Test262: 62 deterministic compiled-mode regressions vs stale baseline (14 compiled-only codegen bugs)

[nickna]

Summary

A clean run of the compiled Test262 baseline (Test262CompiledTests.CompiledBaseline) against the committed baselines/compiled.txt shows 62 regressions and 108 new passes (net +46; suite still improved overall). All 62 regressions are deterministic — the 20 sampled in the run log reproduced 20/20, and the full set was reconstructed and independently re-verified (882 candidate paths re-run → exactly 62, zero flake).

The root context: baselines/compiled.txt is 187 commits stale — last regenerated at 0494a763 (#693, 2026-06-15), while baselines/interpreted.txt is current as of b9d3aa3f (#802, 2026-06-17). So this drift is the accumulated compiled-mode delta across 187 commits, never recorded.

Cross-referencing each regressed path against the (current) interpreted baseline splits the 62 cleanly:

Class	Count	Meaning
Compiled-only (interp still Pass)	14	Genuine compiled/codegen regressions — interpreter gets these right, compiled does not. Priority.
Shared (interp baseline already Fail/RuntimeError)	48	Behavior changes already accepted on the interpreted side at #802; the stale compiled baseline just never recorded them.

Buckets: 37 RuntimeError + 25 Fail.

Reproduce

dotnet build SharpTS.Test262.Worker/SharpTS.Test262.Worker.csproj -c Release   # worker is NOT rebuilt by dotnet test
dotnet test SharpTS.Test262/SharpTS.Test262.csproj -c Release \
  --filter "FullyQualifiedName~CompiledBaseline"        # ~17 min; run backgrounded
# Regenerate (only after the 14 compiled-only are triaged — don't bury them):
#   SHARPTS_TEST262_UPDATE_BASELINE=1 dotnet test ... --filter "FullyQualifiedName~CompiledBaseline"

The 14 compiled-only regressions (priority — codegen bugs)

These pass in the interpreter but fail compiled. Three themes:

• Sloppy-mode this not bound to the global object — Array/prototype/filter/15.4.4.20-5-1, some/15.4.4.17-5-1 (callback reads this._x where this should be global), language/expressions/call/11.2.3-3_8 (this.bar(foo()) at top level). Compiled throws instead of resolving this → globalThis.

• Object.defineProperty/defineProperties/create reading descriptor fields off an attributes object that has an accessor without a getter — 15.2.3.6-3-215..217, 15.2.3.7-5-b-201..203, 15.2.3.5-4-241..243. e.g. Object.defineProperty(attrs, "get", {set(){}}) then defineProperty(obj, "p", attrs); reading attrs.get must yield undefined, compiled throws (RuntimeError).

• eval with exotic whitespace between new and the operand — language/expressions/new/S11.2.2_A1.1, A1.2 evaluate new<WS>Number for WS ∈ {U+0009 tab, U+000B VT, U+000C FF, U+00A0 NBSP, U+2028, U+2029, …} via eval(...) and expect 0.

• test/built-ins/Array/prototype/filter/15.4.4.20-5-1.js Pass -> RuntimeError (interp: Pass)

• test/built-ins/Array/prototype/some/15.4.4.17-5-1.js Pass -> RuntimeError (interp: Pass)

• test/built-ins/Object/create/15.2.3.5-4-241.js Pass -> RuntimeError (interp: Pass)

• test/built-ins/Object/create/15.2.3.5-4-242.js Pass -> RuntimeError (interp: Pass)

• test/built-ins/Object/create/15.2.3.5-4-243.js Pass -> RuntimeError (interp: Pass)

• test/built-ins/Object/defineProperties/15.2.3.7-5-b-201.js Pass -> RuntimeError (interp: Pass)

• test/built-ins/Object/defineProperties/15.2.3.7-5-b-202.js Pass -> RuntimeError (interp: Pass)

• test/built-ins/Object/defineProperties/15.2.3.7-5-b-203.js Pass -> RuntimeError (interp: Pass)

• test/built-ins/Object/defineProperty/15.2.3.6-3-215.js Pass -> RuntimeError (interp: Pass)

