test262 staging/ exposes 109 conformance gaps, grouped and diagnosed

[lahma]

Turning on test262's staging/ directory (#3008, PR #3016) runs 1,483 files that were never generated before, largely SpiderMonkey's own suite contributed upstream. 1,352 of them pass — 91%. Of the 131 that do not:

• 20 hang or exhaust the 30-second budget and are excluded for that reason alone (unbounded recursion, {length: Infinity} generics, the DST-caching family). Nothing conformance-related.
• 2 were gated on sibling PRs and now run. staging/sm/String/replace-math.js was unparked by Give JavaScript strings a maximum length instead of a wrapped array rent #3015 and staging/Temporal/removed-methods.js by Unpark staging/Temporal/removed-methods.js, which #3014 already fixed #3023, so the exclusion list is 129 entries, not 131, and the passing count is 1,354 of 1,483.
• 109 are genuine conformance gaps, and they are what this issue tracks.

Each is excluded in Jint.Tests.Test262/Test262Harness.settings.json with a prose diagnosis of the group it belongs to and what removes it, per the ExcludedFiles convention in AGENTS.md. That convention also says exclusions outside intl402/ stay temporary — 109 comments in a JSON file is a record of the debt, not a plan for paying it down, so this issue is the plan's index.

The groups are independent. Any one can be picked up alone, and the PR that fixes it deletes its block from ExcludedFiles. Nothing here is a regression: every one of these was already failing, invisibly, before staging/ was generated.

The 109, grouped

13 files — Cross-realm error identity.

Cross-realm error identity. $262.createRealm() works, but an error Jint raises while running a function belonging to another realm is constructed from the current realm's intrinsics, so e instanceof otherGlobal.TypeError is false. The realm's global also does not carry its own $262, so otherGlobal.$262.detachArrayBuffer is undefined. Removed by making error creation realm-aware (and by installing $262 into every created realm).

• staging/sm/Array/change-array-by-copy-cross-compartment-create.js
• staging/sm/Array/change-array-by-copy-errors-from-correct-realm.js
• staging/sm/Array/from_realms.js
• staging/sm/Iterator/from/wrap-new-global.js
• staging/sm/Iterator/prototype/every/error-from-correct-realm.js
• staging/sm/Iterator/prototype/find/error-from-correct-realm.js
• staging/sm/Iterator/prototype/forEach/error-from-correct-realm.js
• staging/sm/Iterator/prototype/reduce/error-from-correct-realm.js
• staging/sm/Iterator/prototype/some/error-from-correct-realm.js
• staging/sm/Iterator/prototype/toArray/create-in-current-realm.js
• staging/sm/JSON/parse-with-source.js
• staging/sm/TypedArray/constructor-buffer-sequence.js
• staging/sm/TypedArray/set-wrapped.js

12 files — Legacy Function.

Legacy Function.prototype.caller / arguments.callee.caller. These are Annex B accessors whose sloppy-mode observable behaviour (the calling function, null at the top of the stack, poisoned on strict functions, transparent through eval frames) Jint does not reproduce. Removed by implementing the legacy accessors, not by relaxing the tests.

• staging/sm/Function/function-caller-restrictions.js
• staging/sm/extensions/arguments-property-access-in-function.js
• staging/sm/extensions/censor-strict-caller.js
• staging/sm/extensions/function-caller-skips-eval-frames.js
• staging/sm/extensions/function-properties.js
• staging/sm/regress/regress-577648-1.js
• staging/sm/regress/regress-577648-2.js
• staging/sm/regress/regress-584355.js
• staging/sm/regress/regress-586482-1.js
• staging/sm/regress/regress-586482-2.js
• staging/sm/regress/regress-586482-3.js
• staging/sm/regress/regress-586482-4.js

10 files — Argument validation raising no error, or the wrong one, on built-ins called with a bad this or a bad index.

Argument validation raising no error, or the wrong one, on built-ins called with a bad this or a bad index. Includes SpiderMonkey's own toSource extension, which Jint does not implement at all -- that half is removed by the test moving, not by Jint changing.

