DOM known-fail files fall to six
DOM baseline corrections exposed previously unexecuted subtests, while QuickJS binding milestones passed audit. CSS rendering fixes added 341 passing cases across caching, line clamping and tables.
Overview
539 commits this week. The js-dom track ran 169 rounds (R194–R378, roughly 350 of them code). Mid-week a baseline migration unlocked the real subtests of twelve mega-case files whose setup scripts had not been executing, and the full WPT sweep ends at 55,807 passes; the known-fail file set went from 11 to 6.
M6, the QuickJS native-binding milestone, was audited and confirmed fully closed (DC-7); M3's Vue e2e acceptance now passes under QuickJS too; and a new pending-apply lifecycle RFC opened, with its first audit slice landed. Rendering flipped 341 upstream CSS reftest cases green, 13,237→13,577. One style-key cache fix accounts for 73 of those; the long -webkit-line-clamp tail and table border-collapse/caption work account for most of the rest, and the SVG intrinsic-size cluster closed at 100%.
Service Worker baselines grew again: fetch 15 cases / 34 subtests passing, SW CacheStorage 12 / 157, core 162/162. The CSS parser took a counter-style grammar tightening series plus @namespace support. CI was functionally green on all seven platforms. The two recurring failures are pre-existing: windows-aarch64 ring/clang (round 21), and two benchmarks metrics on the 7763 runner that ran 5% and 8% over budget on a platform that itself drifted about +20%.
Completed
js-dom R194–R378, 169 rounds with roughly 350 code commits. The Range API got a rewrite: extractContents, surroundContents and compareBoundaryPoints, plus live boundary adjustment. The single biggest jump was fresh-document body/head rebinding in R221, worth +1,910 passes across the Range families.
DC-3 coverage advanced through domain imports: dom/lists with DOMTokenList branding (36P/153F → 189P/0F), dom/abort with frame-scoped AbortSignal.timeout, and the dom root-case set with WebIDL interface-global normalization (sweep +115).
The known-fail file count went from 11 to 6; each file closed with a specific fix: cross-realm MutationObserver reporting, per-realm in-flight events, scoped querySelectorAll pending-union, iframe inner-body connectivity, and insertion-time script execution. customElements gained per-realm registries and creation-realm routing; ShadowRoot.prototype.innerHTML landed.
An audit confirmed M6 (QuickJS native bindings) fully closed; the Vue e2e suite joined the QuickJS matrix; and a pending-apply lifecycle RFC (pa1–pa4) opened, with the pa1 mark-lifecycle audit landed.
Rendering flipped 341 upstream reftest cases green, 13,237→13,577 (R3786–R3816). A style-key cache sibling-combinator mismatch fix flipped 73 cases in one round. The -webkit-line-clamp tail covered cross-block clamping, clamp-point retreat, float exclusion and per-run white-space, closing the with-floats subset.
Elsewhere: replaced-element §10.4 constraint tables, border-style:none frames, col/colgroup border widths, caption margins/BFC, atomic inline-box margin-box alignment; legacy -webkit-box display now maps to flex; and the SVG mixed abs+%+viewBox ComputedIntrinsic cluster closed at 100%.
Service Worker WPT baselines grew: fetch/interception runner to 15 cases / 34 subtests, all passing (claim semantics, uncontrolled pages, stopPropagation, throw-after-respondWith, custom response bodies); SW CacheStorage serviceworker runner 12 cases / 157 subtests; core lifecycle 162/162. The storage-cache-api sibling brought its CacheStorage window baseline to 431 subtests, all passing.
CSS parsing: counter-style grammar tightened across systems, symbols, ranges, prefixes and negative/pad descriptors, plus disclosure counter styles and calc integers; @namespace registration with namespace-qualified selectors; math lengths accepted in flex-basis, gap, background-size and more; will-change, keyframe-selector and var() name validation fixes.
Measurement: eight CI-GUARD patrol rounds (20260824–20260830). All seven build-and-test platforms green, including verification of the week's R3804–R3807 fixes. The windows-aarch64 ring/clang clippy failure was recorded for the 21st consecutive round as pre-existing infrastructure. Two 7763-runner benchmarks metrics (transform_chain_100, webview_builder_create) ran over budget by 5% and 8% in two independent runs; attributed to platform drift (~20% across the board), pending a baseline re-capture decision.
Perf baselines and trends were auto-tightened and committed daily.
In progress
Next on js-dom: pa2 (apply generation tokens, wiring a host-apply completion callback into the shim) and the M6 mark fix for replaceWith on sel paths. Turning both engines' native bindings on by default (M5/M7) still needs a user decision; the M7 trigger condition was confirmed met and formally raised this week.
The rendering stream's open clusters: background-root propagation (investigation pinned the defect to which reference box background-position resolves against under body fallback), calc-margins in table cells (blocked on splitting the anon-cell wrapping by child semantics, which needs a LayoutBox flag), and two authorization-gated deep items (webkit-039 R109 box-tree lineage, table cell-height triple interlock).
CI carries the same two pre-existing items as last week: the windows-aarch64 ring/clang environment failure and the pending re-capture of drifted benchmark baselines for the 9V74/7763 runners. The physical-GPU memory retest still has no machine assigned.
Sources
Engineering guides
How ZeroWeb tests AI-autonomous software engineering
A browser gives AI development a long-running test case: architecture, code, debugging and verification must work together.
ZeroWeb Rust browser architecture: from request to pixels
Follow the page pipeline and the boundaries between the browser application, embeddable WebView, engine and rendering infrastructure.
Getting started with ZeroWebView and ZeroBrowser
Prepare a local checkout, launch the desktop browser and inspect the WebView API before embedding the experimental engine.
How to read ZeroWeb WPT and performance results
Understand self-source reftests, assertion-based suites, changing corpus sizes and p95 benchmarks before comparing results.