• test/built-ins/Object/defineProperty/15.2.3.6-3-216.js Pass -> RuntimeError (interp: Pass)

• test/built-ins/Object/defineProperty/15.2.3.6-3-217.js Pass -> RuntimeError (interp: Pass)

• test/language/expressions/call/11.2.3-3_8.js Pass -> Fail (interp: Pass)

• test/language/expressions/new/S11.2.2_A1.1.js Pass -> Fail (interp: Pass)

• test/language/expressions/new/S11.2.2_A1.2.js Pass -> Fail (interp: Pass)

The 48 shared regressions (interp baseline already records these failing)

Lower priority — these mirror behavior already accepted in interpreted.txt at #802; regenerating the compiled baseline will fold them in. They cluster in the same Object.* descriptor area, sloppy-this Promise reaction handlers (*-nonstrict/*-sloppy), sparse-array inherited-accessor iteration (Array/prototype/{reduce,reduceRight,...}), and for-in/propertyIsEnumerable DontEnum attribute tests (Number.MAX_VALUE/MIN_VALUE, Boolean/Error/String prototype).

Full list of all 48 shared regressions

• test/built-ins/Array/prototype/every/15.4.4.16-7-c-i-22.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/filter/15.4.4.20-5-30.js Pass -> RuntimeError (interp: Fail)
• test/built-ins/Array/prototype/filter/15.4.4.20-9-c-i-22.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/forEach/15.4.4.18-5-25.js Pass -> RuntimeError (interp: Fail)
• test/built-ins/Array/prototype/forEach/15.4.4.18-7-c-i-22.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/indexOf/15.4.4.14-9-b-i-21.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/lastIndexOf/15.4.4.15-8-b-i-20.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/lastIndexOf/15.4.4.15-8-b-i-21.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/map/15.4.4.19-8-c-i-22.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/reduce/15.4.4.21-8-b-iii-1-18.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/reduce/15.4.4.21-8-b-iii-1-20.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/reduce/15.4.4.21-8-b-iii-1-22.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/reduce/15.4.4.21-8-c-4.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Array/prototype/reduce/15.4.4.21-9-c-i-18.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/reduce/15.4.4.21-9-c-i-20.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/reduce/15.4.4.21-9-c-i-22.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/reduceRight/15.4.4.22-8-b-iii-1-22.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/reduceRight/15.4.4.22-8-c-4.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Array/prototype/reduceRight/15.4.4.22-9-c-i-18.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/reduceRight/15.4.4.22-9-c-i-20.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/reduceRight/15.4.4.22-9-c-i-22.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Array/prototype/some/15.4.4.17-5-25.js Pass -> RuntimeError (interp: Fail)
• test/built-ins/Array/prototype/some/15.4.4.17-7-c-i-22.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Boolean/prototype/S15.6.3.1_A4.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Error/prototype/S15.11.3.1_A2_T1.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Error/the-initial-value-of-errorprototypemessage-is-the-empty-string.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Number/MAX_VALUE/S15.7.3.2_A4.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Number/MIN_VALUE/S15.7.3.3_A4.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Object/defineProperties/15.2.3.7-6-a-102.js Pass -> RuntimeError (interp: Fail)
• test/built-ins/Object/defineProperties/15.2.3.7-6-a-92.js Pass -> RuntimeError (interp: Fail)
• test/built-ins/Object/defineProperties/15.2.3.7-6-a-93.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Object/defineProperty/15.2.3.6-4-108.js Pass -> RuntimeError (interp: Fail)
• test/built-ins/Object/defineProperty/15.2.3.6-4-470.js Pass -> RuntimeError (interp: Fail)
• test/built-ins/Object/preventExtensions/15.2.3.10-3-10.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Object/preventExtensions/15.2.3.10-3-18.js Pass -> RuntimeError (interp: Fail)
• test/built-ins/Object/preventExtensions/15.2.3.10-3-19.js Pass -> RuntimeError (interp: Fail)
• test/built-ins/Object/preventExtensions/15.2.3.10-3-20.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Object/preventExtensions/15.2.3.10-3-8.js Pass -> RuntimeError (interp: Fail)
• test/built-ins/Object/preventExtensions/15.2.3.10-3-9.js Pass -> RuntimeError (interp: Fail)
• test/built-ins/Object/prototype/S15.2.3.1_A2.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Promise/executor-call-context-sloppy.js Pass -> Fail (interp: RuntimeError)
• test/built-ins/Promise/prototype/finally/rejected-observable-then-calls-argument.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Promise/prototype/then/rxn-handler-fulfilled-invoke-nonstrict.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/Promise/prototype/then/rxn-handler-rejected-invoke-nonstrict.js Pass -> RuntimeError (interp: RuntimeError)
• test/built-ins/String/prototype/S15.5.3.1_A2.js Pass -> RuntimeError (interp: RuntimeError)
• test/language/expressions/new/spread-err-mult-err-expr-throws.js Pass -> Fail (interp: Fail)
• test/language/expressions/new/spread-err-sngl-err-expr-throws.js Pass -> Fail (interp: Fail)
• test/language/expressions/property-accessors/S11.2.1_A4_T1.js Pass -> Fail (interp: RuntimeError)

