png

Additions

• Added Filter::MinEntropy for encoding, which selects the filter that minimizes the Shannon entropy of the filtered row. (#647)
• Made ScaledFloat::SCALING constant public. (#650)
• Support skipping chunks without loading them into memory. (#630)
• Allow full unfiltering for partial data, enabling progressive decoding use cases. (#664)

Changes

• Improved Paeth unfilter auto-vectorization by changing how loop state is maintained. (#635)
• Added portable_simd Paeth unfiltering for 3bpp and 4bpp images on nightly. (#632, #633)
• Add early exit for adaptive filter selection when a perfect filter is found. (#646)
• Ignore duplicate acTL chunks instead of producing errors. (#651)
• Require IDAT chunk(s) before fdAT chunks, as mandated by PNG specification. (#653)
• Enforce acTL frame counts to be valid PNG 4-byte integers (capped at 2^31-1). (#673)

Fixes

• Fixed and improved intra-doc links in documentation comments. (#665)
• Fixed nightly portable_simd build. (#675)

API Breaking Changes

• Removed deprecated Info::encode and Encoder::set_srgb methods.
• Improved the compression settings API for encoding.
• Decoder now requires a reader that implements Seek and BufRead traits.
• Bump bitflags dependency to 2.0.
• StreamingDecoder::update now takes a structured UnfilterBuf argument instead of a direct reference to a vector. This allows in-place decompression. There is a public constructor for UnfilterBuf.
• The methods Decoder::output_buffer_size and output_line_size now return Option<usize> to reflect that these calculations no longer overflow on some targets where the required buffers can not be represented in the address space. They return the mathematically correct size where possible.
• The Decoded enum returned from StreamingDecoder::update was simplified to no longer contains any chunk payload data. Instead, it now contains only chunk events where every chunk that was started will eventually be ended by ChunkComplete, BadAncillaryChunk or SkippedAncillaryChunk.
• Ancillary chunks, i.e. those not critical to decoder interpretation of the file, which fail to parse are now terminated with a BadAncillaryChunk event but no longer returned a DecodingError. This includes text chunks as well as many metadata chunks (except for fcTL that we deem crucial to the parser's understanding of the image sequence in an APNG despite being technically ancillary).

Additions

• Added Reader::read_row method.
• Add support for parsing eXIf chunk.
• Treat most auxiliary chunk errors as benign.
• Added splat_interlaced_row, which implements an alternative method for merging Adam7 interlaced lines into the output buffer that is more suitable for the presentation of progressive states of the buffer.
• Added Adam7Variant documenting the various methods for applying interlaced rows and to prepare an API to progressively read frames through Decoder.

Changes

• The decoding of Adam7 interlaced data is now much faster.
• The acTL chunk is now ignored when it is invalid, instead of producing errors while reading or decoding the following APNG chunks.
• The requirement of the fcTL chunk for the default image to match the IHDR's indicate image size is now enforced.
• More minor format errors in auxiliary chunks are now ignored by the decoder, instead disregarding the malformed chunk.
• Adam7 Interlacing on 32-bit targets now handles some cases correctly that previously wrote some bytes to the wrong pixel indices due to overflows.

Cumulative to the last 0.17 release

• Removed deprecated Info::encode and Encoder::set_srgb methods.

• Improved the compression settings API for encoding.

• Decoder now requires a reader that implements Seek and BufRead traits.

• Bump bitflags dependency to 2.0.

• StreamingDecoder::update now takes a structured UnfilterBuf argument
  instead of a direct reference to a vector. This allows in-place
  decompression. There is a public constructor for UnfilterBuf.

• The methods Decoder::output_buffer_size and output_line_size now return
  Option<usize> to reflect that these calculations no longer overflow on some
  targets where the required buffers can not be represented in the address
  space. They return the mathematically correct size where possible.

• Added Reader::read_row method.

• Add support for parsing eXIf chunk.

• Treat most auxiliary chunk errors as benign.

