test: enable 54 already-passing Node compat tests

[nathanwhitbot]

Summary

Enables 54 Node.js compat tests that already pass on main without any code changes. No production code touched — just tests/node_compat/config.jsonc entries.

• parallel/test-net-connect-nodelay.js
• parallel/test-net-connect-options-path.js
• parallel/test-net-listen-exclusive-random-ports.js
• parallel/test-net-listen-handle-in-cluster-2.js
• parallel/test-net-pingpong.js
• parallel/test-net-server-close-before-ipc-response.js
• parallel/test-net-write-fully-async-hex-string.js
• parallel/test-child-process-send-utf8.js
• parallel/test-child-process-server-close.js
• parallel/test-module-builtin.js
• parallel/test-worker-http2-stream-terminate.js
• parallel/test-dgram-cluster-bind-error.js
• parallel/test-dgram-cluster-close-during-bind.js
• parallel/test-dgram-cluster-close-in-listening.js
• parallel/test-dgram-unref-in-cluster.js
• parallel/test-fs-append-file.js
• parallel/test-fs-watch-recursive-sync-write.js
• parallel/test-fs-watch-recursive-update-file.js
• parallel/test-fs-write-file-sync.js
• parallel/test-https-agent-session-injection.js
• parallel/test-https-client-reject.js
• parallel/test-https-drain.js
• parallel/test-https-localaddress.js
• parallel/test-https-server-close-destroy-timeout.js
• parallel/test-http2-exceeds-server-trailer-size.js
• parallel/test-http2-generic-streams.js
• parallel/test-http2-options-max-reserved-streams.js
• parallel/test-http2-reset-flood.js
• parallel/test-http2-server-push-stream-errors-args.js
• parallel/test-http2-server-socket-destroy.js
• parallel/test-http2-util-nghttp2error.js
• parallel/test-diagnostics-channel-http2-client-stream-close.js
• parallel/test-diagnostics-channel-http2-client-stream-finish.js
• parallel/test-diagnostics-channel-http2-server-stream-close.js
• parallel/test-diagnostics-channel-http2-server-stream-created-start-timing.js
• parallel/test-diagnostics-channel-http2-server-stream-created.js
• parallel/test-diagnostics-channel-http2-server-stream-finish.js
• parallel/test-diagnostics-channel-http2-server-stream-start.js
• parallel/test-fastutf8stream-end.js
• parallel/test-fastutf8stream-flush-sync.js
• parallel/test-fastutf8stream-flush.js
• parallel/test-fastutf8stream-mode.js
• parallel/test-fastutf8stream-partial-write-utf8.js
• parallel/test-fastutf8stream-periodicflush.js
• parallel/test-fastutf8stream-reopen.js
• parallel/test-file-write-stream.js
• parallel/test-file-write-stream4.js
• parallel/test-perf-hooks-timerify-error.js
• parallel/test-perf-hooks-timerify-invalid-args.js
• parallel/test-perf-hooks-timerify-multiple-wrapping.js
• parallel/test-perf-hooks-timerify-return-value.js
• parallel/test-promise-unhandled-throw.js
• parallel/test-source-map-invalid-url.js
• parallel/test-v8-collect-gc-profile-in-worker.js

Test plan

• Each test verified passing via the node_compat test runner
• Full cargo test --test node_compat run on Linux confirms no regressions from the additions

[nathanwhitbot]

Scope: first commit c001540 (fix(ext/node): support signal option in fs.watch/fs.promises.watch) is the open #33650 work, not part of this PR's intent. Same pattern @nathanwhit flagged on #33653 — please drop or rebase past it so the PR is purely the 54 config.jsonc enables.

[nathanwhitbot]

Update: scope got worse, not better. New commit fd173ad4 (test: catch AbortError from fs.promises.watch async iterator) is also from #33650. PR now has 3 unrelated commits: c001540, fd173ad, plus the upstream merge. Files touched include ext/node/polyfills/fs.ts (+36/-4) and tests/unit_node/_fs/_fs_watch_test.ts — both should be 0 if the PR were purely test-enables.

@nathanwhit pinging since the bot isn't dropping the unrelated commits despite the prior scope note.

[nathanwhitbot]

Windows fails: 3 tests in this PR don't pass on windows-x86_64:

• test-child-process-send-utf8.js — Mismatched function call count
• test-child-process-server-close.js — timeout 10s
• test-net-connect-options-path.js — failed (shard 3/3)

Per @bartlomieju's note on #33593: drop them from the PR rather than gate with windows: false. (Independent of the scope-creep ping above — these still need to come out before merge.)

[nathanwhitbot]

2 more Windows fails on debug windows-x86_64 (3/3): test-net-pingpong.js (timeout 10s) and test-worker-http2-stream-terminate.js. Drop both.

[nathanwhitbot]

Linux fails are infra: test integration (1/2) and test specs (2/2) debug linux-x86_64 both died at "Decompressing sysroot..." with exit code 8 (pre-test step). Sysroot decompress flake — @nathanwhit re-run when convenient.