Recommended next steps

1. Triage/fix the 14 compiled-only regressions — start with the sloppy-mode this → globalThis binding in compiled call/codegen, which spans the Array, call, and (in the shared set) Promise-reaction clusters.
2. Spot-check the Object.defineProperty descriptor-extraction path (accessor-without-getter on the attributes object) — contiguous test numbers suggest one codegen path.
3. Then regenerate baselines/compiled.txt (SHARPTS_TEST262_UPDATE_BASELINE=1) so the 108 genuine new passes (incl. parser gains: ParseError 88→10) are recorded and the baseline stops being 187 commits stale.

Provenance

• Branch: main @ 33f266f9; full compiled run: 11,384 tests in 1007.5 s → Pass=7942 Fail=1843 ParseError=10 RuntimeError=690 Timeout=2 HarnessError=18 Skipped=879.
• Subset config: config/subset.json (timeout 15 s). Env: win32, .NET 10, 12 logical cores.
• Regressions independently re-verified deterministic (882-path targeted re-run, recycle-safe).

[nickna]

Status update — 14/14 compiled-only regressions fixed + strict-mode runner gap closed

Branch wrk/882-compiled-sloppy-this-desc-getter (3 commits). All 14 priority compiled-only regressions now pass, across three root causes — note the third was misdiagnosed in the original report.

Fixes

1. Accessor descriptor with get: undefined (9 tests) — Object.{defineProperty,defineProperties,create} 15.2.3.6-3-215..217, 15.2.3.7-5-b-201..203, 15.2.3.5-4-241..243.
Descriptor extraction stores the JS undefined sentinel in the Getter slot to keep the descriptor classified as an accessor, but the emitted TryGetGetter only checked Getter != null — so it returned the sentinel as a real getter and GetProperty Invoke()d undefined → "undefined is not a function". Fix: treat a sentinel getter as "no getter" so reads fall through to the descriptor's undefined value. (TryGetSetter deliberately untouched — set: undefined must keep its no-op path.)

2. Sloppy-mode this not bound to globalThis — Array/prototype/filter/15.4.4.20-5-1, some/15.4.4.17-5-1, call/11.2.3-3_8, plus the shared Promise nonstrict-reaction regressions and ~10 more (forEach, map, find* non-strict, Function.call/apply).
LoadThis coerced CLR-null this→globalThis but not the undefined sentinel that array HOFs stage for a no-thisArg callback. Fix: a strict-aware EmitCoerceSloppyThisToGlobal() — null→globalThis always; undefined→globalThis only when the callee is non-strict (strict keeps this === undefined, matching the interpreter).