• staging/sm/Function/bound-non-constructable.js
• staging/sm/TypedArray/constructor-byteoffsets-bounds.js
• staging/sm/TypedArray/iterator-next-with-detached.js
• staging/sm/class/newTargetDVG.js
• staging/sm/class/superPropOrdering.js
• staging/sm/expressions/computed-property-side-effects.js
• staging/sm/expressions/optional-chain.js
• staging/sm/extensions/extension-methods-reject-null-undefined-this.js
• staging/sm/extensions/quote-string-for-nul-character.js
• staging/sm/misc/builtin-methods-reject-null-undefined-this.js

7 files — RegExp semantics: case-insensitive matching across the Latin-1 boundary, and lastIndex bookkeeping in the Symbol.

RegExp semantics: case-insensitive matching across the Latin-1 boundary, and lastIndex bookkeeping in the Symbol.match / Symbol.replace / matchAll paths (including -0 and the ToLength recomputation the spec performs per iteration).

• staging/sm/RegExp/ignoreCase-non-latin1-to-latin1.js
• staging/sm/RegExp/lastIndex-match-or-replace.js
• staging/sm/RegExp/replace-local-tolength-lastindex.js
• staging/sm/RegExp/replace-local-tolength-recompilation.js
• staging/sm/RegExp/unicode-ignoreCase-ascii.js
• staging/sm/String/matchAll.js
• staging/sm/String/replace-updates-global-lastIndex.js

7 files — Set methods (new Set.

Set methods (new Set.prototype union/intersection/difference/... ) observe the set-like argument in a strictly specified order -- GetSetRecord reads size, then has, then keys, and coerces size exactly once. Jint's order and coercion count differ. Removed by implementing GetSetRecord to the letter.

• staging/sm/Set/difference.js
• staging/sm/Set/intersection.js
• staging/sm/Set/is-disjoint-from.js
• staging/sm/Set/is-subset-of.js
• staging/sm/Set/is-superset-of.js
• staging/sm/Set/symmetric-difference.js
• staging/sm/Set/union.js

7 files — Iterator protocol: IteratorClose is not invoked (or not invoked at the right point) on the abrupt paths of Array.

Iterator protocol: IteratorClose is not invoked (or not invoked at the right point) on the abrupt paths of Array.from, the Map/Set constructors, for-of and array destructuring, and the "done" bookkeeping of a destructuring pattern differs from the spec's.

• staging/sm/Array/for_of_2.js
• staging/sm/Array/from-iterator-close.js
• staging/sm/Map/constructor-iterator-close.js
• staging/sm/destructuring/order.js
• staging/sm/destructuring/order-super.js
• staging/sm/expressions/destructuring-array-done.js
• staging/sm/statements/for-of-iterator-close.js

7 files — Parser early errors that Acornima does not raise, or raises as the wrong error type: an invalid parameter list, `super()` inside an arrow inside a direct eval in a derived constructor, a destructuring pattern with a duplicate binding, a legacy octal literal in strict mode, `yield` as an identifier, and a field initializer's TDZ.

Parser early errors that Acornima does not raise, or raises as the wrong error type: an invalid parameter list, super() inside an arrow inside a direct eval in a derived constructor, a destructuring pattern with a duplicate binding, a legacy octal literal in strict mode, yield as an identifier, and a field initializer's TDZ.

• staging/sm/Function/invalid-parameter-list.js
• staging/sm/class/derivedConstructorArrowEvalNestedSuperCall.js
• staging/sm/class/derivedConstructorArrowEvalSuperCall.js
• staging/sm/destructuring/bug1396261.js
• staging/sm/fields/bug1587574.js
• staging/sm/strict/deprecated-octal-noctal-tokens.js
• staging/sm/syntax/yield-as-identifier.js

6 files — Math precision.

Math precision. Jint delegates to the BCL, whose acosh/asinh/atanh/cbrt/expm1/log1p differ from the reference values in the last few ULPs, and expm1/log1p flush 1e-300 to zero. Removed by implementing these with the spec-grade algorithms rather than the BCL primitives.

