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

Environments: elide binding read locks for private activations #533

Closed

Description

@chrisbbreuer

Motivation

The shared-realm transition publication benchmark in #531 attributes a material part of per-edge synchronization to declarative Environment reads. Fresh tree-walker function activations are owned by one running mutator until a closure captures them, while the concurrent marker only reads their tables. Taking binding_lock for every identifier read therefore adds atomic traffic without protecting against a mutator writer.

Split from #532 so object descriptor synchronization and Environment ownership remain separately reviewable. Parent performance work: #531. Related shared-realm synchronization work: #478, #480.

Requirements

  • Establish an explicit private-activation invariant for fresh tree-walker call environments and nested declarative scopes.
  • Revoke private status before any closure can publish the environment or an ancestor to another mutator.
  • Keep all binding writes synchronized with concurrent marking; only owner-mutator reads may elide the lock.
  • Root, module, captured, VM, suspended, and otherwise shared environments remain on the locked path unless independently proven private.
  • Preserve lexical lookup, aliases, TDZ, const/function-name metadata, direct eval, with, catch, per-iteration bindings, and closure semantics.
  • Do not weaken GC relocation, concurrent marking, or no-GIL publication ordering.

Acceptance

  • Focused tests prove zero Environment lock acquisitions for private reads and immediate locked fallback after capture.
  • Shared closure and concurrent-GC tests pass under no-GIL and TSan.
  • Full unit suite and relevant test262 scopes have zero regressions.
  • Contention attribution records the before/after Environment lock counts under the Shapes: scale shared transition publication beyond two lanes #531 workload.
  • Any public performance claim follows the checked-in benchmark-evidence workflow; counter reductions alone are diagnostic, not completion evidence.

Activity

  1. chrisbbreuer commented on Aug 11, 2026

    @chrisbbreuer
    MemberAuthor

    Implemented and pushed as perf(environment): elide private activation read locks (#533). Exact evidence: nested private/capture unit proof passed; shared-closure and concurrent-marker tests were TSan-clean; threadfuzz ran 400 deterministic programs with 0 failures; full units were 1,587 passed / 1 skipped / 0 failed / 0 leaked. Exact ReleaseFast test262 test/language remained 23,648 pass / 0 fail (0 flips). Diagnostic attribution retained checksums 5,009,000 and 18,026,000 while invocation Environment locks fell 376,876 -> 36,299 at 4 lanes and 753,608 -> 72,531 at 8 lanes. This is a counter result, not a public throughput claim.

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