3. eval()-boundary boxed-primitive coercion — language/expressions/new/S11.2.2_A1.1, A1.2.
These are not a whitespace bug — the exotic-whitespace lexing between new and the operand works fine. The actual failure is that eval("new Number") != 0 returned true: eval runs in the interpreter and returns a SharpTSObject boxed wrapper, but the compiled runtime keys all boxed-primitive handling (==, ToNumber, valueOf) on Isinst $Object, which the interpreter type misses — so == didn't coerce and valueOf() threw (typeof correctly stayed "object"). Fixed at the boundary with $Runtime.NormalizeForeignBoxedPrimitive, which re-wraps a foreign Number/Boolean/String wrapper into a native $Object on the eval result.

Strict-mode runner gap (follow-up, closed)

The harness assembler ran every test sloppy — onlyStrict tests had no strict directive — so strict-only behavior was untestable. Now Assemble() prepends a program-level "use strict" for onlyStrict tests (both runners honor a top-level directive prologue). This also resolves the temporary wash introduced by fix #2: Array find*/predicate-call-this-strict now pass for the right reason (real strict this === undefined) while their noStrict counterparts still pass sloppy.

Verification

• Worker run: 14/14 priority compiled-only tests pass; the 4 find* strict + 1 find* non-strict pairs all pass.
• Before/after diffs (worker): sloppy-this slice (~4,000 this-sensitive tests) +15 net; eval slice (140) +2 / 0 regr; onlyStrict subset compiled +6, interpreted +2, 0 regr; broad 279-test cross-suite onlyStrict sample 0 regr.
• Unit suite: 13,994 pass. The only real failures are the two pre-existing stale-baseline tests (Interpreted/CompiledBaseline) — proven byte-identical with/without these changes; transient compiled-mode failures are harness concurrency flakiness (all pass on isolated re-run).

Remaining

• Regenerate baselines/compiled.txt (SHARPTS_TEST262_UPDATE_BASELINE=1) to fold in the 108 genuine new passes + these fixes and stop the 187-commit drift. The find* strict gains should be re-baselined as passes.
• Out of scope here: running the unflagged sloppy+strict variant pair (currently each test runs once, sloppy) — a separate, larger runner change.

[nickna]

Status update — strict-mode conformance: runner + codegen; variant-pair gap scoped

Branch wrk/882-compiled-sloppy-this-desc-getter now has 4 commits. Continuing from the strict-mode runner fix, I took on the unflagged sloppy+strict variant-pair gap — but it turns out to be a large, maintainer-gated infrastructure change, so I scoped it and instead fixed the concrete strict-mode codegen bug it surfaces.

Variant-pair gap — investigated, not implemented (needs a deliberate call)

Test262's canonical harness runs each unflagged test as both a sloppy and a strict variant; SharpTS runs each test once (sloppy). Implementing variant pairs means two results per path, which:

• changes the path → bucket baseline model (every unflagged test gains a strict entry),
• roughly doubles run time, and
• surfaces a large strict-failure surface: I measured a 495-test unflagged sample run force-strict — ~9% (49/495) pass sloppy but fail strict, 0 strict-only gains. Extrapolated, that's hundreds of new strict-Fail baseline entries.

That's a baseline-semantics + coverage change that should be a planned rollout (consistent with this issue's "regenerate after triage" stance), not an autonomous drop. Flagging it here for a decision rather than implementing it.

Strict-mode codegen fix (the substance behind the gap) — 332cbad8

The measured strict failures are real compiled-mode bugs. The largest clean cluster: dynamic property writes on built-in objects were silently dropped in strict mode — arr.foo = fn; arr.foo() saw foo === undefined under "use strict", and var d = new Date(0); d.enumerable = true (a defineProperty attributes object) lost the field. (Array slice/splice S15.4.4.*, Object.defineProperty 15.2.3.6-3-39, etc.)

Root cause: strict assignment routes through SetPropertyStrict / SetFieldsPropertyStrict, which lacked the array and built-in-exotic store branches their non-strict twins have. Fixed both:

• SetPropertyStrict gained a $Array/List<object?> branch that enforces strict frozen / own-non-writable-data TypeErrors (OrdinarySet → strict PutValue throws), then delegates the store to non-strict SetProperty.
• SetFieldsPropertyStrict gained the Date/RegExp/Promise/Error PDS-store arm mirrored from SetFieldsProperty.