• staging/sm/Math/acosh-approx.js
• staging/sm/Math/asinh-approx.js
• staging/sm/Math/atanh-approx.js
• staging/sm/Math/cbrt-approx.js
• staging/sm/Math/expm1-approx.js
• staging/sm/Math/log1p-approx.js

5 files — Function source text and inferred names.

Function source text and inferred names. Function.prototype.toString does not return the original source for class expressions and some method forms, and NamedEvaluation does not reach every position the spec names (for-in heads, computed method keys, ...). Removed by carrying the source range and completing NamedEvaluation.

• staging/sm/Function/function-name-for.js
• staging/sm/Function/function-name-method.js
• staging/sm/Function/function-name-property.js
• staging/sm/class/parenExprToString.js
• staging/sm/generators/runtime.js

5 files — Date parsing and clipping: the implementation-defined fallback parser accepts (and rejects) different strings than SpiderMonkey's, two-digit years map differently, TimeClip does not reject every out-of-range value, and Symbol.

Date parsing and clipping: the implementation-defined fallback parser accepts (and rejects) different strings than SpiderMonkey's, two-digit years map differently, TimeClip does not reject every out-of-range value, and Symbol.toPrimitive throws a non-object.

• staging/sm/Date/non-iso.js
• staging/sm/Date/timeclip.js
• staging/sm/Date/toISOString.js
• staging/sm/Date/toPrimitive.js
• staging/sm/Date/two-digit-years.js

5 files — Annex B B.

Annex B B.3.3 block-scoped function hoisting: the var-scoped alias a sloppy-mode block function creates is not synthesised the way the spec's FunctionDeclarationInstantiation amendment describes, so the name is visible with the wrong value (or not at all) in the enclosing function, inside a direct eval, inside with, and after a redeclaration.

• staging/sm/lexical-environment/block-scoped-functions-annex-b-arguments.js
• staging/sm/lexical-environment/block-scoped-functions-annex-b-eval.js
• staging/sm/lexical-environment/block-scoped-functions-annex-b-with.js
• staging/sm/lexical-environment/block-scoped-functions-deprecated-redecl.js
• staging/sm/regress/regress-602621.js

5 files — Array exotic [[DefineOwnProperty]] and the generics' element ordering: truncating length past a non-configurable sparse element stops at the wrong index, sort/unshift observe the wrong element sequence, and the strict-mode generics do not report failure to write.

Array exotic [[DefineOwnProperty]] and the generics' element ordering: truncating length past a non-configurable sparse element stops at the wrong index, sort/unshift observe the wrong element sequence, and the strict-mode generics do not report failure to write.

• staging/sm/Array/length-truncate-nonconfigurable-sparse.js
• staging/sm/Array/sort-typedarray-with-own-length.js
• staging/sm/Array/unshift-with-enumeration.js
• staging/sm/strict/15.4.4.9.js
• staging/sm/strict/15.4.4.12.js

4 files — Number conversion and formatting.

Number conversion and formatting. toFixed/toPrecision lose the exact mathematical value for extreme magnitudes and subnormals, and ToPrimitive on a Number wrapper consults the wrong method. Removed by exact-value formatting and a spec-order OrdinaryToPrimitive.

• staging/sm/Number/20.1.3.3-toFixed.js
• staging/sm/Number/defaultvalue.js
• staging/sm/Number/toFixed-values.js
• staging/sm/Number/toPrecision-values.js

4 files — Proxy trap invariants and the exact trap sequence the array built-ins are specified to perform.

Proxy trap invariants and the exact trap sequence the array built-ins are specified to perform. Jint's own invariant check on a defineProperty trap also fires where it should not.

• staging/sm/Array/concat-proxy.js
• staging/sm/Array/from_proxy.js
• staging/sm/Array/species.js
• staging/sm/Reflect/has.js

4 files — Remaining one-offs, each its own gap: Iterator.

Remaining one-offs, each its own gap: Iterator.from's next() result validation, the Map constructor's per-entry type check, Symbol.keyFor over a cross-realm registry, and a module namespace re-exported under a string name.

