |= u32::from(sub_average2(sub_a, sub_b)) << (i * 8);
+    }
+    avg
+}
+
+/// Get average of 2 bytes
+fn sub_average2(a: u8, b: u8) -> u8 {
+    ((u16::from(a) + u16::from(b)) / 2).try_into().unwrap()
+}
+
+/// Get a specific byte from argb pixel
+fn get_byte(val: u32, byte: u8) -> u8 {
+    ((val >> (byte * 8)) & 0xff).try_into().unwrap()
+}
+
+/// Get byte as i32 for convenience
+fn get_byte_i32(val: u32, byte: u8) -> i32 {
+    i32::from(get_byte(val, byte))
+}
+
+/// Select left or top byte
+fn select(left: u32, top: u32, top_left: u32) -> u32 {
+    let predict_alpha = get_byte_i32(left, 3) + get_byte_i32(top, 3) - get_byte_i32(top_left, 3);
+    let predict_red = get_byte_i32(left, 2) + get_byte_i32(top, 2) - get_byte_i32(top_left, 2);
+    let predict_green = get_byte_i32(left, 1) + get_byte_i32(top, 1) - get_byte_i32(top_left, 1);
+    let predict_blue = get_byte_i32(left, 0) + get_byte_i32(top, 0) - get_byte_i32(top_left, 0);
+
+    let predict_left = i32::abs(predict_alpha - get_byte_i32(left, 3))
+        + i32::abs(predict_red - get_byte_i32(left, 2))
+        + i32::abs(predict_green - get_byte_i32(left, 1))
+        + i32::abs(predict_blue - get_byte_i32(left, 0));
+    let predict_top = i32::abs(predict_alpha - get_byte_i32(top, 3))
+        + i32::abs(predict_red - get_byte_i32(top, 2))
+        + i32::abs(predict_green - get_byte_i32(top, 1))
+        + i32::abs(predict_blue - get_byte_i32(top, 0));
+
+    if predict_left < predict_top {
+        left
+    } else {
+        top
+    }
+}
+
+/// Clamp a to [0, 255]
+fn clamp(a: i32) -> i32 {
+    if a < 0 {
+        0
+    } else if a > 255 {
+        255
+    } else {
+        a
+    }
+}
+
+/// Clamp add subtract full on one part
+fn clamp_add_subtract_full_sub(a: i32, b: i32, c: i32) -> i32 {
+    clamp(a + b - c)
+}
+
+/// Clamp add subtract half on one part
+fn clamp_add_subtract_half_sub(a: i32, b: i32) -> i32 {
+    clamp(a + (a - b) / 2)
+}
+
+/// Clamp add subtract full on 3 pixels
+fn clamp_add_subtract_full(a: u32, b: u32, c: u32) -> u32 {
+    let mut value: u32 = 0;
+    for i in 0..4u8 {
+        let sub_a: i32 = ((a >> (i * 8)) & 0xff).try_into().unwrap();
+        let sub_b: i32 = ((b >> (i * 8)) & 0xff).try_into().unwrap();
+        let sub_c: i32 = ((c >> (i * 8)) & 0xff).try_into().unwrap();
+        value |=
+            u32::try_from(clamp_add_subtract_full_sub(sub_a, sub_b, sub_c)).unwrap() << (i * 8);
+    }
+    value
+}
+
+/// Clamp add subtract half on 2 pixels
+fn clamp_add_subtract_half(a: u32, b: u32) -> u32 {
+    let mut value = 0;
+    for i in 0..4u8 {
+        let sub_a: i32 = ((a >> (i * 8)) & 0xff).try_into().unwrap();
+        let sub_b: i32 = ((b >> (i * 8)) & 0xff).try_into().unwrap();
+        value |= u32::try_from(clamp_add_subtract_half_sub(sub_a, sub_b)).unwrap() << (i * 8);
+    }
+
+    value
+}
+
+//color transform
+
+#[derive(Debug, Clone, Copy)]
+struct ColorTransformElement {
+    green_to_red: u8,
+    green_to_blue: u8,
+    red_to_blue: u8,
+}
+
+impl ColorTransformElement {
+    fn from_color_code(color_code: u32) -> ColorTransformElement {
+        ColorTransformElement {
+            green_to_red: (color_code & 0xff).try_into().unwrap(),
+            green_to_blue: ((color_code >> 8) & 0xff).try_into().unwrap(),
+            red_to_blue: ((color_code >> 16) & 0xff).try_into().unwrap(),
+        }
+    }
+}
+
+/// Does color transform on red and blue transformed by green
+fn color_transform(red: u8, blue: u8, green: u8, trans: &ColorTransformElement) -> (u8, u8) {
+    let mut temp_red = u32::from(red);
+    let mut temp_blue = u32::from(blue);
+
+    //as does the conversion from u8 to signed two's complement i8 required
+    temp_red += color_transform_delta(trans.green_to_red as i8, green as i8);
+    temp_blue += color_transform_delta(trans.green_to_blue as i8, green as i8);
+    temp_blue += color_transform_delta(trans.red_to_blue as i8, temp_red as i8);
+
+    (
+        (temp_red & 0xff).try_into().unwrap(),
+        (temp_blue & 0xff).try_into().unwrap(),
+    )
+}
+
+/// Does color transform on 2 numbers
+fn color_transform_delta(t: i8, c: i8) -> u32 {
+    ((i16::from(t) * i16::from(c)) as u32) >> 5
+}
+
+// Does color transform on a pixel with a color transform element
+fn transform_color(multiplier: &ColorTransformElement, color_value: u32) -> u32 {
+    let alpha = get_byte(color_value, 3);
+    let red = get_byte(color_value, 2);
+    let green = get_byte(color_value, 1);
+    let blue = get_byte(color_value, 0);
+
+    let (new_red, new_blue) = color_transform(red, blue, green, multiplier);
+
+    (u32::from(alpha) << 24)
+        + (u32::from(new_red) << 16)
+        + (u32::from(green) << 8)
+        + u32::from(new_blue)
+}
+
+//subtract green function
+
+/// Adds green to red and blue of a pixel
+fn add_green(argb: u32) -> u32 {
+    let red = (argb >> 16) & 0xff;
+    let green = (argb >> 8) & 0xff;
+    let blue = argb & 0xff;
+
+    let new_red = (red + green) & 0xff;
+    let new_blue = (blue + green) & 0xff;
+
+    (argb & 0xff00ff00) | (new_red << 16) | (new_blue)
+}

vendor/image/src/codecs/webp/mod.rs

@@ -0,0 +1,28 @@
+//! Decoding and Encoding of WebP Images
+
+#[cfg(feature = "webp-encoder")]
+pub use self::encoder::{WebPEncoder, WebPQuality};
+
+#[cfg(feature = "webp-encoder")]
+mod encoder;
+
+#[cfg(feature = "webp")]
+pub use self::decoder::WebPDecoder;
+
+#[cfg(feature = "webp")]
+mod decoder;
+#[cfg(feature = "webp")]
+mod extended;
+#[cfg(feature = "webp")]
+mod huffman;
+#[cfg(feature = "webp")]
+mod loop_filter;
+#[cfg(feature = "webp")]
+mod lossless;
+#[cfg(feature = "webp")]
+mod lossless_transform;
+#[cfg(feature = "webp")]
+mod transform;
+
+#[cfg(feature = "webp")]
+pub mod vp8;

vendor/image/src/codecs/webp/transform.rs

@@ -0,0 +1,77 @@
+static CONST1: i64 = 20091;
+static CONST2: i64 = 35468;
+
+pub(crate) fn idct4x4(block: &mut [i32]) {
+    // The intermediate results may overflow the types, so we stretch the type.
+    fn fetch(block: &mut [i32], idx: usize) -> i64 {
+        i64::from(block[idx])
+    }
+
+    for i in 0usize..4 {
+        let a1 = fetch(block, i) + fetch(block, 8 + i);
+        let b1 = fetch(block, i) - fetch(block, 8 + i);
+
+        let t1 = (fetch(block, 4 + i) * CONST2) >> 16;
+        let t2 = fetch(block, 12 + i) + ((fetch(block, 12 + i) * CONST1) >> 16);
+        let c1 = t1 - t2;
+
+        let t1 = fetch(block, 4 + i) + ((fetch(block, 4 + i) * CONST1) >> 16);
+        let t2 = (fetch(block, 12 + i) * CONST2) >> 16;
+        let d1 = t1 + t2;
+
+        block[i] = (a1 + d1) as i32;
+        block[4 + i] = (b1 + c1) as i32;
+        block[4 * 3 + i] = (a1 - d1) as i32;
+        block[4 * 2 + i] = (b1 - c1) as i32;
+    }
+
+    for i in 0usize..4 {
+        let a1 = fetch(block, 4 * i) + fetch(block, 4 * i + 2);
+        let b1 = fetch(block, 4 * i) - fetch(block, 4 * i + 2);
+
+        let t1 = (fetch(block, 4 * i + 1) * CONST2) >> 16;
+        let t2 = fetch(block, 4 * i + 3) + ((fetch(block, 4 * i + 3) * CONST1) >> 16);
+        let c1 = t1 - t2;
+
+        let t1 = fetch(block, 4 * i + 1) + ((fetch(block, 4 * i + 1) * CONST1) >> 16);
+        let t2 = (fetch(block, 4 * i + 3) * CONST2) >> 16;
+        let d1 = t1 + t2;
+
+        block[4 * i] = ((a1 + d1 + 4) >> 3) as i32;
+        block[4 * i + 3] = ((a1 - d1 + 4) >> 3) as i32;
+        block[4 * i + 1] = ((b1 + c1 + 4) >> 3) as i32;
+        block[4 * i + 2] = ((b1 - c1 + 4) >> 3) as i32;
+    }
+}
+
+// 14.3
+pub(crate) fn iwht4x4(block: &mut [i32]) {
+    for i in 0usize..4 {
+        let a1 = block[i] + block[12 + i];
+        let b1 = block[4 + i] + block[8 + i];
+        let c1 = block[4 + i] - block[8 + i];
+        let d1 = block[i] - block[12 + i];
+
+        block[i] = a1 + b1;
+        block[4 + i] = c1 + d1;
+        block[8 + i] = a1 - b1;
+        block[12 + i] = d1 - c1;
+    }
+
+    for i in 0usize..4 {
+        let a1 = block[4 * i] + block[4 * i + 3];
+        let b1 = block[4 * i + 1] + block[4 * i + 2];
+        let c1 = block[4 * i + 1] - block[4 * i + 2];
+        let d1 = block[4 * i] - block[4 * i + 3];
+
+        let a2 = a1 + b1;
+        let b2 = c1 + d1;
+        let c2 = a1 - b1;
+        let d2 = d1 - c1;
+
+        block[4 * i] = (a2 + 3) >> 3;
+        block[4 * i + 1] = (b2 + 3) >> 3;
+        block[4 * i + 2] = (c2 + 3) >> 3;
+        block[4 * i + 3] = (d2 + 3) >> 3;
+    }
+}

