- Lower asynchronous ESM initialization with Node-compatible dependency ordering and caching.
- Finish top-level promises consistently across the C and LLVM runtimes.
- Add differential coverage for imports, rejection, and unsettled evaluation.
- Lower Number.prototype.toFixed through the compiler IR and native backends.
- Match JavaScript rounding, range errors, and numeric edge cases in the runtime.
- Add differential coverage and update surface metadata and diagnostics.
- Add Agent connection accounting and HTTP/socket compatibility surfaces
- Support dynamic Buffer at and slice access
- Pin behavior with differential corpus coverage
- The written-binding decline yields when the declared type has no mapping or the write's RHS roots at an ambient-undefined name
- A class extending a collection-poisoned base flushes the base's diagnostics eagerly: the heritage evaluates at module init, so deferring the compiler's own fence manufactured a silent wrong output
- for-of, destructuring, and spread over engine-held dyn values ride the engine's own iterator protocol with V8's exact non-iterable spellings; for-of over dyn lowers generally
- Static promises cross into the engine as thenables; Promise.all over mixed and non-literal entries runs the engine's own combinator; engine-array fulfillments exit by reference
- The re-minted silent-wrong-answer path is closed: island-literal members whose values poison fence at the read with the captured diagnostic
- The checked-dynamic tree replaces the old nickname across comments and internal docs
- trapDeclRootOf's written-binding decline now requires a mappable declared type: ordinary storage only buys anything when it compiles — an unmappable recursive call shape traded the compiling no-storage trap for a guaranteed SC2009 on a value that never exists
- bindingEverWritten skips statement-position writes whose RHS chain roots at an ambient-undefined name: that statement IS the RHS root's throw (lowerExprStatement's first claim) and never touches the target's storage, so it cannot justify the decline
- A poisoned base a top-level class extends flushes its deferred diagnostics eagerly (collectClassShape): the heritage evaluates at module init and the base's fence is the compiler's own — Node defines the base fine — so the deferred trap compiled a binary that refused at startup where Node runs; leaf and derived-only poisoned classes keep the JS deferral
- Fixtures: corpus 2632 (unmappable written trap binding) and 2633 (ambient-rooted rewrite), diagnostics js-extends-poisoned-base (the eager report names the base's real blocker), errors.test pins for both deferral edges that must NOT report eagerly; order-parity baselines recorded for the new entries
- text, json, and buffer settle real promises over the readable machinery with Node's settle timing and rejection set; arrayBuffer and blob fence with their representation reasons
- createRequire over this file's URL is compile-time plumbing: builtin specs become namespace imports, relative JSON bakes as validated documents, npm packages ride the require-condition graph, unresolvable names throw Node's catchable MODULE_NOT_FOUND
- module.builtinModules bakes as the compat target's exact list
# Conflicts:
# packages/compiler/src/frontend/lowering/lower-modules.ts
# packages/compiler/src/frontend/lowering/lowerer.ts
- The forwarding shape — emit(event, ...args) whose rest appears only as super.emit(event, ...args) — lowers per dispatching event as a real hierarchy method, vtable-participating
- error events keep the throw-when-unhandled path through overrides; super.emit resolves to the nearest override ancestor, JS's chain
- Receivers above every override route through interned instanceOf dispatch; other emitter members and stream-rooted classes keep their fences with reasons stated
# Conflicts:
# packages/compiler/test/ts7/baselines/order-parity.json
- An interval-and-wholeness inference over the lowered IR proves integrality and range for every value reaching a declared i64/u64 slot, with loop-header widening precise enough that the bounded counter loop proves its exact range
- Refusals carry the slot path, the failed obligation with evidence, and the author's concrete fix as SC4021, SC4022, and SC4023
- Declared slots are both obligation and assumption: call sites check arguments, callee bodies assume their classes; integer returns emit exact machine-integer wrappers and inbound params trap through the structured host-contract family
- Math.trunc and Math.ceil compile statically
# Conflicts:
# packages/compiler/src/library/profile.ts
- The ambient-trap probe resolves as a side-effect-free probe: deferred collection diagnostics stay deferred, and the merged-namespace fence lands as a diagnostic instead of crashing the collect walk
- deadUnmappableBinding and the nullish family claim statement-position declarators only; for-init heads and catch bindings stay locals
- Trap classification declines written bindings; the seven 0.0.10 pass-to-fail fences restore byte-identical to 0.0.9 behavior
- library-int suite snapshots SC4021/SC4022/SC4023 rendered end to end (slot path, evidence, span, fix) — the diagnostics-corpus discipline for codes the executable lane cannot reach
- dynamic-surface diagnostics fixture swaps Math.ceil for Math.sqrt (trunc/ceil compile statically now) and its snapshot regenerates
- collectGlobals' trap/nullish/dead classification probes catch PoisonError per the per-probe recovery discipline: the cross-block merged-namespace SC1090 fence reached through the ambient-root walk recorded its diagnostic and then crashed the CLI mid-discovery (5 merged-namespace corpus tests, ice bucket)
- On a poisoned probe, JS files defer the recorded diagnostics to runtime fences (deferJsCollectionDiags) and the binding falls through to ordinary registration; the statement lowering re-fences at its own site
- The no-storage binding families (nullish, dead-unmappable) only claim statement-position declarators: an uninitialized for-init let (`for (let x; false;) {}`) was claimed as dead-unmappable, violating lowerVarDeclList's contract that a for-init declarator always lowers ("lowerer bug: for-init declarator resolved to a global", 2 corpus tests)
- Fixtures: corpus 2608 (for-init heads with the masking variants), diagnostics merged-namespace-collect-probe (the collection-reached fence stays an honest SC1090)
- The ratified rename rider (world-unification §7 lane 5): "DOM" collided with the browser DOM and, post-lane-1, the structure is no longer a document tree. Comments and doc prose now say "the checked-dynamic tree" for the noun and "dyn" adjectivally; runtime symbol names (scr_dyn_*), the error.domClone libCall, and the real web DOMException family are unchanged.
