conformance: test_gap_6301_event_target_subclass regressed on main (untriaged, inside required conformance-smoke-complete) — fixed 07-13, red again by 07-31

[proggeramlug]

test_gap_6301_event_target_subclass is failing on main (pass -> parity_fail per the harness log) and is not in test-parity/known_failures.json — an untriaged regression of a test that was fixed on 2026-07-13 by 81ab8e5d6 (issue #6301, native-base EventTarget subclassing).

Why it matters more than a normal red

It fails inside conformance-smoke (4), which feeds conformance-smoke-complete — a required status check. An untriaged regression in a required context blocks every PR (or trains people to admin-bypass, which is worse).

Independent confirmations

• The fix(gc): re-derive a string-literal operand below the collection point (#7114) #7116 session's gap run hit it at 166/438 and verified the failure is byte-identical on pristine base (ff85fd483) — i.e. it pre-dates all of the 2026-07-31/08-01 merges.
• The fix(runtime): make WebAssembly.Module instanceof recognize constructed modules #7124/test(compile): lock in auto-opt freshness on runtime/stdlib source edits #7155 triage reproduced it on both PRs' conformance-smoke (4) runs (disjoint diffs — inherited by construction).

What's needed

1. Bisect between 81ab8e5d6 (fix landed, green) and ff85fd483 (confirmed red) — the window is ~2.5 weeks of merges.
2. Native-base subclassing is a documented weak family (5 defects, keyed on literal extends names) — the regression likely rode in on one of the globalThis/proto/builtin-install changes in that window.
3. Until fixed: either triage into known_failures.json with provenance (issue # + date, per the test-parity: audit known_failures.json — every entry needs an issue # and date #797 skip-list policy) to unblock the required context honestly, or fix it. Leaving it untriaged-and-red in a required context is the worst of both.

[proggeramlug]

Already fixed — no further code change needed

This is the same regression as #7518 (same failing gap test, same class Bus extends EventTarget shape). It was fixed and closed by #7520 on 2026-08-06 (6a051ec18), five days after this issue was opened. #7520 closed #7518 but never closed this one.

Root cause (from #7520 / #7518)

c6ed8175d (#6853, 2026-07-30) added "EventTarget" to builtin_constructor_spec_length so EventTarget.length === 0 like Node. That flipped an incidental gate in try_dispatch_value_called_proto_method: constructors that share the globalThis no-op thunk were only excluded because they had no recorded .length. With a length recorded, js_fetch_or_value_super(<EventTarget>, this, …) re-dispatched as bus.EventTarget() and threw TypeError: EventTarget is not a function.

Fix on main

#7520 makes the exclusion explicit and table-driven via is_global_this_builtin_constructor_name / GLOBAL_THIS_BUILTIN_CONSTRUCTORS, and adds a per-PR unit test (global_builtin_constructor_values_are_not_redispatched_by_name) so a future .length addition cannot re-open the hole.

Re-verified on current origin/main (6fb58cdea)

• test_gap_6301_event_target_subclass is byte-identical to node --experimental-strip-types (Node 26.5.1), exit 0 on both sides.
• Unit test global_builtin_constructor_values_are_not_redispatched_by_name passes.
• Not present in test-parity/known_failures.json.

Closing as completed. No PR required.