vendor/image/src/color.rs

@@ -0,0 +1,985 @@
+use std::ops::{Index, IndexMut};
+
+use num_traits::{NumCast, ToPrimitive, Zero};
+
+use crate::traits::{Enlargeable, Pixel, Primitive};
+
+/// An enumeration over supported color types and bit depths
+#[derive(Copy, PartialEq, Eq, Debug, Clone, Hash)]
+#[non_exhaustive]
+pub enum ColorType {
+    /// Pixel is 8-bit luminance
+    L8,
+    /// Pixel is 8-bit luminance with an alpha channel
+    La8,
+    /// Pixel contains 8-bit R, G and B channels
+    Rgb8,
+    /// Pixel is 8-bit RGB with an alpha channel
+    Rgba8,
+
+    /// Pixel is 16-bit luminance
+    L16,
+    /// Pixel is 16-bit luminance with an alpha channel
+    La16,
+    /// Pixel is 16-bit RGB
+    Rgb16,
+    /// Pixel is 16-bit RGBA
+    Rgba16,
+
+    /// Pixel is 32-bit float RGB
+    Rgb32F,
+    /// Pixel is 32-bit float RGBA
+    Rgba32F,
+}
+
+impl ColorType {
+    /// Returns the number of bytes contained in a pixel of `ColorType` ```c```
+    pub fn bytes_per_pixel(self) -> u8 {
+        match self {
+            ColorType::L8 => 1,
+            ColorType::L16 | ColorType::La8 => 2,
+            ColorType::Rgb8 => 3,
+            ColorType::Rgba8 | ColorType::La16 => 4,
+            ColorType::Rgb16 => 6,
+            ColorType::Rgba16 => 8,
+            ColorType::Rgb32F => 3 * 4,
+            ColorType::Rgba32F => 4 * 4,
+        }
+    }
+
+    /// Returns if there is an alpha channel.
+    pub fn has_alpha(self) -> bool {
+        use ColorType::*;
+        match self {
+            L8 | L16 | Rgb8 | Rgb16 | Rgb32F => false,
+            La8 | Rgba8 | La16 | Rgba16 | Rgba32F => true,
+        }
+    }
+
+    /// Returns false if the color scheme is grayscale, true otherwise.
+    pub fn has_color(self) -> bool {
+        use ColorType::*;
+        match self {
+            L8 | L16 | La8 | La16 => false,
+            Rgb8 | Rgb16 | Rgba8 | Rgba16 | Rgb32F | Rgba32F => true,
+        }
+    }
+
+    /// Returns the number of bits contained in a pixel of `ColorType` ```c``` (which will always be
+    /// a multiple of 8).
+    pub fn bits_per_pixel(self) -> u16 {
+        <u16 as From<u8>>::from(self.bytes_per_pixel()) * 8
+    }
+
+    /// Returns the number of color channels that make up this pixel
+    pub fn channel_count(self) -> u8 {
+        let e: ExtendedColorType = self.into();
+        e.channel_count()
+    }
+}
+
+/// An enumeration of color types encountered in image formats.
+///
+/// This is not exhaustive over all existing image formats but should be granular enough to allow
+/// round tripping of decoding and encoding as much as possible. The variants will be extended as
+/// necessary to enable this.
+///
+/// Another purpose is to advise users of a rough estimate of the accuracy and effort of the
+/// decoding from and encoding to such an image format.
+#[derive(Copy, PartialEq, Eq, Debug, Clone, Hash)]
+#[non_exhaustive]
+pub enum ExtendedColorType {
+    /// Pixel is 8-bit alpha
+    A8,
+    /// Pixel is 1-bit luminance
+    L1,
+    /// Pixel is 1-bit luminance with an alpha channel
+    La1,
+    /// Pixel contains 1-bit R, G and B channels
+    Rgb1,
+    /// Pixel is 1-bit RGB with an alpha channel
+    Rgba1,
+    /// Pixel is 2-bit luminance
+    L2,
+    /// Pixel is 2-bit luminance with an alpha channel
+    La2,
+    /// Pixel contains 2-bit R, G and B channels
+    Rgb2,
+    /// Pixel is 2-bit RGB with an alpha channel
+    Rgba2,
+    /// Pixel is 4-bit luminance
+    L4,
+    /// Pixel is 4-bit luminance with an alpha channel
+    La4,
+    /// Pixel contains 4-bit R, G and B channels
+    Rgb4,
+    /// Pixel is 4-bit RGB with an alpha channel
+    Rgba4,
+    /// Pixel is 8-bit luminance
+    L8,
+    /// Pixel is 8-bit luminance with an alpha channel
+    La8,
+    /// Pixel contains 8-bit R, G and B channels
+    Rgb8,
+    /// Pixel is 8-bit RGB with an alpha channel
+    Rgba8,
+    /// Pixel is 16-bit luminance
+    L16,
+    /// Pixel is 16-bit luminance with an alpha channel
+    La16,
+    /// Pixel contains 16-bit R, G and B channels
+    Rgb16,
+    /// Pixel is 16-bit RGB with an alpha channel
+    Rgba16,
+    /// Pixel contains 8-bit B, G and R channels
+    Bgr8,
+    /// Pixel is 8-bit BGR with an alpha channel
+    Bgra8,
+
+    // TODO f16 types?
+    /// Pixel is 32-bit float RGB
+    Rgb32F,
+    /// Pixel is 32-bit float RGBA
+    Rgba32F,
+
+    /// Pixel is of unknown color type with the specified bits per pixel. This can apply to pixels
+    /// which are associated with an external palette. In that case, the pixel value is an index
+    /// into the palette.
+    Unknown(u8),
+}
+
+impl ExtendedColorType {
+    /// Get the number of channels for colors of this type.
+    ///
+    /// Note that the `Unknown` variant returns a value of `1` since pixels can only be treated as
+    /// an opaque datum by the library.
+    pub fn channel_count(self) -> u8 {
+        match self {
+            ExtendedColorType::A8
+            | ExtendedColorType::L1
+            | ExtendedColorType::L2
+            | ExtendedColorType::L4
+            | ExtendedColorType::L8
+            | ExtendedColorType::L16
+            | ExtendedColorType::Unknown(_) => 1,
+            ExtendedColorType::La1
+            | ExtendedColorType::La2
+            | ExtendedColorType::La4
+            | ExtendedColorType::La8
+            | ExtendedColorType::La16 => 2,
+            ExtendedColorType::Rgb1
+            | ExtendedColorType::Rgb2
+            | ExtendedColorType::Rgb4
+            | ExtendedColorType::Rgb8
+            | ExtendedColorType::Rgb16
+            | ExtendedColorType::Rgb32F
+            | ExtendedColorType::Bgr8 => 3,
+            ExtendedColorType::Rgba1
+            | ExtendedColorType::Rgba2
+            | ExtendedColorType::Rgba4
+            | ExtendedColorType::Rgba8
+            | ExtendedColorType::Rgba16
+            | ExtendedColorType::Rgba32F
+            | ExtendedColorType::Bgra8 => 4,
+        }
+    }
+}
+impl From<ColorType> for ExtendedColorType {
+    fn from(c: ColorType) -> Self {
+        match c {
+            ColorType::L8 => ExtendedColorType::L8,
+            ColorType::La8 => ExtendedColorType::La8,
+            ColorType::Rgb8 => ExtendedColorType::Rgb8,
+            ColorType::Rgba8 => ExtendedColorType::Rgba8,
+            ColorType::L16 => ExtendedColorType::L16,
+            ColorType::La16 => ExtendedColorType::La16,
+            ColorType::Rgb16 => ExtendedColorType::Rgb16,
+            ColorType::Rgba16 => ExtendedColorType::Rgba16,
+            ColorType::Rgb32F => ExtendedColorType::Rgb32F,
+            ColorType::Rgba32F => ExtendedColorType::Rgba32F,
+        }
+    }
+}
+
+macro_rules! define_colors {
+    {$(
+        $(#[$doc:meta])*
+        pub struct $ident:ident<T: $($bound:ident)*>([T; $channels:expr, $alphas:expr])
+            = $interpretation:literal;
+    )*} => {
+
+$( // START Structure definitions
+
+$(#[$doc])*
+#[derive(PartialEq, Eq, Clone, Debug, Copy, Hash)]
+#[repr(C)]
+#[allow(missing_docs)]
+pub struct $ident<T> (pub [T; $channels]);
+
+impl<T: $($bound+)*> Pixel for $ident<T> {
+    type Subpixel = T;
+
+    const CHANNEL_COUNT: u8 = $channels;
+
+    #[inline(always)]
+    fn channels(&self) -> &[T] {
+        &self.0
+    }
+
+    #[inline(always)]
+    fn channels_mut(&mut self) -> &mut [T] {
+        &mut self.0
+    }
+
+    const COLOR_MODEL: &'static str = $interpretation;
+
+    fn channels4(&self) -> (T, T, T, T) {
+        const CHANNELS: usize = $channels;
+        let mut channels = [T::DEFAULT_MAX_VALUE; 4];
+        channels[0..CHANNELS].copy_from_slice(&self.0);
+        (channels[0], channels[1], channels[2], channels[3])
+    }
+
+    fn from_channels(a: T, b: T, c: T, d: T,) -> $ident<T> {
+        const CHANNELS: usize = $channels;
+        *<$ident<T> as Pixel>::from_slice(&[a, b, c, d][..CHANNELS])
+    }
+
+    fn from_slice(slice: &[T]) -> &$ident<T> {
+        assert_eq!(slice.len(), $channels);
+        unsafe { &*(slice.as_ptr() as *const $ident<T>) }
+    }
+    fn from_slice_mut(slice: &mut [T]) -> &mut $ident<T> {
+        assert_eq!(slice.len(), $channels);
+        unsafe { &mut *(slice.as_mut_ptr() as *mut $ident<T>) }
+    }
+
+    fn to_rgb(&self) -> Rgb<T> {
+        let mut pix = Rgb([Zero::zero(), Zero::zero(), Zero::zero()]);
+        pix.from_color(self);
+        pix
+    }
+
+    fn to_rgba(&self) -> Rgba<T> {
+        let mut pix = Rgba([Zero::zero(), Zero::zero(), Zero::zero(), Zero::zero()]);
+        pix.from_color(self);
+        pix
+    }
+
+    fn to_luma(&self) -> Luma<T> {
+        let mut pix = Luma([Zero::zero()]);
+        pix.from_color(self);
+        pix
+    }
+
+    fn to_luma_alpha(&self) -> LumaA<T> {
+        let mut pix = LumaA([Zero::zero(), Zero::zero()]);
+        pix.from_color(self);
+        pix
+    }
+
+    fn map<F>(& self, f: F) -> $ident<T> where F: FnMut(T) -> T {
+        let mut this = (*self).clone();
+        this.apply(f);
+        this
+    }
+
+    fn apply<F>(&mut self, mut f: F) where F: FnMut(T) -> T {
+        for v in &mut self.0 {
+            *v = f(*v)
+        }
+    }
+
+    fn map_with_alpha<F, G>(&self, f: F, g: G) -> $ident<T> where F: FnMut(T) -> T, G: FnMut(T) -> T {
+        let mut this = (*self).clone();
+        this.apply_with_alpha(f, g);
+        this
+    }
+
+    fn apply_with_alpha<F, G>(&mut self, mut f: F, mut g: G) where F: FnMut(T) -> T, G: FnMut(T) -> T {
+        const ALPHA: usize = $channels - $alphas;
+        for v in self.0[..ALPHA].iter_mut() {
+            *v = f(*v)
+        }
+        // The branch of this match is `const`. This way ensures that no subexpression fails the
+        // `const_err` lint (the expression `self.0[ALPHA]` would).
+        if let Some(v) = self.0.get_mut(ALPHA) {
+            *v = g(*v)
+        }
+    }
+
+    fn map2<F>(&self, other: &Self, f: F) -> $ident<T> where F: FnMut(T, T) -> T {
+        let mut this = (*self).clone();
+        this.apply2(other, f);
+        this
+    }
+
+    fn apply2<F>(&mut self, other: &$ident<T>, mut f: F) where F: FnMut(T, T) -> T {
+        for (a, &b) in self.0.iter_mut().zip(other.0.iter()) {
+            *a = f(*a, b)
+        }
+    }
+
+    fn invert(&mut self) {
+        Invert::invert(self)
+    }
+
+    fn blend(&mut self, other: &$ident<T>) {
+        Blend::blend(self, other)
+    }
+}
+
+impl<T> Index<usize> for $ident<T> {
+    type Output = T;
+    #[inline(always)]
+    fn index(&self, _index: usize) -> &T {
+        &self.0[_index]
+    }
+}
+
+impl<T> IndexMut<usize> for $ident<T> {
+    #[inline(always)]
+    fn index_mut(&mut self, _index: usize) -> &mut T {
+        &mut self.0[_index]
+    }
+}
+
+impl<T> From<[T; $channels]> for $ident<T> {
+    fn from(c: [T; $channels]) -> Self {
+        Self(c)
+    }
+}
+
+)* // END Structure definitions
+
+    }
+}
+
+define_colors! {
+    /// RGB colors.
+    ///
+    /// For the purpose of color conversion, as well as blending, the implementation of `Pixel`
+    /// assumes an `sRGB` color space of its data.
+    pub struct Rgb<T: Primitive Enlargeable>([T; 3, 0]) = "RGB";
+    /// Grayscale colors.
+    pub struct Luma<T: Primitive>([T; 1, 0]) = "Y";
+    /// RGB colors + alpha channel
+    pub struct Rgba<T: Primitive Enlargeable>([T; 4, 1]) = "RGBA";
+    /// Grayscale colors + alpha channel
+    pub struct LumaA<T: Primitive>([T; 2, 1]) = "YA";
+}
+
+/// Convert from one pixel component type to another. For example, convert from `u8` to `f32` pixel values.
+pub trait FromPrimitive<Component> {
+    /// Converts from any pixel component type to this type.
+    fn from_primitive(component: Component) -> Self;
+}
+
+impl<T: Primitive> FromPrimitive<T> for T {
+    fn from_primitive(sample: T) -> Self {
+        sample
+    }
+}
+
+// from f32:
+// Note that in to-integer-conversion we are performing rounding but NumCast::from is implemented
+// as truncate towards zero. We emulate rounding by adding a bias.
+
+impl FromPrimitive<f32> for u8 {
+    fn from_primitive(float: f32) -> Self {
+        let inner = (float.clamp(0.0, 1.0) * u8::MAX as f32).round();
+        NumCast::from(inner).unwrap()
+    }
+}
+
+impl FromPrimitive<f32> for u16 {
+    fn from_primitive(float: f32) -> Self {
+        let inner = (float.clamp(0.0, 1.0) * u16::MAX as f32).round();
+        NumCast::from(inner).unwrap()
+    }
+}
+
+// from u16:
+
+impl FromPrimitive<u16> for u8 {
+    fn from_primitive(c16: u16) -> Self {
+        fn from(c: impl Into<u32>) -> u32 {
+            c.into()
+        }
+        // The input c is the numerator of `c / u16::MAX`.
+        // Derive numerator of `num / u8::MAX`, with rounding.
+        //
+        // This method is based on the inverse (see FromPrimitive<u8> for u16) and was tested
+        // exhaustively in Python. It's the same as the reference function:
+        //  round(c * (2**8 - 1) / (2**16 - 1))
+        NumCast::from((from(c16) + 128) / 257).unwrap()
+    }
+}
+
+impl FromPrimitive<u16> for f32 {
+    fn from_primitive(int: u16) -> Self {
+        (int as f32 / u16::MAX as f32).clamp(0.0, 1.0)
+    }
+}
+
+// from u8:
+
+impl FromPrimitive<u8> for f32 {
+    fn from_primitive(int: u8) -> Self {
+        (int as f32 / u8::MAX as f32).clamp(0.0, 1.0)
+    }
+}
+
+impl FromPrimitive<u8> for u16 {
+    fn from_primitive(c8: u8) -> Self {
+        let x = c8.to_u64().unwrap();
+        NumCast::from((x << 8) | x).unwrap()
+    }
+}
+
+/// Provides color conversions for the different pixel types.
+pub trait FromColor<Other> {
+    /// Changes `self` to represent `Other` in the color space of `Self`
+    fn from_color(&mut self, _: &Other);
+}
+
+/// Copy-based conversions to target pixel types using `FromColor`.
+// FIXME: this trait should be removed and replaced with real color space models
+// rather than assuming sRGB.
+pub(crate) trait IntoColor<Other> {
+    /// Constructs a pixel of the target type and converts this pixel into it.
+    fn into_color(&self) -> Other;
+}
+
+impl<O, S> IntoColor<O> for S
+where
+    O: Pixel + FromColor<S>,
+{
+    fn into_color(&self) -> O {
+        // Note we cannot use Pixel::CHANNELS_COUNT here to directly construct
+        // the pixel due to a current bug/limitation of consts.
+        #[allow(deprecated)]
+        let mut pix = O::from_channels(Zero::zero(), Zero::zero(), Zero::zero(), Zero::zero());
+        pix.from_color(self);
+        pix
+    }
+}
+
+/// Coefficients to transform from sRGB to a CIE Y (luminance) value.
+const SRGB_LUMA: [u32; 3] = [2126, 7152, 722];
+const SRGB_LUMA_DIV: u32 = 10000;
+
+#[inline]
+fn rgb_to_luma<T: Primitive + Enlargeable>(rgb: &[T]) -> T {
+    let l = <T::Larger as NumCast>::from(SRGB_LUMA[0]).unwrap() * rgb[0].to_larger()
+        + <T::Larger as NumCast>::from(SRGB_LUMA[1]).unwrap() * rgb[1].to_larger()
+        + <T::Larger as NumCast>::from(SRGB_LUMA[2]).unwrap() * rgb[2].to_larger();
+    T::clamp_from(l / <T::Larger as NumCast>::from(SRGB_LUMA_DIV).unwrap())
+}
+
+// `FromColor` for Luma
+impl<S: Primitive, T: Primitive> FromColor<Luma<S>> for Luma<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &Luma<S>) {
+        let own = self.channels_mut();
+        let other = other.channels();
+        own[0] = T::from_primitive(other[0]);
+    }
+}
+
+impl<S: Primitive, T: Primitive> FromColor<LumaA<S>> for Luma<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &LumaA<S>) {
+        self.channels_mut()[0] = T::from_primitive(other.channels()[0])
+    }
+}
+
+impl<S: Primitive + Enlargeable, T: Primitive> FromColor<Rgb<S>> for Luma<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &Rgb<S>) {
+        let gray = self.channels_mut();
+        let rgb = other.channels();
+        gray[0] = T::from_primitive(rgb_to_luma(rgb));
+    }
+}
+
+impl<S: Primitive + Enlargeable, T: Primitive> FromColor<Rgba<S>> for Luma<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &Rgba<S>) {
+        let gray = self.channels_mut();
+        let rgb = other.channels();
+        let l = rgb_to_luma(rgb);
+        gray[0] = T::from_primitive(l);
+    }
+}
+
+// `FromColor` for LumaA
+
+impl<S: Primitive, T: Primitive> FromColor<LumaA<S>> for LumaA<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &LumaA<S>) {
+        let own = self.channels_mut();
+        let other = other.channels();
+        own[0] = T::from_primitive(other[0]);
+        own[1] = T::from_primitive(other[1]);
+    }
+}
+
+impl<S: Primitive + Enlargeable, T: Primitive> FromColor<Rgb<S>> for LumaA<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &Rgb<S>) {
+        let gray_a = self.channels_mut();
+        let rgb = other.channels();
+        gray_a[0] = T::from_primitive(rgb_to_luma(rgb));
+        gray_a[1] = T::DEFAULT_MAX_VALUE;
+    }
+}
+
+impl<S: Primitive + Enlargeable, T: Primitive> FromColor<Rgba<S>> for LumaA<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &Rgba<S>) {
+        let gray_a = self.channels_mut();
+        let rgba = other.channels();
+        gray_a[0] = T::from_primitive(rgb_to_luma(rgba));
+        gray_a[1] = T::from_primitive(rgba[3]);
+    }
+}
+
+impl<S: Primitive, T: Primitive> FromColor<Luma<S>> for LumaA<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &Luma<S>) {
+        let gray_a = self.channels_mut();
+        gray_a[0] = T::from_primitive(other.channels()[0]);
+        gray_a[1] = T::DEFAULT_MAX_VALUE;
+    }
+}
+
+// `FromColor` for RGBA
+
+impl<S: Primitive, T: Primitive> FromColor<Rgba<S>> for Rgba<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &Rgba<S>) {
+        let own = &mut self.0;
+        let other = &other.0;
+        own[0] = T::from_primitive(other[0]);
+        own[1] = T::from_primitive(other[1]);
+        own[2] = T::from_primitive(other[2]);
+        own[3] = T::from_primitive(other[3]);
+    }
+}
+
+impl<S: Primitive, T: Primitive> FromColor<Rgb<S>> for Rgba<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &Rgb<S>) {
+        let rgba = &mut self.0;
+        let rgb = &other.0;
+        rgba[0] = T::from_primitive(rgb[0]);
+        rgba[1] = T::from_primitive(rgb[1]);
+        rgba[2] = T::from_primitive(rgb[2]);
+        rgba[3] = T::DEFAULT_MAX_VALUE;
+    }
+}
+
+impl<S: Primitive, T: Primitive> FromColor<LumaA<S>> for Rgba<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, gray: &LumaA<S>) {
+        let rgba = &mut self.0;
+        let gray = &gray.0;
+        rgba[0] = T::from_primitive(gray[0]);
+        rgba[1] = T::from_primitive(gray[0]);
+        rgba[2] = T::from_primitive(gray[0]);
+        rgba[3] = T::from_primitive(gray[1]);
+    }
+}
+
+impl<S: Primitive, T: Primitive> FromColor<Luma<S>> for Rgba<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, gray: &Luma<S>) {
+        let rgba = &mut self.0;
+        let gray = gray.0[0];
+        rgba[0] = T::from_primitive(gray);
+        rgba[1] = T::from_primitive(gray);
+        rgba[2] = T::from_primitive(gray);
+        rgba[3] = T::DEFAULT_MAX_VALUE;
+    }
+}
+
+// `FromColor` for RGB
+
+impl<S: Primitive, T: Primitive> FromColor<Rgb<S>> for Rgb<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &Rgb<S>) {
+        let own = &mut self.0;
+        let other = &other.0;
+        own[0] = T::from_primitive(other[0]);
+        own[1] = T::from_primitive(other[1]);
+        own[2] = T::from_primitive(other[2]);
+    }
+}
+
+impl<S: Primitive, T: Primitive> FromColor<Rgba<S>> for Rgb<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &Rgba<S>) {
+        let rgb = &mut self.0;
+        let rgba = &other.0;
+        rgb[0] = T::from_primitive(rgba[0]);
+        rgb[1] = T::from_primitive(rgba[1]);
+        rgb[2] = T::from_primitive(rgba[2]);
+    }
+}
+
+impl<S: Primitive, T: Primitive> FromColor<LumaA<S>> for Rgb<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &LumaA<S>) {
+        let rgb = &mut self.0;
+        let gray = other.0[0];
+        rgb[0] = T::from_primitive(gray);
+        rgb[1] = T::from_primitive(gray);
+        rgb[2] = T::from_primitive(gray);
+    }
+}
+
+impl<S: Primitive, T: Primitive> FromColor<Luma<S>> for Rgb<T>
+where
+    T: FromPrimitive<S>,
+{
+    fn from_color(&mut self, other: &Luma<S>) {
+        let rgb = &mut self.0;
+        let gray = other.0[0];
+        rgb[0] = T::from_primitive(gray);
+        rgb[1] = T::from_primitive(gray);
+        rgb[2] = T::from_primitive(gray);
+    }
+}
+
+/// Blends a color inter another one
+pub(crate) trait Blend {
+    /// Blends a color in-place.
+    fn blend(&mut self, other: &Self);
+}
+
+impl<T: Primitive> Blend for LumaA<T> {
+    fn blend(&mut self, other: &LumaA<T>) {
+        let max_t = T::DEFAULT_MAX_VALUE;
+        let max_t = max_t.to_f32().unwrap();
+        let (bg_luma, bg_a) = (self.0[0], self.0[1]);
+        let (fg_luma, fg_a) = (other.0[0], other.0[1]);
+
+        let (bg_luma, bg_a) = (
+            bg_luma.to_f32().unwrap() / max_t,
+            bg_a.to_f32().unwrap() / max_t,
+        );
+        let (fg_luma, fg_a) = (
+            fg_luma.to_f32().unwrap() / max_t,
+            fg_a.to_f32().unwrap() / max_t,
+        );
+
+        let alpha_final = bg_a + fg_a - bg_a * fg_a;
+        if alpha_final == 0.0 {
+            return;
+        };
+        let bg_luma_a = bg_luma * bg_a;
+        let fg_luma_a = fg_luma * fg_a;
+
+        let out_luma_a = fg_luma_a + bg_luma_a * (1.0 - fg_a);
+        let out_luma = out_luma_a / alpha_final;
+
+        *self = LumaA([
+            NumCast::from(max_t * out_luma).unwrap(),
+            NumCast::from(max_t * alpha_final).unwrap(),
+        ])
+    }
+}
+
+impl<T: Primitive> Blend for Luma<T> {
+    fn blend(&mut self, other: &Luma<T>) {
+        *self = *other
+    }
+}
+
+impl<T: Primitive> Blend for Rgba<T> {
+    fn blend(&mut self, other: &Rgba<T>) {
+        // http://stackoverflow.com/questions/7438263/alpha-compositing-algorithm-blend-modes#answer-11163848
+
+        if other.0[3].is_zero() {
+            return;
+        }
+        if other.0[3] == T::DEFAULT_MAX_VALUE {
+            *self = *other;
+            return;
+        }
+
+        // First, as we don't know what type our pixel is, we have to convert to floats between 0.0 and 1.0
+        let max_t = T::DEFAULT_MAX_VALUE;
+        let max_t = max_t.to_f32().unwrap();
+        let (bg_r, bg_g, bg_b, bg_a) = (self.0[0], self.0[1], self.0[2], self.0[3]);
+        let (fg_r, fg_g, fg_b, fg_a) = (other.0[0], other.0[1], other.0[2], other.0[3]);
+        let (bg_r, bg_g, bg_b, bg_a) = (
+            bg_r.to_f32().unwrap() / max_t,
+            bg_g.to_f32().unwrap() / max_t,
+            bg_b.to_f32().unwrap() / max_t,
+            bg_a.to_f32().unwrap() / max_t,
+        );
+        let (fg_r, fg_g, fg_b, fg_a) = (
+            fg_r.to_f32().unwrap() / max_t,
+            fg_g.to_f32().unwrap() / max_t,
+            fg_b.to_f32().unwrap() / max_t,
+            fg_a.to_f32().unwrap() / max_t,
+        );
+
+        // Work out what the final alpha level will be
+        let alpha_final = bg_a + fg_a - bg_a * fg_a;
+        if alpha_final == 0.0 {
+            return;
+        };
+
+        // We premultiply our channels by their alpha, as this makes it easier to calculate
+        let (bg_r_a, bg_g_a, bg_b_a) = (bg_r * bg_a, bg_g * bg_a, bg_b * bg_a);
+        let (fg_r_a, fg_g_a, fg_b_a) = (fg_r * fg_a, fg_g * fg_a, fg_b * fg_a);
+
+        // Standard formula for src-over alpha compositing
+        let (out_r_a, out_g_a, out_b_a) = (
+            fg_r_a + bg_r_a * (1.0 - fg_a),
+            fg_g_a + bg_g_a * (1.0 - fg_a),
+            fg_b_a + bg_b_a * (1.0 - fg_a),
+        );
+
+        // Unmultiply the channels by our resultant alpha channel
+        let (out_r, out_g, out_b) = (
+            out_r_a / alpha_final,
+            out_g_a / alpha_final,
+            out_b_a / alpha_final,
+        );
+
+        // Cast back to our initial type on return
+        *self = Rgba([
+            NumCast::from(max_t * out_r).unwrap(),
+            NumCast::from(max_t * out_g).unwrap(),
+            NumCast::from(max_t * out_b).unwrap(),
+            NumCast::from(max_t * alpha_final).unwrap(),
+        ])
+    }
+}
+
+impl<T: Primitive> Blend for Rgb<T> {
+    fn blend(&mut self, other: &Rgb<T>) {
+        *self = *other
+    }
+}
+
+/// Invert a color
+pub(crate) trait Invert {
+    /// Inverts a color in-place.
+    fn invert(&mut self);
+}
+
+impl<T: Primitive> Invert for LumaA<T> {
+    fn invert(&mut self) {
+        let l = self.0;
+        let max = T::DEFAULT_MAX_VALUE;
+
+        *self = LumaA([max - l[0], l[1]])
+    }
+}
+
+impl<T: Primitive> Invert for Luma<T> {
+    fn invert(&mut self) {
+        let l = self.0;
+
+        let max = T::DEFAULT_MAX_VALUE;
+        let l1 = max - l[0];
+
+        *self = Luma([l1])
+    }
+}
+
+impl<T: Primitive> Invert for Rgba<T> {
+    fn invert(&mut self) {
+        let rgba = self.0;
+
+        let max = T::DEFAULT_MAX_VALUE;
+
+        *self = Rgba([max - rgba[0], max - rgba[1], max - rgba[2], rgba[3]])
+    }
+}
+
+impl<T: Primitive> Invert for Rgb<T> {
+    fn invert(&mut self) {
+        let rgb = self.0;
+
+        let max = T::DEFAULT_MAX_VALUE;
+
+        let r1 = max - rgb[0];
+        let g1 = max - rgb[1];
+        let b1 = max - rgb[2];
+
+        *self = Rgb([r1, g1, b1])
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{Luma, LumaA, Pixel, Rgb, Rgba};
+
+    #[test]
+    fn test_apply_with_alpha_rgba() {
+        let mut rgba = Rgba([0, 0, 0, 0]);
+        rgba.apply_with_alpha(|s| s, |_| 0xFF);
+        assert_eq!(rgba, Rgba([0, 0, 0, 0xFF]));
+    }
+
+    #[test]
+    fn test_apply_with_alpha_rgb() {
+        let mut rgb = Rgb([0, 0, 0]);
+        rgb.apply_with_alpha(|s| s, |_| panic!("bug"));
+        assert_eq!(rgb, Rgb([0, 0, 0]));
+    }
+
+    #[test]
+    fn test_map_with_alpha_rgba() {
+        let rgba = Rgba([0, 0, 0, 0]).map_with_alpha(|s| s, |_| 0xFF);
+        assert_eq!(rgba, Rgba([0, 0, 0, 0xFF]));
+    }
+
+    #[test]
+    fn test_map_with_alpha_rgb() {
+        let rgb = Rgb([0, 0, 0]).map_with_alpha(|s| s, |_| panic!("bug"));
+        assert_eq!(rgb, Rgb([0, 0, 0]));
+    }
+
+    #[test]
+    fn test_blend_luma_alpha() {
+        let ref mut a = LumaA([255 as u8, 255]);
+        let b = LumaA([255 as u8, 255]);
+        a.blend(&b);
+        assert_eq!(a.0[0], 255);
+        assert_eq!(a.0[1], 255);
+
+        let ref mut a = LumaA([255 as u8, 0]);
+        let b = LumaA([255 as u8, 255]);
+        a.blend(&b);
+        assert_eq!(a.0[0], 255);
+        assert_eq!(a.0[1], 255);
+
+        let ref mut a = LumaA([255 as u8, 255]);
+        let b = LumaA([255 as u8, 0]);
+        a.blend(&b);
+        assert_eq!(a.0[0], 255);
+        assert_eq!(a.0[1], 255);
+
+        let ref mut a = LumaA([255 as u8, 0]);
+        let b = LumaA([255 as u8, 0]);
+        a.blend(&b);
+        assert_eq!(a.0[0], 255);
+        assert_eq!(a.0[1], 0);
+    }
+
+    #[test]
+    fn test_blend_rgba() {
+        let ref mut a = Rgba([255 as u8, 255, 255, 255]);
+        let b = Rgba([255 as u8, 255, 255, 255]);
+        a.blend(&b);
+        assert_eq!(a.0, [255, 255, 255, 255]);
+
+        let ref mut a = Rgba([255 as u8, 255, 255, 0]);
+        let b = Rgba([255 as u8, 255, 255, 255]);
+        a.blend(&b);
+        assert_eq!(a.0, [255, 255, 255, 255]);
+
+        let ref mut a = Rgba([255 as u8, 255, 255, 255]);
+        let b = Rgba([255 as u8, 255, 255, 0]);
+        a.blend(&b);
+        assert_eq!(a.0, [255, 255, 255, 255]);
+
+        let ref mut a = Rgba([255 as u8, 255, 255, 0]);
+        let b = Rgba([255 as u8, 255, 255, 0]);
+        a.blend(&b);
+        assert_eq!(a.0, [255, 255, 255, 0]);
+    }
+
+    #[test]
+    fn test_apply_without_alpha_rgba() {
+        let mut rgba = Rgba([0, 0, 0, 0]);
+        rgba.apply_without_alpha(|s| s + 1);
+        assert_eq!(rgba, Rgba([1, 1, 1, 0]));
+    }
+
+    #[test]
+    fn test_apply_without_alpha_rgb() {
+        let mut rgb = Rgb([0, 0, 0]);
+        rgb.apply_without_alpha(|s| s + 1);
+        assert_eq!(rgb, Rgb([1, 1, 1]));
+    }
+
+    #[test]
+    fn test_map_without_alpha_rgba() {
+        let rgba = Rgba([0, 0, 0, 0]).map_without_alpha(|s| s + 1);
+        assert_eq!(rgba, Rgba([1, 1, 1, 0]));
+    }
+
+    #[test]
+    fn test_map_without_alpha_rgb() {
+        let rgb = Rgb([0, 0, 0]).map_without_alpha(|s| s + 1);
+        assert_eq!(rgb, Rgb([1, 1, 1]));
+    }
+
+    macro_rules! test_lossless_conversion {
+        ($a:ty, $b:ty, $c:ty) => {
+            let a: $a = [<$a as Pixel>::Subpixel::DEFAULT_MAX_VALUE >> 2;
+                <$a as Pixel>::CHANNEL_COUNT as usize]
+                .into();
+            let b: $b = a.into_color();
+            let c: $c = b.into_color();
+            assert_eq!(a.channels(), c.channels());
+        };
+    }
+
+    #[test]
+    fn test_lossless_conversions() {
+        use super::IntoColor;
+        use crate::traits::Primitive;
+
+        test_lossless_conversion!(Luma<u8>, Luma<u16>, Luma<u8>);
+        test_lossless_conversion!(LumaA<u8>, LumaA<u16>, LumaA<u8>);
+        test_lossless_conversion!(Rgb<u8>, Rgb<u16>, Rgb<u8>);
+        test_lossless_conversion!(Rgba<u8>, Rgba<u16>, Rgba<u8>);
+    }
+
+    #[test]
+    fn accuracy_conversion() {
+        use super::{Luma, Pixel, Rgb};
+        let pixel = Rgb::from([13, 13, 13]);
+        let Luma([luma]) = pixel.to_luma();
+        assert_eq!(luma, 13);
+    }
+}