- The few user-facing strings carrying the nickname move with it (the structuredClone cyclic message, the dc subscriber/publish fences, two emitter ICE texts); diagnostics snapshots regenerated (96 pass). One private compiler constant (DOM_PROTO_METHOD_NAMES -> DYN_PROTO_METHOD_NAMES) rides along.
- Mechanical sweep, 142 files, comments/prose only — no behavior change.
- The pattern fence lifts: the vendored engine already executes named groups and named backreferences
- Match results project .groups at compile time as the checker's own record shape, with Node's key order, undefined-when-none, and null-read error texts
- Replace and replaceAll resolve $<name> templates with Node's exact edge behaviors; matchAll rows carry groups through for-of, drains, and destructuring
- Destructuring over index-signature records reads through the overflow map generally
- lowerRegexLiteral drops the (?<name> fence: patterns and \k<name> backreferences run natively in libregexp, and .groups on exec/match/matchAll results desugars into a compile-time record projection over the honest slice — the checker's own Record<string,string> hybrid with overflow entries in group declaration order — whenever the regex is statically known (literals, const bindings, new RegExp("..."), matchAll rows via for-of bindings, stored drains, spread rows, element access)
- Node's exact shapes: a group-less regex answers .groups === undefined, nullable receivers throw Node's TypeError on the unit arms, ES2025 duplicate names project the participating alternative, and destructuring rides — flat, aliased, and the nested { groups: { year } } pattern — via the groups wrap plus a general overflow read for undeclared keys of index-signature shapes
- scr_put_substitution resolves $<name> against lre_get_groupnames with Node's edge answers: nonexistent and nonparticipating names substitute empty, $< stays literal when the pattern has no named groups, an unterminated $<name stays literal
- s.match(re) also claims a nullable string receiver (the Dict<string> member spelling) through the checked narrow; an untraceable .groups read keeps the SC2020 fence with a hint naming the traceable shapes
- Differential corpus 2604-2607 (both backends, SAN), runtime C substitution/backreference cases, refreshed diagnostics snapshots and surface manifest; corpus 1614's deliberate deferred fence moves to a Proxy wrap now that hybrid record reads compile
- createRequire(import.meta.url) (or __filename / import.meta.filename) in a const binding is compile-time plumbing that emits nothing; each require call through it resolves per site against the binding's own file, and the inline createRequire(import.meta.url)("spec") spelling erases identically
- The erasure per target: a builtin spec makes the binding a namespace import in const clothing (namespace, destructured, and inline-member spellings all key the existing module tables, with the fallback-typed 'as typeof import(...)' cast idiom stripped); a relative .json document validates at compile time and lowers to json.parse over its baked text (JSON.parse's checked-dynamic unknown stance — exactly Node's value); a bare name nothing installed resolves compiles to Node's catchable MODULE_NOT_FOUND throw (the optional-dependency try/require pattern); an installed npm package embeds under --dynamic through the island's NEW require-condition entry (NpmGraphBuilder.addRequire — a dual package answers its CJS arm, exactly Node's createRequire) and reports the requires-dynamic diagnostic in a static build
- Dynamic specifiers, '#' project imports, relative non-.json targets, --npm-static opt-ins, and non-this-file bases all fence by name; the island-handle rescue skips the plumbing decls so --dynamic builds erase them too
- module.builtinModules bakes as Node v24's exact list (a fresh string[] per read, pinned to the compat target so emitted programs stay host-independent); isBuiltin and syncBuiltinESMExports fence per site with pointed hints, and the createRequire hint, its diagnostics snapshot, the surface manifest, and the npm-static offender wording follow
- Corpus 2606 pins json/builtin/builtinModules/MODULE_NOT_FOUND differentially on both backends and under ASan; the npm create-require fixture case pins cjszoo's module.exports, dual's require-condition CJS arm, and the missing-package rejection against Node
- class C extends EventEmitter may re-declare emit(event: string, ...args: unknown[]): boolean when the rest parameter only forwards through super.emit(event, ...args) in the method's own body; the body lowers once per event name as the real hierarchy method emit:<event> (vtable participation, (this, ...tuple) => bool ABI), with event bound to the name and the forward lowered to the runtime emit — or to the nearest override ancestor's specialization, JS's super chain.
