Moving an audio or video element to another Document reset paused even when
playback was active. Retain playback intent while resource selection restarts
under the new owner, allowing the clock and decoded frames to resume.
Add a 49-case realm/transport/media matrix and a 100-adoption decoder release
check. Correct two owned adoption expectations and retain their metadata,
Promise-realm and cancellation checks. Original WPT assertions stay unchanged.
Full fmt/clippy and 22,345 native tests pass with zero automatic retries.
The expanded original WPT cohort gains two PASS results with no prior losses.
Check initial same-value writes before any playback or muted setter primes the
control state. Check that reentrant numeric conversion compares against the
volume written during conversion and queues only one notification.
Keep the existing normative default/true/false muted-state assertions intact.
The frozen parent already passes all 208 media control cases. Full fmt/clippy
and 22,344 native tests pass with 32-way execution and zero automatic retries.
Map destination pixel centers to source texel centers and isolate interpolation
inside the requested crop. Sample padded crops on a transparent-black grid
without allocating a crop-sized intermediate image. Preserve premultiplied
filtering, HDR precision and exact nearest or unscaled pixel copies.
Add pixel, crop and cross-realm regressions. Full fmt/clippy and 22,343 native
tests pass; original video-resize WPT passes over HTTP and HTTPS, with no prior
PASS losses in the 1,874-subtest cohort.
Keep pending play requests in native element state and settle them through
owned media tasks. Preserve media object Promise realms and complete queued
settlements before load discards events. Reject failed requests in the error
task, and reject stale playback epochs before any task-end checkpoint.
Defer resource selection and removal checks to stable state and restart
media when its Document changes. Cover main/iframe calls, windowless media,
shadow-tree adoption, and captured-task cancellation without draining timers
or unrelated script work.
Replace the gradient shell with branded native geometry and mutable color
stops. Share paint sampling across fills, strokes and text, using the drawing
CTM and receiving context color space while preserving assigned object
identity and registered native proxies.
Use WebIDL double conversion and receiver validation before callbacks, and
allocate factory results in the receiver realm with callee-realm exceptions.
Keep uncovered geometry and transparent nonfinite colors out of sampling.
Validation: fmt, workspace clippy with all targets/features and warnings
denied, and nextest (22269 passed, 16 skipped). Original gradient/style WPT
improves from 890 to 1168 of 1192; all existing Canvas PASS cases are retained.
Retain float16/unorm8, sRGB/Display-P3 and alpha representation through
Canvas, OffscreenCanvas, ImageData and ImageBitmap operations. Composite
in premultiplied coordinates, convert at native format boundaries and
preserve untouched pixel bytes across partial writes and bitmap copies.
Keep parsed drawing colors independent from CSS serialization and use
the shared native receiver bindings for color attributes. Correct the
old trailing-zero serialization expectation against Chromium output.
Validation: fmt, full workspace clippy and nextest (22234 passed,
16 skipped; 32 jobs/threads, zero retries). Original WPT: 914/1116 (+12), all 902 prior passes retained. Sources,
harnesses and assertions are unchanged; browser evidence is retained.
Track opt-in original tag positions in html5ever, including UTF-16,
CRLF, character-reference rewinds and chunk boundaries. Carry completed
script start positions through the existing parser handoff and use them
for child Document CSP reports. Register synchronous document.write
handoffs before CSP so unknown provenance remains unknown.
Validation: full workspace fmt, clippy and nextest (22,156 passed,
16 skipped; 32 jobs/threads, no retries), plus 24 upstream/local
html5ever tokenizer and driver tests.