• staging/sm/Iterator/from/wrap-next-not-object-throws.js
• staging/sm/Map/iterable.js
• staging/sm/Symbol/keyFor.js
• staging/sm/module/module-export-name-star.js

2 files — Array.

Array.from applied to an arbitrary constructor: the spec constructs this when it is a constructor and falls back to ArrayCreate when it is not, and each element goes in through CreateDataPropertyOrThrow, so a read-only element or a non-extensible target has to throw.

• staging/sm/Array/from_constructor.js
• staging/sm/Array/from_errors.js

2 files — String case mapping is done with the BCL's culture-aware tables rather than the Unicode Default Case Conversion the spec requires, so a handful of code points (dotless i, the supplementary planes) map differently.

String case mapping is done with the BCL's culture-aware tables rather than the Unicode Default Case Conversion the spec requires, so a handful of code points (dotless i, the supplementary planes) map differently. Removed by using the Unicode data directly.

• staging/sm/String/string-code-point-upper-lower-mapping.js
• staging/sm/String/string-upper-lower-mapping.js

2 files — Mapped arguments: the exotic object's link to the formal parameters (and what an object created *from* it inherits) does not match the spec's mapping.

Mapped arguments: the exotic object's link to the formal parameters (and what an object created from it inherits) does not match the spec's mapping.

• staging/sm/Function/arguments-parameter-shadowing.js
• staging/sm/regress/regress-552432.js

2 files — Source phase imports (stage 3): Jint supports the phase for WebAssembly-shaped sources only, so `import source` of a JavaScript module is refused outright and the prototype chain of an AbstractModuleSource is not the one the proposal describes.

Source phase imports (stage 3): Jint supports the phase for WebAssembly-shaped sources only, so import source of a JavaScript module is refused outright and the prototype chain of an AbstractModuleSource is not the one the proposal describes.

• staging/source-phase-imports/import-source-source-text-module.js
• staging/source-phase-imports/module-source-prototype-chain.js

Notes

• These are not ranked by value. Size is a proxy for how much of the suite a fix unblocks, not for how much an embedder cares — the parseInt and Proxy-as-global defects that justified test262 staging/ directory is never generated, and it had coverage for two 4.16.0 defects #3008 in the first place were one file each.
• Several groups are plainly out of scope for a conformance push and may stay excluded indefinitely: legacy Function.prototype.caller is Annex B behaviour Jint has never implemented, and Math precision is the BCL's transcendental accuracy rather than anything Jint controls. Those want a decision, not a fix.
• Discovered while enabling staging/ for the 4.16.0 follow-up review. See test262 staging/ directory is never generated, and it had coverage for two 4.16.0 defects #3008 for why the directory was never generated and what turning it on cost (~2,800 cases, +12s of suite time).

---

Scope note (2026-08-15)

This issue tracks conformance gaps that staging/ exposed, and nothing else. Three defects found during the same 4.16.0 follow-up work are deliberately not folded in here, because none of them is a staging exclusion and each is actionable on its own:

• JSON.stringify can build a string longer than the maximum string length the engine enforces everywhere else #3024 — JSON.stringify can build a string longer than JsString.MaxLength. A follow-up to A string past the implementation limit overflows the builder instead of throwing RangeError #3011, not a conformance gap; no staging file covers it.
• An Atomics.waitAsync that is never notified keeps its engine alive for the life of the process #3025 — an Atomics.waitAsync that is never notified keeps its engine alive for the life of the process. A memory-retention defect; the test262 Atomics suite passes today.
• Delete the two unreferenced private Islamic date conversions in TemporalHelpers #3026 — two unreferenced private Islamic date conversions in TemporalHelpers. Dead-code cleanup.

Folding those in would make this issue "everything found in August", which is the failure mode an umbrella issue has: it stops being a work index and becomes a diary. The 109 below share a property that makes the grouping useful — each is a block of ExcludedFiles that a single PR can delete — and that is the boundary worth keeping.