- Emit sites dispatch by the whole-program rule: receivers at-or-below an override call the specialization directly (virtually when a deeper override exists); receivers above every override route through an interned helper that instanceOf-tests the topmost override subtrees and falls through to the runtime emit, so plain EventEmitter bindings holding override instances still dispatch like Node.
- super.<member> over an emitter-rooted base now lowers through the emitter spoke with static dispatch (an override at-or-below the lexical class never answers a super call); registering 'newListener'/'removeListener' listeners fences while a comparable class overrides emit (the runtime fires meta events internally, past the override), and every other emitter member keeps the surface fence with the forwarding-shape rule spelled out for emit.
- Corpus 2604-2609 pin both backends against Node: the logging override with upcast dispatch, two-level override chains with literal super.emit, once/off/prepend ordering and handled errors under an override, the unhandled-'error' throw through an override, the filtering override, and an emitter-subclass job queue; the diagnostics snapshot pins the three fences.
- Extract preflight's benign-cycle admission (the cheap per-edge rule plus the declaration-only init window analysis, with Tarjan SCC and cluster-verdict memos) into makeCycleAdmission; preflight's static walk and appendDynamicImportModules now judge back edges identically, so benign cycles behind dynamic import() compile as Node cache hits instead of blanket SC1016.
- appendDynamicImportModules runs once in lowerToIr before any pass constructs: the old per-constructor run minted cycle refusals into the discovery pass, whose diagnostics are discarded by design, so inadmissible import() cycles silently compiled; refusals now land on the emit pass with the same narrowed reason (binding and line, or the offending top-level statement) the static walk mints.
- SC1016 gains a registry hint naming the admitted shape and the fix; diagnostics snapshots, order-parity baselines, and the surface manifest repin with it.
- Corpus 2604 (mutual class cycle across two modules), 2605 (three-module function ring with two back edges and a function-body-only value use), 2606 (import() into an admitted mutual-recursion cycle) — all byte-identical to Node through both backends; diagnostics fixture dyn-import-cycle-tdz pins the dynamic-path TDZ refusal.
- declare-erased roots make every chain rooted there a catchable ReferenceError at first touch, exactly Node's erasure semantics
- Never-read unmappable bindings emit nothing; nullish-cast bindings answer Node's exact TypeError on member access
- Assertion-supplied generic signatures register like annotations and monomorphize per call
- Stdlib generic members take the lib fence naming the member
# Conflicts:
# packages/compiler/src/frontend/lowering/lower-stmts.ts
# packages/compiler/test/ts7/baselines/order-parity.json
- Object.create(null) is a real null-prototype dictionary: inspect, toString, and deepStrictEqual answer Node's forms; the dynamic tier routes to the engine's own Object.create for live delegation
- Object prototypes fence by name in static builds — an own-copy stance is observably wrong under Node's empty-own-set answers and live proto mutations
- entries() and keys() walk any static array live in for-of heads and iterator-helper chains
- Object.assign(target, ...sources) packs spreads with JS's evaluation order before any copy, with V8's position-dependent error texts
# Conflicts:
# packages/compiler/test/ts7/baselines/order-parity.json
# packages/runtime/src/scr_json.c
- mapType: a union with an 'object'/'unknown'/'{}'-flavored (dyn) arm beside only dyn-subsumable siblings — scalars, units, dynFrom-convertible records/arrays — maps to DYN wholesale; class/Map/Set/func/promise/regex arms keep their typed homes, and the SC2009 classifier now names the sibling that cannot ride the collapse instead of blaming the dyn arm.
- A dyn ELEMENT makes the whole array the DOM value (unknown[], object[], the plugins-slot (string | object)[]): the DOM array literal builds from TS annotations, .length on collapsed-array receivers is a DOM keyed read, element writes ride dyn.keySet (ARR index set/extend with undefined-hole padding), and for-of over dyn gets a pointed operation fence instead of the stale component story.