• Added splat_interlaced_row, which implements an alternative method for
  merging Adam7 interlaced lines into the output buffer that is more suitable
  for the presentation of progressive states of the buffer.

• Added Adam7Variant documenting the various methods for applying interlaced
  rows and to prepare an API to progressively read frames through Decoder.

• The decoding of Adam7 interlaced data is now much faster.

• The acTL chunk is now ignored when it is invalid, instead of producing
  errors while reading or decoding the following APNG chunks.

• The requirement of the fcTL chunk for the default image to match the IHDR's
  indicate image size is now enforced.

• More minor format errors in auxiliary chunks are now ignored by the decoder,
  instead disregarding the malformed chunk.

• Adam7 Interlacing on 32-bit targets now handles some cases correctly that
  previously wrote some bytes to the wrong pixel indices due to overflows.

• Make gAMA and cHRM fallback optional for sRGB (#547)
• Pass through nightly feature to crc32fast crate to get SIMD crc32 on Aarch64 (#545)
• Fix bug in iCCP chunk encoding (#458)
• Deprecate Info::encode (#550)

Added

• Add a public API to advance to the next frame in APNG decoder (#518)
• Add APIs to write ICC and EXIF chunks (#526)
• Add support for parsing the sBIT chunk (#524)
• Add support for parsing the bKGD chunk (#538)
• Add support for parsing the cICP chunk (#529)
• Add support for parsing mDCV and cLLI chunks (#528, (#543))

Changed

• Improve performance of Paeth filter during decoding (#512, #539)

Fixed

• Avoid an infinite loop when retrying after a fatal error using StreamingDecoder (#520)
• Fixed chunk order in encoded files (#526)

• Updated to miniz_oxide 0.8.0.
• Added public API to consume interlaced rows one by one (#495)
• Improved support for resuming decoding after an UnexpectedEof, which lets you start parsing a file before it's fully received over the network (#496)
• Fixed some broken links in documentation, improved some documentation comments

• Updated to miniz_oxide 0.8.0.
• Added public API to consume interlaced rows one by one (#495)
• Improved support for resuming decoding after an UnexpectedEof, which lets you start parsing a file before it's fully received over the network (#496)
• Fixed some broken links in documentation, improved some documentation comments

Release 0.17.13 (#472)

• Fix Send bound on Reader.

Release 0.17.12 (#468)

• Reject zero-sized frames.
• Optimized decoding of paletted images.
• Removed remaining uses of miniz_oxide for decoding.
• Correct lifetime used for Info struct.
• Fix build issue with -Z minimal-versions.

Release 0.17.11 (#455)

• Ignore subsequent iCCP chunks to match libpng behavior.
• Added an option to ignore ancillary chunks with invalid CRC.
• Added new_with_info constructor for encoder.
• Removed hard-coded memory limits.
• No longer allow zero sized images.
• Added Reader::finish to read all the auxiliary chunks that comes after the image.

• Added Transformations::ALPHA
• Enable encoding pixel dimensions

• Fixed a bug in ICC profile decompression.
• Improved unfilter performance.

• Increased MSRV to 1.57.0.
• Substantially optimized encoding and decoding:
  • Autovectorize filtering and unfiltering.
  • Make the "fast" compression preset use fdeflate.
  • Switch decompression to always use fdeflate.
  • Updated to miniz_oxide 0.7.
  • Added an option to ignore checksums.
• Added corpus-bench example which measures the compression ratio and time to re-encode and subsequently decode a corpus of images.
• More fuzz testing.

• Fixed handling broken tRNS chunk.
• Updated to miniz_oxide 0.6.

• Added Decoder::read_header_info to query the information contained in the PNG header.
• Switched to using the flate2 crate for encoding.

• Fixed a regression, introduced by chunk validation, that made the decoder sensitive to the order of gAMA, cHRM, and sRGB chunks.