Both paths are strict-only; sloppy execution is structurally unchanged.

Verification

• Force-strict 495-test unflagged sample: sloppy-Pass → strict-Fail 49 → 34; 0 new failures; frozen-array strict writes now throw TypeError.
• Committed onlyStrict subset: 0 change (no regression); the priority Test262: 62 deterministic compiled-mode regressions vs stale baseline (14 compiled-only codegen bugs) #882 tests + find* strict/non-strict pairs remain green (19/19).
• Unit suite: only the 2 pre-existing stale-baseline failures (Fs_LinkSync is harness flakiness — passes on isolated re-run).

Remaining

• Decide on the variant-pair rollout (above).
• Other strict-variant clusters still open: boxed-wrapper $Object strict set (new Number(-9).configurable = true), strict delete, array-HOF strict callbacks, Object.defineProperties strict.
• Baseline regen (SHARPTS_TEST262_UPDATE_BASELINE=1) still pending to fold in all gains.

[nickna]

Status update — strict-mode cluster: delete of Object.defineProperty properties (378e7eef)

Continuing to pick off the strict-variant codegen clusters one at a time.

Bug: In compiled strict mode, delete obj.p for a property installed via Object.defineProperty left it in place. DeletePropertyStrict's dictionary handler only called Dictionary.Remove and never consulted the PropertyDescriptorStore (where defineProperty-installed properties live), and it ignored configurability entirely. The non-strict DeleteProperty already did this correctly — only the strict path diverged.

Surfaced by running unflagged tests strict: Object/create 15.2.3.5-4-116/146/..., Object/defineProperty 15.2.3.6-3-80/90/110/... (each does Object.defineProperty(o, p, {configurable:true}); delete o.p; assert(!hasOwn)).

Fix: mirror the non-strict dict handler — read the PDS descriptor, then:

• non-configurable → strict throws TypeError (ECMA-262 §13.5.1.2), sloppy returns false;
• configurable → PDSDeleteProperty + Dictionary.Remove, return true.

Strict-only path; sloppy delete is unchanged.

Verification

• delete of a configurable defineProperty property now removes it under strict; non-configurable strict delete throws TypeError.
• Force-strict 495-test unflagged sample: sloppy-Pass → strict-Fail 34 → 27, 0 new failures.
• Committed onlyStrict subset: 0 change; priority Test262: 62 deterministic compiled-mode regressions vs stale baseline (14 compiled-only codegen bugs) #882 + find* pairs green (19/19).
• Unit suite: only the 2 pre-existing stale-baseline failures (4 others were harness concurrency flakiness — all pass on isolated re-run).

Progress on the strict-variant surface

Sample sloppy-Pass→strict-Fail: 49 → 40 → 34 → 27 across the last three commits (built-in dynamic writes, Date/RegExp/Promise/Error writes, delete of PDS props).

Remaining strict-variant clusters (~27 in the sample)

• Object.defineProperties 15.2.3.7-* (~10) and Object.defineProperty 15.2.3.6-3/4-* (~7) — descriptor-attribute reads off various object types;
• Object.create 15.2.3.5-4-177/268/282/298;
• array-HOF strict callbacks (every/filter/indexOf/reduce);
• Array.from iterator-error, getOwnPropertySymbols.

Still pending (unchanged): the variant-pair architecture decision and the baseline regen.

[nickna]

Status update — strict-mode cluster: non-writable / accessor property writes (7fe8a237)

Next strict-variant cluster picked off — and the biggest single drop so far.

Bug: In compiled strict mode, both obj.p = v and obj[k] = v ignored an existing PropertyDescriptorStore descriptor: a writable:false data property was silently overwritten (no TypeError), and an accessor setter was bypassed and clobbered with a data value. Under "use strict", propertyHelper.js's isWritable probe (which writes via obj[name] = …) then saw the write succeed and reported writable:true, so verifyProperty failed across the Object.defineProperty / defineProperties / create suites (e.g. 15.2.3.6-3-181, 15.2.3.7-5-b-*, 15.2.3.5-4-*). The non-strict SetProperty already honored these — only the strict paths diverged.