- Appendix-A repro 4 stops SC2009-fencing: an island plugins array wraps at the collapsed slot (dynFromJsval) and the element read reaches the loud keyed-read ladder until the routing lane arms it (dyncheck pin jsval-collapsed-union-slot).
- Corpus 2584/2585 differential both backends (guard ladder, option tables, unknown[] first-class; SAN clean); json-dyn snapshot repinned where the unknown[] declaration fence retires; conformance movers verified locally: discriminateObjectTypesOnly, unknownPropertiesAreAssignableToObjectUnion, unknownLikeUnionObjectFlagsNotPropagated.
- SEMANTICS 398 documents the collapse rule, the exact subsumable-arm domain, and the width/identity consequences (deep-copy entry, full-width JSON, per-site fences for non-subsumable values).
- ambientUndefVarRootOf walks calls, `new`, optional chains, assertions, instantiation expressions, and tagged templates, and accepts `declare function` and trap-binding roots: Node erases the declaration, so the root read IS the crash — lowerCall, member reads, assignments, and variable declarations lower the whole statement to the catchable ReferenceError, arguments never evaluate, and the never-initialized binding registers as a trap whose references can never run
- never-read unmappable bindings with value-only initializers and writes emit nothing: Node builds the value and drops it, so `var xs2: typeof Array;` and the write-only `var f2: { <T, U>(x: T, y: U): T }` compile instead of fencing on a type the program never consumes
- null!/as-any bindings whose types keep generic signatures (or intern empty record shapes) register no storage; generic-method calls and member reads through them compile to Node's exact TypeError, nullish-to-nullish flows to nothing, heritage-clause uses decline the family so mixin resolution keeps its fence, and checked-dynamic (`any`) bindings keep their dyn story untouched
- assertion-shaped generic arrows (`< <T>(x: T) => T >((x) => x)`) register like annotated bindings and monomorphize per call; stdlib generic members (Promise.then, Object.defineProperty) take the SC2020 lib fence, and Function.prototype.apply/call/bind on compiled function values name the direct-call spelling
- diagnostics fixtures that used `declare` to conjure unconstructible values now conjure them by cast so each fence family keeps its pin; snapshots re-recorded, the deferred checker-panic diagnostic now flushes at its first reference as designed
- Static builds mint a DOM object flagged prototype-free: own-member walks are already Node's null-proto answers; inspect prefixes [Object: null prototype] (nested/empty/beyond-depth forms), o.toString() throws Node's "is not a function" (dyn.toString carries the source spelling), and deepStrictEqual gains the prototype gate, diff rendering included
- Under --dynamic Object.create routes through the engine's own Object.create for null and engine-held prototypes — live delegation, shadowing writes, and mutation-through-proto all Node-exact
- DOM and static-record prototypes fence with the reason instead of the design doc's own-copy stance: Object.keys/inspect/JSON answer the EMPTY own set in Node and prototype mutations show through live, which no copy answers honestly
- Element writes on plain dyn receivers ride dyn.keySet generally (the result[key] = value memo idiom; unmapped static-any receivers probe their lowered world), unblocking the dictionary's use
- Corpus 2581 (static, byte-for-byte both backends) and 2582 (@dynamic engine route); diagnostics object-create-fences pins the three named fences
- Documented residue: String()/template/join over a null-proto DOM value answer [object Object] where Node throws — the emitted String(unknown) walkers are never-throw by contract
- assert-fences drops its non-literal expected-object pin (that form lowers now through the errValue key walk — corpus 2556 pins the behavior); the diagnostics snapshot re-records.
- order-parity baselines re-record for the five new corpus fixtures.