vendor/image/src/error.rs

@@ -0,0 +1,506 @@
+//! Contains detailed error representation.
+//!
+//! See the main [`ImageError`] which contains a variant for each specialized error type. The
+//! subtypes used in each variant are opaque by design. They can be roughly inspected through their
+//! respective `kind` methods which work similar to `std::io::Error::kind`.
+//!
+//! The error interface makes it possible to inspect the error of an underlying decoder or encoder,
+//! through the `Error::source` method. Note that this is not part of the stable interface and you
+//! may not rely on a particular error value for a particular operation. This means mainly that
+//! `image` does not promise to remain on a particular version of its underlying decoders but if
+//! you ensure to use the same version of the dependency (or at least of the error type) through
+//! external means then you could inspect the error type in slightly more detail.
+//!
+//! [`ImageError`]: enum.ImageError.html
+
+use std::error::Error;
+use std::{fmt, io};
+
+use crate::color::ExtendedColorType;
+use crate::image::ImageFormat;
+
+/// The generic error type for image operations.
+///
+/// This high level enum allows, by variant matching, a rough separation of concerns between
+/// underlying IO, the caller, format specifications, and the `image` implementation.
+#[derive(Debug)]
+pub enum ImageError {
+    /// An error was encountered while decoding.
+    ///
+    /// This means that the input data did not conform to the specification of some image format,
+    /// or that no format could be determined, or that it did not match format specific
+    /// requirements set by the caller.
+    Decoding(DecodingError),
+
+    /// An error was encountered while encoding.
+    ///
+    /// The input image can not be encoded with the chosen format, for example because the
+    /// specification has no representation for its color space or because a necessary conversion
+    /// is ambiguous. In some cases it might also happen that the dimensions can not be used with
+    /// the format.
+    Encoding(EncodingError),
+
+    /// An error was encountered in input arguments.
+    ///
+    /// This is a catch-all case for strictly internal operations such as scaling, conversions,
+    /// etc. that involve no external format specifications.
+    Parameter(ParameterError),
+
+    /// Completing the operation would have required more resources than allowed.
+    ///
+    /// Errors of this type are limits set by the user or environment, *not* inherent in a specific
+    /// format or operation that was executed.
+    Limits(LimitError),
+
+    /// An operation can not be completed by the chosen abstraction.
+    ///
+    /// This means that it might be possible for the operation to succeed in general but
+    /// * it requires a disabled feature,
+    /// * the implementation does not yet exist, or
+    /// * no abstraction for a lower level could be found.
+    Unsupported(UnsupportedError),
+
+    /// An error occurred while interacting with the environment.
+    IoError(io::Error),
+}
+
+/// The implementation for an operation was not provided.
+///
+/// See the variant [`Unsupported`] for more documentation.
+///
+/// [`Unsupported`]: enum.ImageError.html#variant.Unsupported
+#[derive(Debug)]
+pub struct UnsupportedError {
+    format: ImageFormatHint,
+    kind: UnsupportedErrorKind,
+}
+
+/// Details what feature is not supported.
+#[derive(Clone, Debug, Hash, PartialEq)]
+#[non_exhaustive]
+pub enum UnsupportedErrorKind {
+    /// The required color type can not be handled.
+    Color(ExtendedColorType),
+    /// An image format is not supported.
+    Format(ImageFormatHint),
+    /// Some feature specified by string.
+    /// This is discouraged and is likely to get deprecated (but not removed).
+    GenericFeature(String),
+}
+
+/// An error was encountered while encoding an image.
+///
+/// This is used as an opaque representation for the [`ImageError::Encoding`] variant. See its
+/// documentation for more information.
+///
+/// [`ImageError::Encoding`]: enum.ImageError.html#variant.Encoding
+#[derive(Debug)]
+pub struct EncodingError {
+    format: ImageFormatHint,
+    underlying: Option<Box<dyn Error + Send + Sync>>,
+}
+
+/// An error was encountered in inputs arguments.
+///
+/// This is used as an opaque representation for the [`ImageError::Parameter`] variant. See its
+/// documentation for more information.
+///
+/// [`ImageError::Parameter`]: enum.ImageError.html#variant.Parameter
+#[derive(Debug)]
+pub struct ParameterError {
+    kind: ParameterErrorKind,
+    underlying: Option<Box<dyn Error + Send + Sync>>,
+}
+
+/// Details how a parameter is malformed.
+#[derive(Clone, Debug, Hash, PartialEq)]
+#[non_exhaustive]
+pub enum ParameterErrorKind {
+    /// The dimensions passed are wrong.
+    DimensionMismatch,
+    /// Repeated an operation for which error that could not be cloned was emitted already.
+    FailedAlready,
+    /// A string describing the parameter.
+    /// This is discouraged and is likely to get deprecated (but not removed).
+    Generic(String),
+    /// The end of the image has been reached.
+    NoMoreData,
+}
+
+/// An error was encountered while decoding an image.
+///
+/// This is used as an opaque representation for the [`ImageError::Decoding`] variant. See its
+/// documentation for more information.
+///
+/// [`ImageError::Decoding`]: enum.ImageError.html#variant.Decoding
+#[derive(Debug)]
+pub struct DecodingError {
+    format: ImageFormatHint,
+    underlying: Option<Box<dyn Error + Send + Sync>>,
+}
+
+/// Completing the operation would have required more resources than allowed.
+///
+/// This is used as an opaque representation for the [`ImageError::Limits`] variant. See its
+/// documentation for more information.
+///
+/// [`ImageError::Limits`]: enum.ImageError.html#variant.Limits
+#[derive(Debug)]
+pub struct LimitError {
+    kind: LimitErrorKind,
+    // do we need an underlying error?
+}
+
+/// Indicates the limit that prevented an operation from completing.
+///
+/// Note that this enumeration is not exhaustive and may in the future be extended to provide more
+/// detailed information or to incorporate other resources types.
+#[derive(Clone, Debug, Hash, PartialEq, Eq)]
+#[non_exhaustive]
+#[allow(missing_copy_implementations)] // Might be non-Copy in the future.
+pub enum LimitErrorKind {
+    /// The resulting image exceed dimension limits in either direction.
+    DimensionError,
+    /// The operation would have performed an allocation larger than allowed.
+    InsufficientMemory,
+    /// The specified strict limits are not supported for this operation
+    Unsupported {
+        /// The given limits
+        limits: crate::io::Limits,
+        /// The supported strict limits
+        supported: crate::io::LimitSupport,
+    },
+}
+
+/// A best effort representation for image formats.
+#[derive(Clone, Debug, Hash, PartialEq)]
+#[non_exhaustive]
+pub enum ImageFormatHint {
+    /// The format is known exactly.
+    Exact(ImageFormat),
+
+    /// The format can be identified by a name.
+    Name(String),
+
+    /// A common path extension for the format is known.
+    PathExtension(std::path::PathBuf),
+
+    /// The format is not known or could not be determined.
+    Unknown,
+}
+
+impl UnsupportedError {
+    /// Create an `UnsupportedError` for an image with details on the unsupported feature.
+    ///
+    /// If the operation was not connected to a particular image format then the hint may be
+    /// `Unknown`.
+    pub fn from_format_and_kind(format: ImageFormatHint, kind: UnsupportedErrorKind) -> Self {
+        UnsupportedError { format, kind }
+    }
+
+    /// Returns the corresponding `UnsupportedErrorKind` of the error.
+    pub fn kind(&self) -> UnsupportedErrorKind {
+        self.kind.clone()
+    }
+
+    /// Returns the image format associated with this error.
+    pub fn format_hint(&self) -> ImageFormatHint {
+        self.format.clone()
+    }
+}
+
+impl DecodingError {
+    /// Create a `DecodingError` that stems from an arbitrary error of an underlying decoder.
+    pub fn new(format: ImageFormatHint, err: impl Into<Box<dyn Error + Send + Sync>>) -> Self {
+        DecodingError {
+            format,
+            underlying: Some(err.into()),
+        }
+    }
+
+    /// Create a `DecodingError` for an image format.
+    ///
+    /// The error will not contain any further information but is very easy to create.
+    pub fn from_format_hint(format: ImageFormatHint) -> Self {
+        DecodingError {
+            format,
+            underlying: None,
+        }
+    }
+
+    /// Returns the image format associated with this error.
+    pub fn format_hint(&self) -> ImageFormatHint {
+        self.format.clone()
+    }
+}
+
+impl EncodingError {
+    /// Create an `EncodingError` that stems from an arbitrary error of an underlying encoder.
+    pub fn new(format: ImageFormatHint, err: impl Into<Box<dyn Error + Send + Sync>>) -> Self {
+        EncodingError {
+            format,
+            underlying: Some(err.into()),
+        }
+    }
+
+    /// Create an `EncodingError` for an image format.
+    ///
+    /// The error will not contain any further information but is very easy to create.
+    pub fn from_format_hint(format: ImageFormatHint) -> Self {
+        EncodingError {
+            format,
+            underlying: None,
+        }
+    }
+
+    /// Return the image format associated with this error.
+    pub fn format_hint(&self) -> ImageFormatHint {
+        self.format.clone()
+    }
+}
+
+impl ParameterError {
+    /// Construct a `ParameterError` directly from a corresponding kind.
+    pub fn from_kind(kind: ParameterErrorKind) -> Self {
+        ParameterError {
+            kind,
+            underlying: None,
+        }
+    }
+
+    /// Returns the corresponding `ParameterErrorKind` of the error.
+    pub fn kind(&self) -> ParameterErrorKind {
+        self.kind.clone()
+    }
+}
+
+impl LimitError {
+    /// Construct a generic `LimitError` directly from a corresponding kind.
+    pub fn from_kind(kind: LimitErrorKind) -> Self {
+        LimitError { kind }
+    }
+
+    /// Returns the corresponding `LimitErrorKind` of the error.
+    pub fn kind(&self) -> LimitErrorKind {
+        self.kind.clone()
+    }
+}
+
+impl From<io::Error> for ImageError {
+    fn from(err: io::Error) -> ImageError {
+        ImageError::IoError(err)
+    }
+}
+
+impl From<ImageFormat> for ImageFormatHint {
+    fn from(format: ImageFormat) -> Self {
+        ImageFormatHint::Exact(format)
+    }
+}
+
+impl From<&'_ std::path::Path> for ImageFormatHint {
+    fn from(path: &'_ std::path::Path) -> Self {
+        match path.extension() {
+            Some(ext) => ImageFormatHint::PathExtension(ext.into()),
+            None => ImageFormatHint::Unknown,
+        }
+    }
+}
+
+impl From<ImageFormatHint> for UnsupportedError {
+    fn from(hint: ImageFormatHint) -> Self {
+        UnsupportedError {
+            format: hint.clone(),
+            kind: UnsupportedErrorKind::Format(hint),
+        }
+    }
+}
+
+/// Result of an image decoding/encoding process
+pub type ImageResult<T> = Result<T, ImageError>;
+
+impl fmt::Display for ImageError {
+    fn fmt(&self, fmt: &mut fmt::Formatter) -> Result<(), fmt::Error> {
+        match self {
+            ImageError::IoError(err) => err.fmt(fmt),
+            ImageError::Decoding(err) => err.fmt(fmt),
+            ImageError::Encoding(err) => err.fmt(fmt),
+            ImageError::Parameter(err) => err.fmt(fmt),
+            ImageError::Limits(err) => err.fmt(fmt),
+            ImageError::Unsupported(err) => err.fmt(fmt),
+        }
+    }
+}
+
+impl Error for ImageError {
+    fn source(&self) -> Option<&(dyn Error + 'static)> {
+        match self {
+            ImageError::IoError(err) => err.source(),
+            ImageError::Decoding(err) => err.source(),
+            ImageError::Encoding(err) => err.source(),
+            ImageError::Parameter(err) => err.source(),
+            ImageError::Limits(err) => err.source(),
+            ImageError::Unsupported(err) => err.source(),
+        }
+    }
+}
+
+impl fmt::Display for UnsupportedError {
+    fn fmt(&self, fmt: &mut fmt::Formatter) -> Result<(), fmt::Error> {
+        match &self.kind {
+            UnsupportedErrorKind::Format(ImageFormatHint::Unknown) => {
+                write!(fmt, "The image format could not be determined",)
+            }
+            UnsupportedErrorKind::Format(format @ ImageFormatHint::PathExtension(_)) => write!(
+                fmt,
+                "The file extension {} was not recognized as an image format",
+                format,
+            ),
+            UnsupportedErrorKind::Format(format) => {
+                write!(fmt, "The image format {} is not supported", format,)
+            }
+            UnsupportedErrorKind::Color(color) => write!(
+                fmt,
+                "The decoder for {} does not support the color type `{:?}`",
+                self.format, color,
+            ),
+            UnsupportedErrorKind::GenericFeature(message) => match &self.format {
+                ImageFormatHint::Unknown => write!(
+                    fmt,
+                    "The decoder does not support the format feature {}",
+                    message,
+                ),
+                other => write!(
+                    fmt,
+                    "The decoder for {} does not support the format features {}",
+                    other, message,
+                ),
+            },
+        }
+    }
+}
+
+impl Error for UnsupportedError {}
+
+impl fmt::Display for ParameterError {
+    fn fmt(&self, fmt: &mut fmt::Formatter) -> Result<(), fmt::Error> {
+        match &self.kind {
+            ParameterErrorKind::DimensionMismatch => write!(
+                fmt,
+                "The Image's dimensions are either too \
+                 small or too large"
+            ),
+            ParameterErrorKind::FailedAlready => write!(
+                fmt,
+                "The end the image stream has been reached due to a previous error"
+            ),
+            ParameterErrorKind::Generic(message) => {
+                write!(fmt, "The parameter is malformed: {}", message,)
+            }
+            ParameterErrorKind::NoMoreData => write!(fmt, "The end of the image has been reached",),
+        }?;
+
+        if let Some(underlying) = &self.underlying {
+            write!(fmt, "\n{}", underlying)?;
+        }
+
+        Ok(())
+    }
+}
+
+impl Error for ParameterError {
+    fn source(&self) -> Option<&(dyn Error + 'static)> {
+        match &self.underlying {
+            None => None,
+            Some(source) => Some(&**source),
+        }
+    }
+}
+
+impl fmt::Display for EncodingError {
+    fn fmt(&self, fmt: &mut fmt::Formatter) -> Result<(), fmt::Error> {
+        match &self.underlying {
+            Some(underlying) => write!(
+                fmt,
+                "Format error encoding {}:\n{}",
+                self.format, underlying,
+            ),
+            None => write!(fmt, "Format error encoding {}", self.format,),
+        }
+    }
+}
+
+impl Error for EncodingError {
+    fn source(&self) -> Option<&(dyn Error + 'static)> {
+        match &self.underlying {
+            None => None,
+            Some(source) => Some(&**source),
+        }
+    }
+}
+
+impl fmt::Display for DecodingError {
+    fn fmt(&self, fmt: &mut fmt::Formatter) -> Result<(), fmt::Error> {
+        match &self.underlying {
+            None => match self.format {
+                ImageFormatHint::Unknown => write!(fmt, "Format error"),
+                _ => write!(fmt, "Format error decoding {}", self.format),
+            },
+            Some(underlying) => {
+                write!(fmt, "Format error decoding {}: {}", self.format, underlying)
+            }
+        }
+    }
+}
+
+impl Error for DecodingError {
+    fn source(&self) -> Option<&(dyn Error + 'static)> {
+        match &self.underlying {
+            None => None,
+            Some(source) => Some(&**source),
+        }
+    }
+}
+
+impl fmt::Display for LimitError {
+    fn fmt(&self, fmt: &mut fmt::Formatter) -> Result<(), fmt::Error> {
+        match self.kind {
+            LimitErrorKind::InsufficientMemory => write!(fmt, "Insufficient memory"),
+            LimitErrorKind::DimensionError => write!(fmt, "Image is too large"),
+            LimitErrorKind::Unsupported { .. } => {
+                write!(fmt, "The following strict limits are specified but not supported by the opertation: ")?;
+                Ok(())
+            }
+        }
+    }
+}
+
+impl Error for LimitError {}
+
+impl fmt::Display for ImageFormatHint {
+    fn fmt(&self, fmt: &mut fmt::Formatter) -> Result<(), fmt::Error> {
+        match self {
+            ImageFormatHint::Exact(format) => write!(fmt, "{:?}", format),
+            ImageFormatHint::Name(name) => write!(fmt, "`{}`", name),
+            ImageFormatHint::PathExtension(ext) => write!(fmt, "`.{:?}`", ext),
+            ImageFormatHint::Unknown => write!(fmt, "`Unknown`"),
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::mem;
+
+    #[allow(dead_code)]
+    // This will fail to compile if the size of this type is large.
+    const ASSERT_SMALLISH: usize = [0][(mem::size_of::<ImageError>() >= 200) as usize];
+
+    #[test]
+    fn test_send_sync_stability() {
+        fn assert_send_sync<T: Send + Sync>() {}
+
+        assert_send_sync::<ImageError>();
+    }
+}