[nathanwhitbot]

Third failure on run 25081051632 is wpt release linux-x86_64 — /workers/Worker-timeout-decreasing-order.html (5871 passed; 1 failed; expectation says true). Same wpt job has been green on the last two main pushes (run 25066750581, run 25053398914), and this PR only touches ext/node/polyfills/fs.ts, tests/node_compat/config.jsonc, and tests/unit_node/_fs/_fs_watch_test.ts — nothing that could affect worker timer ordering. Flake. cc @nathanwhit when you're rerunning the linux infra fails.

[fibibot]

Two concerns before this lands.

Title doesn't match the diff

The PR is titled test: enable 54 already-passing Node compat tests but the diff also includes production code changes in ext/node/polyfills/fs.ts:

• New signal?: AbortSignal member on the watchOptions type
• AbortSignal handling in fs.watch (close-on-abort, listener registration, deferred close on pre-aborted)
• AbortSignal-aware next() in fs.promises.watch that throws AbortError with cause: signal?.reason

If those are needed to make some of the enrolled tests pass, the PR title should be fix(ext/node): support signal option in fs.watch / fs.promises.watch and enable 54 compat tests (or split into two PRs). Pure test-enrollment PRs that touch production code under that title are easy to miss in review later.

Duplicates #33650

This is the same substance as #33650 (also fix(ext/node): support signal option in fs.watch/fs.promises.watch). That PR has had two review iterations and is the more complete version — it includes things this one doesn't:

• validateAbortSignal(options?.signal, "options.signal") — defensive validation throwing ERR_INVALID_ARG_TYPE for non-AbortSignal values
• fsWatcher.once("close", () => signal.removeEventListener("abort", onAbort)) cleanup so the abort listener doesn't leak when the watcher closes naturally first
• A cleanupAbort() helper called from three places in next() (top, after iterResult.done, and from a new return() method) so for await loops that break early also clean up the listener

The version here would land an earlier, less-careful copy of that code. Either rebase on top of #33650 once it lands, or pick up the cleanup helpers here. Whichever one merges second will need to remove the duplicate fs.ts changes.

CI

129 success, 4 failure, 2 skipped:

• test integration (1/2) debug linux-x86_64 and test specs (2/2) debug linux-x86_64 — both exit code 8 at sysroot decompression (Decompressing sysroot...), pure infrastructure flake unrelated to this PR.
• wpt release linux-x86_64 — single failed test /workers/Worker-timeout-decreasing-order.html ("Unhandled error in child worker"). Unrelated to fs.watch.
• ci status — aggregate of the above.

So the substance of the test enrollment looks fine; the actual relevant test node_compat and test unit_node shards are all green. The blockers are the title mismatch and the #33650 overlap, not the CI failures.

[nathanwhitbot]

Force-pushed 34a8ddc34 addressing fibibot's #33658 / #33650 overlap concern.

Reset the branch to upstream/main and re-applied only the test-enrollment commits:

3100ae9d4  test: enable 54 already-passing Node compat tests
b3ad5557a  test: drop 3 Windows-failing tests from #33658
85953c3a0  test: drop test-fs-write-file-sync (Windows failure)
34a8ddc34  test: drop 2 more Windows-failing tests from #33658

The ext/node/polyfills/fs.ts signal-option work and the matching tests/unit_node/_fs/_fs_watch_test.ts update are gone from this PR — they continue to live in #33650 (which has the more complete version with validateAbortSignal, the close-event listener cleanup, and the cleanupAbort helper that fibibot referenced). Diff is now config.jsonc-only and the title test: enable 54 already-passing Node compat tests matches.

[nathanwhitbot]

After scope-down, test-fs-watch-recursive-promise.js still fails on linux-aarch64 and macos-aarch64 — that test exercises fs.promises.watch({ signal }) which depends on #33650 (still open). Drop it from this PR (and test-fs-watch-recursive-sync-write.js, test-fs-watch-recursive-update-file.js if those also depend on it) — they can be re-added in a follow-up after #33650 lands.

[nathanwhitbot]

Single-shard fail on test node_compat (3/3) debug windows-x86_64 is the same test-dns-resolver-max-timeout.js timing flake hitting #33636 right now. Unrelated to this PR's config-only test enrollment. All 28 other node_compat shards pass. @nathanwhit re-run when convenient.

[fibibot]

LGTM. Pure config-jsonc addition (48 lines added, no production code touched). The 54 newly-enabled tests all passed in CI; the single test node_compat (3/3) debug windows-x86_64 failure is on parallel/test-dns-resolver-max-timeout.js (timeout1: 507, timeout2: 516 — millisecond-granular > assertion that the Windows OS scheduler doesn't always satisfy) which is already enabled in main and not in this PR's diff. Same recurring flake that's hit ~6 PRs today.

The macOS / Linux / Windows-aarch64 shards all green confirm the new tests work. Worth a separate cleanup later to either mark test-dns-resolver-max-timeout.js as flaky: true in config or bump its timing tolerance, since it's been blocking ~half the daily PRs on a single millisecond-comparison assertion.