Repository navigation
test262 class tail — ~211 language/statements + expressions class fails #5345
Description
The largest language gap. language/statements/class = 211 fails (+ a parallel cluster in language/expressions/class).
Reasons (statements)
- 45×
Expected a ReferenceError(TDZ / binding-resolution in class bodies) - 37×
Expected a TypeError(missed validation negatives) - 19× async class-method test failures
- 12× (no output), 11× missed-negatives
Likely areas
Class static/instance field TDZ, private-field brand checks, super edge cases, async-generator methods, decorator-adjacent semantics. Deep semantics — diminishing returns per fix, but the single biggest language bucket.
Measure: scripts/test262_subset.py --dir language/statements/class language/expressions/class (node v26).
Activity
codegen: enforce derived-ctor this-TDZ on the standalone constructor-…
fix(runtime): hasOwnProperty recognizes class accessors/methods on th…
fix(runtime): class accessor deletion + static-accessor hasOwnProperty (
style: rustfmt the class-reflection fixes (#5345)
fix(class): super-in-static resolution + accessor .name reflection (#…
proggeramlug commentedon Jun 18, 2026
Status update (2026-06-18)
Progress on the class tail since this was filed:
super-in-static resolution + accessor.namereflection.this-TDZ on the standalone constructor-symbol path.No class-tail PR is currently in flight.
Remaining, unchanged:
Expected a ReferenceError 45— TDZ / binding-resolution in class bodiesExpected a TypeError 37— missed validation negativessuperedge cases, async-generator methods, decorator-adjacent semanticsDeep semantics with diminishing returns per fix, but the single biggest language bucket. Keeping open as the class tracker; authoritative numbers from
scripts/test262_subset.py --dir language/statements/class language/expressions/class.proggeramlug commentedon Jun 21, 2026
Re-measured the class tail (
scripts/test262_subset.py --dir language/statements/class language/expressions/class --jobs 7, test2624249661): parity 95.0% — 5099 pass, 56 diff, 199 runtime-fail, 16 compile-fail.Opened #5508 for one concrete cluster from the compile-fail bucket: duplicate static field names (
static f = …; static f = …;) emitted a redefined@perry_static_…__C__fLLVM global, so clang rejected the whole module. That sankelements/static-field-redeclarationand contributed tostatic-init-sequence. Fixed by deduping the defining-global emission while still running every initializer in order; added a parity gap test.Remaining buckets (unchanged, deep semantics):
Expected a TypeError … not thrown(29, missed validation negatives),value is not a function(12), TDZExpected a ReferenceError(5+), private brand-check negatives, async-method ordering.grammar-static-ctor-*-valid,class x extends xTDZ),name-binding/in-extends-expression*codegen.static-init-sequenceneeds interleaved field/static-block ordering (currently all fields then all blocks) — a separate, larger change, not in fix(codegen): #5345 — dedup duplicate static-field LLVM global emission #5508.fix(codegen): #5345 — dedup duplicate static-field LLVM global emissi…
proggeramlug commentedon Jun 23, 2026
Re-measured 2026-06-23 on an internal Linux sweep host (perry 0.5.1204): the persistent class tail is 260 (statements 161, expressions 99).⚠️ A separate regression (#5579) re-broke ~313 more class tests on current main () — that is tracked in #5579, not here. The true standing gap is refiled concretely as #5592 (accessor-name computed-
in, async-gen static methods, private brand checks). Closing in favor of #5592 + #5579.proggeramlug commentedon Jun 23, 2026
Correction to the note above: the regression signature in #5579 is the test262 message "m descriptor should be configurable" (class-method property descriptor).