vendor/image/src/imageops/affine.rs

@@ -0,0 +1,410 @@
+//! Functions for performing affine transformations.
+
+use crate::error::{ImageError, ParameterError, ParameterErrorKind};
+use crate::image::{GenericImage, GenericImageView};
+use crate::traits::Pixel;
+use crate::ImageBuffer;
+
+/// Rotate an image 90 degrees clockwise.
+pub fn rotate90<I: GenericImageView>(
+    image: &I,
+) -> ImageBuffer<I::Pixel, Vec<<I::Pixel as Pixel>::Subpixel>>
+where
+    I::Pixel: 'static,
+{
+    let (width, height) = image.dimensions();
+    let mut out = ImageBuffer::new(height, width);
+    let _ = rotate90_in(image, &mut out);
+    out
+}
+
+/// Rotate an image 180 degrees clockwise.
+pub fn rotate180<I: GenericImageView>(
+    image: &I,
+) -> ImageBuffer<I::Pixel, Vec<<I::Pixel as Pixel>::Subpixel>>
+where
+    I::Pixel: 'static,
+{
+    let (width, height) = image.dimensions();
+    let mut out = ImageBuffer::new(width, height);
+    let _ = rotate180_in(image, &mut out);
+    out
+}
+
+/// Rotate an image 270 degrees clockwise.
+pub fn rotate270<I: GenericImageView>(
+    image: &I,
+) -> ImageBuffer<I::Pixel, Vec<<I::Pixel as Pixel>::Subpixel>>
+where
+    I::Pixel: 'static,
+{
+    let (width, height) = image.dimensions();
+    let mut out = ImageBuffer::new(height, width);
+    let _ = rotate270_in(image, &mut out);
+    out
+}
+
+/// Rotate an image 90 degrees clockwise and put the result into the destination [`ImageBuffer`].
+pub fn rotate90_in<I, Container>(
+    image: &I,
+    destination: &mut ImageBuffer<I::Pixel, Container>,
+) -> crate::ImageResult<()>
+where
+    I: GenericImageView,
+    I::Pixel: 'static,
+    Container: std::ops::DerefMut<Target = [<I::Pixel as Pixel>::Subpixel]>,
+{
+    let ((w0, h0), (w1, h1)) = (image.dimensions(), destination.dimensions());
+    if w0 != h1 || h0 != w1 {
+        return Err(ImageError::Parameter(ParameterError::from_kind(
+            ParameterErrorKind::DimensionMismatch,
+        )));
+    }
+
+    for y in 0..h0 {
+        for x in 0..w0 {
+            let p = image.get_pixel(x, y);
+            destination.put_pixel(h0 - y - 1, x, p);
+        }
+    }
+    Ok(())
+}
+
+/// Rotate an image 180 degrees clockwise and put the result into the destination [`ImageBuffer`].
+pub fn rotate180_in<I, Container>(
+    image: &I,
+    destination: &mut ImageBuffer<I::Pixel, Container>,
+) -> crate::ImageResult<()>
+where
+    I: GenericImageView,
+    I::Pixel: 'static,
+    Container: std::ops::DerefMut<Target = [<I::Pixel as Pixel>::Subpixel]>,
+{
+    let ((w0, h0), (w1, h1)) = (image.dimensions(), destination.dimensions());
+    if w0 != w1 || h0 != h1 {
+        return Err(ImageError::Parameter(ParameterError::from_kind(
+            ParameterErrorKind::DimensionMismatch,
+        )));
+    }
+
+    for y in 0..h0 {
+        for x in 0..w0 {
+            let p = image.get_pixel(x, y);
+            destination.put_pixel(w0 - x - 1, h0 - y - 1, p);
+        }
+    }
+    Ok(())
+}
+
+/// Rotate an image 270 degrees clockwise and put the result into the destination [`ImageBuffer`].
+pub fn rotate270_in<I, Container>(
+    image: &I,
+    destination: &mut ImageBuffer<I::Pixel, Container>,
+) -> crate::ImageResult<()>
+where
+    I: GenericImageView,
+    I::Pixel: 'static,
+    Container: std::ops::DerefMut<Target = [<I::Pixel as Pixel>::Subpixel]>,
+{
+    let ((w0, h0), (w1, h1)) = (image.dimensions(), destination.dimensions());
+    if w0 != h1 || h0 != w1 {
+        return Err(ImageError::Parameter(ParameterError::from_kind(
+            ParameterErrorKind::DimensionMismatch,
+        )));
+    }
+
+    for y in 0..h0 {
+        for x in 0..w0 {
+            let p = image.get_pixel(x, y);
+            destination.put_pixel(y, w0 - x - 1, p);
+        }
+    }
+    Ok(())
+}
+
+/// Flip an image horizontally
+pub fn flip_horizontal<I: GenericImageView>(
+    image: &I,
+) -> ImageBuffer<I::Pixel, Vec<<I::Pixel as Pixel>::Subpixel>>
+where
+    I::Pixel: 'static,
+{
+    let (width, height) = image.dimensions();
+    let mut out = ImageBuffer::new(width, height);
+    let _ = flip_horizontal_in(image, &mut out);
+    out
+}
+
+/// Flip an image vertically
+pub fn flip_vertical<I: GenericImageView>(
+    image: &I,
+) -> ImageBuffer<I::Pixel, Vec<<I::Pixel as Pixel>::Subpixel>>
+where
+    I::Pixel: 'static,
+{
+    let (width, height) = image.dimensions();
+    let mut out = ImageBuffer::new(width, height);
+    let _ = flip_vertical_in(image, &mut out);
+    out
+}
+
+/// Flip an image horizontally and put the result into the destination [`ImageBuffer`].
+pub fn flip_horizontal_in<I, Container>(
+    image: &I,
+    destination: &mut ImageBuffer<I::Pixel, Container>,
+) -> crate::ImageResult<()>
+where
+    I: GenericImageView,
+    I::Pixel: 'static,
+    Container: std::ops::DerefMut<Target = [<I::Pixel as Pixel>::Subpixel]>,
+{
+    let ((w0, h0), (w1, h1)) = (image.dimensions(), destination.dimensions());
+    if w0 != w1 || h0 != h1 {
+        return Err(ImageError::Parameter(ParameterError::from_kind(
+            ParameterErrorKind::DimensionMismatch,
+        )));
+    }
+
+    for y in 0..h0 {
+        for x in 0..w0 {
+            let p = image.get_pixel(x, y);
+            destination.put_pixel(w0 - x - 1, y, p);
+        }
+    }
+    Ok(())
+}
+
+/// Flip an image vertically and put the result into the destination [`ImageBuffer`].
+pub fn flip_vertical_in<I, Container>(
+    image: &I,
+    destination: &mut ImageBuffer<I::Pixel, Container>,
+) -> crate::ImageResult<()>
+where
+    I: GenericImageView,
+    I::Pixel: 'static,
+    Container: std::ops::DerefMut<Target = [<I::Pixel as Pixel>::Subpixel]>,
+{
+    let ((w0, h0), (w1, h1)) = (image.dimensions(), destination.dimensions());
+    if w0 != w1 || h0 != h1 {
+        return Err(ImageError::Parameter(ParameterError::from_kind(
+            ParameterErrorKind::DimensionMismatch,
+        )));
+    }
+
+    for y in 0..h0 {
+        for x in 0..w0 {
+            let p = image.get_pixel(x, y);
+            destination.put_pixel(x, h0 - 1 - y, p);
+        }
+    }
+    Ok(())
+}
+
+/// Rotate an image 180 degrees clockwise in place.
+pub fn rotate180_in_place<I: GenericImage>(image: &mut I) {
+    let (width, height) = image.dimensions();
+
+    for y in 0..height / 2 {
+        for x in 0..width {
+            let p = image.get_pixel(x, y);
+
+            let x2 = width - x - 1;
+            let y2 = height - y - 1;
+
+            let p2 = image.get_pixel(x2, y2);
+            image.put_pixel(x, y, p2);
+            image.put_pixel(x2, y2, p);
+        }
+    }
+
+    if height % 2 != 0 {
+        let middle = height / 2;
+
+        for x in 0..width / 2 {
+            let p = image.get_pixel(x, middle);
+            let x2 = width - x - 1;
+
+            let p2 = image.get_pixel(x2, middle);
+            image.put_pixel(x, middle, p2);
+            image.put_pixel(x2, middle, p);
+        }
+    }
+}
+
+/// Flip an image horizontally in place.
+pub fn flip_horizontal_in_place<I: GenericImage>(image: &mut I) {
+    let (width, height) = image.dimensions();
+
+    for y in 0..height {
+        for x in 0..width / 2 {
+            let x2 = width - x - 1;
+            let p2 = image.get_pixel(x2, y);
+            let p = image.get_pixel(x, y);
+            image.put_pixel(x2, y, p);
+            image.put_pixel(x, y, p2);
+        }
+    }
+}
+
+/// Flip an image vertically in place.
+pub fn flip_vertical_in_place<I: GenericImage>(image: &mut I) {
+    let (width, height) = image.dimensions();
+
+    for y in 0..height / 2 {
+        for x in 0..width {
+            let y2 = height - y - 1;
+            let p2 = image.get_pixel(x, y2);
+            let p = image.get_pixel(x, y);
+            image.put_pixel(x, y2, p);
+            image.put_pixel(x, y, p2);
+        }
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::{
+        flip_horizontal, flip_horizontal_in_place, flip_vertical, flip_vertical_in_place,
+        rotate180, rotate180_in_place, rotate270, rotate90,
+    };
+    use crate::image::GenericImage;
+    use crate::traits::Pixel;
+    use crate::{GrayImage, ImageBuffer};
+
+    macro_rules! assert_pixels_eq {
+        ($actual:expr, $expected:expr) => {{
+            let actual_dim = $actual.dimensions();
+            let expected_dim = $expected.dimensions();
+
+            if actual_dim != expected_dim {
+                panic!(
+                    "dimensions do not match. \
+                     actual: {:?}, expected: {:?}",
+                    actual_dim, expected_dim
+                )
+            }
+
+            let diffs = pixel_diffs($actual, $expected);
+
+            if !diffs.is_empty() {
+                let mut err = "".to_string();
+
+                let diff_messages = diffs
+                    .iter()
+                    .take(5)
+                    .map(|d| format!("\nactual: {:?}, expected {:?} ", d.0, d.1))
+                    .collect::<Vec<_>>()
+                    .join("");
+
+                err.push_str(&diff_messages);
+                panic!("pixels do not match. {:?}", err)
+            }
+        }};
+    }
+
+    #[test]
+    fn test_rotate90() {
+        let image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(2, 3, vec![10u8, 00u8, 11u8, 01u8, 12u8, 02u8]).unwrap();
+
+        assert_pixels_eq!(&rotate90(&image), &expected);
+    }
+
+    #[test]
+    fn test_rotate180() {
+        let image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![12u8, 11u8, 10u8, 02u8, 01u8, 00u8]).unwrap();
+
+        assert_pixels_eq!(&rotate180(&image), &expected);
+    }
+
+    #[test]
+    fn test_rotate270() {
+        let image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(2, 3, vec![02u8, 12u8, 01u8, 11u8, 00u8, 10u8]).unwrap();
+
+        assert_pixels_eq!(&rotate270(&image), &expected);
+    }
+
+    #[test]
+    fn test_rotate180_in_place() {
+        let mut image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![12u8, 11u8, 10u8, 02u8, 01u8, 00u8]).unwrap();
+
+        rotate180_in_place(&mut image);
+
+        assert_pixels_eq!(&image, &expected);
+    }
+
+    #[test]
+    fn test_flip_horizontal() {
+        let image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![02u8, 01u8, 00u8, 12u8, 11u8, 10u8]).unwrap();
+
+        assert_pixels_eq!(&flip_horizontal(&image), &expected);
+    }
+
+    #[test]
+    fn test_flip_vertical() {
+        let image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![10u8, 11u8, 12u8, 00u8, 01u8, 02u8]).unwrap();
+
+        assert_pixels_eq!(&flip_vertical(&image), &expected);
+    }
+
+    #[test]
+    fn test_flip_horizontal_in_place() {
+        let mut image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![02u8, 01u8, 00u8, 12u8, 11u8, 10u8]).unwrap();
+
+        flip_horizontal_in_place(&mut image);
+
+        assert_pixels_eq!(&image, &expected);
+    }
+
+    #[test]
+    fn test_flip_vertical_in_place() {
+        let mut image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![10u8, 11u8, 12u8, 00u8, 01u8, 02u8]).unwrap();
+
+        flip_vertical_in_place(&mut image);
+
+        assert_pixels_eq!(&image, &expected);
+    }
+
+    fn pixel_diffs<I, J, P>(left: &I, right: &J) -> Vec<((u32, u32, P), (u32, u32, P))>
+    where
+        I: GenericImage<Pixel = P>,
+        J: GenericImage<Pixel = P>,
+        P: Pixel + Eq,
+    {
+        left.pixels()
+            .zip(right.pixels())
+            .filter(|&(p, q)| p != q)
+            .collect::<Vec<_>>()
+    }
+}

vendor/image/src/imageops/colorops.rs

@@ -0,0 +1,646 @@
+//! Functions for altering and converting the color of pixelbufs
+
+use num_traits::NumCast;
+use std::f64::consts::PI;
+
+use crate::color::{FromColor, IntoColor, Luma, LumaA, Rgba};
+use crate::image::{GenericImage, GenericImageView};
+use crate::traits::{Pixel, Primitive};
+use crate::utils::clamp;
+use crate::ImageBuffer;
+
+type Subpixel<I> = <<I as GenericImageView>::Pixel as Pixel>::Subpixel;
+
+/// Convert the supplied image to grayscale. Alpha channel is discarded.
+pub fn grayscale<I: GenericImageView>(
+    image: &I,
+) -> ImageBuffer<Luma<Subpixel<I>>, Vec<Subpixel<I>>> {
+    grayscale_with_type(image)
+}
+
+/// Convert the supplied image to grayscale. Alpha channel is preserved.
+pub fn grayscale_alpha<I: GenericImageView>(
+    image: &I,
+) -> ImageBuffer<LumaA<Subpixel<I>>, Vec<Subpixel<I>>> {
+    grayscale_with_type_alpha(image)
+}
+
+/// Convert the supplied image to a grayscale image with the specified pixel type. Alpha channel is discarded.
+pub fn grayscale_with_type<NewPixel, I: GenericImageView>(
+    image: &I,
+) -> ImageBuffer<NewPixel, Vec<NewPixel::Subpixel>>
+where
+    NewPixel: Pixel + FromColor<Luma<Subpixel<I>>>,
+{
+    let (width, height) = image.dimensions();
+    let mut out = ImageBuffer::new(width, height);
+
+    for (x, y, pixel) in image.pixels() {
+        let grayscale = pixel.to_luma();
+        let new_pixel = grayscale.into_color(); // no-op for luma->luma
+
+        out.put_pixel(x, y, new_pixel);
+    }
+
+    out
+}
+
+/// Convert the supplied image to a grayscale image with the specified pixel type. Alpha channel is preserved.
+pub fn grayscale_with_type_alpha<NewPixel, I: GenericImageView>(
+    image: &I,
+) -> ImageBuffer<NewPixel, Vec<NewPixel::Subpixel>>
+where
+    NewPixel: Pixel + FromColor<LumaA<Subpixel<I>>>,
+{
+    let (width, height) = image.dimensions();
+    let mut out = ImageBuffer::new(width, height);
+
+    for (x, y, pixel) in image.pixels() {
+        let grayscale = pixel.to_luma_alpha();
+        let new_pixel = grayscale.into_color(); // no-op for luma->luma
+
+        out.put_pixel(x, y, new_pixel);
+    }
+
+    out
+}
+
+/// Invert each pixel within the supplied image.
+/// This function operates in place.
+pub fn invert<I: GenericImage>(image: &mut I) {
+    // TODO find a way to use pixels?
+    let (width, height) = image.dimensions();
+
+    for y in 0..height {
+        for x in 0..width {
+            let mut p = image.get_pixel(x, y);
+            p.invert();
+
+            image.put_pixel(x, y, p);
+        }
+    }
+}
+
+/// Adjust the contrast of the supplied image.
+/// ```contrast``` is the amount to adjust the contrast by.
+/// Negative values decrease the contrast and positive values increase the contrast.
+///
+/// *[See also `contrast_in_place`.][contrast_in_place]*
+pub fn contrast<I, P, S>(image: &I, contrast: f32) -> ImageBuffer<P, Vec<S>>
+where
+    I: GenericImageView<Pixel = P>,
+    P: Pixel<Subpixel = S> + 'static,
+    S: Primitive + 'static,
+{
+    let (width, height) = image.dimensions();
+    let mut out = ImageBuffer::new(width, height);
+
+    let max = S::DEFAULT_MAX_VALUE;
+    let max: f32 = NumCast::from(max).unwrap();
+
+    let percent = ((100.0 + contrast) / 100.0).powi(2);
+
+    for (x, y, pixel) in image.pixels() {
+        let f = pixel.map(|b| {
+            let c: f32 = NumCast::from(b).unwrap();
+
+            let d = ((c / max - 0.5) * percent + 0.5) * max;
+            let e = clamp(d, 0.0, max);
+
+            NumCast::from(e).unwrap()
+        });
+        out.put_pixel(x, y, f);
+    }
+
+    out
+}
+
+/// Adjust the contrast of the supplied image in place.
+/// ```contrast``` is the amount to adjust the contrast by.
+/// Negative values decrease the contrast and positive values increase the contrast.
+///
+/// *[See also `contrast`.][contrast]*
+pub fn contrast_in_place<I>(image: &mut I, contrast: f32)
+where
+    I: GenericImage,
+{
+    let (width, height) = image.dimensions();
+
+    let max = <I::Pixel as Pixel>::Subpixel::DEFAULT_MAX_VALUE;
+    let max: f32 = NumCast::from(max).unwrap();
+
+    let percent = ((100.0 + contrast) / 100.0).powi(2);
+
+    // TODO find a way to use pixels?
+    for y in 0..height {
+        for x in 0..width {
+            let f = image.get_pixel(x, y).map(|b| {
+                let c: f32 = NumCast::from(b).unwrap();
+
+                let d = ((c / max - 0.5) * percent + 0.5) * max;
+                let e = clamp(d, 0.0, max);
+
+                NumCast::from(e).unwrap()
+            });
+
+            image.put_pixel(x, y, f);
+        }
+    }
+}
+
+/// Brighten the supplied image.
+/// ```value``` is the amount to brighten each pixel by.
+/// Negative values decrease the brightness and positive values increase it.
+///
+/// *[See also `brighten_in_place`.][brighten_in_place]*
+pub fn brighten<I, P, S>(image: &I, value: i32) -> ImageBuffer<P, Vec<S>>
+where
+    I: GenericImageView<Pixel = P>,
+    P: Pixel<Subpixel = S> + 'static,
+    S: Primitive + 'static,
+{
+    let (width, height) = image.dimensions();
+    let mut out = ImageBuffer::new(width, height);
+
+    let max = S::DEFAULT_MAX_VALUE;
+    let max: i32 = NumCast::from(max).unwrap();
+
+    for (x, y, pixel) in image.pixels() {
+        let e = pixel.map_with_alpha(
+            |b| {
+                let c: i32 = NumCast::from(b).unwrap();
+                let d = clamp(c + value, 0, max);
+
+                NumCast::from(d).unwrap()
+            },
+            |alpha| alpha,
+        );
+        out.put_pixel(x, y, e);
+    }
+
+    out
+}
+
+/// Brighten the supplied image in place.
+/// ```value``` is the amount to brighten each pixel by.
+/// Negative values decrease the brightness and positive values increase it.
+///
+/// *[See also `brighten`.][brighten]*
+pub fn brighten_in_place<I>(image: &mut I, value: i32)
+where
+    I: GenericImage,
+{
+    let (width, height) = image.dimensions();
+
+    let max = <I::Pixel as Pixel>::Subpixel::DEFAULT_MAX_VALUE;
+    let max: i32 = NumCast::from(max).unwrap(); // TODO what does this do for f32? clamp at 1??
+
+    // TODO find a way to use pixels?
+    for y in 0..height {
+        for x in 0..width {
+            let e = image.get_pixel(x, y).map_with_alpha(
+                |b| {
+                    let c: i32 = NumCast::from(b).unwrap();
+                    let d = clamp(c + value, 0, max);
+
+                    NumCast::from(d).unwrap()
+                },
+                |alpha| alpha,
+            );
+
+            image.put_pixel(x, y, e);
+        }
+    }
+}
+
+/// Hue rotate the supplied image.
+/// `value` is the degrees to rotate each pixel by.
+/// 0 and 360 do nothing, the rest rotates by the given degree value.
+/// just like the css webkit filter hue-rotate(180)
+///
+/// *[See also `huerotate_in_place`.][huerotate_in_place]*
+pub fn huerotate<I, P, S>(image: &I, value: i32) -> ImageBuffer<P, Vec<S>>
+where
+    I: GenericImageView<Pixel = P>,
+    P: Pixel<Subpixel = S> + 'static,
+    S: Primitive + 'static,
+{
+    let (width, height) = image.dimensions();
+    let mut out = ImageBuffer::new(width, height);
+
+    let angle: f64 = NumCast::from(value).unwrap();
+
+    let cosv = (angle * PI / 180.0).cos();
+    let sinv = (angle * PI / 180.0).sin();
+    let matrix: [f64; 9] = [
+        // Reds
+        0.213 + cosv * 0.787 - sinv * 0.213,
+        0.715 - cosv * 0.715 - sinv * 0.715,
+        0.072 - cosv * 0.072 + sinv * 0.928,
+        // Greens
+        0.213 - cosv * 0.213 + sinv * 0.143,
+        0.715 + cosv * 0.285 + sinv * 0.140,
+        0.072 - cosv * 0.072 - sinv * 0.283,
+        // Blues
+        0.213 - cosv * 0.213 - sinv * 0.787,
+        0.715 - cosv * 0.715 + sinv * 0.715,
+        0.072 + cosv * 0.928 + sinv * 0.072,
+    ];
+    for (x, y, pixel) in out.enumerate_pixels_mut() {
+        let p = image.get_pixel(x, y);
+
+        #[allow(deprecated)]
+        let (k1, k2, k3, k4) = p.channels4();
+        let vec: (f64, f64, f64, f64) = (
+            NumCast::from(k1).unwrap(),
+            NumCast::from(k2).unwrap(),
+            NumCast::from(k3).unwrap(),
+            NumCast::from(k4).unwrap(),
+        );
+
+        let r = vec.0;
+        let g = vec.1;
+        let b = vec.2;
+
+        let new_r = matrix[0] * r + matrix[1] * g + matrix[2] * b;
+        let new_g = matrix[3] * r + matrix[4] * g + matrix[5] * b;
+        let new_b = matrix[6] * r + matrix[7] * g + matrix[8] * b;
+        let max = 255f64;
+
+        #[allow(deprecated)]
+        let outpixel = Pixel::from_channels(
+            NumCast::from(clamp(new_r, 0.0, max)).unwrap(),
+            NumCast::from(clamp(new_g, 0.0, max)).unwrap(),
+            NumCast::from(clamp(new_b, 0.0, max)).unwrap(),
+            NumCast::from(clamp(vec.3, 0.0, max)).unwrap(),
+        );
+        *pixel = outpixel;
+    }
+    out
+}
+
+/// Hue rotate the supplied image in place.
+/// `value` is the degrees to rotate each pixel by.
+/// 0 and 360 do nothing, the rest rotates by the given degree value.
+/// just like the css webkit filter hue-rotate(180)
+///
+/// *[See also `huerotate`.][huerotate]*
+pub fn huerotate_in_place<I>(image: &mut I, value: i32)
+where
+    I: GenericImage,
+{
+    let (width, height) = image.dimensions();
+
+    let angle: f64 = NumCast::from(value).unwrap();
+
+    let cosv = (angle * PI / 180.0).cos();
+    let sinv = (angle * PI / 180.0).sin();
+    let matrix: [f64; 9] = [
+        // Reds
+        0.213 + cosv * 0.787 - sinv * 0.213,
+        0.715 - cosv * 0.715 - sinv * 0.715,
+        0.072 - cosv * 0.072 + sinv * 0.928,
+        // Greens
+        0.213 - cosv * 0.213 + sinv * 0.143,
+        0.715 + cosv * 0.285 + sinv * 0.140,
+        0.072 - cosv * 0.072 - sinv * 0.283,
+        // Blues
+        0.213 - cosv * 0.213 - sinv * 0.787,
+        0.715 - cosv * 0.715 + sinv * 0.715,
+        0.072 + cosv * 0.928 + sinv * 0.072,
+    ];
+
+    // TODO find a way to use pixels?
+    for y in 0..height {
+        for x in 0..width {
+            let pixel = image.get_pixel(x, y);
+
+            #[allow(deprecated)]
+            let (k1, k2, k3, k4) = pixel.channels4();
+
+            let vec: (f64, f64, f64, f64) = (
+                NumCast::from(k1).unwrap(),
+                NumCast::from(k2).unwrap(),
+                NumCast::from(k3).unwrap(),
+                NumCast::from(k4).unwrap(),
+            );
+
+            let r = vec.0;
+            let g = vec.1;
+            let b = vec.2;
+
+            let new_r = matrix[0] * r + matrix[1] * g + matrix[2] * b;
+            let new_g = matrix[3] * r + matrix[4] * g + matrix[5] * b;
+            let new_b = matrix[6] * r + matrix[7] * g + matrix[8] * b;
+            let max = 255f64;
+
+            #[allow(deprecated)]
+            let outpixel = Pixel::from_channels(
+                NumCast::from(clamp(new_r, 0.0, max)).unwrap(),
+                NumCast::from(clamp(new_g, 0.0, max)).unwrap(),
+                NumCast::from(clamp(new_b, 0.0, max)).unwrap(),
+                NumCast::from(clamp(vec.3, 0.0, max)).unwrap(),
+            );
+
+            image.put_pixel(x, y, outpixel);
+        }
+    }
+}
+
+/// A color map
+pub trait ColorMap {
+    /// The color type on which the map operates on
+    type Color;
+    /// Returns the index of the closest match of `color`
+    /// in the color map.
+    fn index_of(&self, color: &Self::Color) -> usize;
+    /// Looks up color by index in the color map.  If `idx` is out of range for the color map, or
+    /// ColorMap doesn't implement `lookup` `None` is returned.
+    fn lookup(&self, index: usize) -> Option<Self::Color> {
+        let _ = index;
+        None
+    }
+    /// Determine if this implementation of ColorMap overrides the default `lookup`.
+    fn has_lookup(&self) -> bool {
+        false
+    }
+    /// Maps `color` to the closest color in the color map.
+    fn map_color(&self, color: &mut Self::Color);
+}
+
+/// A bi-level color map
+///
+/// # Examples
+/// ```
+/// use image::imageops::colorops::{index_colors, BiLevel, ColorMap};
+/// use image::{ImageBuffer, Luma};
+///
+/// let (w, h) = (16, 16);
+/// // Create an image with a smooth horizontal gradient from black (0) to white (255).
+/// let gray = ImageBuffer::from_fn(w, h, |x, y| -> Luma<u8> { [(255 * x / w) as u8].into() });
+/// // Mapping the gray image through the `BiLevel` filter should map gray pixels less than half
+/// // intensity (127) to black (0), and anything greater to white (255).
+/// let cmap = BiLevel;
+/// let palletized = index_colors(&gray, &cmap);
+/// let mapped = ImageBuffer::from_fn(w, h, |x, y| {
+///     let p = palletized.get_pixel(x, y);
+///     cmap.lookup(p.0[0] as usize)
+///         .expect("indexed color out-of-range")
+/// });
+/// // Create an black and white image of expected output.
+/// let bw = ImageBuffer::from_fn(w, h, |x, y| -> Luma<u8> {
+///     if x <= (w / 2) {
+///         [0].into()
+///     } else {
+///         [255].into()
+///     }
+/// });
+/// assert_eq!(mapped, bw);
+/// ```
+#[derive(Clone, Copy)]
+pub struct BiLevel;
+
+impl ColorMap for BiLevel {
+    type Color = Luma<u8>;
+
+    #[inline(always)]
+    fn index_of(&self, color: &Luma<u8>) -> usize {
+        let luma = color.0;
+        if luma[0] > 127 {
+            1
+        } else {
+            0
+        }
+    }
+
+    #[inline(always)]
+    fn lookup(&self, idx: usize) -> Option<Self::Color> {
+        match idx {
+            0 => Some([0].into()),
+            1 => Some([255].into()),
+            _ => None,
+        }
+    }
+
+    /// Indicate NeuQuant implements `lookup`.
+    fn has_lookup(&self) -> bool {
+        true
+    }
+
+    #[inline(always)]
+    fn map_color(&self, color: &mut Luma<u8>) {
+        let new_color = 0xFF * self.index_of(color) as u8;
+        let luma = &mut color.0;
+        luma[0] = new_color;
+    }
+}
+
+impl ColorMap for color_quant::NeuQuant {
+    type Color = Rgba<u8>;
+
+    #[inline(always)]
+    fn index_of(&self, color: &Rgba<u8>) -> usize {
+        self.index_of(color.channels())
+    }
+
+    #[inline(always)]
+    fn lookup(&self, idx: usize) -> Option<Self::Color> {
+        self.lookup(idx).map(|p| p.into())
+    }
+
+    /// Indicate NeuQuant implements `lookup`.
+    fn has_lookup(&self) -> bool {
+        true
+    }
+
+    #[inline(always)]
+    fn map_color(&self, color: &mut Rgba<u8>) {
+        self.map_pixel(color.channels_mut())
+    }
+}
+
+/// Floyd-Steinberg error diffusion
+fn diffuse_err<P: Pixel<Subpixel = u8>>(pixel: &mut P, error: [i16; 3], factor: i16) {
+    for (e, c) in error.iter().zip(pixel.channels_mut().iter_mut()) {
+        *c = match <i16 as From<_>>::from(*c) + e * factor / 16 {
+            val if val < 0 => 0,
+            val if val > 0xFF => 0xFF,
+            val => val as u8,
+        }
+    }
+}
+
+macro_rules! do_dithering(
+    ($map:expr, $image:expr, $err:expr, $x:expr, $y:expr) => (
+        {
+            let old_pixel = $image[($x, $y)];
+            let new_pixel = $image.get_pixel_mut($x, $y);
+            $map.map_color(new_pixel);
+            for ((e, &old), &new) in $err.iter_mut()
+                                        .zip(old_pixel.channels().iter())
+                                        .zip(new_pixel.channels().iter())
+            {
+                *e = <i16 as From<_>>::from(old) - <i16 as From<_>>::from(new)
+            }
+        }
+    )
+);
+
+/// Reduces the colors of the image using the supplied `color_map` while applying
+/// Floyd-Steinberg dithering to improve the visual conception
+pub fn dither<Pix, Map>(image: &mut ImageBuffer<Pix, Vec<u8>>, color_map: &Map)
+where
+    Map: ColorMap<Color = Pix> + ?Sized,
+    Pix: Pixel<Subpixel = u8> + 'static,
+{
+    let (width, height) = image.dimensions();
+    let mut err: [i16; 3] = [0; 3];
+    for y in 0..height - 1 {
+        let x = 0;
+        do_dithering!(color_map, image, err, x, y);
+        diffuse_err(image.get_pixel_mut(x + 1, y), err, 7);
+        diffuse_err(image.get_pixel_mut(x, y + 1), err, 5);
+        diffuse_err(image.get_pixel_mut(x + 1, y + 1), err, 1);
+        for x in 1..width - 1 {
+            do_dithering!(color_map, image, err, x, y);
+            diffuse_err(image.get_pixel_mut(x + 1, y), err, 7);
+            diffuse_err(image.get_pixel_mut(x - 1, y + 1), err, 3);
+            diffuse_err(image.get_pixel_mut(x, y + 1), err, 5);
+            diffuse_err(image.get_pixel_mut(x + 1, y + 1), err, 1);
+        }
+        let x = width - 1;
+        do_dithering!(color_map, image, err, x, y);
+        diffuse_err(image.get_pixel_mut(x - 1, y + 1), err, 3);
+        diffuse_err(image.get_pixel_mut(x, y + 1), err, 5);
+    }
+    let y = height - 1;
+    let x = 0;
+    do_dithering!(color_map, image, err, x, y);
+    diffuse_err(image.get_pixel_mut(x + 1, y), err, 7);
+    for x in 1..width - 1 {
+        do_dithering!(color_map, image, err, x, y);
+        diffuse_err(image.get_pixel_mut(x + 1, y), err, 7);
+    }
+    let x = width - 1;
+    do_dithering!(color_map, image, err, x, y);
+}
+
+/// Reduces the colors using the supplied `color_map` and returns an image of the indices
+pub fn index_colors<Pix, Map>(
+    image: &ImageBuffer<Pix, Vec<u8>>,
+    color_map: &Map,
+) -> ImageBuffer<Luma<u8>, Vec<u8>>
+where
+    Map: ColorMap<Color = Pix> + ?Sized,
+    Pix: Pixel<Subpixel = u8> + 'static,
+{
+    let mut indices = ImageBuffer::new(image.width(), image.height());
+    for (pixel, idx) in image.pixels().zip(indices.pixels_mut()) {
+        *idx = Luma([color_map.index_of(pixel) as u8])
+    }
+    indices
+}
+
+#[cfg(test)]
+mod test {
+
+    use super::*;
+    use crate::{GrayImage, ImageBuffer};
+
+    macro_rules! assert_pixels_eq {
+        ($actual:expr, $expected:expr) => {{
+            let actual_dim = $actual.dimensions();
+            let expected_dim = $expected.dimensions();
+
+            if actual_dim != expected_dim {
+                panic!(
+                    "dimensions do not match. \
+                     actual: {:?}, expected: {:?}",
+                    actual_dim, expected_dim
+                )
+            }
+
+            let diffs = pixel_diffs($actual, $expected);
+
+            if !diffs.is_empty() {
+                let mut err = "".to_string();
+
+                let diff_messages = diffs
+                    .iter()
+                    .take(5)
+                    .map(|d| format!("\nactual: {:?}, expected {:?} ", d.0, d.1))
+                    .collect::<Vec<_>>()
+                    .join("");
+
+                err.push_str(&diff_messages);
+                panic!("pixels do not match. {:?}", err)
+            }
+        }};
+    }
+
+    #[test]
+    fn test_dither() {
+        let mut image = ImageBuffer::from_raw(2, 2, vec![127, 127, 127, 127]).unwrap();
+        let cmap = BiLevel;
+        dither(&mut image, &cmap);
+        assert_eq!(&*image, &[0, 0xFF, 0xFF, 0]);
+        assert_eq!(index_colors(&image, &cmap).into_raw(), vec![0, 1, 1, 0])
+    }
+
+    #[test]
+    fn test_grayscale() {
+        let mut image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        assert_pixels_eq!(&grayscale(&mut image), &expected);
+    }
+
+    #[test]
+    fn test_invert() {
+        let mut image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![255u8, 254u8, 253u8, 245u8, 244u8, 243u8]).unwrap();
+
+        invert(&mut image);
+        assert_pixels_eq!(&image, &expected);
+    }
+    #[test]
+    fn test_brighten() {
+        let image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![10u8, 11u8, 12u8, 20u8, 21u8, 22u8]).unwrap();
+
+        assert_pixels_eq!(&brighten(&image, 10), &expected);
+    }
+
+    #[test]
+    fn test_brighten_place() {
+        let mut image: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![00u8, 01u8, 02u8, 10u8, 11u8, 12u8]).unwrap();
+
+        let expected: GrayImage =
+            ImageBuffer::from_raw(3, 2, vec![10u8, 11u8, 12u8, 20u8, 21u8, 22u8]).unwrap();
+
+        brighten_in_place(&mut image, 10);
+        assert_pixels_eq!(&image, &expected);
+    }
+
+    fn pixel_diffs<I, J, P>(left: &I, right: &J) -> Vec<((u32, u32, P), (u32, u32, P))>
+    where
+        I: GenericImage<Pixel = P>,
+        J: GenericImage<Pixel = P>,
+        P: Pixel + Eq,
+    {
+        left.pixels()
+            .zip(right.pixels())
+            .filter(|&(p, q)| p != q)
+            .collect::<Vec<_>>()
+    }
+}