- ServerResponse/IncomingMessage/Socket members that fenced on checked-dynamic values now dispatch: flushHeaders, getHeaders, setTimeout, cork/uncork with a real writableCorked counter and coalescing flush, writeHead's raw-array form, write/end encodings and deferred callbacks, req.pause/resume with buffered drain, destroyed/readable/bytesWritten/res.req reads, setNoDelay, destroySoon, end(cb), and dyn Buffer at()/slice()
- new http.Agent/https.Agent lowers to a checked-dynamic handle: literal option parsing, getName, destroy, sockets/requests/freeSockets snapshots, settable defaultPort, and REAL maxSockets accounting (over-limit requests defer their dial and queue); keepAlive: true throws the named pooling fence
- the request path threads the agent (http.requestAgent/https.requestAgent rows, both backends); dyn keys now stringify for indexing and the in-operator over dyn receivers (dyn.hasKey)
- differential fixtures 2573-2577 pin both families against Node
- pipeline and finished settle real promises over the existing watcher machinery, on both backends
- domain, v8, inspector, node:sqlite, and the underscore-prefixed internals each name why a compiled binary cannot provide them
- stream/web, the Console class, and dns/promises stay honest not-yets with documented paths
- The SC1010 hint gains stream/promises and the long-missing querystring
- The DataView setter family lands end-to-end; fresh ArrayBuffers erase into zero-filled typed-array views; resizable/shared surfaces fence by name
- Object.is compiles with the spec's SameValue across every kind, including union operands
- Composed en-US Intl.NumberFormat and toLocaleString print byte-exact through the Ryu digit core's shortest-round-trip rounding, 30k-double fuzz-verified; the rest of Intl fences by name with reasons
- DurationFormat deliberately fences: Node throws V8-internal Temporal errors with embedded source paths on plain inputs
# Conflicts:
# packages/compiler/test/ts7/baselines/order-parity.json
- SC1010's wording routes through one helper (unsupportedModuleFeatureOf): pending modules keep the plain not-yet story, out-of-scope ones append the reason
- v8/inspector name the missing V8 engine, node:sqlite the unbundled SQLite library, domain its deprecation and engine-level interception, and the fourteen underscore-prefixed legacy modules their Node-internal implementation objects
- All five fence sites reword together: the four import/require edges in program.ts and the fenced-binding use site in lower-exprs.ts
- New diagnostics fixture module-out-of-scope.ts pins the reasons (and that stream/web stays plain not-yet); the coverage import-fences snapshot repins with the v8 reason
- recordWidthPlan narrows index-signature sources into declared-only targets (declared fields copy off their struct slots, the overflow drops with the width); absent-field completion declines there — the overflow could hold the key at runtime, and a completed undefined would drop the value Node keeps
- the overflow capture initializes same-named declared target fields directly off the source, admitting REQUIRED declared members; literal writes to names the target does not declare go overflowOnly, and a dispatch-writes gate keeps the validator's collision rule for the rest
- widthLiftPlan gains dynIn (a typed field into an 'unknown' slot — the static→DOM deep copy), upcast (derived into base, the prefix-layout reinterpret), and funcAdapt (funcCoerceAdapter's CLEAN subset — the stranded trap-only dispositions stay top-level, never inside a width plan)
- SC2002 now names the first blocking rule per record pair (describeRecordWidthBlocker mirrors the planners' gates) and its message/hint/registry entry speak the width-copy truth — the manifest flips the one entry name
- corpus 2556/2557 pin both hybrid directions and the new lifts byte-identical vs Node on both backends; diagnostics fixture record-width pins the residue's pointed messages
- isSupportedIndexValue admits func/map/set: the overflow map's RC and trace machinery is already fully type-directed (scr_closure/scr_map adapters, the cycle fixpoint's map rule recursing on the value), and nested index-signature records rode the record kind all along
- the C keyed-read miss trap can now describe Map/Set-typed slots (dynDesc), reachable only through that message — no dynCheck ever expects a container
- SC2006's message and hint name the real value domain (functions, Maps, Sets, RegExps, Promises, ...) and the honest blockers left: symbol keys, dual signatures with unequal value types, and values with no representation at all
- diagnostics fixtures pivot to still-fenced shapes (Uint8Array values, unequal duals, a Date-valued bag); corpus 2558/2559 pin the command-registry and container-table patterns — declared+overflow hybrids, enumeration, delete, registry-capturing cycles, container churn — byte-identical vs Node on both backends
- new Intl.NumberFormat("en-US").format(x), the callable spelling, and x.toLocaleString("en-US") lower to one runtime formatter with default options: decimal notation, 0-3 fraction digits rounded half-up on the SHORTEST round-tripping decimal (ICU's rounding input, probed against Node — format(1.0005) is "1.001" though toFixed(3) answers "1.000"; 1e23 prints the shortest form's zeros, not the exact expansion), "," grouping, "∞"/"NaN" texts, and "-0" for negatives rounding to zero. The en-US/latn symbols are the whole embedded locale surface — no ICU is linked.
- The Ryū digit core factors out of scr_f64_to_str as scr_f64_digits (same bytes; the 1M-double oracle gate stays green) and feeds the formatter; a 30,000-double ad-hoc fuzz across magnitude regimes matched Node byte-for-byte on both backends.
- Every unlowered Intl form fences by NAME with its own reason: the environment-default locale a compiled binary cannot carry (no-locale forms), other locales and the rest of the namespace (DurationFormat, DateTimeFormat, PluralRules, getCanonicalLocales — ICU data the binary does not embed; Node 24's DurationFormat additionally throws V8-internal Temporal errors with embedded source paths, so byte parity there is a version-dependent landmine), options bags, formatter values (the composed form is the lowered shape), and non-number format arguments.
