Self-contained handoff doc. Generated 2026-07-17 from the latest committed run.

This repo (elide-dev conformance testsuite) runs upstream conformance suites against pinned Elide builds and publishes versioned reports. The wpt-wintertc suite is a sparse Web Platform Tests slice covering WinterTC / ECMA-429 JavaScript-facing APIs: encoding (40 files), url (28 files), and fetch (245 files). The manifest is manifests/wintertc-wpt-2025.toml; the WPT checkout is a sparse submodule under suites/wpt.

This suite has the single largest and most concentrated compat wins across all five suites in the repo.

Counts are WPT subtests (file :: subtest name). Latest reports:

• reports/1.4.1+20260716.b8b6531/3d3ea83ed640/wpt-wintertc/impact.md — failures clustered by root-cause signature, largest first
• .../wpt-wintertc/impact.json — same data, machine-readable (bySignature)
• .../wpt-wintertc/wpt-wintertc.md — full run report
• expectations/wpt-wintertc.ratchet.toml — the known-failure baseline (3,438 lines)

bun run setup                       # once: deps + submodules + sparse checkouts
# smoke (seconds):
bun run testsuite --elide nightly --suite wpt-wintertc --include 'url/urlsearchparams-constructor.any.js' --log
# a target area:
bun run testsuite --elide nightly --suite wpt-wintertc --include 'encoding/textdecoder-fatal-single-byte.any.js' --log
# by name, any file whose path contains 'encodeInto' (case-insensitive glob, see docs/running.md 'Targeted runs'):
bun run testsuite --elide nightly --log --filter 'wpt-wintertc:*encodeinto*'
# full suite (~12s of harness runtime inside Docker):
bun run testsuite --elide nightly --suite wpt-wintertc --threads 8
# after fixes land in an Elide nightly, re-baseline:
bun run testsuite --elide nightly --suite wpt-wintertc --ratchet

Green = zero regressions vs the expectations baseline; new passes show up in changes.md and are locked in by --ratchet.

Failures are extremely concentrated. Directory rollup:

Signature-level highlights (see impact.md for the full list):

• TextDecoder: encoding ISO-8859-10 / -14 not available on this JVM — 524
• TextDecoder: encoding macintosh not available on this JVM — 262
• TextDecoder: encoding ISO-8859-8-I not available on this JVM — 226
• decoding error (fatal mode): UNMAPPABLE[n] (should NOT throw) — 68
• throws RangeError where spec wants TypeError (fatal-flag tests) — 47
• sabConstructor is not a constructor (no SharedArrayBuffer) — 71
• URL: assert_equals: property expected "https:" but got "http:" etc. — 80; Invalid URL port — 24

Root cause for the encoding mass: Elide's TextDecoder delegates to JVM Charset. The JVM lacks several WHATWG-required encodings (ISO-8859-10, ISO-8859-14, ISO-8859-8-I, macintosh, x-mac-cyrillic, x-user-defined; label aliases like unicode11utf8) and diverges from the Encoding Standard on error handling (fatal mode, replacement-character emission, ISO-2022-JP state machine).

1. WHATWG single-byte encodings (Elide runtime) — unlocks ~1,400 subtests, complexity low-medium. Implement the Encoding Standard's single-byte decoders directly from its index tables (128 entries each; https://encoding.spec.whatwg.org/indexes.json) instead of JVM Charset. Include the full label→encoding table and correct fatal/replacement behavior. Verify with encoding/textdecoder-fatal-single-byte.any.js (1,092 failing), textdecoder-labels.any.js, single-byte-decoder.window.js.
2. URL setter stripping (Elide runtime) — unlocks ~130 subtests, complexity low-medium. URL component setters must strip tab/newline and handle U+0000 per the URL Standard; port setter must ignore invalid input rather than throw. Verify with url/url-setters-stripping.any.js.
3. Legacy multi-byte decoders: gbk/gb18030, ISO-2022-JP (Elide runtime) — unlocks ~450 subtests, complexity medium (stateful decoders, larger index tables). Natural follow-on to item 1 once the index-table infrastructure exists.
4. WPT server support in the harness (THIS repo, not Elide) — un-gates up to ~1,300 subtests (fetch/api, fetch/http-cache, …), complexity medium-high. The runner executes .any.js files without a wptserve instance or base URL, so anything fetching a relative resource dies in setup. Standing up wptserve (or a minimal substitute honoring WPT's {{host}}/{{ports}} substitution) inside the Docker run converts these from permanently-failing into a real fetch-conformance signal. Expect real fetch bugs to surface after; that's the point.
5. Park / skip candidates: fetch/fetch-later (136 — browser-only, needs window/iframe/BFCache), .window.js tests requiring document/ XMLHttpRequest (~90). Consider adding these to [skip] in expectations/wpt-wintertc.toml with a comment. Note this only cleans up the "vs expectations" rate; the overall pass rate keeps skipped tests in its denominator on purpose.

Items 1–3 alone should move the suite from 75.3% to roughly 88–90%.

• Elide fixes happen in the Elide runtime repo; this repo only pins nightlies and measures. Use --elide nightly to pick up a fresh build.
• Failure text like not available on this JVM comes from Elide's own error messages — grep the Elide codebase for it to find the TextDecoder impl fast.
• The suite runs in ~12s, so iterate with --include 'encoding/**' --log.
• After a fix lands: run full suite → check changes.md for new passes and no regressions → --ratchet → commit reports + expectations together.