vendor/image/src/imageops/mod.rs

@@ -0,0 +1,485 @@
+//! Image Processing Functions
+use std::cmp;
+
+use crate::image::{GenericImage, GenericImageView, SubImage};
+use crate::traits::{Lerp, Pixel, Primitive};
+
+pub use self::sample::FilterType;
+
+pub use self::sample::FilterType::{CatmullRom, Gaussian, Lanczos3, Nearest, Triangle};
+
+/// Affine transformations
+pub use self::affine::{
+    flip_horizontal, flip_horizontal_in, flip_horizontal_in_place, flip_vertical, flip_vertical_in,
+    flip_vertical_in_place, rotate180, rotate180_in, rotate180_in_place, rotate270, rotate270_in,
+    rotate90, rotate90_in,
+};
+
+/// Image sampling
+pub use self::sample::{
+    blur, filter3x3, interpolate_bilinear, interpolate_nearest, resize, sample_bilinear,
+    sample_nearest, thumbnail, unsharpen,
+};
+
+/// Color operations
+pub use self::colorops::{
+    brighten, contrast, dither, grayscale, grayscale_alpha, grayscale_with_type,
+    grayscale_with_type_alpha, huerotate, index_colors, invert, BiLevel, ColorMap,
+};
+
+mod affine;
+// Public only because of Rust bug:
+// https://github.com/rust-lang/rust/issues/18241
+pub mod colorops;
+mod sample;
+
+/// Return a mutable view into an image
+/// The coordinates set the position of the top left corner of the crop.
+pub fn crop<I: GenericImageView>(
+    image: &mut I,
+    x: u32,
+    y: u32,
+    width: u32,
+    height: u32,
+) -> SubImage<&mut I> {
+    let (x, y, width, height) = crop_dimms(image, x, y, width, height);
+    SubImage::new(image, x, y, width, height)
+}
+
+/// Return an immutable view into an image
+/// The coordinates set the position of the top left corner of the crop.
+pub fn crop_imm<I: GenericImageView>(
+    image: &I,
+    x: u32,
+    y: u32,
+    width: u32,
+    height: u32,
+) -> SubImage<&I> {
+    let (x, y, width, height) = crop_dimms(image, x, y, width, height);
+    SubImage::new(image, x, y, width, height)
+}
+
+fn crop_dimms<I: GenericImageView>(
+    image: &I,
+    x: u32,
+    y: u32,
+    width: u32,
+    height: u32,
+) -> (u32, u32, u32, u32) {
+    let (iwidth, iheight) = image.dimensions();
+
+    let x = cmp::min(x, iwidth);
+    let y = cmp::min(y, iheight);
+
+    let height = cmp::min(height, iheight - y);
+    let width = cmp::min(width, iwidth - x);
+
+    (x, y, width, height)
+}
+
+/// Calculate the region that can be copied from top to bottom.
+///
+/// Given image size of bottom and top image, and a point at which we want to place the top image
+/// onto the bottom image, how large can we be? Have to wary of the following issues:
+/// * Top might be larger than bottom
+/// * Overflows in the computation
+/// * Coordinates could be completely out of bounds
+///
+/// The main idea is to make use of inequalities provided by the nature of `saturating_add` and
+/// `saturating_sub`. These intrinsically validate that all resulting coordinates will be in bounds
+/// for both images.
+///
+/// We want that all these coordinate accesses are safe:
+/// 1. `bottom.get_pixel(x + [0..x_range), y + [0..y_range))`
+/// 2. `top.get_pixel([0..x_range), [0..y_range))`
+///
+/// Proof that the function provides the necessary bounds for width. Note that all unaugmented math
+/// operations are to be read in standard arithmetic, not integer arithmetic. Since no direct
+/// integer arithmetic occurs in the implementation, this is unambiguous.
+///
+/// ```text
+/// Three short notes/lemmata:
+/// - Iff `(a - b) <= 0` then `a.saturating_sub(b) = 0`
+/// - Iff `(a - b) >= 0` then `a.saturating_sub(b) = a - b`
+/// - If  `a <= c` then `a.saturating_sub(b) <= c.saturating_sub(b)`
+///
+/// 1.1 We show that if `bottom_width <= x`, then `x_range = 0` therefore `x + [0..x_range)` is empty.
+///
+/// x_range
+///  = (top_width.saturating_add(x).min(bottom_width)).saturating_sub(x)
+/// <= bottom_width.saturating_sub(x)
+///
+/// bottom_width <= x
+/// <==> bottom_width - x <= 0
+/// <==> bottom_width.saturating_sub(x) = 0
+///  ==> x_range <= 0
+///  ==> x_range  = 0
+///
+/// 1.2 If `x < bottom_width` then `x + x_range < bottom_width`
+///
+/// x + x_range
+/// <= x + bottom_width.saturating_sub(x)
+///  = x + (bottom_width - x)
+///  = bottom_width
+///
+/// 2. We show that `x_range <= top_width`
+///
+/// x_range
+///  = (top_width.saturating_add(x).min(bottom_width)).saturating_sub(x)
+/// <= top_width.saturating_add(x).saturating_sub(x)
+/// <= (top_wdith + x).saturating_sub(x)
+///  = top_width (due to `top_width >= 0` and `x >= 0`)
+/// ```
+///
+/// Proof is the same for height.
+pub fn overlay_bounds(
+    (bottom_width, bottom_height): (u32, u32),
+    (top_width, top_height): (u32, u32),
+    x: u32,
+    y: u32,
+) -> (u32, u32) {
+    let x_range = top_width
+        .saturating_add(x) // Calculate max coordinate
+        .min(bottom_width) // Restrict to lower width
+        .saturating_sub(x); // Determinate length from start `x`
+    let y_range = top_height
+        .saturating_add(y)
+        .min(bottom_height)
+        .saturating_sub(y);
+    (x_range, y_range)
+}
+
+/// Calculate the region that can be copied from top to bottom.
+///
+/// Given image size of bottom and top image, and a point at which we want to place the top image
+/// onto the bottom image, how large can we be? Have to wary of the following issues:
+/// * Top might be larger than bottom
+/// * Overflows in the computation
+/// * Coordinates could be completely out of bounds
+///
+/// The returned value is of the form:
+///
+/// `(origin_bottom_x, origin_bottom_y, origin_top_x, origin_top_y, x_range, y_range)`
+///
+/// The main idea is to do computations on i64's and then clamp to image dimensions.
+/// In particular, we want to ensure that all these coordinate accesses are safe:
+/// 1. `bottom.get_pixel(origin_bottom_x + [0..x_range), origin_bottom_y + [0..y_range))`
+/// 2. `top.get_pixel(origin_top_y + [0..x_range), origin_top_y + [0..y_range))`
+///
+fn overlay_bounds_ext(
+    (bottom_width, bottom_height): (u32, u32),
+    (top_width, top_height): (u32, u32),
+    x: i64,
+    y: i64,
+) -> (u32, u32, u32, u32, u32, u32) {
+    // Return a predictable value if the two images don't overlap at all.
+    if x > i64::from(bottom_width)
+        || y > i64::from(bottom_height)
+        || x.saturating_add(i64::from(top_width)) <= 0
+        || y.saturating_add(i64::from(top_height)) <= 0
+    {
+        return (0, 0, 0, 0, 0, 0);
+    }
+
+    // Find the maximum x and y coordinates in terms of the bottom image.
+    let max_x = x.saturating_add(i64::from(top_width));
+    let max_y = y.saturating_add(i64::from(top_height));
+
+    // Clip the origin and maximum coordinates to the bounds of the bottom image.
+    // Casting to a u32 is safe because both 0 and `bottom_{width,height}` fit
+    // into 32-bits.
+    let max_inbounds_x = max_x.clamp(0, i64::from(bottom_width)) as u32;
+    let max_inbounds_y = max_y.clamp(0, i64::from(bottom_height)) as u32;
+    let origin_bottom_x = x.clamp(0, i64::from(bottom_width)) as u32;
+    let origin_bottom_y = y.clamp(0, i64::from(bottom_height)) as u32;
+
+    // The range is the difference between the maximum inbounds coordinates and
+    // the clipped origin. Unchecked subtraction is safe here because both are
+    // always positive and `max_inbounds_{x,y}` >= `origin_{x,y}` due to
+    // `top_{width,height}` being >= 0.
+    let x_range = max_inbounds_x - origin_bottom_x;
+    let y_range = max_inbounds_y - origin_bottom_y;
+
+    // If x (or y) is negative, then the origin of the top image is shifted by -x (or -y).
+    let origin_top_x = x.saturating_mul(-1).clamp(0, i64::from(top_width)) as u32;
+    let origin_top_y = y.saturating_mul(-1).clamp(0, i64::from(top_height)) as u32;
+
+    (
+        origin_bottom_x,
+        origin_bottom_y,
+        origin_top_x,
+        origin_top_y,
+        x_range,
+        y_range,
+    )
+}
+
+/// Overlay an image at a given coordinate (x, y)
+pub fn overlay<I, J>(bottom: &mut I, top: &J, x: i64, y: i64)
+where
+    I: GenericImage,
+    J: GenericImageView<Pixel = I::Pixel>,
+{
+    let bottom_dims = bottom.dimensions();
+    let top_dims = top.dimensions();
+
+    // Crop our top image if we're going out of bounds
+    let (origin_bottom_x, origin_bottom_y, origin_top_x, origin_top_y, range_width, range_height) =
+        overlay_bounds_ext(bottom_dims, top_dims, x, y);
+
+    for y in 0..range_height {
+        for x in 0..range_width {
+            let p = top.get_pixel(origin_top_x + x, origin_top_y + y);
+            let mut bottom_pixel = bottom.get_pixel(origin_bottom_x + x, origin_bottom_y + y);
+            bottom_pixel.blend(&p);
+
+            bottom.put_pixel(origin_bottom_x + x, origin_bottom_y + y, bottom_pixel);
+        }
+    }
+}
+
+/// Tile an image by repeating it multiple times
+///
+/// # Examples
+/// ```no_run
+/// use image::{RgbaImage};
+///
+/// let mut img = RgbaImage::new(1920, 1080);
+/// let tile = image::open("tile.png").unwrap();
+///
+/// image::imageops::tile(&mut img, &tile);
+/// img.save("tiled_wallpaper.png").unwrap();
+/// ```
+pub fn tile<I, J>(bottom: &mut I, top: &J)
+where
+    I: GenericImage,
+    J: GenericImageView<Pixel = I::Pixel>,
+{
+    for x in (0..bottom.width()).step_by(top.width() as usize) {
+        for y in (0..bottom.height()).step_by(top.height() as usize) {
+            overlay(bottom, top, i64::from(x), i64::from(y));
+        }
+    }
+}
+
+/// Fill the image with a linear vertical gradient
+///
+/// This function assumes a linear color space.
+///
+/// # Examples
+/// ```no_run
+/// use image::{Rgba, RgbaImage, Pixel};
+///
+/// let mut img = RgbaImage::new(100, 100);
+/// let start = Rgba::from_slice(&[0, 128, 0, 0]);
+/// let end = Rgba::from_slice(&[255, 255, 255, 255]);
+///
+/// image::imageops::vertical_gradient(&mut img, start, end);
+/// img.save("vertical_gradient.png").unwrap();
+pub fn vertical_gradient<S, P, I>(img: &mut I, start: &P, stop: &P)
+where
+    I: GenericImage<Pixel = P>,
+    P: Pixel<Subpixel = S> + 'static,
+    S: Primitive + Lerp + 'static,
+{
+    for y in 0..img.height() {
+        let pixel = start.map2(stop, |a, b| {
+            let y = <S::Ratio as num_traits::NumCast>::from(y).unwrap();
+            let height = <S::Ratio as num_traits::NumCast>::from(img.height() - 1).unwrap();
+            S::lerp(a, b, y / height)
+        });
+
+        for x in 0..img.width() {
+            img.put_pixel(x, y, pixel);
+        }
+    }
+}
+
+/// Fill the image with a linear horizontal gradient
+///
+/// This function assumes a linear color space.
+///
+/// # Examples
+/// ```no_run
+/// use image::{Rgba, RgbaImage, Pixel};
+///
+/// let mut img = RgbaImage::new(100, 100);
+/// let start = Rgba::from_slice(&[0, 128, 0, 0]);
+/// let end = Rgba::from_slice(&[255, 255, 255, 255]);
+///
+/// image::imageops::horizontal_gradient(&mut img, start, end);
+/// img.save("horizontal_gradient.png").unwrap();
+pub fn horizontal_gradient<S, P, I>(img: &mut I, start: &P, stop: &P)
+where
+    I: GenericImage<Pixel = P>,
+    P: Pixel<Subpixel = S> + 'static,
+    S: Primitive + Lerp + 'static,
+{
+    for x in 0..img.width() {
+        let pixel = start.map2(stop, |a, b| {
+            let x = <S::Ratio as num_traits::NumCast>::from(x).unwrap();
+            let width = <S::Ratio as num_traits::NumCast>::from(img.width() - 1).unwrap();
+            S::lerp(a, b, x / width)
+        });
+
+        for y in 0..img.height() {
+            img.put_pixel(x, y, pixel);
+        }
+    }
+}
+
+/// Replace the contents of an image at a given coordinate (x, y)
+pub fn replace<I, J>(bottom: &mut I, top: &J, x: i64, y: i64)
+where
+    I: GenericImage,
+    J: GenericImageView<Pixel = I::Pixel>,
+{
+    let bottom_dims = bottom.dimensions();
+    let top_dims = top.dimensions();
+
+    // Crop our top image if we're going out of bounds
+    let (origin_bottom_x, origin_bottom_y, origin_top_x, origin_top_y, range_width, range_height) =
+        overlay_bounds_ext(bottom_dims, top_dims, x, y);
+
+    for y in 0..range_height {
+        for x in 0..range_width {
+            let p = top.get_pixel(origin_top_x + x, origin_top_y + y);
+            bottom.put_pixel(origin_bottom_x + x, origin_bottom_y + y, p);
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+
+    use super::{overlay, overlay_bounds_ext};
+    use crate::color::Rgb;
+    use crate::ImageBuffer;
+    use crate::RgbaImage;
+
+    #[test]
+    fn test_overlay_bounds_ext() {
+        assert_eq!(
+            overlay_bounds_ext((10, 10), (10, 10), 0, 0),
+            (0, 0, 0, 0, 10, 10)
+        );
+        assert_eq!(
+            overlay_bounds_ext((10, 10), (10, 10), 1, 0),
+            (1, 0, 0, 0, 9, 10)
+        );
+        assert_eq!(
+            overlay_bounds_ext((10, 10), (10, 10), 0, 11),
+            (0, 0, 0, 0, 0, 0)
+        );
+        assert_eq!(
+            overlay_bounds_ext((10, 10), (10, 10), -1, 0),
+            (0, 0, 1, 0, 9, 10)
+        );
+        assert_eq!(
+            overlay_bounds_ext((10, 10), (10, 10), -10, 0),
+            (0, 0, 0, 0, 0, 0)
+        );
+        assert_eq!(
+            overlay_bounds_ext((10, 10), (10, 10), 1i64 << 50, 0),
+            (0, 0, 0, 0, 0, 0)
+        );
+        assert_eq!(
+            overlay_bounds_ext((10, 10), (10, 10), -(1i64 << 50), 0),
+            (0, 0, 0, 0, 0, 0)
+        );
+        assert_eq!(
+            overlay_bounds_ext((10, 10), (u32::MAX, 10), 10 - i64::from(u32::MAX), 0),
+            (0, 0, u32::MAX - 10, 0, 10, 10)
+        );
+    }
+
+    #[test]
+    /// Test that images written into other images works
+    fn test_image_in_image() {
+        let mut target = ImageBuffer::new(32, 32);
+        let source = ImageBuffer::from_pixel(16, 16, Rgb([255u8, 0, 0]));
+        overlay(&mut target, &source, 0, 0);
+        assert!(*target.get_pixel(0, 0) == Rgb([255u8, 0, 0]));
+        assert!(*target.get_pixel(15, 0) == Rgb([255u8, 0, 0]));
+        assert!(*target.get_pixel(16, 0) == Rgb([0u8, 0, 0]));
+        assert!(*target.get_pixel(0, 15) == Rgb([255u8, 0, 0]));
+        assert!(*target.get_pixel(0, 16) == Rgb([0u8, 0, 0]));
+    }
+
+    #[test]
+    /// Test that images written outside of a frame doesn't blow up
+    fn test_image_in_image_outside_of_bounds() {
+        let mut target = ImageBuffer::new(32, 32);
+        let source = ImageBuffer::from_pixel(32, 32, Rgb([255u8, 0, 0]));
+        overlay(&mut target, &source, 1, 1);
+        assert!(*target.get_pixel(0, 0) == Rgb([0, 0, 0]));
+        assert!(*target.get_pixel(1, 1) == Rgb([255u8, 0, 0]));
+        assert!(*target.get_pixel(31, 31) == Rgb([255u8, 0, 0]));
+    }
+
+    #[test]
+    /// Test that images written to coordinates out of the frame doesn't blow up
+    /// (issue came up in #848)
+    fn test_image_outside_image_no_wrap_around() {
+        let mut target = ImageBuffer::new(32, 32);
+        let source = ImageBuffer::from_pixel(32, 32, Rgb([255u8, 0, 0]));
+        overlay(&mut target, &source, 33, 33);
+        assert!(*target.get_pixel(0, 0) == Rgb([0, 0, 0]));
+        assert!(*target.get_pixel(1, 1) == Rgb([0, 0, 0]));
+        assert!(*target.get_pixel(31, 31) == Rgb([0, 0, 0]));
+    }
+
+    #[test]
+    /// Test that images written to coordinates with overflow works
+    fn test_image_coordinate_overflow() {
+        let mut target = ImageBuffer::new(16, 16);
+        let source = ImageBuffer::from_pixel(32, 32, Rgb([255u8, 0, 0]));
+        // Overflows to 'sane' coordinates but top is larger than bot.
+        overlay(
+            &mut target,
+            &source,
+            i64::from(u32::max_value() - 31),
+            i64::from(u32::max_value() - 31),
+        );
+        assert!(*target.get_pixel(0, 0) == Rgb([0, 0, 0]));
+        assert!(*target.get_pixel(1, 1) == Rgb([0, 0, 0]));
+        assert!(*target.get_pixel(15, 15) == Rgb([0, 0, 0]));
+    }
+
+    use super::{horizontal_gradient, vertical_gradient};
+
+    #[test]
+    /// Test that horizontal gradients are correctly generated
+    fn test_image_horizontal_gradient_limits() {
+        let mut img = ImageBuffer::new(100, 1);
+
+        let start = Rgb([0u8, 128, 0]);
+        let end = Rgb([255u8, 255, 255]);
+
+        horizontal_gradient(&mut img, &start, &end);
+
+        assert_eq!(img.get_pixel(0, 0), &start);
+        assert_eq!(img.get_pixel(img.width() - 1, 0), &end);
+    }
+
+    #[test]
+    /// Test that vertical gradients are correctly generated
+    fn test_image_vertical_gradient_limits() {
+        let mut img = ImageBuffer::new(1, 100);
+
+        let start = Rgb([0u8, 128, 0]);
+        let end = Rgb([255u8, 255, 255]);
+
+        vertical_gradient(&mut img, &start, &end);
+
+        assert_eq!(img.get_pixel(0, 0), &start);
+        assert_eq!(img.get_pixel(0, img.height() - 1), &end);
+    }
+
+    #[test]
+    /// Test blur doesn't panick when passed 0.0
+    fn test_blur_zero() {
+        let image = RgbaImage::new(50, 50);
+        let _ = super::blur(&image, 0.0);
+    }
+}