- Differential fixture 2562 pins the rounding discriminators, grouping, ±0/NaN/∞, tiny and huge edges, and both spellings byte-identically vs Node on both backends; the new intl diagnostics fixture pins every named fence, and the Intl hints across the identifier/member/type chokepoints now teach what lowers.
- Number pairs take a new scr_num_same_value runtime primitive (NaN equals NaN, +0 differs from -0 — the two divergences from ===); strings, booleans, unit literals, and every reference kind (arrays, records, class instances with hierarchy widening, functions, class values, Maps/Sets, symbols, typed arrays, promises) ride exactly the strict-equality machinery, whose answers SameValue shares.
- Union operands lower through unionEq, which grows a sameValue flag: the f64 arm's payload compare upgrades to SameValue (a separate interned C helper per union; the LLVM lane calls the runtime primitive inline), every other arm is shared between the two semantics.
- Statically disjoint kind pairs answer the constant false with the operands still evaluated for their effects (Object.is is (any, any) — tsc admits every pair); dyn/jsval operands and partially-overlapping unions fence by name with strict equality's validate/narrow-first stance.
- Differential fixture 2561 pins the NaN/±0 edges, mixed-kind calls with effect counting, reference identity across kinds, and union payloads (including NaN/±0 inside a union and union-vs-plain wraps) byte-identically vs Node on both backends; stdlib-fence pins the named fences.
- Found pre-existing (reproduced at the base commit): the C backend SIGSEGVs on console.log(asBase === d, asBase === new Base()) — an upcast identity compare plus a constructing compare in one call; the fixture avoids the shape and the bug is reported for a follow-up lane.
- new Uint8Array/Uint32Array/Int32Array/Float32Array(new ArrayBuffer(n)) erases the buffer into a zero-filled view exactly like the SharedArrayBuffer sleep idiom (literal byte length divisible by the element size), and new DataView(new ArrayBuffer(n), off?, len?) views a fresh zeroed allocation — nothing else can reference the buffer, so the erasure is exact.
- The DataView setter family (setUint8/Int8/Uint16/Int16/Uint32/Int32/Float32/Float64) lowers end to end: one scr_dataview_set runtime mirror of the getter (ToIndex offsets, Node's constant RangeError, modular integer truncation, double→float rounding, both endiannesses) wired through both backends; the bigint and Float16 setters keep the member fence.
- The resizable surface fences by NAME everywhere it can appear: new ArrayBuffer/resize/transfer/maxByteLength and SharedArrayBuffer.grow now state the reason (no free-standing ArrayBuffer value exists; grow has no threads to share with) at the constructor, identifier, member, and declared-type chokepoints, including the maxByteLength options bag inside an otherwise-erasable position.
- Differential fixture 2560 pins the erasure forms, every setter width/endianness/coercion, aliasing through x.buffer views, and the RangeError edges byte-identically vs Node on both backends; es2025-fences/dataview/stdlib-fence snapshots pin the new named hints.
- keyof-constrained generics specialize per literal key; keyed record writes take the literal-key type rule
- Generic methods called through interface-typed receivers monomorphize against the proven class
- Generic-signature annotations monomorphize their function initializers; generic-fn aliases register their target
- Fixtures 2550-2555 byte-identical on both backends
# Conflicts:
# packages/compiler/src/diagnostics/diagnostic.ts
# packages/compiler/src/frontend/lowering/lower-modules.ts
# packages/compiler/src/frontend/lowering/lower-stmts.ts
# packages/compiler/src/frontend/types.ts
# packages/compiler/test/ts7/baselines/order-parity.json
# tests/harness/__snapshots__/generic-methods.ts.txt
# tests/harness/__snapshots__/generic-value-bindings.ts.txt
# tests/harness/__snapshots__/mixins.ts.txt
# tests/harness/__snapshots__/stdlib.ts.txt
- Baselines added for corpus 2550-2555 (plus the previously unrecorded 2490-2500 entries the sweep now sees); no recorded entry changed.
- tests/diagnostics/generics.ts keeps its alias/annotation sections as reached COMPILING bodies (their absence from the snapshot is the point), fixing the dangling calls the earlier edit left.
- Nested patterns and member targets in assignment position, with JS's reference-before-value evaluation order pinned
- Class-instance sources call getters exactly once per element; a latent double-call in the accessor-default path is fixed
- Class rest packs the inheritance chain in Node's key order, fencing where enumeration can't be reproduced
- Fixtures 2536-2540 for the promotions; 2530-2535 and 2541 pin the already-static declaration-position surface
# Conflicts:
# packages/compiler/test/ts7/baselines/order-parity.json
- Destructuring assignment now takes nested patterns (`({ p: { q } } = src)`, `[[a], b] = t`) and property/element targets (`[c.x, box.vals[i]] = arr`, `({ a: rec.f } = o)`, rest into members) — receivers temp at their pattern position, JS's reference-before-read order; defaults on member targets test against the target's own type. Defaults on nested targets keep a named fence (no single binding type to test against).
