pure_jpeg 0.3.3 → 0.4.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 7c252d678f3b8c916a2430b802569dc5099a19636359e9864637ed9fe01c90a3
-  data.tar.gz: 8feec524d8c41ca74be1e9bc47cf41b762cb14dc3500dc03c9240680d5877933
+  metadata.gz: 41b11ef04efbfb35b257ce2e93e40ae4b4fd07fec3fd05918b87eacba9a517eb
+  data.tar.gz: 460755287366135f012fca8975822c9c50b52fa9e55144c3d1afc3c00538af37
 SHA512:
-  metadata.gz: c8d26b4d4bed6da87f048036b06bc80eddd6c7ff8318032fb9cfccc15f5aea961e62fefe9965dde33c3d33cea5b611b4c2f6cb2397efa2f43effd8eb456fcf93
-  data.tar.gz: 63a9fe6bdf1136c8ea6c8e38a6b6641e6be2975fef4fea9394f0b48de22b01862996a87557b5c0bf7d7915020630f08c9285ebf8b8e8482f38f228c31b2d08e7
+  metadata.gz: 2d9c6cb32fb7197cd2fb78bf391e970729036aab6c37b4215c8fca9aef687ca2079d209367956618e6951aaf9f6e809030a64e724b8bc34c070f0de1ad373680
+  data.tar.gz: 9d88a0c1e1d15068afe4827f838fa6736c1fde4403720edb32ac4c6e840619ccacf9ac7788e6b1ebe9df6dd8b61cc096bcac07424cd03b8ce9e89f9770ad1977

data/CHANGELOG.md

@@ -1,5 +1,52 @@
 # Changelog
 
+## 0.4.1
+
+This is a robustness and compatibility release. It makes decoding more reliable
+across valid JPEG variants, improves handling of malformed input, and strengthens
+Spinel support while preserving the existing public API and encoder output.
+
+Fixes:
+
+- Decode progressive images correctly when component edges require padded blocks
+- Preserve each scan's restart interval and each component's original quantization table
+- Decode baseline images with separate component scans and handle subsampled luminance
+- Handle JPEG marker fill bytes correctly between scans and at restart boundaries
+- Validate restart markers without recursion; reject malformed frame/scan parameters,
+  invalid Huffman tables, excessive coefficient runs, and incomplete scan data with `DecodeError`
+- Reject unsupported precision and RGB/Adobe color transforms explicitly
+- Read ICC headers after the frame header in `PureJPEG.info`; ignore incomplete or
+  inconsistent multipart profiles instead of returning partial profile bytes
+- Treat binary garbage containing null bytes as data rather than a filesystem path
+- Avoid a Spinel compiler bug in conditional hash assignment during decoding
+
+Testing and packaging:
+
+- Add small libjpeg-turbo fixtures with trusted decoded-pixel hashes and malformed-input regressions
+- Add Spinel pixel-parity checks for the decoder fixes; all four package checks and
+  all eight selected RawSource tests pass on Spinel revision `2406a5649`
+- Keep all existing tests, public API signatures, source-adapter behavior, and encoder output unchanged
+- Run CI on pushes to `main` and smoke-test the built, installed gem without development dependencies
+- Include creative-options documentation and IJG notices in the gem; correct the DCT algorithm attribution
+
+## 0.4.0
+
+Performance:
+
+- Faster Huffman coding and decoding, with scan references resolved once per scan
+- Combined quantization and zigzag passes, including direct reconstruction from progressive coefficient buffers
+- Color encoding downsamples chroma two rows at a time instead of allocating full-resolution chroma images
+- Block-based `RawSource` construction avoids allocating pixels that would immediately be replaced
+- Measured encode/decode times reduced by roughly 12–28% on Ruby 4.0.2 with YJIT; baseline decode allocations fell from 76,994 to 114 on the bundled 1024x1024 benchmark
+- Encoding array storage allocated for that image fell from 29.4 MB to 12.6 MB, with encoded bytes and decoded pixels preserved in equivalence checks
+
+Compatibility and testing:
+
+- Experimental Spinel compatibility, tested at revision `7f08abd9`, with benchmarks and a `rake test:spinel` proof-of-concept runner
+- Spinel currently passes seven of the eight selected RawSource tests; the remaining test fails because `respond_to?` is unavailable on its pixel struct. See `test/SPINEL.md` for details
+- Added regression tests for quantization rounding, progressive coefficient offsets, and chroma downsampling edges
+- Updated performance comparisons for CRuby, TruffleRuby, JRuby, and Spinel
+
 ## 0.3.3
 