• Added {Decoder,StreamDecoder}::set_ignore_text_chunk to disable decoding of ancillary text chunks during the decoding process (chunks decoded by default).
• Added duplicate chunk checks. The decoder now enforces that standard chunks such as palette, gamma, … occur at most once as specified.
• Added #[forbid(unsafe_code)] again. This may come at a minor performance cost when decoding ASCII text for now.
• Fixed a bug where decoding of large chunks (>32kB) failed to produce the correct result, or fail the image decoding. As new chunk types are decoded this introduced regressions relative to previous versions.

• Fixed a bug where Writer::finish would not drop the underlying writer. This would fail to flush and leak memory when using a buffered file writers.
• Calling Writer::finish will now eagerly flush the underlying writer, returning any error that this operation may result in.
• Errors in inflate are now diagnosed with more details.
• The color and depth combination is now checked in stream decoder.

• Added support for encoding and decoding tEXt/zTXt/iTXt chunks.
• Added Encoder::validate_sequence to enable validation of the written frame sequence, that is, if the number of written images is consistent with the animation state.
• Validation is now off by default. The basis of the new validation had been introduced in 0.17 but this fixes some cases where this validation was too aggressive compared to previous versions.
• Added Writer::finish to fully check the write of the end of an image instead of silently ignoring potential errors in Drop.
• The Writer::write_chunk method now validates that the computed chunk length does not overflow the limit set by PNG.
• Fix an issue where the library would panic or even abort the process when flush or write of an underlying writer panicked, or in some other uses of StreamWriter.

• Fix panic in adaptive filter method sum_buffer

• Increased MSRV to 1.46.0
• Rework output info usage
• Implement APNG encoding
• Improve ergonomics of encoder set_palette and set_trns methods
• Make Info struct non-exhaustive
• Make encoder a core feature
• Default Transformations to Identity
• Add Adaptive filtering method for encoding
• Fix SCREAM_CASE on ColorType variants
• Forbid unsafe code

• Added Encoder::set_trns to register a transparency table to be written.

• Fixed silent integer overflows in buffer size calculation, resulting in panics from assertions and out-of-bounds accesses when actually decoding. This improves the stability of 32-bit and 16-bit targets and make decoding run as stable as on 64-bit.
• Reject invalid color/depth combinations. Some would lead to mismatched output buffer size and panics during decoding.
• Add Clone impl for Info struct.

• Decoding of APNG subframes is now officially supported and specified. Note that dispose ops and positioning in the image need to be done by the caller.
• Added encoding of indexed data.
• Switched to miniz_oxide for decompressing image data, with 30%-50% speedup in common cases and up to 200% in special ones.
• Fix accepting images only with consecutive IDAT chunks, rules out data loss.

• The fdAT frames are no longer inspected when the main image is read. This would previously be the case for non-interlaced images. This would lead to incorrect failure and, e.g. an error of the form "invalid filter method".
• Fix always validating the last IDAT-chunks checksum, was sometimes ignored.
• Prevent encoding color/bit-depth combinations forbidden by the specification.
• The fixes for APNG/fdAT enable further implementation. The next release is expected to officially support APNG.

• Fix encoding with filtering methods Up, Avg, Paeth
• Optimize decoding throughput by up to +30%

• Added method constructing an owned stream encoder.

• Addressed files bloating the packed crate

• Fix a bug compressing images with deflate
• Address use of deprecated error interfaces

• Fix panic while trying to encode empty images. Such images are no longer accepted and error when calling write_header before any data has been written. The specification does not permit empty images.

• Fix EXPAND transformation to leave bit depths above 8 unchanged

• Fix encoding writing invalid chunks. Images written can be corrected: see https://github.com/image-rs/image/issues/1074 for a recovery.
• Fix a panic in bit unpacking with checked arithmetic (e.g. in debug builds)
• Added better fuzzer integration
• Update term, rand dev-dependency
• Note: The show example program requires a newer compiler than 1.34.2 on some targets due to depending on glium. This is not considered a breaking bug.