- Class instances destructure in assignment position like they already did in declarations: field slots and getter calls per element via the same fieldGetExpr dispatch, defaults, renames, and member targets; methods and undeclared names keep their pointed fences.
- Rest over class-instance sources packs the remaining instance FIELDS into a fresh record (JS's CopyDataProperties copies own enumerable properties — never prototype methods/accessors, matching the checker's rest type), in declarations and assignments; non-public fields, runtime-provided chains (Error/EventEmitter/stream), and key orders the packed shape cannot reproduce fence instead of diverging.
- Defaults on destructured getter results now lower (record and class accessors alike), with the getter result temped so the default's undefined test never calls it twice; numeric-literal pattern keys (`const { 2: x } = o`) fold to JS's canonical property name.
- Corpus 2536-2540 pin the new forms byte-identically against Node on both backends; the destructuring-fences diagnostics program keeps the honest residue (method extraction, non-public rest fields, nested-target defaults, index-signature rest).
- The ts7 order-parity baseline re-record backfills the entries the released file was missing (2490-2500 included) alongside the new fixtures; no existing entry changed.
- Instantiations carry ts-level bindings alongside the mapped IrTypes: a `K extends keyof T` parameter keys its instance on the bound literal (no cross-literal sharing — the bodies read different fields), and literals forward through generic-to-generic calls.
- `T[K]` inside instance bodies resolves through the bound checker types (mapBoundIndexedAccess via the new resolveTypeParamTs hook), and `o[k]` reads whose key TYPE proves one literal — a keyof-bound parameter or a literal-typed const — lower to static field reads.
- `const h = id` registers the target's GenericFnInfo under the alias's symbol (chains, block scope, pinned values, and calls all resolve like the target's own name); non-const/reassigned aliases fence by name, and unpinned VALUE uses keep their fence at the reference.
- Fixtures 2550 (keyof/pick shapes, forwarding, runtime keys, literal-typed key vars) and 2551 (alias chains, contextual pinning, annotated generic aliases), byte-identical vs Node on both backends.
- SC2005 generic call signatures, SC2006 index-signature shapes, SC2007 overloads, SC2008 unresolved intersections, SC2009 supported shapes over an unsupported component with the component named
- Stdlib nominal types reroute to the SC2020 story instead of a duplicate code
- SC2001 keeps only bigint/symbol, constructor objects, and unresolved library generics — its registry name now says so
- Five new diagnostics programs pin one frame per code; manifest entries added; zero behavior change to accepted programs
# Conflicts:
# tests/harness/__snapshots__/regex.ts.txt
- New codes, minted in badType's classification order: SC2005 values whose type keeps a generic call signature, SC2006 index-signature shapes outside the supported set, SC2007 overloaded function types, SC2008 intersections with no resolved lowering, SC2009 supported container/function/record shapes over a component outside its slot — the SC2009 message names the offending component (Map/Set key-value domains, array/tuple elements, union arms, function rest/params/returns, record members) via describeComponentBlocker/describeRecordMemberBlocker in frontend/types.ts, which mirror mapType's own rules on the failure path only.
- Standard-library nominal types with no lowering at all (Date, ArrayBuffer, WeakMap, the iterator and constructor interfaces) now route to the existing SC2020 story instead of the supported-set recitation; stdlib provenance is checked after the container claims so Map/Set/Promise instantiation failures keep their component story.
- SC2001 keeps the honest remainder — bigint and symbol primitives, constructor objects, and library-derived or unresolved generic shapes — and its registry name now says so; across the diagnostics corpus its rendered sites drop from 57 to 5.
- Every new code flows through the diagnostics registry, the regenerated surface manifest, and coverage-report grouping; five new tests/diagnostics programs pin one rendered frame per code, and the ts7 order-parity baselines re-record to include them.
- Zero behavior change to accepted programs: the differential corpus (900 programs) is untouched and green, and the full suite passes.
- staticCallOn and lowerClassValueProperty fenced every #private static access with a through-a-VALUE receiver, so C.#m()/C.#f on a const-bound class expression, A.#f through a const alias of the class, and Example.#m() under a replacing decorator all died with SC1090 while the direct class-name spelling compiled.
- The static-WRITE path already knew the sound rule (exactClassOfReceiver plus the leaf-class arm); classValueIsExactlyOwn now shares it with the call and read paths: a receiver that provably IS the class, or a class with no strict descendants, always carries the declaring class's brand — the fence stays for slots a subclass can inhabit, where Node throws TypeError.