vendor/image/src/io/free_functions.rs

@@ -0,0 +1,312 @@
+use std::fs::File;
+use std::io::{BufRead, BufReader, BufWriter, Seek};
+use std::path::Path;
+use std::u32;
+
+use crate::codecs::*;
+
+use crate::dynimage::DynamicImage;
+use crate::error::{ImageError, ImageFormatHint, ImageResult};
+use crate::image;
+use crate::image::ImageFormat;
+#[allow(unused_imports)] // When no features are supported
+use crate::image::{ImageDecoder, ImageEncoder};
+use crate::{
+    color,
+    error::{UnsupportedError, UnsupportedErrorKind},
+    ImageOutputFormat,
+};
+
+pub(crate) fn open_impl(path: &Path) -> ImageResult<DynamicImage> {
+    let buffered_read = BufReader::new(File::open(path).map_err(ImageError::IoError)?);
+
+    load(buffered_read, ImageFormat::from_path(path)?)
+}
+
+/// Create a new image from a Reader.
+///
+/// Assumes the reader is already buffered. For optimal performance,
+/// consider wrapping the reader with a `BufReader::new()`.
+///
+/// Try [`io::Reader`] for more advanced uses.
+///
+/// [`io::Reader`]: io/struct.Reader.html
+#[allow(unused_variables)]
+// r is unused if no features are supported.
+pub fn load<R: BufRead + Seek>(r: R, format: ImageFormat) -> ImageResult<DynamicImage> {
+    load_inner(r, super::Limits::default(), format)
+}
+
+pub(crate) trait DecoderVisitor {
+    type Result;
+    fn visit_decoder<'a, D: ImageDecoder<'a>>(self, decoder: D) -> ImageResult<Self::Result>;
+}
+
+pub(crate) fn load_decoder<R: BufRead + Seek, V: DecoderVisitor>(
+    r: R,
+    format: ImageFormat,
+    limits: super::Limits,
+    visitor: V,
+) -> ImageResult<V::Result> {
+    #[allow(unreachable_patterns)]
+    // Default is unreachable if all features are supported.
+    match format {
+        #[cfg(feature = "avif-decoder")]
+        image::ImageFormat::Avif => visitor.visit_decoder(avif::AvifDecoder::new(r)?),
+        #[cfg(feature = "png")]
+        image::ImageFormat::Png => visitor.visit_decoder(png::PngDecoder::with_limits(r, limits)?),
+        #[cfg(feature = "gif")]
+        image::ImageFormat::Gif => visitor.visit_decoder(gif::GifDecoder::new(r)?),
+        #[cfg(feature = "jpeg")]
+        image::ImageFormat::Jpeg => visitor.visit_decoder(jpeg::JpegDecoder::new(r)?),
+        #[cfg(feature = "webp")]
+        image::ImageFormat::WebP => visitor.visit_decoder(webp::WebPDecoder::new(r)?),
+        #[cfg(feature = "tiff")]
+        image::ImageFormat::Tiff => visitor.visit_decoder(tiff::TiffDecoder::new(r)?),
+        #[cfg(feature = "tga")]
+        image::ImageFormat::Tga => visitor.visit_decoder(tga::TgaDecoder::new(r)?),
+        #[cfg(feature = "dds")]
+        image::ImageFormat::Dds => visitor.visit_decoder(dds::DdsDecoder::new(r)?),
+        #[cfg(feature = "bmp")]
+        image::ImageFormat::Bmp => visitor.visit_decoder(bmp::BmpDecoder::new(r)?),
+        #[cfg(feature = "ico")]
+        image::ImageFormat::Ico => visitor.visit_decoder(ico::IcoDecoder::new(r)?),
+        #[cfg(feature = "hdr")]
+        image::ImageFormat::Hdr => visitor.visit_decoder(hdr::HdrAdapter::new(BufReader::new(r))?),
+        #[cfg(feature = "exr")]
+        image::ImageFormat::OpenExr => visitor.visit_decoder(openexr::OpenExrDecoder::new(r)?),
+        #[cfg(feature = "pnm")]
+        image::ImageFormat::Pnm => visitor.visit_decoder(pnm::PnmDecoder::new(r)?),
+        #[cfg(feature = "farbfeld")]
+        image::ImageFormat::Farbfeld => visitor.visit_decoder(farbfeld::FarbfeldDecoder::new(r)?),
+        #[cfg(feature = "qoi")]
+        image::ImageFormat::Qoi => visitor.visit_decoder(qoi::QoiDecoder::new(r)?),
+        _ => Err(ImageError::Unsupported(
+            ImageFormatHint::Exact(format).into(),
+        )),
+    }
+}
+
+pub(crate) fn load_inner<R: BufRead + Seek>(
+    r: R,
+    limits: super::Limits,
+    format: ImageFormat,
+) -> ImageResult<DynamicImage> {
+    struct LoadVisitor(super::Limits);
+
+    impl DecoderVisitor for LoadVisitor {
+        type Result = DynamicImage;
+
+        fn visit_decoder<'a, D: ImageDecoder<'a>>(
+            self,
+            mut decoder: D,
+        ) -> ImageResult<Self::Result> {
+            let mut limits = self.0;
+            // Check that we do not allocate a bigger buffer than we are allowed to
+            // FIXME: should this rather go in `DynamicImage::from_decoder` somehow?
+            limits.reserve(decoder.total_bytes())?;
+            decoder.set_limits(limits)?;
+            DynamicImage::from_decoder(decoder)
+        }
+    }
+
+    load_decoder(r, format, limits.clone(), LoadVisitor(limits))
+}
+
+pub(crate) fn image_dimensions_impl(path: &Path) -> ImageResult<(u32, u32)> {
+    let format = image::ImageFormat::from_path(path)?;
+    let reader = BufReader::new(File::open(path)?);
+    image_dimensions_with_format_impl(reader, format)
+}
+
+#[allow(unused_variables)]
+// fin is unused if no features are supported.
+pub(crate) fn image_dimensions_with_format_impl<R: BufRead + Seek>(
+    buffered_read: R,
+    format: ImageFormat,
+) -> ImageResult<(u32, u32)> {
+    struct DimVisitor;
+
+    impl DecoderVisitor for DimVisitor {
+        type Result = (u32, u32);
+        fn visit_decoder<'a, D: ImageDecoder<'a>>(self, decoder: D) -> ImageResult<Self::Result> {
+            Ok(decoder.dimensions())
+        }
+    }
+
+    load_decoder(buffered_read, format, super::Limits::default(), DimVisitor)
+}
+
+#[allow(unused_variables)]
+// Most variables when no features are supported
+pub(crate) fn save_buffer_impl(
+    path: &Path,
+    buf: &[u8],
+    width: u32,
+    height: u32,
+    color: color::ColorType,
+) -> ImageResult<()> {
+    let format = ImageFormat::from_path(path)?;
+    save_buffer_with_format_impl(path, buf, width, height, color, format)
+}
+
+#[allow(unused_variables)]
+// Most variables when no features are supported
+pub(crate) fn save_buffer_with_format_impl(
+    path: &Path,
+    buf: &[u8],
+    width: u32,
+    height: u32,
+    color: color::ColorType,
+    format: ImageFormat,
+) -> ImageResult<()> {
+    let buffered_file_write = &mut BufWriter::new(File::create(path)?); // always seekable
+
+    let format = match format {
+        #[cfg(feature = "pnm")]
+        image::ImageFormat::Pnm => {
+            let ext = path
+                .extension()
+                .and_then(|s| s.to_str())
+                .map_or("".to_string(), |s| s.to_ascii_lowercase());
+            ImageOutputFormat::Pnm(match &*ext {
+                "pbm" => pnm::PnmSubtype::Bitmap(pnm::SampleEncoding::Binary),
+                "pgm" => pnm::PnmSubtype::Graymap(pnm::SampleEncoding::Binary),
+                "ppm" => pnm::PnmSubtype::Pixmap(pnm::SampleEncoding::Binary),
+                "pam" => pnm::PnmSubtype::ArbitraryMap,
+                _ => {
+                    return Err(ImageError::Unsupported(
+                        ImageFormatHint::Exact(format).into(),
+                    ))
+                } // Unsupported Pnm subtype.
+            })
+        }
+        // #[cfg(feature = "hdr")]
+        // image::ImageFormat::Hdr => hdr::HdrEncoder::new(fout).encode(&[Rgb<f32>], width, height), // usize
+        format => format.into(),
+    };
+
+    write_buffer_impl(buffered_file_write, buf, width, height, color, format)
+}
+
+#[allow(unused_variables)]
+// Most variables when no features are supported
+pub(crate) fn write_buffer_impl<W: std::io::Write + Seek>(
+    buffered_write: &mut W,
+    buf: &[u8],
+    width: u32,
+    height: u32,
+    color: color::ColorType,
+    format: ImageOutputFormat,
+) -> ImageResult<()> {
+    match format {
+        #[cfg(feature = "png")]
+        ImageOutputFormat::Png => {
+            png::PngEncoder::new(buffered_write).write_image(buf, width, height, color)
+        }
+        #[cfg(feature = "jpeg")]
+        ImageOutputFormat::Jpeg(quality) => {
+            jpeg::JpegEncoder::new_with_quality(buffered_write, quality)
+                .write_image(buf, width, height, color)
+        }
+        #[cfg(feature = "pnm")]
+        ImageOutputFormat::Pnm(subtype) => pnm::PnmEncoder::new(buffered_write)
+            .with_subtype(subtype)
+            .write_image(buf, width, height, color),
+        #[cfg(feature = "gif")]
+        ImageOutputFormat::Gif => {
+            gif::GifEncoder::new(buffered_write).encode(buf, width, height, color)
+        }
+        #[cfg(feature = "ico")]
+        ImageOutputFormat::Ico => {
+            ico::IcoEncoder::new(buffered_write).write_image(buf, width, height, color)
+        }
+        #[cfg(feature = "bmp")]
+        ImageOutputFormat::Bmp => {
+            bmp::BmpEncoder::new(buffered_write).write_image(buf, width, height, color)
+        }
+        #[cfg(feature = "farbfeld")]
+        ImageOutputFormat::Farbfeld => {
+            farbfeld::FarbfeldEncoder::new(buffered_write).write_image(buf, width, height, color)
+        }
+        #[cfg(feature = "tga")]
+        ImageOutputFormat::Tga => {
+            tga::TgaEncoder::new(buffered_write).write_image(buf, width, height, color)
+        }
+        #[cfg(feature = "exr")]
+        ImageOutputFormat::OpenExr => {
+            openexr::OpenExrEncoder::new(buffered_write).write_image(buf, width, height, color)
+        }
+        #[cfg(feature = "tiff")]
+        ImageOutputFormat::Tiff => {
+            tiff::TiffEncoder::new(buffered_write).write_image(buf, width, height, color)
+        }
+        #[cfg(feature = "avif-encoder")]
+        ImageOutputFormat::Avif => {
+            avif::AvifEncoder::new(buffered_write).write_image(buf, width, height, color)
+        }
+        #[cfg(feature = "qoi")]
+        ImageOutputFormat::Qoi => {
+            qoi::QoiEncoder::new(buffered_write).write_image(buf, width, height, color)
+        }
+        #[cfg(feature = "webp-encoder")]
+        ImageOutputFormat::WebP => {
+            webp::WebPEncoder::new(buffered_write).write_image(buf, width, height, color)
+        }
+
+        image::ImageOutputFormat::Unsupported(msg) => Err(ImageError::Unsupported(
+            UnsupportedError::from_format_and_kind(
+                ImageFormatHint::Unknown,
+                UnsupportedErrorKind::Format(ImageFormatHint::Name(msg)),
+            ),
+        )),
+    }
+}
+
+static MAGIC_BYTES: [(&[u8], ImageFormat); 23] = [
+    (b"\x89PNG\r\n\x1a\n", ImageFormat::Png),
+    (&[0xff, 0xd8, 0xff], ImageFormat::Jpeg),
+    (b"GIF89a", ImageFormat::Gif),
+    (b"GIF87a", ImageFormat::Gif),
+    (b"RIFF", ImageFormat::WebP), // TODO: better magic byte detection, see https://github.com/image-rs/image/issues/660
+    (b"MM\x00*", ImageFormat::Tiff),
+    (b"II*\x00", ImageFormat::Tiff),
+    (b"DDS ", ImageFormat::Dds),
+    (b"BM", ImageFormat::Bmp),
+    (&[0, 0, 1, 0], ImageFormat::Ico),
+    (b"#?RADIANCE", ImageFormat::Hdr),
+    (b"P1", ImageFormat::Pnm),
+    (b"P2", ImageFormat::Pnm),
+    (b"P3", ImageFormat::Pnm),
+    (b"P4", ImageFormat::Pnm),
+    (b"P5", ImageFormat::Pnm),
+    (b"P6", ImageFormat::Pnm),
+    (b"P7", ImageFormat::Pnm),
+    (b"farbfeld", ImageFormat::Farbfeld),
+    (b"\0\0\0 ftypavif", ImageFormat::Avif),
+    (b"\0\0\0\x1cftypavif", ImageFormat::Avif),
+    (&[0x76, 0x2f, 0x31, 0x01], ImageFormat::OpenExr), // = &exr::meta::magic_number::BYTES
+    (b"qoif", ImageFormat::Qoi),
+];
+
+/// Guess image format from memory block
+///
+/// Makes an educated guess about the image format based on the Magic Bytes at the beginning.
+/// TGA is not supported by this function.
+/// This is not to be trusted on the validity of the whole memory block
+pub fn guess_format(buffer: &[u8]) -> ImageResult<ImageFormat> {
+    match guess_format_impl(buffer) {
+        Some(format) => Ok(format),
+        None => Err(ImageError::Unsupported(ImageFormatHint::Unknown.into())),
+    }
+}
+
+pub(crate) fn guess_format_impl(buffer: &[u8]) -> Option<ImageFormat> {
+    for &(signature, format) in &MAGIC_BYTES {
+        if buffer.starts_with(signature) {
+            return Some(format);
+        }
+    }
+
+    None
+}

vendor/image/src/io/mod.rs

@@ -0,0 +1,166 @@
+//! Input and output of images.
+
+use std::convert::TryFrom;
+
+use crate::{error, ImageError, ImageResult};
+
+pub(crate) mod free_functions;
+mod reader;
+
+pub use self::reader::Reader;
+
+/// Set of supported strict limits for a decoder.
+#[derive(Clone, Debug, Eq, PartialEq, Hash)]
+#[allow(missing_copy_implementations)]
+#[allow(clippy::manual_non_exhaustive)]
+pub struct LimitSupport {
+    _non_exhaustive: (),
+}
+
+#[allow(clippy::derivable_impls)]
+impl Default for LimitSupport {
+    fn default() -> LimitSupport {
+        LimitSupport {
+            _non_exhaustive: (),
+        }
+    }
+}
+
+/// Resource limits for decoding.
+///
+/// Limits can be either *strict* or *non-strict*. Non-strict limits are best-effort
+/// limits where the library does not guarantee that limit will not be exceeded. Do note
+/// that it is still considered a bug if a non-strict limit is exceeded, however as
+/// some of the underlying decoders do not support not support such limits one cannot
+/// rely on these limits being supported. For stric limits the library makes a stronger
+/// guarantee that the limit will not be exceeded. Exceeding a strict limit is considered
+/// a critical bug. If a decoder cannot guarantee that it will uphold a strict limit it
+/// *must* fail with `image::error::LimitErrorKind::Unsupported`.
+///
+/// Currently the only strict limits supported are the `max_image_width` and `max_image_height`
+/// limits, however more will be added in the future. [`LimitSupport`] will default to support
+/// being false and decoders should enable support for the limits they support in
+/// [`ImageDecoder::set_limits`].
+///
+/// The limit check should only ever fail if a limit will be exceeded or an unsupported strict
+/// limit is used.
+///
+/// [`LimitSupport`]: ./struct.LimitSupport.html
+/// [`ImageDecoder::set_limits`]: ../trait.ImageDecoder.html#method.set_limits
+#[derive(Clone, Debug, Eq, PartialEq, Hash)]
+#[allow(missing_copy_implementations)]
+#[allow(clippy::manual_non_exhaustive)]
+pub struct Limits {
+    /// The maximum allowed image width. This limit is strict. The default is no limit.
+    pub max_image_width: Option<u32>,
+    /// The maximum allowed image height. This limit is strict. The default is no limit.
+    pub max_image_height: Option<u32>,
+    /// The maximum allowed sum of allocations allocated by the decoder at any one time excluding
+    /// allocator overhead. This limit is non-strict by default and some decoders may ignore it.
+    /// The default is 512MiB.
+    pub max_alloc: Option<u64>,
+    _non_exhaustive: (),
+}
+
+impl Default for Limits {
+    fn default() -> Limits {
+        Limits {
+            max_image_width: None,
+            max_image_height: None,
+            max_alloc: Some(512 * 1024 * 1024),
+            _non_exhaustive: (),
+        }
+    }
+}
+
+impl Limits {
+    /// Disable all limits.
+    pub fn no_limits() -> Limits {
+        Limits {
+            max_image_width: None,
+            max_image_height: None,
+            max_alloc: None,
+            _non_exhaustive: (),
+        }
+    }
+
+    /// This function checks that all currently set strict limits are supported.
+    pub fn check_support(&self, _supported: &LimitSupport) -> ImageResult<()> {
+        Ok(())
+    }
+
+    /// This function checks the `max_image_width` and `max_image_height` limits given
+    /// the image width and height.
+    pub fn check_dimensions(&self, width: u32, height: u32) -> ImageResult<()> {
+        if let Some(max_width) = self.max_image_width {
+            if width > max_width {
+                return Err(ImageError::Limits(error::LimitError::from_kind(
+                    error::LimitErrorKind::DimensionError,
+                )));
+            }
+        }
+
+        if let Some(max_height) = self.max_image_height {
+            if height > max_height {
+                return Err(ImageError::Limits(error::LimitError::from_kind(
+                    error::LimitErrorKind::DimensionError,
+                )));
+            }
+        }
+
+        Ok(())
+    }
+
+    /// This function checks that the current limit allows for reserving the set amount
+    /// of bytes, it then reduces the limit accordingly.
+    pub fn reserve(&mut self, amount: u64) -> ImageResult<()> {
+        if let Some(max_alloc) = self.max_alloc.as_mut() {
+            if *max_alloc < amount {
+                return Err(ImageError::Limits(error::LimitError::from_kind(
+                    error::LimitErrorKind::InsufficientMemory,
+                )));
+            }
+
+            *max_alloc -= amount;
+        }
+
+        Ok(())
+    }
+
+    /// This function acts identically to [`reserve`], but takes a `usize` for convenience.
+    pub fn reserve_usize(&mut self, amount: usize) -> ImageResult<()> {
+        match u64::try_from(amount) {
+            Ok(n) => self.reserve(n),
+            Err(_) if self.max_alloc.is_some() => Err(ImageError::Limits(
+                error::LimitError::from_kind(error::LimitErrorKind::InsufficientMemory),
+            )),
+            Err(_) => {
+                // Out of bounds, but we weren't asked to consider any limit.
+                Ok(())
+            }
+        }
+    }
+
+    /// This function increases the `max_alloc` limit with amount. Should only be used
+    /// together with [`reserve`].
+    ///
+    /// [`reserve`]: #method.reserve
+    pub fn free(&mut self, amount: u64) {
+        if let Some(max_alloc) = self.max_alloc.as_mut() {
+            *max_alloc = max_alloc.saturating_add(amount);
+        }
+    }
+
+    /// This function acts identically to [`free`], but takes a `usize` for convenience.
+    pub fn free_usize(&mut self, amount: usize) {
+        match u64::try_from(amount) {
+            Ok(n) => self.free(n),
+            Err(_) if self.max_alloc.is_some() => {
+                panic!("max_alloc is set, we should have exited earlier when the reserve failed");
+            }
+            Err(_) => {
+                // Out of bounds, but we weren't asked to consider any limit.
+            }
+        }
+    }
+}