Fix (two strict-only paths):

• SetPropertyStrict's dict doSet arm now mirrors non-strict SetProperty: invoke a PDS accessor setter; otherwise a PDSIsWritable guard — non-writable (a writable:false data property or a getter-only accessor) throws TypeError in strict / silently returns in sloppy (ECMA-262 §6.2.5.6 PutValue); otherwise store.
• SetIndexStrict's dictionary arm routes obj[key] = v through SetPropertyStrict(obj, ToString(key), value, strictMode) instead of a raw dict.set_Item, so indexed writes get the same setter / writable / frozen / sealed handling as named writes (the probe uses bracket access).

Strict-only; sloppy writes unchanged.

Verification

• Non-writable write now throws TypeError and leaves the value unchanged (named and indexed); accessor setters still fire under strict; normal writes unaffected.
• Force-strict 495-test unflagged sample: sloppy-Pass → strict-Fail 27 → 8.
• onlyStrict subset: +1 (Object/freeze/frozen-object-contains-symbol-properties-strict Fail→Pass), 0 regressions; priority Test262: 62 deterministic compiled-mode regressions vs stale baseline (14 compiled-only codegen bugs) #882 + find* pairs green (19/19).
• Unit suite: only the 2 pre-existing stale baselines (3 others were harness concurrency flakiness — all pass on isolated re-run).

Cumulative strict-variant surface (sample)

49 → 40 → 34 → 27 → 8 across the strict-mode codegen fixes (built-in dynamic writes; Date/RegExp/Promise/Error writes; delete of PDS props; non-writable/accessor writes).

Remaining (smaller, separate clusters)

• array-HOF strict callbacks (every/filter/indexOf/reduce 15.4.4.*-2-*),
• Array.from iterator-value error, Object.create 15.2.3.5-4-177, Object.defineProperty 15.2.3.6-3-230, getOwnPropertySymbols.

Still pending (unchanged): the variant-pair architecture decision and the baseline regen.

[nickna]

Status update — strict-mode cluster: indexed writes on $Object receivers (d67261bf)

Array-HOF strict-callback cluster picked off.

Bug: In compiled strict mode, obj[k] = v on a $Object / class-instance / $TSFunction receiver was silently dropped — SetIndexStrict dispatched $Array/List/Dictionary, then fell through to a bare ret for everything else. So child[0] = 12 on a new Con() instance stored nothing, and Array.prototype.{reduce,every,filter,indexOf}.call(child, …) over such an arraylike saw a length but no elements and never iterated (Test262 Array/prototype/*/15.4.4.*-2-*). The non-strict SetIndex already routed these receivers to SetProperty.

Fix (two strict-only paths):

• SetIndexStrict: route the non-array/list/dict fallback to SetPropertyStrict(obj, ToString(index), value, strictMode), mirroring non-strict SetIndex → SetProperty.
• That routing exposed a latent gap: TSObjectSetPropertyStrict honored frozen and sealed but not preventExtensions, so a write of a new property to a non-extensible $Object (e.g. new Boolean(true)) was added instead of throwing. Added the non-extensible guard (strict → TypeError, sloppy → no-op), mirroring TSObjectSetProperty (Object/preventExtensions 15.2.3.10-3-6).

Strict-only; sloppy writes unchanged.

Verification

• Indexed writes on a $Object now persist and iterate (forEach/reduce.call over a new Con() arraylike); non-extensible $Object writes throw TypeError in strict and aren't added.
• Force-strict 495-test unflagged sample: sloppy-Pass → strict-Fail 8 → 4 (the 4 array-HOF 15.4.4.*-2-* fixed), 0 new regressions (the preventExtensions case I briefly regressed mid-cluster is resolved by the second part of this fix).
• onlyStrict subset: 0 change; priority Test262: 62 deterministic compiled-mode regressions vs stale baseline (14 compiled-only codegen bugs) #882 + find* green (19/19).
• Unit suite: only the 2 pre-existing stale baselines (2 others were harness flakiness — pass on isolated re-run).