- Rebindable decorated names ride the leaf arm for free: replacement decorators fence any class with subclasses before minting the rebindable global, so the decoration result is always the declaring class object.
- The read path additionally gains the subclass-spelling fence (reading an inherited private static through a subclass-named binding) that the call and direct-read paths already had.
- Fixtures 2520-2523 pin the four call forms (class-name calls including static field initializers and subclassed declarers, const-bound class expressions with reads/writes/calls, decorated classes under both replacing and void decorators, aliased class values local/module/param) byte-identical vs Node on both backends; the private-members diagnostics fixture repins the residual fence on a descendant-capable slot.
- mapType back-references mint placeholder shapes/unions finalized at frame exit; identity is per checker type, and context-sensitive (generic/mixin) knots stay fenced
- every type walk survives cyclic shape graphs: coinductive JSON/dyn-safety predicates, guarded formatIrType/jsvalLiftable, cycle headers via the existing collector fixpoint on both backends
- JSON.stringify throws V8's byte-exact circular TypeError (hop lines, ellipsis rule); cyclic values crossing into checked-dynamic slots trap instead of hanging
- console.log/util.inspect render <ref *N>/[Circular *N] exactly (SEMANTICS 129 resolved; depth:null lowers) and deepStrictEqual terminates through Node's pair memo
- corpus 2480-2488 green on C, LLVM, and the sanitized RC-audit lane, which proves child->parent cycles collect
- console.log routes non-scalar arguments through the static inspect machinery
- Union arguments dispatch per arm at runtime; string arms print raw
- Object literals widen per-field into a uniquely-fitting union arm
- The limitations page documents the type surface
- Island values exit validated into intrinsic slots; checker-any dispatches to the DOM machinery
- World-mixing backstops make jsval/dyn confusion structurally impossible
- Regex arrays compile; dyn-root JSON.stringify matches across backends
- A silent invalid-C escape and a giant-TU stack overflow fixed
- Lexically-bound private calls devirtualize by construction; statics and static blocks included
- Private accessors with observable get-then-set order; private generator methods
- Brand checks fold statically or ride the vtable interval test
- util.inspect omits private members exactly like Node
- The performance global routes through the perf_hooks lowering in all spellings
- Response declares its implemented headers/arrayBuffer/bytes surface
- Math.max and Math.min lower at any arity with NaN and negative-zero exactness
- Copy-installed workspace members classify like symlinked ones
- The performance global routes through the perf_hooks lowering in all spellings
- Response declares its implemented headers/arrayBuffer/bytes surface
- Math.max and Math.min lower at any arity
- Copy-installed workspace members classify like symlinked ones
- island ('any') arguments at intrinsic/libCall/callValue slots take the validated exit (dynCheck's stance); non-exit-safe slots fence with the slot named, and the boundary pass backstops every jsOp/dynCall/dynInvoke argument so world-mixing sites can never ICE
- a checker-'any' receiver/callee that LOWERED checked-dynamic dispatches to the DOM machinery: ambient-this reads are dynKeyGet, rest-args map/forEach ride the runtime dispatch, dyn callees take dynCall
- Object.entries/values push unit-armed (null | undefined) fields as the null literal; new Set(seed) exits an island-handle seed as a validated copy
- fence-bodied functions emit even when their ABI names an unregistered class, and a run()-end sweep rewrites leftover unregistered-class type slots to the inert f64 placeholder — the clang invalid-C escape closes with the ICE
- RegExp[] compiles (REF elements, both backends; map callbacks returning other unrepresentable kinds meet a named fence), and the C emitter appends program-sized line arrays by loop, not spread (500k-line TUs overflowed the stack)
- Privates ride the ordinary member maps under their unspellable spelled names; a subclass redeclaring an inherited private name is fenced, so base-chain walks ARE lexical resolution and privates never enter vtables — JS's no-dynamic-dispatch semantics by construction.
- Private GENERATOR methods compose the generator machinery with this as param 0 through the gen-spawn wrapper (the vtable problem that keeps public generator methods fenced does not exist for privates); async and generic privates ride their existing static-dispatch machinery.
- Private statics lower as module globals/functions through the declaring class's own name; subclass-named and class-value receivers fence (JS brands the declaring constructor alone).
- '#x in obj' lowers as instanceof-the-declarer: static folds at/below and disjoint, the vtable interval test when the declarer sits below the receiver's static class, and unions of decided class arms collapse to tag tests; unknown/generic/static brands fence with pointed hints.
- util.inspect omits private fields exactly like Node (an all-private class prints 'C {}'); corpus 2450-2457 pin every form differentially on both backends, diagnostics private-members pins the fences.
Chris Tate