 New features:

data/CREATIVE.md

@@ -0,0 +1,100 @@
+# Creative encoding with PureJPEG
+
+PureJPEG exposes access to several 'knobs' to manipulate the JPEG encoding pipeline. These produce a range of aesthetic effects by controlling how the DCT frequency coefficients are quantized.
+
+All examples assume:
+
+```ruby
+require "chunky_png"
+require_relative "lib/pure_jpeg"
+
+image = ChunkyPNG::Image.from_file("input.png")
+source = PureJPEG::Source::ChunkyPNGSource.new(image)
+```
+
+## Scrambled quantization ("early digicam" look)
+
+The JPEG spec stores quantization tables in zigzag order. Enabling `scramble_quantization` writes them in raster order instead, which means any standard viewer applies the wrong quantization values to the wrong frequency coefficients. The result is a chaotic, crunchy artifact pattern reminiscent of early 2000s digital cameras (what I really wanted in the first place).
+
+```ruby
+PureJPEG.encode(source, quality: 20, scramble_quantization: true).write("scrambled.jpg")
+```
+
+Lower quality values amplify the effect.
+
+## Chroma crush (sharp detail, broken color)
+
+Set high luminance quality but aggressively compress the chrominance channels. Detail and edges stay sharp but the color information collapses into large blocky patches with hue shifts.
+
+```ruby
+PureJPEG.encode(source, quality: 90, chroma_quality: 5).write("chromacrush.jpg")
+```
+
+## Luma crush (soft image, faithful color)
+
+The opposite: crush the luminance channel while preserving chrominance. Produces a soft, almost oil-painting quality where the image is blocky and low-detail but the colors remain surprisingly accurate.
+
+```ruby
+PureJPEG.encode(source, quality: 10, chroma_quality: 95).write("lumacrush.jpg")
+```
+
+## DC-only (8x8 mosaic)
+
+A custom quantization table that preserves only the DC coefficient (the average brightness of each 8x8 block) and kills all 63 AC coefficients. Every block becomes a single flat color, producing a hard mosaic effect.
+
+```ruby
+dc_only = [1] + [255] * 63
+PureJPEG.encode(source, luminance_table: dc_only, chrominance_table: dc_only).write("dconly.jpg")
+```
+
+You can soften the mosaic by letting a few low-frequency AC coefficients through:
+
+```ruby
+# Keep DC and the first few AC coefficients
+gentle_mosaic = [1, 1, 1, 2, 4, 8] + [255] * 58
+PureJPEG.encode(source, luminance_table: gentle_mosaic, chrominance_table: gentle_mosaic).write("gentle_mosaic.jpg")
+```
+
+## Inverted frequency emphasis
+
+Normally JPEG preserves low frequencies (smooth gradients) and discards high frequencies (fine detail). This modifier inverts that, preserving edges and texture while crushing smooth areas. The result is an embossed or etched look.
+
+```ruby
+inverter = ->(table, _channel) {
+  max = table.max
+  table.map { |v| [max + 1 - v, 1].max }
+}
+PureJPEG.encode(source, quality: 30, quantization_modifier: inverter).write("inverted.jpg")
+```
+
+## Combining effects
+
+The `quantization_modifier` proc receives the table and a channel identifier (`:luminance` or `:chrominance`), so you can apply different transformations per channel:
+
+```ruby
+modifier = ->(table, channel) {
+  case channel
+  when :luminance
+    # Sharpen luma by halving quantization values
+    table.map { |v| [v / 2, 1].max }
+  when :chrominance
+    # Posterize color by doubling quantization values
+    table.map { |v| [v * 2, 255].min }
+  end
+}
+PureJPEG.encode(source, quality: 50, quantization_modifier: modifier).write("combined.jpg")
+```
+
+## Full option reference
+
+```ruby
+PureJPEG.encode(source,
+  quality: 85,                    # 1-100, overall compression level
+  grayscale: false,               # encode as single-channel grayscale
+  chroma_quality: nil,            # 1-100, independent Cb/Cr quality (defaults to quality)
+  luminance_table: nil,           # custom 64-element quantization table for Y (raster order)
+  chrominance_table: nil,         # custom 64-element quantization table for Cb/Cr (raster order)
+  quantization_modifier: nil,     # proc(table, :luminance/:chrominance) -> modified table
+  scramble_quantization: false    # write quant tables in wrong order (digicam effect)
+)
+```

data/LICENSE

@@ -19,3 +19,10 @@ 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.
+
+The integer DCT implementation in lib/pure_jpeg/dct.rb is adapted from the
+Independent JPEG Group's jfdctint.c and jidctint.c. Its upstream copyright,
+disclaimer, and distribution terms are included in LICENSE-IJG. PureJPEG's
+adaptation ports the transforms to Ruby arrays, removes the C integration
+and scaled-output variants, and normalizes forward coefficients before
+quantization. The IJG-derived portions remain subject to those terms.

data/LICENSE-IJG

@@ -0,0 +1,260 @@
+libjpeg-turbo note:  This file has been modified by The libjpeg-turbo Project
+to include only information relevant to libjpeg-turbo, to wordsmith certain
+sections, and to remove impolitic language that existed in the libjpeg v8
+README.  It is included only for reference.  Please see README.md for
+information specific to libjpeg-turbo.
+
+
+The Independent JPEG Group's JPEG software
+==========================================
+
+This distribution contains a release of the Independent JPEG Group's free JPEG
+software.  You are welcome to redistribute this software and to use it for any
+purpose, subject to the conditions under LEGAL ISSUES, below.
+
+This software is the work of Tom Lane, Guido Vollbeding, Philip Gladstone,
+Bill Allombert, Jim Boucher, Lee Crocker, Bob Friesenhahn, Ben Jackson,
+Julian Minguillon, Luis Ortiz, George Phillips, Davide Rossi, Ge' Weijers,
+and other members of the Independent JPEG Group.
+
+IJG is not affiliated with the ISO/IEC JTC1/SC29/WG1 standards committee
+(also known as JPEG, together with ITU-T SG16).
+
+
+DOCUMENTATION ROADMAP
+=====================
+
+This file contains the following sections:
+
+OVERVIEW            General description of JPEG and the IJG software.
+LEGAL ISSUES        Copyright, lack of warranty, terms of distribution.
+REFERENCES          Where to learn more about JPEG.
+ARCHIVE LOCATIONS   Where to find newer versions of this software.
+FILE FORMAT WARS    Software *not* to get.
+TO DO               Plans for future IJG releases.
+
+Other documentation files in the distribution are:
+
+User documentation:
+  doc/usage.txt         Usage instructions for cjpeg, djpeg, jpegtran,
+                        rdjpgcom, and wrjpgcom.
+  doc/*.1               Unix-style man pages for programs (same info as
+                        usage.txt).
+  doc/wizard.txt        Advanced usage instructions for JPEG wizards only.
+  doc/change.log        Version-to-version change highlights.
+Programmer and internal documentation:
+  doc/libjpeg.txt       How to use the JPEG library in your own programs.
+  src/example.c         Sample code for calling the JPEG library.
+  doc/structure.txt     Overview of the JPEG library's internal structure.
+  doc/coderules.txt     Coding style rules --- please read if you contribute
+                        code.
+
+Please read at least usage.txt.  Some information can also be found in the JPEG
+FAQ (Frequently Asked Questions) article.  See ARCHIVE LOCATIONS below to find
+out where to obtain the FAQ article.
+
+If you want to understand how the JPEG code works, we suggest reading one or
+more of the REFERENCES, then looking at the documentation files (in roughly
+the order listed) before diving into the code.
+
+
+OVERVIEW
+========
+
+This package contains C software to implement JPEG image encoding, decoding,
+and transcoding.  JPEG (pronounced "jay-peg") is a standardized compression
+method for full-color and grayscale images.  JPEG's strong suit is compressing
+photographic images or other types of images that have smooth color and
+brightness transitions between neighboring pixels.  Images with sharp lines or
+other abrupt features may not compress well with JPEG, and a higher JPEG
+quality may have to be used to avoid visible compression artifacts with such
+images.
+
+JPEG is normally lossy, meaning that the output pixels are not necessarily
+identical to the input pixels.  However, on photographic content and other
+"smooth" images, very good compression ratios can be obtained with no visible
+compression artifacts, and extremely high compression ratios are possible if
+you are willing to sacrifice image quality (by reducing the "quality" setting
+in the compressor.)
+
+This software implements JPEG baseline, extended-sequential, progressive, and
+lossless compression processes.  Provision is made for supporting all variants
+of these processes, although some uncommon parameter settings aren't
+implemented yet.  We have made no provision for supporting the hierarchical
+processes defined in the standard.
+
+We provide a set of library routines for reading and writing JPEG image files,
+plus two sample applications "cjpeg" and "djpeg", which use the library to
+perform conversion between JPEG and some other popular image file formats.
+The library is intended to be reused in other applications.
+
+In order to support file conversion and viewing software, we have included
+considerable functionality beyond the bare JPEG coding/decoding capability;
+for example, the color quantization modules are not strictly part of JPEG
+decoding, but they are essential for output to colormapped file formats.  These
+extra functions can be compiled out of the library if not required for a
+particular application.
+
+We have also included "jpegtran", a utility for lossless transcoding between
+different JPEG processes, and "rdjpgcom" and "wrjpgcom", two simple
+applications for inserting and extracting textual comments in JFIF files.
+
+The emphasis in designing this software has been on achieving portability and
+flexibility, while also making it fast enough to be useful.  In particular,
+the software is not intended to be read as a tutorial on JPEG.  (See the
+REFERENCES section for introductory material.)  Rather, it is intended to
+be reliable, portable, industrial-strength code.  We do not claim to have
+achieved that goal in every aspect of the software, but we strive for it.
+
+We welcome the use of this software as a component of commercial products.
+No royalty is required, but we do ask for an acknowledgement in product
+documentation, as described under LEGAL ISSUES.
+
+
+LEGAL ISSUES
+============
+
+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) 1991-2020, 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.
+
+
+REFERENCES
+==========
+
+We recommend reading one or more of these references before trying to
+understand the innards of the JPEG software.
+
+The best short technical introduction to the JPEG compression algorithm is
+        Wallace, Gregory K.  "The JPEG Still Picture Compression Standard",
+        Communications of the ACM, April 1991 (vol. 34 no. 4), pp. 30-44.
+(Adjacent articles in that issue discuss MPEG motion picture compression,
+applications of JPEG, and related topics.)  If you don't have the CACM issue
+handy, a PDF file containing a revised version of Wallace's article is
+available at http://www.ijg.org/files/Wallace.JPEG.pdf.  The file (actually
+a preprint for an article that appeared in IEEE Trans. Consumer Electronics)
+omits the sample images that appeared in CACM, but it includes corrections
+and some added material.  Note: the Wallace article is copyright ACM and IEEE,
+and it may not be used for commercial purposes.
+
+A somewhat less technical, more leisurely introduction to JPEG can be found in
+"The Data Compression Book" by Mark Nelson and Jean-loup Gailly, published by
+M&T Books (New York), 2nd ed. 1996, ISBN 1-55851-434-1.  This book provides
+good explanations and example C code for a multitude of compression methods
+including JPEG.  It is an excellent source if you are comfortable reading C
+code but don't know much about data compression in general.  The book's JPEG
+sample code is far from industrial-strength, but when you are ready to look
+at a full implementation, you've got one here...
+
+The best currently available description of JPEG is the textbook "JPEG Still
+Image Data Compression Standard" by William B. Pennebaker and Joan L.
+Mitchell, published by Van Nostrand Reinhold, 1993, ISBN 0-442-01272-1.
+Price US$59.95, 638 pp.  The book includes the complete text of the ISO JPEG
+standards (DIS 10918-1 and draft DIS 10918-2).
+
+The original JPEG standard is divided into two parts, Part 1 being the actual
+specification, while Part 2 covers compliance testing methods.  Part 1 is
+titled "Digital Compression and Coding of Continuous-tone Still Images,
+Part 1: Requirements and guidelines" and has document numbers ISO/IEC IS
+10918-1, ITU-T T.81.  Part 2 is titled "Digital Compression and Coding of
+Continuous-tone Still Images, Part 2: Compliance testing" and has document
+numbers ISO/IEC IS 10918-2, ITU-T T.83.
+
+The JPEG standard does not specify all details of an interchangeable file
+format.  For the omitted details, we follow the "JFIF" conventions, revision
+1.02.  JFIF version 1 has been adopted as ISO/IEC 10918-5 (05/2013) and
+Recommendation ITU-T T.871 (05/2011): Information technology - Digital
+compression and coding of continuous-tone still images: JPEG File Interchange
+Format (JFIF).  It is available as a free download in PDF file format from
+https://www.iso.org/standard/54989.html and http://www.itu.int/rec/T-REC-T.871.
+A PDF file of the older JFIF 1.02 specification is available at
+http://www.w3.org/Graphics/JPEG/jfif3.pdf.
+
+The TIFF 6.0 file format specification can be obtained from
+http://mirrors.ctan.org/graphics/tiff/TIFF6.ps.gz.  The JPEG incorporation
+scheme found in the TIFF 6.0 spec of 3-June-92 has a number of serious
+problems.  IJG does not recommend use of the TIFF 6.0 design (TIFF Compression
+tag 6).  Instead, we recommend the JPEG design proposed by TIFF Technical Note
+#2 (Compression tag 7).  Copies of this Note can be obtained from
+http://www.ijg.org/files/.  It is expected that the next revision
+of the TIFF spec will replace the 6.0 JPEG design with the Note's design.
+Although IJG's own code does not support TIFF/JPEG, the free libtiff library
+uses our library to implement TIFF/JPEG per the Note.
+
+
+ARCHIVE LOCATIONS
+=================
+
+The "official" archive site for this software is www.ijg.org.
+The most recent released version can always be found there in
+directory "files".
+
+The JPEG FAQ (Frequently Asked Questions) article is a source of some
+general information about JPEG.  It is available at
+http://www.faqs.org/faqs/jpeg-faq.
+
+
+FILE FORMAT COMPATIBILITY
+=========================
+
+This software implements ITU T.81 | ISO/IEC 10918 with some extensions from
+ITU T.871 | ISO/IEC 10918-5 (JPEG File Interchange Format-- see REFERENCES).
+Informally, the term "JPEG image" or "JPEG file" most often refers to JFIF or
+a subset thereof, but there are other formats containing the name "JPEG" that
+are incompatible with the original JPEG standard or with JFIF (for instance,
+JPEG 2000 and JPEG XR).  This software therefore does not support these
+formats.  Indeed, one of the original reasons for developing this free software
+was to help force convergence on a common, interoperable format standard for
+JPEG files.
+
+JFIF is a minimal or "low end" representation.  TIFF/JPEG (TIFF revision 6.0 as
+modified by TIFF Technical Note #2) can be used for "high end" applications
+that need to record a lot of additional data about an image.
+
+
+TO DO
+=====
+
+Please send bug reports, offers of help, etc. to jpeg-info@jpegclub.org.

data/README.md

@@ -6,7 +6,7 @@
 
 Convert PNG or other pixel data to JPEG. Or the other way! Implements baseline JPEG encoding (DCT, Huffman, 4:2:0 chroma subsampling) and decodes both baseline and progressive JPEGs. Exposes a variety of encoding options to adjust parts of the JPEG pipeline not normally available (I needed this to recreate the JPEG compression styles of older digital cameras - don't ask..)
 
-It works on CRuby 3.0+, TruffleRuby 33.0, and JRuby 10.0. There's *almost* 100% test coverage - I need to find some "broken" JPEGs to do the rest (hit me up if you have any sources..)
+It works on CRuby 3.0+, TruffleRuby 33.0, JRuby 10.0, and [Spinel](https://github.com/matz/spinel) (tested at `7f08abd9`). Tests include external JPEG fixtures with pixel expectations generated by libjpeg-turbo, alongside round-trip and malformed-input checks.
 
 > [!NOTE]
 > Rubyists might find the [AI Disclosure](#ai-disclosure) section below of interest.
@@ -23,6 +23,15 @@ gem "pure_jpeg"
 gem install pure_jpeg
 ```
 
+> [!NOTE]
+> Using [Spinel](https://github.com/matz/spinel)? Add PureJPEG from your Spin project directory:
+>
+> ```sh
+> spin add pure_jpeg --git https://github.com/peterc/pure_jpeg
+> ```
+>
+> Then use `require "pure_jpeg"` in your code as usual.
+
 There are no runtime dependencies. [ChunkyPNG](https://github.com/wvanbergen/chunky_png) is optional (though quite useful) if you want to use `from_chunky_png`.
 
 `examples/` contains some useful example scripts for basic JPEG to PNG and PNG to JPEG conversion if you want to do some quick tests without writing code.
@@ -151,8 +160,13 @@ info.width           # => 1024
 info.height          # => 768
 info.component_count # => 3
 info.progressive     # => false
+info.icc_profile     # => binary profile bytes, or nil
 ```
 
+Metadata queries read headers through the first scan marker, without decoding
+pixels. Incomplete or inconsistent multipart ICC profiles are ignored and
+reported as `nil`. ICC extraction does not perform color-profile conversion.
+
 ### Iterating pixels
 
 ```ruby
@@ -198,22 +212,33 @@ Decoding:
 - 1-component (grayscale) and 3-component (YCbCr) images
 - Any chroma subsampling factor (4:4:4, 4:2:2, 4:2:0, etc.)
 - Restart markers (DRI/RST)
+- Separate component scans in baseline JPEGs
 
-Not supported: arithmetic coding, 12-bit precision, EXIF/ICC profile preservation, adding a default background for transparent sources (see what happens above!). Largely because I don't need these, but they are all do-able, especially with how loosely coupled this library is internally. Raise an issue if you really care about them!
+Encoding and decoding allow dimensions up to 8192 pixels on each axis. The
+decoder allocates whole-image buffers; this dimension limit is not a memory
+budget. `PureJPEG.info` can inspect dimensions without allocating pixel buffers.
+Invalid or unsupported JPEG data raises `PureJPEG::DecodeError`.
+
+Not supported: arithmetic coding, 12-bit precision, RGB/CMYK JPEG color spaces,
+or EXIF/ICC profile preservation when re-encoding. Transparent PNG sources can
+be composited with the explicit `background:` option above. Raise an issue if
+you need one of the unsupported formats!
 
 Possible future improvements: ICC profile rendering/conversion.
 
 ## Performance
 
-On a 1024x1024 image (Apple M5, 5 runs after warmup):
+On the bundled 1024x1024 images (Apple M5, median of 15 runs after 20 seconds of warmup per operation):
+
+| Operation | CRuby 4.0.2 (YJIT) | TruffleRuby 34.0.1 | JRuby 9.4.14.0 | Spinel (`7f08abd9`) |
+|-----------|-------------------|-------------------|---------------|--------------------|
+| Encode (color, q85) | ~0.131s | ~0.088s | ~0.577s | ~0.104s |
+| Decode (baseline) | ~0.112s | ~0.027s | ~0.554s | ~0.078s |
+| Decode (progressive) | ~0.140s | ~0.037s | ~0.617s | ~0.092s |
 
-| Operation | CRuby 4.0.2 (YJIT) | TruffleRuby 33.0.1 |
-|-----------|---------------------|---------------------|
-| Encode (color, q85) | ~0.16s | ~0.08s |
-| Decode (baseline) | ~0.14s | ~0.05s |
-| Decode (progressive) | ~0.18s | ~0.09s |
+Inputs were loaded before timing, and encoded output was returned in memory; file I/O and PNG loading are excluded. Runtimes were benchmarked sequentially. JRuby used its default JIT on OpenJDK 21.0.10. Spinel was measured separately with the same inputs and timing protocol, using a portable harness with PNG pixels preloaded; its encoded JPEG and decoded pixels matched CRuby byte-for-byte. Results vary with image content, garbage collection, and JIT warmup.
 
-The encoder and decoder use an integer-scaled AAN (Arai-Agui-Nakajima) DCT with fixed-point arithmetic throughout — no Float operations in the hot path. Color space conversion uses fixed-point integer math, and pixel data is stored as packed integers to avoid per-pixel object allocation. TruffleRuby's Graal JIT compiler can optimize these tight integer loops particularly well, resulting in 2-3x faster performance once warmed up.
+The encoder and decoder use the IJG accurate integer DCT, based on the Loeffler factorization, with fixed-point arithmetic throughout — no Float operations in the hot path. Color space conversion uses fixed-point integer math, and pixel data is stored as packed integers to avoid per-pixel object allocation. TruffleRuby's Graal JIT compiler can optimize these tight integer loops particularly well once warmed up.
 
 ## Example scripts
 
@@ -264,6 +289,19 @@ rake benchmark   # basic benchmark of encoding and decoding (5 runs after warmup
 rake profile     # CPU profile with StackProf (requires the stackprof gem)
 ```
 
+The Minitest suite lives in `test/minitest/`; `rake` and `rake test` run it
+with SimpleCov as before. To run one file:
+
+```sh
+bundle exec rake test TEST=test/minitest/test_raw_source.rb
+```
+
+With Spinel and Spin installed, `spin test` runs the standalone package checks
+in `test/*.rb`, comparing their output with CRuby. These cover color/grayscale
+round trips, metadata, and invalid input; they supplement the full Minitest
+suite. See [test/SPINEL.md](test/SPINEL.md) for details and the existing
+`rake test:spinel` runner.
+
 ## Full benchmark script
 
 `benchmark/run.rb` is a more thorough benchmark that exercises the encode and decode paths in several ways. It auto-enables YJIT, warms up before measuring, and reports object allocations, throughput (iterations/second via `benchmark-ips`), best-of-N wall-clock times, and a sustained mixed workload across encode (q85, q95 optimized, grayscale) and decode (baseline and progressive).
@@ -299,4 +337,6 @@ ruby benchmark/run.rb --profile-alloc  # retained-object profile with Vernier
 
 ## License
 
-MIT
+MIT, with IJG terms for the adapted integer DCT implementation. This software
+is based in part on the work of the Independent JPEG Group. See [LICENSE](LICENSE)
+and [LICENSE-IJG](LICENSE-IJG) for the notices and adaptation details.

data/lib/pure_jpeg/bit_reader.rb

@@ -35,6 +35,26 @@ module PureJPEG
       @buffer = 0
     end
 
+    # Consume exactly the restart expected at an MCU boundary. Marker fill
+    # bytes are allowed; entropy bytes and out-of-sequence restarts are not.
+    def read_restart(expected)
+      reset
+      unless @data.getbyte(@pos) == 0xFF
+        raise DecodeError, "Missing JPEG restart marker"
+      end
+      @pos += 1 while @pos < @length && @data.getbyte(@pos) == 0xFF
+      unless @data.getbyte(@pos) == 0xD0 + expected
+        raise DecodeError, "Unexpected JPEG restart marker (expected RST#{expected})"
+      end
+      @pos += 1
+    end
+
+    def finish_scan!
+      # Remaining bits in the last byte are padding, but whole unread bytes
+      # indicate extra entropy data or an unexpected trailing restart marker.
+      raise DecodeError, "Extra data after JPEG scan" unless @pos == @length
+    end
+
     # Read additional bits and sign-extend per JPEG spec (receive/extend).
     def receive_extend(size)
       return 0 if size == 0
@@ -56,10 +76,7 @@ module PureJPEG
         raise PureJPEG::DecodeError, "Unexpected end of scan data" if @pos >= @length
         next_byte = @data.getbyte(@pos)
         @pos += 1
-        # 0xFF 0x00 is a stuffed 0xFF byte
-        # Skip restart markers (0xD0-0xD7)
-        return fill_buffer if next_byte != 0x00 && next_byte >= 0xD0 && next_byte <= 0xD7
-        # For 0x00, the byte is 0xFF (stuffing removed)
+        raise DecodeError, "Unexpected marker in JPEG entropy data" unless next_byte == 0x00
       end
       @buffer = byte
       @bits_in_buffer = 8

data/lib/pure_jpeg/dct.rb

@@ -2,8 +2,10 @@
 
 module PureJPEG
   # Integer-scaled DCT based on the IJG (Independent JPEG Group) reference
-  # implementation (jfdctint.c / jidctint.c). Uses the Arai-Agui-Nakajima
-  # factorization with 13-bit fixed-point constants.
+  # implementation (jfdctint.c / jidctint.c). Uses the Loeffler-based accurate
+  # integer factorization with 13-bit fixed-point constants. The upstream
+  # copyright and distribution terms are included in LICENSE-IJG; adaptation
+  # details are recorded in LICENSE.
   #
   # All arithmetic is pure Integer (additions, shifts, multiplies) — no Float
   # operations. This is ~3x faster than the matrix-multiply float DCT under