Skip to content

Research: prototype faster selected-record source binding - #2793

Merged
Mister-Meeseeks merged 1 commit into
CrocSwap:mainfrom
Arham-Tahir64:codex/reviewable-result-01a1418b
Oct 16, 2026
Merged

Mister-Meeseeks merged 1 commit into
CrocSwap:mainfrom
Arham-Tahir64:codex/reviewable-result-01a1418b

Conversation

@Arham-Tahir64

Copy link
Copy Markdown
Contributor

The selected source-binding preflight resolves all 1,579 inventory paths a second time when binding the primary proof. Add an opt-in prototype that reuses the canonical paths from the same invocation’s containment checks. Every source is still enumerated and hashed on every call, and both inventory formats retain their alias and proof-membership guards.

Nine alternating fresh-process pairs per interpreter measured complete check(record) medians of 289.6 → 208.8 ms on CPython 3.12.14 (1.39×) and 231.4 → 173.6 ms on 3.14.7 (1.33×) on macOS arm64. Raw samples, exact source/code bindings, scope and reproduction commands are in research/selected-preflight-path-reuse/. This saves about 58–81 ms per preflight on this host; it is not a whole cold-replay measurement or a new exponent.

The production checker and its two byte-pinned consumer audits stay unchanged. Default adoption needs a separate successor-provenance migration for those audits. This package uses explicit manual replay commands and does not register a generic CI replay.

Validation:

  • Both Python versions pass all 25 inherited selected-result, inventory-link and primary-proof groups plus four new groups against each engine: 58 behavioral groups per interpreter, with zero skips.
  • Controls cover same-size source edits with restored mtime, same-content alias retargeting after success, unhashed-manifest proof selection and overlapping roots. Both actual optimized CLIs refuse execution.
  • Original and candidate CLI output agrees on the selected Conditional kappa 0.001672660197076918: rare-pattern T16 exchanges #1094 record and historical Conditional κ = 0.001672188119205421: improved T16 owner packing on the seven-slot k8 word #1042 checkpoint. The original source-binding check and both preserved audits’ metadata/source pins pass.
  • Full selected construction replay, other operating systems and hosted CI were not run. Stable filesystems during each check are an inherited premise; concurrent mutation is outside scope.

Task: revision-5 replay-runtime, refined to selected preflight path reuse. Coordination: Discussion #2787. Started from canonical main 7788b59946ebf3803dc38b6a19b47b4acb27f40b; exact sources and freshness checkpoints are recorded in TASK.json/SOURCES.json.

Credit Douglas Colkitt’s checker, Mihir Srivastava’s PR841 primary-proof binding, yssite’s PR959 symlink closure, and Aaron Wang’s PR1864 source-binding audit and the retained dependency-head audit. Prepared by Arham Tahir with substantial OpenAI Codex assistance; no independent human review is claimed. Apache-2.0.

Final overlap refresh covered 723 open PRs including drafts and 1,434 Discussion records with complete pagination and no warnings, plus local notes. Exact-scope results include the canonical claim and credited predecessor audits; no competing path-reuse runtime claim was found within that coverage.

@Mister-Meeseeks
Mister-Meeseeks merged commit 30825d9 into CrocSwap:main Oct 16, 2026
28 of 61 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants