镜像站点 · 本页由第三方 GitHub 只读镜像提供,非 GitHub 官方站点,不接受任何登录或凭据输入。前往 github.com
Skip to content

VM: use canonical lowering for parameter expressions and patterns #806

Closed

Description

@chrisbbreuer

Parent: #465. Related: #798, #799, #803, #771, #493.

At f83c419, the parameter-only supportedPlainParameterDefault/supportedPlainParameterPattern whitelist still rejects array/object/RegExp/class literals, optional chains, spread, delete and destructuring assignment even where canonical expression/pattern lowering already exists. The exact-parent 13-subtree paired matrix has 672 explicit parameter-prologue rejections, 124 TypeError results requiring causal diagnosis, and five other mismatches (801 failures / 9,643 scored files). Rejection is not evidence of an unsupported VM opcode.

For example ([[x,y,z]=[4,5,6]])=>x+y+z is rejected in required mode before its existing array/pattern lowering can run. Nested defaults and aggregate parameter defaults reproduce this across arrow/function/method/class forms.

Scope:

  • Make canonical expression/pattern compilation the admission authority for parameter evaluation, retaining stable parameter-prologue rejection for genuinely unsupported lowering and never swallowing OOM.
  • Preserve left-to-right binding initialization, exact undefined default selection, later/self TDZ, separate parameter/body records, live closures/eval, private and super context, inferred names, optional/spread evaluation order, iterator close and abrupt completion.
  • Audit newly reachable existing lowering for real semantic/ownership defects instead of treating removal of a rejection as completion.
  • Add structural/OOM, paired semantic, actual moving-GC and shared no-GIL/TSan checks, with nonzero exact-parent corpus accounting.

No parallel syntax whitelist, blanket acceptance of unsupported nodes, AST fallback counted as bytecode, detached binding snapshots, source rewrite, suppressed failures, disabled GC, or native stack/queue/job/Promise/Worker/Map/Set lifecycle redesign.

Acceptance:

  • Parameter admission follows canonical lowering and reports unsupported/OOM causally.
  • All newly admitted expression families and nested patterns pass exact paired controls without fallback.
  • TDZ/scope/context/iterator/GC/OOM/no-GIL invariants pass targeted and full integration checks.
  • Exact-parent corpus gains/regressions and structural/allocation effects are reported; conventional commit and pushed CI linked separately.

Completed integration (2026-08-28 UTC)

Implementation: f2340f27. The measured local evidence above and in the linked implementation comments remains unchanged.

  • Complete pushed integration: CI 33144524467 at descendant 81978f7b completed successfully: 53 jobs passed, one scheduled-only job skipped, zero failures. This supersedes earlier pending, cancelled, or failed integration reports; those earlier runs are not counted as successful gates.

Closing this implemented issue as completed. Parent-roadmap work and independently tracked follow-ups remain open; this closure does not claim those broader scopes are finished.

Activity

  1. chrisbbreuer commented on Aug 28, 2026

    @chrisbbreuer
    MemberAuthor

    Initial canonical-lowering candidate passes 37/38 broad paired controls (baseline 2/38); its only failure is a separately reproduced parser defect in valid new this.#Ctor(), now tracked as #807. The old parameter whitelist is removed and unsupported/OOM classification is retained at actual parameter compilation. A same-inventory 13-subtree corpus comparison is running. Combined tests additionally cover private construction, 20 actual-moving cases, compiler allocation replay and concurrent shared no-GIL calls. No native activation/job lifecycle changes.

  2. chrisbbreuer commented on Aug 28, 2026

    @chrisbbreuer
    MemberAuthor

    Initial compiler-only exact-parent measurement is complete: identical 9,643 eligible positive upstream files, 8,842 -> 9,633 required-bytecode passes, 791 gains and zero regressions. Gains by subtree: expressions/arrow-function 78, expressions/class 278, expressions/function 78, expressions/object/method-definition 1, statements/class 278, statements/function 78. All 13 default-tier reports remain byte-identical (11,494 passes). Stronger validation exposed prerequisite parser #807 and aggregate ownership #808; combined Debug now passes 47 paired controls, 28 actual-moving cases, shared no-GIL calls and compiler/parser allocation replay. Final combined corpus and full/sanitized checks remain pending; no closure or performance claim yet.

  3. chrisbbreuer commented on Aug 28, 2026

    @chrisbbreuer
    MemberAuthor

    Landed on origin/main as f2340f27, with prerequisite parser d3170836 (#807) and aggregate ownership #808. Chris is author/committer; no trailers.

    Final same 9,643 eligible positive upstream files: required-bytecode passes 8,842 -> 9,633, failures 801 -> 10, 791 gains / zero regressions. Per-subtree pass counts: expressions/arrow-function 172 -> 250, expressions/class 3,165 -> 3,443, expressions/function 147 -> 225, expressions/object/method-definition 197 -> 198, statements/class 3,424 -> 3,702, statements/function 313 -> 391. The other seven required-tier subtrees are unchanged.

    All 13 default-tier reports remain byte-identical (11,494 passes). Additional object 1,170, array 52, new 59 and optional-chaining 38 controls are unchanged too; the object run overlaps the method-definition subtree and is not added twice to a claimed denominator.

    Retained paired controls improve 2/47 -> 47/47; Node independently agrees with every expected result. Debug, ReleaseSafe and unsuppressed TSan pass 47 semantic cases in both tiers, 28 actual-moving cases, shared four-worker no-GIL calls, compiler/parser allocation replay and existing reentrant runtime OOM restoration. Full Debug: 2,042 passed, one platform skip, zero failures/leaks, 604.5 seconds / four shards. Seed-one fuzz: 400 default + 300 verify, zero failures.

    No whitelist, fallback coverage substitution, source rewrite, disabled GC or native lifecycle redesign. The separate Proxy metadata operand audit is #810; these tests do not claim to resolve it. Ten remaining required-tier files remain visible under parent #465.

    All task build caches and intermediate binaries are retired after retaining exact baseline/final binaries, source hashes, witness files and logs in /Users/chris/.cache/zig-js-parameter.jIiaXW. Full pushed integration is still pending; this issue remains open until that gate completes. No performance claim is made.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions