Review test coverage and report score to Interop team

[foolip]

@hjanuschka added a lot of tests in web-platform-tests/wpt#58481 and we should review which of the features and web integration points in the README are now sufficiently tested.

Features of JPEG XL that should be tested:

• High bit depth: Test for 8, 10, 12, and 16-bit depths.
• Wide gamut: Test Display P3, Adobe RGB, Rec.2020, ProPhoto RGB, etc.
• HDR: Test tone mapping (PQ/HLG) and gain maps (ISO 21496-1)
• ICC profiles: Test exotic/uncommon use of ICC profiles.
• CMYK: Test using the de-facto standard conversion between sRGB and CMYK.
• VarDCT: Test at least the largest and smallest block sizes, and smaller/larger than allowed by spec if the bitstream can express it.
• Modular mode: Test both lossy and lossless mode.
• Alpha channel: Test with VarDCT and modular mode.
• Progressive Rendering: Align on how browsers should render just the DC data, and whether to support progressive_dc which is even further downsampled.
• Lossless Transcoded JPEG: Test basic support.
• Orientation: Test that orientation in codestream is respected and EXIF orientation is ignored.
• Animations: Test basic support.
• Layers: Test basic support.
• JPEG XL art: Support for minimal JPEG XL files using modular mode to produce JPEG XL art (may be sensitive to minor differences in decoders)

Web integration points to test:

• <img>, <picture> & <source>
• <video poster>
• <input type=image>
• <embed>, <object> & <iframe>
• <link rel=preload as=image>
• SVG <image>
• CSS background-image, content, image-set(), list-style-image
• Canvas (2D, WebGL, WebGPU) including createImageBitmap() and Float16Array
• ImageDecoder API

[hjanuschka]

not sure if i am qualified to self sign off what i sent tests for 🙈

Features - covered:

• High bit depth -- 8, 10, 12, 16-bit all have reftests
• ICC profiles -- icc-uncommon-profile-reftest
• CMYK -- cmyk-basic-conversion-reftest
• VarDCT -- large + small block reftests
• Modular mode -- lossy + lossless
• Alpha channel -- VarDCT + modular
• Progressive Rendering -- progressive-dc-observable-reftest
• Lossless Transcoded JPEG -- lossless-jpeg-transcode
• Orientation -- identity, 90 CCW, codestream-vs-EXIF reftests
• Animations -- imagedecoder-animation (multi-frame via ImageDecoder)
• Layers -- layers-basic-reftest
• JPEG XL art -- jxl-art-basic-reftest

Features - gaps remaining:

• Wide gamut -- Display P3 & Rec.2020 are decode-only scaffolds (no actual gamut verification, blocked on WPT color-management infra per Add the ability to test HDR and wide color rendering wpt#44320). ProPhoto RGB is not covered at all. Adobe RGB has a reftest.
• HDR -- PQ/HLG only have .manual.html tests (not automated). Gain map is just a placeholder paragraph with no actual test image.

Web integration:

• <img>, <picture> & <source> -- html-img.html
• <video poster> -- html-video-poster.html
• <input type=image> -- html-input-image.html
• <embed>, <object> & <iframe> -- html-embed-object-iframe.html
• <link rel=preload as=image> -- html-preload-image.html
• SVG <image> -- svg-image-element.html
• CSS background-image, content, image-set(), list-style-image -- css-image-usage.html
• ImageDecoder API -- imagedecoder.https.html + animation variant

Web integration - gaps remaining:

• Canvas -- canvas-decode-paths.html covers 2D drawImage and createImageBitmap(), but WebGL, WebGPU, and Float16Array are not tested yet.

i'll do follow up for Canvas (missed that one)

[stof]

It would be great to actually compute a score based on this update and report it to the Interop dashboard. See web-platform-tests/wpt.fyi#4846 for an example of PR updating the score of an investigation area.

[hjanuschka]

@stof you have any hint/samples on how to generate the score? or is this a human-made-up number? sorry if that might be a dumb question, first time in this position/task

and the canvas cases should also be done already with the recent CL

[stof]

@hjanuschka defining the scoring criteria is something that is under the responsibility of the investigation area team.
The readme mentions scoring but I don't know whether this is precise enough to be able to compute the score. The computation is manual (if scoring could be automated based on passing existing tests, it would be a focus area, not an investigation area).

For reference, see this issue for the scoring of the accessibility area (their scoring is split in different parts, with only misc using linear scoring). See web-platform-tests/interop-accessibility#148 (comment) for an example of computation done in 2025.

One potential computation could be to consider the scoring linear based on the 23 checkboxes in the issue description (assuming that this list is exhaustive). However, as some of those checkboxes cover multiple things to tests, another option might be to split them. I'll let the investigation team give precision on the scoring methodology (I'm not involved directly here, but I became the guy reminding investigation areas to update the dashboard in the past, due to curiosity about the state of interop and frustration of not having the data on the dashboard)