vendor/image/src/io/reader.rs

@@ -0,0 +1,239 @@
+use std::fs::File;
+use std::io::{self, BufRead, BufReader, Cursor, Read, Seek, SeekFrom};
+use std::path::Path;
+
+use crate::dynimage::DynamicImage;
+use crate::error::{ImageFormatHint, UnsupportedError, UnsupportedErrorKind};
+use crate::image::ImageFormat;
+use crate::{ImageError, ImageResult};
+
+use super::free_functions;
+
+/// A multi-format image reader.
+///
+/// Wraps an input reader to facilitate automatic detection of an image's format, appropriate
+/// decoding method, and dispatches into the set of supported [`ImageDecoder`] implementations.
+///
+/// ## Usage
+///
+/// Opening a file, deducing the format based on the file path automatically, and trying to decode
+/// the image contained can be performed by constructing the reader and immediately consuming it.
+///
+/// ```no_run
+/// # use image::ImageError;
+/// # use image::io::Reader;
+/// # fn main() -> Result<(), ImageError> {
+/// let image = Reader::open("path/to/image.png")?
+///     .decode()?;
+/// # Ok(()) }
+/// ```
+///
+/// It is also possible to make a guess based on the content. This is especially handy if the
+/// source is some blob in memory and you have constructed the reader in another way. Here is an
+/// example with a `pnm` black-and-white subformat that encodes its pixel matrix with ascii values.
+///
+/// ```
+/// # use image::ImageError;
+/// # use image::io::Reader;
+/// # fn main() -> Result<(), ImageError> {
+/// use std::io::Cursor;
+/// use image::ImageFormat;
+///
+/// let raw_data = b"P1 2 2\n\
+///     0 1\n\
+///     1 0\n";
+///
+/// let mut reader = Reader::new(Cursor::new(raw_data))
+///     .with_guessed_format()
+///     .expect("Cursor io never fails");
+/// assert_eq!(reader.format(), Some(ImageFormat::Pnm));
+///
+/// # #[cfg(feature = "pnm")]
+/// let image = reader.decode()?;
+/// # Ok(()) }
+/// ```
+///
+/// As a final fallback or if only a specific format must be used, the reader always allows manual
+/// specification of the supposed image format with [`set_format`].
+///
+/// [`set_format`]: #method.set_format
+/// [`ImageDecoder`]: ../trait.ImageDecoder.html
+pub struct Reader<R: Read> {
+    /// The reader. Should be buffered.
+    inner: R,
+    /// The format, if one has been set or deduced.
+    format: Option<ImageFormat>,
+    /// Decoding limits
+    limits: super::Limits,
+}
+
+impl<R: Read> Reader<R> {
+    /// Create a new image reader without a preset format.
+    ///
+    /// Assumes the reader is already buffered. For optimal performance,
+    /// consider wrapping the reader with a `BufReader::new()`.
+    ///
+    /// It is possible to guess the format based on the content of the read object with
+    /// [`with_guessed_format`], or to set the format directly with [`set_format`].
+    ///
+    /// [`with_guessed_format`]: #method.with_guessed_format
+    /// [`set_format`]: method.set_format
+    pub fn new(buffered_reader: R) -> Self {
+        Reader {
+            inner: buffered_reader,
+            format: None,
+            limits: super::Limits::default(),
+        }
+    }
+
+    /// Construct a reader with specified format.
+    ///
+    /// Assumes the reader is already buffered. For optimal performance,
+    /// consider wrapping the reader with a `BufReader::new()`.
+    pub fn with_format(buffered_reader: R, format: ImageFormat) -> Self {
+        Reader {
+            inner: buffered_reader,
+            format: Some(format),
+            limits: super::Limits::default(),
+        }
+    }
+
+    /// Get the currently determined format.
+    pub fn format(&self) -> Option<ImageFormat> {
+        self.format
+    }
+
+    /// Supply the format as which to interpret the read image.
+    pub fn set_format(&mut self, format: ImageFormat) {
+        self.format = Some(format);
+    }
+
+    /// Remove the current information on the image format.
+    ///
+    /// Note that many operations require format information to be present and will return e.g. an
+    /// `ImageError::Unsupported` when the image format has not been set.
+    pub fn clear_format(&mut self) {
+        self.format = None;
+    }
+
+    /// Disable all decoding limits.
+    pub fn no_limits(&mut self) {
+        self.limits = super::Limits::no_limits();
+    }
+
+    /// Set a custom set of decoding limits.
+    pub fn limits(&mut self, limits: super::Limits) {
+        self.limits = limits;
+    }
+
+    /// Unwrap the reader.
+    pub fn into_inner(self) -> R {
+        self.inner
+    }
+}
+
+impl Reader<BufReader<File>> {
+    /// Open a file to read, format will be guessed from path.
+    ///
+    /// This will not attempt any io operation on the opened file.
+    ///
+    /// If you want to inspect the content for a better guess on the format, which does not depend
+    /// on file extensions, follow this call with a call to [`with_guessed_format`].
+    ///
+    /// [`with_guessed_format`]: #method.with_guessed_format
+    pub fn open<P>(path: P) -> io::Result<Self>
+    where
+        P: AsRef<Path>,
+    {
+        Self::open_impl(path.as_ref())
+    }
+
+    fn open_impl(path: &Path) -> io::Result<Self> {
+        Ok(Reader {
+            inner: BufReader::new(File::open(path)?),
+            format: ImageFormat::from_path(path).ok(),
+            limits: super::Limits::default(),
+        })
+    }
+}
+
+impl<R: BufRead + Seek> Reader<R> {
+    /// Make a format guess based on the content, replacing it on success.
+    ///
+    /// Returns `Ok` with the guess if no io error occurs. Additionally, replaces the current
+    /// format if the guess was successful. If the guess was unable to determine a format then
+    /// the current format of the reader is unchanged.
+    ///
+    /// Returns an error if the underlying reader fails. The format is unchanged. The error is a
+    /// `std::io::Error` and not `ImageError` since the only error case is an error when the
+    /// underlying reader seeks.
+    ///
+    /// When an error occurs, the reader may not have been properly reset and it is potentially
+    /// hazardous to continue with more io.
+    ///
+    /// ## Usage
+    ///
+    /// This supplements the path based type deduction from [`open`](Reader::open) with content based deduction.
+    /// This is more common in Linux and UNIX operating systems and also helpful if the path can
+    /// not be directly controlled.
+    ///
+    /// ```no_run
+    /// # use image::ImageError;
+    /// # use image::io::Reader;
+    /// # fn main() -> Result<(), ImageError> {
+    /// let image = Reader::open("image.unknown")?
+    ///     .with_guessed_format()?
+    ///     .decode()?;
+    /// # Ok(()) }
+    /// ```
+    pub fn with_guessed_format(mut self) -> io::Result<Self> {
+        let format = self.guess_format()?;
+        // Replace format if found, keep current state if not.
+        self.format = format.or(self.format);
+        Ok(self)
+    }
+
+    fn guess_format(&mut self) -> io::Result<Option<ImageFormat>> {
+        let mut start = [0; 16];
+
+        // Save current offset, read start, restore offset.
+        let cur = self.inner.stream_position()?;
+        let len = io::copy(
+            // Accept shorter files but read at most 16 bytes.
+            &mut self.inner.by_ref().take(16),
+            &mut Cursor::new(&mut start[..]),
+        )?;
+        self.inner.seek(SeekFrom::Start(cur))?;
+
+        Ok(free_functions::guess_format_impl(&start[..len as usize]))
+    }
+
+    /// Read the image dimensions.
+    ///
+    /// Uses the current format to construct the correct reader for the format.
+    ///
+    /// If no format was determined, returns an `ImageError::Unsupported`.
+    pub fn into_dimensions(mut self) -> ImageResult<(u32, u32)> {
+        let format = self.require_format()?;
+        free_functions::image_dimensions_with_format_impl(self.inner, format)
+    }
+
+    /// Read the image (replaces `load`).
+    ///
+    /// Uses the current format to construct the correct reader for the format.
+    ///
+    /// If no format was determined, returns an `ImageError::Unsupported`.
+    pub fn decode(mut self) -> ImageResult<DynamicImage> {
+        let format = self.require_format()?;
+        free_functions::load_inner(self.inner, self.limits, format)
+    }
+
+    fn require_format(&mut self) -> ImageResult<ImageFormat> {
+        self.format.ok_or_else(|| {
+            ImageError::Unsupported(UnsupportedError::from_format_and_kind(
+                ImageFormatHint::Unknown,
+                UnsupportedErrorKind::Format(ImageFormatHint::Unknown),
+            ))
+        })
+    }
+}

vendor/image/src/lib.rs

@@ -0,0 +1,310 @@
+//! # Overview
+//!
+//! This crate provides native rust implementations of image encoding and decoding as well as some
+//! basic image manipulation functions. Additional documentation can currently also be found in the
+//! [README.md file which is most easily viewed on
+//! github](https://github.com/image-rs/image/blob/master/README.md).
+//!
+//! There are two core problems for which this library provides solutions: a unified interface for image
+//! encodings and simple generic buffers for their content. It's possible to use either feature
+//! without the other. The focus is on a small and stable set of common operations that can be
+//! supplemented by other specialized crates. The library also prefers safe solutions with few
+//! dependencies.
+//!
+//! # High level API
+//!
+//! Load images using [`io::Reader`]:
+//!
+//! ```rust,no_run
+//! use std::io::Cursor;
+//! use image::io::Reader as ImageReader;
+//! # fn main() -> Result<(), image::ImageError> {
+//! # let bytes = vec![0u8];
+//!
+//! let img = ImageReader::open("myimage.png")?.decode()?;
+//! let img2 = ImageReader::new(Cursor::new(bytes)).with_guessed_format()?.decode()?;
+//! # Ok(())
+//! # }
+//! ```
+//!
+//! And save them using [`save`] or [`write_to`] methods:
+//!
+//! ```rust,no_run
+//! # use std::io::{Write, Cursor};
+//! # use image::{DynamicImage, ImageOutputFormat};
+//! # #[cfg(feature = "png")]
+//! # fn main() -> Result<(), image::ImageError> {
+//! # let img: DynamicImage = unimplemented!();
+//! # let img2: DynamicImage = unimplemented!();
+//! img.save("empty.jpg")?;
+//!
+//! let mut bytes: Vec<u8> = Vec::new();
+//! img2.write_to(&mut Cursor::new(&mut bytes), image::ImageOutputFormat::Png)?;
+//! # Ok(())
+//! # }
+//! # #[cfg(not(feature = "png"))] fn main() {}
+//! ```
+//!
+//! With default features, the crate includes support for [many common image formats](codecs/index.html#supported-formats).
+//!
+//! [`save`]: enum.DynamicImage.html#method.save
+//! [`write_to`]: enum.DynamicImage.html#method.write_to
+//! [`io::Reader`]: io/struct.Reader.html
+//!
+//! # Image buffers
+//!
+//! The two main types for storing images:
+//! * [`ImageBuffer`] which holds statically typed image contents.
+//! * [`DynamicImage`] which is an enum over the supported ImageBuffer formats
+//!     and supports conversions between them.
+//!
+//! As well as a few more specialized options:
+//! * [`GenericImage`] trait for a mutable image buffer.
+//! * [`GenericImageView`] trait for read only references to a GenericImage.
+//! * [`flat`] module containing types for interoperability with generic channel
+//!     matrices and foreign interfaces.
+//!
+//! [`GenericImageView`]: trait.GenericImageView.html
+//! [`GenericImage`]: trait.GenericImage.html
+//! [`ImageBuffer`]: struct.ImageBuffer.html
+//! [`DynamicImage`]: enum.DynamicImage.html
+//! [`flat`]: flat/index.html
+//!
+//! # Low level encoding/decoding API
+//!
+//! Implementations of [`ImageEncoder`] provides low level control over encoding:
+//! ```rust,no_run
+//! # use std::io::Write;
+//! # use image::DynamicImage;
+//! # use image::ImageEncoder;
+//! # #[cfg(feature = "jpeg")]
+//! # fn main() -> Result<(), image::ImageError> {
+//! # use image::codecs::jpeg::JpegEncoder;
+//! # let img: DynamicImage = unimplemented!();
+//! # let writer: Box<dyn Write> = unimplemented!();
+//! let encoder = JpegEncoder::new_with_quality(&mut writer, 95);
+//! img.write_with_encoder(encoder)?;
+//! # Ok(())
+//! # }
+//! # #[cfg(not(feature = "jpeg"))] fn main() {}
+//! ```
+//! While [`ImageDecoder`] and [`ImageDecoderRect`] give access to more advanced decoding options:
+//!
+//! ```rust,no_run
+//! # use std::io::Read;
+//! # use image::DynamicImage;
+//! # use image::ImageDecoder;
+//! # #[cfg(feature = "png")]
+//! # fn main() -> Result<(), image::ImageError> {
+//! # use image::codecs::png::PngDecoder;
+//! # let img: DynamicImage = unimplemented!();
+//! # let reader: Box<dyn Read> = unimplemented!();
+//! let decoder = PngDecoder::new(&mut reader)?;
+//! let icc = decoder.icc_profile();
+//! let img = DynamicImage::from_decoder(decoder)?;
+//! # Ok(())
+//! # }
+//! # #[cfg(not(feature = "png"))] fn main() {}
+//! ```
+//!
+//! [`DynamicImage::from_decoder`]: enum.DynamicImage.html#method.from_decoder
+//! [`ImageDecoderRect`]: trait.ImageDecoderRect.html
+//! [`ImageDecoder`]: trait.ImageDecoder.html
+//! [`ImageEncoder`]: trait.ImageEncoder.html
+#![warn(missing_docs)]
+#![warn(unused_qualifications)]
+#![deny(unreachable_pub)]
+#![deny(deprecated)]
+#![deny(missing_copy_implementations)]
+#![cfg_attr(all(test, feature = "benchmarks"), feature(test))]
+// it's a bit of a pain otherwise
+#![allow(clippy::many_single_char_names)]
+// it's a backwards compatibility break
+#![allow(clippy::wrong_self_convention, clippy::enum_variant_names)]
+
+#[cfg(all(test, feature = "benchmarks"))]
+extern crate test;
+
+#[cfg(test)]
+#[macro_use]
+extern crate quickcheck;
+
+pub use crate::color::{ColorType, ExtendedColorType};
+
+pub use crate::color::{Luma, LumaA, Rgb, Rgba};
+
+pub use crate::error::{ImageError, ImageResult};
+
+pub use crate::image::{
+    AnimationDecoder,
+    GenericImage,
+    GenericImageView,
+    ImageDecoder,
+    ImageDecoderRect,
+    ImageEncoder,
+    ImageFormat,
+    ImageOutputFormat,
+    // Iterators
+    Pixels,
+    Progress,
+    SubImage,
+};
+
+pub use crate::buffer_::{
+    GrayAlphaImage,
+    GrayImage,
+    // Image types
+    ImageBuffer,
+    Rgb32FImage,
+    RgbImage,
+    Rgba32FImage,
+    RgbaImage,
+};
+
+pub use crate::flat::FlatSamples;
+
+// Traits
+pub use crate::traits::{EncodableLayout, Pixel, PixelWithColorType, Primitive};
+
+// Opening and loading images
+pub use crate::dynimage::{
+    image_dimensions, load_from_memory, load_from_memory_with_format, open, save_buffer,
+    save_buffer_with_format, write_buffer_with_format,
+};
+pub use crate::io::free_functions::{guess_format, load};
+
+pub use crate::dynimage::DynamicImage;
+
+pub use crate::animation::{Delay, Frame, Frames};
+
+// More detailed error type
+pub mod error;
+
+/// Iterators and other auxiliary structure for the `ImageBuffer` type.
+pub mod buffer {
+    // Only those not exported at the top-level
+    pub use crate::buffer_::{
+        ConvertBuffer, EnumeratePixels, EnumeratePixelsMut, EnumerateRows, EnumerateRowsMut,
+        Pixels, PixelsMut, Rows, RowsMut,
+    };
+}
+
+// Math utils
+pub mod math;
+
+// Image processing functions
+pub mod imageops;
+
+// Io bindings
+pub mod io;
+
+// Buffer representations for ffi.
+pub mod flat;
+
+/// Encoding and decoding for various image file formats.
+///
+/// # Supported formats
+///
+/// <!--- NOTE: Make sure to keep this table in sync with the README -->
+///
+/// | Format | Decoding | Encoding |
+/// | ------ | -------- | -------- |
+/// | AVIF   | Only 8-bit | Lossy |
+/// | BMP    | Yes | Rgb8, Rgba8, Gray8, GrayA8 |
+/// | DDS    | DXT1, DXT3, DXT5 | No |
+/// | Farbfeld | Yes | Yes |
+/// | GIF    | Yes | Yes |
+/// | ICO    | Yes | Yes |
+/// | JPEG   | Baseline and progressive | Baseline JPEG |
+/// | OpenEXR  | Rgb32F, Rgba32F (no dwa compression) | Rgb32F, Rgba32F (no dwa compression) |
+/// | PNG    | All supported color types | Same as decoding |
+/// | PNM    | PBM, PGM, PPM, standard PAM | Yes |
+/// | QOI    | Yes | Yes |
+/// | TGA    | Yes | Rgb8, Rgba8, Bgr8, Bgra8, Gray8, GrayA8 |
+/// | TIFF   | Baseline(no fax support) + LZW + PackBits | Rgb8, Rgba8, Gray8 |
+/// | WebP   | Yes | Rgb8, Rgba8 |
+///
+/// ## A note on format specific features
+///
+/// One of the main goals of `image` is stability, in runtime but also for programmers. This
+/// ensures that performance as well as safety fixes reach a majority of its user base with little
+/// effort. Re-exporting all details of its dependencies would run counter to this goal as it
+/// linked _all_ major version bumps between them and `image`. As such, we are wary of exposing too
+/// many details, or configuration options, that are not shared between different image formats.
+///
+/// Nevertheless, the advantage of precise control is hard to ignore. We will thus consider
+/// _wrappers_, not direct re-exports, in either of the following cases:
+///
+/// 1. A standard specifies that configuration _x_ is required for decoders/encoders and there
+///    exists an essentially canonical way to control it.
+/// 2. At least two different implementations agree on some (sub-)set of features in practice.
+/// 3. A technical argument including measurements of the performance, space benefits, or otherwise
+///    objectively quantified benefits can be made, and the added interface is unlikely to require
+///    breaking changes.
+///
+/// Features that fulfill two or more criteria are preferred.
+///
+/// Re-exports of dependencies that reach version `1` will be discussed when it happens.
+pub mod codecs {
+    #[cfg(any(feature = "avif-encoder", feature = "avif-decoder"))]
+    pub mod avif;
+    #[cfg(feature = "bmp")]
+    pub mod bmp;
+    #[cfg(feature = "dds")]
+    pub mod dds;
+    #[cfg(feature = "dxt")]
+    #[deprecated = "DXT support will be removed or reworked in a future version. Prefer the `squish` crate instead. See https://github.com/image-rs/image/issues/1623"]
+    pub mod dxt;
+    #[cfg(feature = "farbfeld")]
+    pub mod farbfeld;
+    #[cfg(feature = "gif")]
+    pub mod gif;
+    #[cfg(feature = "hdr")]
+    pub mod hdr;
+    #[cfg(feature = "ico")]
+    pub mod ico;
+    #[cfg(feature = "jpeg")]
+    pub mod jpeg;
+    #[cfg(feature = "exr")]
+    pub mod openexr;
+    #[cfg(feature = "png")]
+    pub mod png;
+    #[cfg(feature = "pnm")]
+    pub mod pnm;
+    #[cfg(feature = "qoi")]
+    pub mod qoi;
+    #[cfg(feature = "tga")]
+    pub mod tga;
+    #[cfg(feature = "tiff")]
+    pub mod tiff;
+    #[cfg(any(feature = "webp", feature = "webp-encoder"))]
+    pub mod webp;
+}
+
+mod animation;
+#[path = "buffer.rs"]
+mod buffer_;
+mod color;
+mod dynimage;
+mod image;
+mod traits;
+mod utils;
+
+// Can't use the macro-call itself within the `doc` attribute. So force it to eval it as part of
+// the macro invocation.
+//
+// The inspiration for the macro and implementation is from
+// <https://github.com/GuillaumeGomez/doc-comment>
+//
+// MIT License
+//
+// Copyright (c) 2018 Guillaume Gomez
+macro_rules! insert_as_doc {
+    { $content:expr } => {
+        #[allow(unused_doc_comments)]
+        #[doc = $content] extern { }
+    }
+}
+
+// Provides the README.md as doc, to ensure the example works!
+insert_as_doc!(include_str!("../README.md"));

vendor/image/src/math/mod.rs

@@ -0,0 +1,6 @@
+//! Mathematical helper functions and types.
+mod rect;
+mod utils;
+
+pub use self::rect::Rect;
+pub(super) use utils::resize_dimensions;

vendor/image/src/math/rect.rs

@@ -0,0 +1,12 @@
+/// A Rectangle defined by its top left corner, width and height.
+#[derive(Copy, Clone, Debug, PartialEq, Eq, Hash)]
+pub struct Rect {
+    /// The x coordinate of the top left corner.
+    pub x: u32,
+    /// The y coordinate of the top left corner.
+    pub y: u32,
+    /// The rectangle's width.
+    pub width: u32,
+    /// The rectangle's height.
+    pub height: u32,
+}