Cumulative strict-variant surface (sample)

49 → 40 → 34 → 27 → 8 → 4 across the strict-mode codegen fixes.

Remaining (4, distinct one-offs)

Array.from iterator-value error, Object.create 15.2.3.5-4-177, Object.defineProperty 15.2.3.6-3-230, getOwnPropertySymbols. Still pending (unchanged): variant-pair architecture decision + baseline regen.

[nickna]

Status update — last 4 strict one-offs cleared; sampled unflagged-strict surface now 0 (87e2550d)

The 4 remaining one-offs turned out to be two root causes; fixing them brings the 495-test force-strict sample to zero sloppy-Pass→strict-Fail.

1. Symbol-keyed indexed writes. SetIndexStrict dispatched only by receiver type ($Array/List/Dictionary) with no Symbol-key branch, so obj[sym] = v and obj[Symbol.iterator] = fn under "use strict" ToString'd the symbol into a bogus string key and were lost from the symbol table (Object/getOwnPropertySymbols object-contains-symbol-property-without-description, Array/from iter-get-iter-val-err). Fix: a Symbol-key branch that delegates the store to the non-strict SetIndex symbol handler, but first enforces the strict invariants its sloppy handler skips — frozen → TypeError always; sealed/non-extensible → TypeError when the key is new; otherwise store (Object/freeze frozen-object-contains-symbol-properties-strict).

2. globalThis-sentinel writes. SetPropertyStrict lacked the globalThis-sentinel branch that SetProperty has, so a strict top-level this.foo = v (compiled this resolves to the globalThis sentinel) fell through to SetFieldsPropertyStrict and was dropped — breaking tests that set this.value / this.get and reuse this as a descriptor (Object/create 15.2.3.5-4-177, defineProperty 15.2.3.6-3-230). Fix: route the sentinel to GlobalThisSetProperty.

Strict-only paths; sloppy writes unchanged.

Verification

• All 4 one-offs pass; extensible symbol writes store, frozen symbol writes throw TypeError; top-level this.x = v persists.
• Force-strict 495-test unflagged sample: sloppy-Pass → strict-Fail 4 → 0.
• onlyStrict subset: +1 (Array/from iter-map-fn-this-strict), 0 regressions (the frozen-symbol case I briefly regressed mid-fix is resolved by the strict guard); priority Test262: 62 deterministic compiled-mode regressions vs stale baseline (14 compiled-only codegen bugs) #882 + find* green (19/19).
• Unit suite: only the 2 pre-existing stale baselines (1 other was harness flakiness — passes on isolated re-run).

Cumulative unflagged-strict surface (495-test sample)

49 → 40 → 34 → 27 → 8 → 4 → 0. The sampled systematic strict-mode codegen clusters are resolved.

Where this leaves #882

• 8 commits on wrk/882-compiled-sloppy-this-desc-getter: the original 14 compiled-only regressions + the strict-mode runner gap + a full sweep of strict-mode write/delete codegen (built-in dynamic writes, Date/RegExp/Promise/Error, delete, non-writable/accessor, $Object-index/preventExtensions, symbol-key, globalThis-sentinel).
• Still pending (the two standing decisions): the variant-pair architecture rollout and the baseline regeneration to bank all these gains. The sample is a slice (every 4th unflagged built-ins test); a full regen would confirm suite-wide.

[nickna]

Opened #884 with all 10 commits (the 14 compiled-only fixes + the strict-mode write/delete codegen sweep + the function-expr "use strict" detection fix + the regenerated standalone baselines).

Baseline regen delta: compiled net +98 (135 new passes, 37 regressions — all in this issue's verified-deterministic "48 shared" set, none new from the branch); interpreted net +408 (437 new passes, 22 pre-existing-drift regressions). Regression sets re-verified solo (the parallel regen flakes heavy tests; flaky Timeouts/Fails were corrected to their deterministic value). See the PR for the full breakdown.