vendor/image/src/math/utils.rs

@@ -0,0 +1,123 @@
+//! Shared mathematical utility functions.
+
+use std::cmp::max;
+
+/// Calculates the width and height an image should be resized to.
+/// This preserves aspect ratio, and based on the `fill` parameter
+/// will either fill the dimensions to fit inside the smaller constraint
+/// (will overflow the specified bounds on one axis to preserve
+/// aspect ratio), or will shrink so that both dimensions are
+/// completely contained within the given `width` and `height`,
+/// with empty space on one axis.
+pub(crate) fn resize_dimensions(
+    width: u32,
+    height: u32,
+    nwidth: u32,
+    nheight: u32,
+    fill: bool,
+) -> (u32, u32) {
+    let wratio = nwidth as f64 / width as f64;
+    let hratio = nheight as f64 / height as f64;
+
+    let ratio = if fill {
+        f64::max(wratio, hratio)
+    } else {
+        f64::min(wratio, hratio)
+    };
+
+    let nw = max((width as f64 * ratio).round() as u64, 1);
+    let nh = max((height as f64 * ratio).round() as u64, 1);
+
+    if nw > u64::from(u32::MAX) {
+        let ratio = u32::MAX as f64 / width as f64;
+        (u32::MAX, max((height as f64 * ratio).round() as u32, 1))
+    } else if nh > u64::from(u32::MAX) {
+        let ratio = u32::MAX as f64 / height as f64;
+        (max((width as f64 * ratio).round() as u32, 1), u32::MAX)
+    } else {
+        (nw as u32, nh as u32)
+    }
+}
+
+#[cfg(test)]
+mod test {
+    quickcheck! {
+        fn resize_bounds_correctly_width(old_w: u32, new_w: u32) -> bool {
+            if old_w == 0 || new_w == 0 { return true; }
+            // In this case, the scaling is limited by scaling of height.
+            // We could check that case separately but it does not conform to the same expectation.
+            if new_w as u64 * 400u64 >= old_w as u64 * u64::from(u32::MAX) { return true; }
+
+            let result = super::resize_dimensions(old_w, 400, new_w, ::std::u32::MAX, false);
+            let exact = (400 as f64 * new_w as f64 / old_w as f64).round() as u32;
+            result.0 == new_w && result.1 == exact.max(1)
+        }
+    }
+
+    quickcheck! {
+        fn resize_bounds_correctly_height(old_h: u32, new_h: u32) -> bool {
+            if old_h == 0 || new_h == 0 { return true; }
+            // In this case, the scaling is limited by scaling of width.
+            // We could check that case separately but it does not conform to the same expectation.
+            if 400u64 * new_h as u64 >= old_h as u64 * u64::from(u32::MAX) { return true; }
+
+            let result = super::resize_dimensions(400, old_h, ::std::u32::MAX, new_h, false);
+            let exact = (400 as f64 * new_h as f64 / old_h as f64).round() as u32;
+            result.1 == new_h && result.0 == exact.max(1)
+        }
+    }
+
+    #[test]
+    fn resize_handles_fill() {
+        let result = super::resize_dimensions(100, 200, 200, 500, true);
+        assert!(result.0 == 250);
+        assert!(result.1 == 500);
+
+        let result = super::resize_dimensions(200, 100, 500, 200, true);
+        assert!(result.0 == 500);
+        assert!(result.1 == 250);
+    }
+
+    #[test]
+    fn resize_never_rounds_to_zero() {
+        let result = super::resize_dimensions(1, 150, 128, 128, false);
+        assert!(result.0 > 0);
+        assert!(result.1 > 0);
+    }
+
+    #[test]
+    fn resize_handles_overflow() {
+        let result = super::resize_dimensions(100, ::std::u32::MAX, 200, ::std::u32::MAX, true);
+        assert!(result.0 == 100);
+        assert!(result.1 == ::std::u32::MAX);
+
+        let result = super::resize_dimensions(::std::u32::MAX, 100, ::std::u32::MAX, 200, true);
+        assert!(result.0 == ::std::u32::MAX);
+        assert!(result.1 == 100);
+    }
+
+    #[test]
+    fn resize_rounds() {
+        // Only truncation will result in (3840, 2229) and (2160, 3719)
+        let result = super::resize_dimensions(4264, 2476, 3840, 2160, true);
+        assert_eq!(result, (3840, 2230));
+
+        let result = super::resize_dimensions(2476, 4264, 2160, 3840, false);
+        assert_eq!(result, (2160, 3720));
+    }
+
+    #[test]
+    fn resize_handles_zero() {
+        let result = super::resize_dimensions(0, 100, 100, 100, false);
+        assert_eq!(result, (1, 100));
+
+        let result = super::resize_dimensions(100, 0, 100, 100, false);
+        assert_eq!(result, (100, 1));
+
+        let result = super::resize_dimensions(100, 100, 0, 100, false);
+        assert_eq!(result, (1, 1));
+
+        let result = super::resize_dimensions(100, 100, 100, 0, false);
+        assert_eq!(result, (1, 1));
+    }
+}

vendor/image/src/traits.rs

@@ -0,0 +1,370 @@
+//! This module provides useful traits that were deprecated in rust
+
+// Note copied from the stdlib under MIT license
+
+use num_traits::{Bounded, Num, NumCast};
+use std::ops::AddAssign;
+
+use crate::color::{ColorType, Luma, LumaA, Rgb, Rgba};
+
+/// Types which are safe to treat as an immutable byte slice in a pixel layout
+/// for image encoding.
+pub trait EncodableLayout: seals::EncodableLayout {
+    /// Get the bytes of this value.
+    fn as_bytes(&self) -> &[u8];
+}
+
+impl EncodableLayout for [u8] {
+    fn as_bytes(&self) -> &[u8] {
+        bytemuck::cast_slice(self)
+    }
+}
+
+impl EncodableLayout for [u16] {
+    fn as_bytes(&self) -> &[u8] {
+        bytemuck::cast_slice(self)
+    }
+}
+
+impl EncodableLayout for [f32] {
+    fn as_bytes(&self) -> &[u8] {
+        bytemuck::cast_slice(self)
+    }
+}
+
+/// The type of each channel in a pixel. For example, this can be `u8`, `u16`, `f32`.
+// TODO rename to `PixelComponent`? Split up into separate traits? Seal?
+pub trait Primitive: Copy + NumCast + Num + PartialOrd<Self> + Clone + Bounded {
+    /// The maximum value for this type of primitive within the context of color.
+    /// For floats, the maximum is `1.0`, whereas the integer types inherit their usual maximum values.
+    const DEFAULT_MAX_VALUE: Self;
+
+    /// The minimum value for this type of primitive within the context of color.
+    /// For floats, the minimum is `0.0`, whereas the integer types inherit their usual minimum values.
+    const DEFAULT_MIN_VALUE: Self;
+}
+
+macro_rules! declare_primitive {
+    ($base:ty: ($from:expr)..$to:expr) => {
+        impl Primitive for $base {
+            const DEFAULT_MAX_VALUE: Self = $to;
+            const DEFAULT_MIN_VALUE: Self = $from;
+        }
+    };
+}
+
+declare_primitive!(usize: (0)..Self::MAX);
+declare_primitive!(u8: (0)..Self::MAX);
+declare_primitive!(u16: (0)..Self::MAX);
+declare_primitive!(u32: (0)..Self::MAX);
+declare_primitive!(u64: (0)..Self::MAX);
+
+declare_primitive!(isize: (Self::MIN)..Self::MAX);
+declare_primitive!(i8: (Self::MIN)..Self::MAX);
+declare_primitive!(i16: (Self::MIN)..Self::MAX);
+declare_primitive!(i32: (Self::MIN)..Self::MAX);
+declare_primitive!(i64: (Self::MIN)..Self::MAX);
+declare_primitive!(f32: (0.0)..1.0);
+declare_primitive!(f64: (0.0)..1.0);
+
+/// An Enlargable::Larger value should be enough to calculate
+/// the sum (average) of a few hundred or thousand Enlargeable values.
+pub trait Enlargeable: Sized + Bounded + NumCast {
+    type Larger: Copy + NumCast + Num + PartialOrd<Self::Larger> + Clone + Bounded + AddAssign;
+
+    fn clamp_from(n: Self::Larger) -> Self {
+        if n > Self::max_value().to_larger() {
+            Self::max_value()
+        } else if n < Self::min_value().to_larger() {
+            Self::min_value()
+        } else {
+            NumCast::from(n).unwrap()
+        }
+    }
+
+    fn to_larger(self) -> Self::Larger {
+        NumCast::from(self).unwrap()
+    }
+}
+
+impl Enlargeable for u8 {
+    type Larger = u32;
+}
+impl Enlargeable for u16 {
+    type Larger = u32;
+}
+impl Enlargeable for u32 {
+    type Larger = u64;
+}
+impl Enlargeable for u64 {
+    type Larger = u128;
+}
+impl Enlargeable for usize {
+    // Note: On 32-bit architectures, u64 should be enough here.
+    type Larger = u128;
+}
+impl Enlargeable for i8 {
+    type Larger = i32;
+}
+impl Enlargeable for i16 {
+    type Larger = i32;
+}
+impl Enlargeable for i32 {
+    type Larger = i64;
+}
+impl Enlargeable for i64 {
+    type Larger = i128;
+}
+impl Enlargeable for isize {
+    // Note: On 32-bit architectures, i64 should be enough here.
+    type Larger = i128;
+}
+impl Enlargeable for f32 {
+    type Larger = f64;
+}
+impl Enlargeable for f64 {
+    type Larger = f64;
+}
+
+/// Linear interpolation without involving floating numbers.
+pub trait Lerp: Bounded + NumCast {
+    type Ratio: Primitive;
+
+    fn lerp(a: Self, b: Self, ratio: Self::Ratio) -> Self {
+        let a = <Self::Ratio as NumCast>::from(a).unwrap();
+        let b = <Self::Ratio as NumCast>::from(b).unwrap();
+
+        let res = a + (b - a) * ratio;
+
+        if res > NumCast::from(Self::max_value()).unwrap() {
+            Self::max_value()
+        } else if res < NumCast::from(0).unwrap() {
+            NumCast::from(0).unwrap()
+        } else {
+            NumCast::from(res).unwrap()
+        }
+    }
+}
+
+impl Lerp for u8 {
+    type Ratio = f32;
+}
+
+impl Lerp for u16 {
+    type Ratio = f32;
+}
+
+impl Lerp for u32 {
+    type Ratio = f64;
+}
+
+impl Lerp for f32 {
+    type Ratio = f32;
+
+    fn lerp(a: Self, b: Self, ratio: Self::Ratio) -> Self {
+        a + (b - a) * ratio
+    }
+}
+
+/// The pixel with an associated `ColorType`.
+/// Not all possible pixels represent one of the predefined `ColorType`s.
+pub trait PixelWithColorType: Pixel + self::private::SealedPixelWithColorType {
+    /// This pixel has the format of one of the predefined `ColorType`s,
+    /// such as `Rgb8`, `La16` or `Rgba32F`.
+    /// This is needed for automatically detecting
+    /// a color format when saving an image as a file.
+    const COLOR_TYPE: ColorType;
+}
+
+impl PixelWithColorType for Rgb<u8> {
+    const COLOR_TYPE: ColorType = ColorType::Rgb8;
+}
+impl PixelWithColorType for Rgb<u16> {
+    const COLOR_TYPE: ColorType = ColorType::Rgb16;
+}
+impl PixelWithColorType for Rgb<f32> {
+    const COLOR_TYPE: ColorType = ColorType::Rgb32F;
+}
+
+impl PixelWithColorType for Rgba<u8> {
+    const COLOR_TYPE: ColorType = ColorType::Rgba8;
+}
+impl PixelWithColorType for Rgba<u16> {
+    const COLOR_TYPE: ColorType = ColorType::Rgba16;
+}
+impl PixelWithColorType for Rgba<f32> {
+    const COLOR_TYPE: ColorType = ColorType::Rgba32F;
+}
+
+impl PixelWithColorType for Luma<u8> {
+    const COLOR_TYPE: ColorType = ColorType::L8;
+}
+impl PixelWithColorType for Luma<u16> {
+    const COLOR_TYPE: ColorType = ColorType::L16;
+}
+impl PixelWithColorType for LumaA<u8> {
+    const COLOR_TYPE: ColorType = ColorType::La8;
+}
+impl PixelWithColorType for LumaA<u16> {
+    const COLOR_TYPE: ColorType = ColorType::La16;
+}
+
+/// Prevents down-stream users from implementing the `Primitive` trait
+mod private {
+    use crate::color::*;
+
+    pub trait SealedPixelWithColorType {}
+    impl SealedPixelWithColorType for Rgb<u8> {}
+    impl SealedPixelWithColorType for Rgb<u16> {}
+    impl SealedPixelWithColorType for Rgb<f32> {}
+
+    impl SealedPixelWithColorType for Rgba<u8> {}
+    impl SealedPixelWithColorType for Rgba<u16> {}
+    impl SealedPixelWithColorType for Rgba<f32> {}
+
+    impl SealedPixelWithColorType for Luma<u8> {}
+    impl SealedPixelWithColorType for LumaA<u8> {}
+
+    impl SealedPixelWithColorType for Luma<u16> {}
+    impl SealedPixelWithColorType for LumaA<u16> {}
+}
+
+/// A generalized pixel.
+///
+/// A pixel object is usually not used standalone but as a view into an image buffer.
+pub trait Pixel: Copy + Clone {
+    /// The scalar type that is used to store each channel in this pixel.
+    type Subpixel: Primitive;
+
+    /// The number of channels of this pixel type.
+    const CHANNEL_COUNT: u8;
+
+    /// Returns the components as a slice.
+    fn channels(&self) -> &[Self::Subpixel];
+
+    /// Returns the components as a mutable slice
+    fn channels_mut(&mut self) -> &mut [Self::Subpixel];
+
+    /// A string that can help to interpret the meaning each channel
+    /// See [gimp babl](http://gegl.org/babl/).
+    const COLOR_MODEL: &'static str;
+
+    /// Returns the channels of this pixel as a 4 tuple. If the pixel
+    /// has less than 4 channels the remainder is filled with the maximum value
+    #[deprecated(since = "0.24.0", note = "Use `channels()` or `channels_mut()`")]
+    fn channels4(
+        &self,
+    ) -> (
+        Self::Subpixel,
+        Self::Subpixel,
+        Self::Subpixel,
+        Self::Subpixel,
+    );
+
+    /// Construct a pixel from the 4 channels a, b, c and d.
+    /// If the pixel does not contain 4 channels the extra are ignored.
+    #[deprecated(
+        since = "0.24.0",
+        note = "Use the constructor of the pixel, for example `Rgba([r,g,b,a])` or `Pixel::from_slice`"
+    )]
+    fn from_channels(
+        a: Self::Subpixel,
+        b: Self::Subpixel,
+        c: Self::Subpixel,
+        d: Self::Subpixel,
+    ) -> Self;
+
+    /// Returns a view into a slice.
+    ///
+    /// Note: The slice length is not checked on creation. Thus the caller has to ensure
+    /// that the slice is long enough to prevent panics if the pixel is used later on.
+    fn from_slice(slice: &[Self::Subpixel]) -> &Self;
+
+    /// Returns mutable view into a mutable slice.
+    ///
+    /// Note: The slice length is not checked on creation. Thus the caller has to ensure
+    /// that the slice is long enough to prevent panics if the pixel is used later on.
+    fn from_slice_mut(slice: &mut [Self::Subpixel]) -> &mut Self;
+
+    /// Convert this pixel to RGB
+    fn to_rgb(&self) -> Rgb<Self::Subpixel>;
+
+    /// Convert this pixel to RGB with an alpha channel
+    fn to_rgba(&self) -> Rgba<Self::Subpixel>;
+
+    /// Convert this pixel to luma
+    fn to_luma(&self) -> Luma<Self::Subpixel>;
+
+    /// Convert this pixel to luma with an alpha channel
+    fn to_luma_alpha(&self) -> LumaA<Self::Subpixel>;
+
+    /// Apply the function ```f``` to each channel of this pixel.
+    fn map<F>(&self, f: F) -> Self
+    where
+        F: FnMut(Self::Subpixel) -> Self::Subpixel;
+
+    /// Apply the function ```f``` to each channel of this pixel.
+    fn apply<F>(&mut self, f: F)
+    where
+        F: FnMut(Self::Subpixel) -> Self::Subpixel;
+
+    /// Apply the function ```f``` to each channel except the alpha channel.
+    /// Apply the function ```g``` to the alpha channel.
+    fn map_with_alpha<F, G>(&self, f: F, g: G) -> Self
+    where
+        F: FnMut(Self::Subpixel) -> Self::Subpixel,
+        G: FnMut(Self::Subpixel) -> Self::Subpixel;
+
+    /// Apply the function ```f``` to each channel except the alpha channel.
+    /// Apply the function ```g``` to the alpha channel. Works in-place.
+    fn apply_with_alpha<F, G>(&mut self, f: F, g: G)
+    where
+        F: FnMut(Self::Subpixel) -> Self::Subpixel,
+        G: FnMut(Self::Subpixel) -> Self::Subpixel;
+
+    /// Apply the function ```f``` to each channel except the alpha channel.
+    fn map_without_alpha<F>(&self, f: F) -> Self
+    where
+        F: FnMut(Self::Subpixel) -> Self::Subpixel,
+    {
+        let mut this = *self;
+        this.apply_with_alpha(f, |x| x);
+        this
+    }
+
+    /// Apply the function ```f``` to each channel except the alpha channel.
+    /// Works in place.
+    fn apply_without_alpha<F>(&mut self, f: F)
+    where
+        F: FnMut(Self::Subpixel) -> Self::Subpixel,
+    {
+        self.apply_with_alpha(f, |x| x);
+    }
+
+    /// Apply the function ```f``` to each channel of this pixel and
+    /// ```other``` pairwise.
+    fn map2<F>(&self, other: &Self, f: F) -> Self
+    where
+        F: FnMut(Self::Subpixel, Self::Subpixel) -> Self::Subpixel;
+
+    /// Apply the function ```f``` to each channel of this pixel and
+    /// ```other``` pairwise. Works in-place.
+    fn apply2<F>(&mut self, other: &Self, f: F)
+    where
+        F: FnMut(Self::Subpixel, Self::Subpixel) -> Self::Subpixel;
+
+    /// Invert this pixel
+    fn invert(&mut self);
+
+    /// Blend the color of a given pixel into ourself, taking into account alpha channels
+    fn blend(&mut self, other: &Self);
+}
+
+/// Private module for supertraits of sealed traits.
+mod seals {
+    pub trait EncodableLayout {}
+
+    impl EncodableLayout for [u8] {}
+    impl EncodableLayout for [u16] {}
+    impl EncodableLayout for [f32] {}
+}

vendor/image/src/utils/mod.rs

@@ -0,0 +1,128 @@
+//!  Utilities
+
+use std::iter::repeat;
+
+#[inline(always)]
+pub(crate) fn expand_packed<F>(buf: &mut [u8], channels: usize, bit_depth: u8, mut func: F)
+where
+    F: FnMut(u8, &mut [u8]),
+{
+    let pixels = buf.len() / channels * bit_depth as usize;
+    let extra = pixels % 8;
+    let entries = pixels / 8
+        + match extra {
+            0 => 0,
+            _ => 1,
+        };
+    let mask = ((1u16 << bit_depth) - 1) as u8;
+    let i = (0..entries)
+        .rev() // Reverse iterator
+        .flat_map(|idx|
+            // This has to be reversed to
+            (0..8/bit_depth).map(|i| i*bit_depth).zip(repeat(idx)))
+        .skip(extra);
+    let buf_len = buf.len();
+    let j_inv = (channels..buf_len).step_by(channels);
+    for ((shift, i), j_inv) in i.zip(j_inv) {
+        let j = buf_len - j_inv;
+        let pixel = (buf[i] & (mask << shift)) >> shift;
+        func(pixel, &mut buf[j..(j + channels)])
+    }
+}
+
+/// Expand a buffer of packed 1, 2, or 4 bits integers into u8's. Assumes that
+/// every `row_size` entries there are padding bits up to the next byte boundary.
+#[allow(dead_code)]
+// When no image formats that use it are enabled
+pub(crate) fn expand_bits(bit_depth: u8, row_size: u32, buf: &[u8]) -> Vec<u8> {
+    // Note: this conversion assumes that the scanlines begin on byte boundaries
+    let mask = (1u8 << bit_depth as usize) - 1;
+    let scaling_factor = 255 / ((1 << bit_depth as usize) - 1);
+    let bit_width = row_size * u32::from(bit_depth);
+    let skip = if bit_width % 8 == 0 {
+        0
+    } else {
+        (8 - bit_width % 8) / u32::from(bit_depth)
+    };
+    let row_len = row_size + skip;
+    let mut p = Vec::new();
+    let mut i = 0;
+    for v in buf {
+        for shift_inv in 1..=8 / bit_depth {
+            let shift = 8 - bit_depth * shift_inv;
+            // skip the pixels that can be neglected because scanlines should
+            // start at byte boundaries
+            if i % (row_len as usize) < (row_size as usize) {
+                let pixel = (v & mask << shift as usize) >> shift as usize;
+                p.push(pixel * scaling_factor);
+            }
+            i += 1;
+        }
+    }
+    p
+}
+
+/// Checks if the provided dimensions would cause an overflow.
+#[allow(dead_code)]
+// When no image formats that use it are enabled
+pub(crate) fn check_dimension_overflow(width: u32, height: u32, bytes_per_pixel: u8) -> bool {
+    width as u64 * height as u64 > std::u64::MAX / bytes_per_pixel as u64
+}
+
+#[allow(dead_code)]
+// When no image formats that use it are enabled
+pub(crate) fn vec_copy_to_u8<T>(vec: &[T]) -> Vec<u8>
+where
+    T: bytemuck::Pod,
+{
+    bytemuck::cast_slice(vec).to_owned()
+}
+
+#[inline]
+pub(crate) fn clamp<N>(a: N, min: N, max: N) -> N
+where
+    N: PartialOrd,
+{
+    if a < min {
+        min
+    } else if a > max {
+        max
+    } else {
+        a
+    }
+}
+
+#[cfg(test)]
+mod test {
+    #[test]
+    fn gray_to_luma8_skip() {
+        let check = |bit_depth, w, from, to| {
+            assert_eq!(super::expand_bits(bit_depth, w, from), to);
+        };
+        // Bit depth 1, skip is more than half a byte
+        check(
+            1,
+            10,
+            &[0b11110000, 0b11000000, 0b00001111, 0b11000000],
+            vec![
+                255, 255, 255, 255, 0, 0, 0, 0, 255, 255, 0, 0, 0, 0, 255, 255, 255, 255, 255, 255,
+            ],
+        );
+        // Bit depth 2, skip is more than half a byte
+        check(
+            2,
+            5,
+            &[0b11110000, 0b11000000, 0b00001111, 0b11000000],
+            vec![255, 255, 0, 0, 255, 0, 0, 255, 255, 255],
+        );
+        // Bit depth 2, skip is 0
+        check(
+            2,
+            4,
+            &[0b11110000, 0b00001111],
+            vec![255, 255, 0, 0, 0, 0, 255, 255],
+        );
+        // Bit depth 4, skip is half a byte
+        check(4, 1, &[0b11110011, 0b00001100], vec![255, 0]);
+    }
+}