Repository navigation
Tooling: filter default reclamation candidates before sorting - #797
Conversation
|
100-PR batch review at This remains open after scoped triage. Next review step: Production compiler change and regenerated historical certificates require code-diff/regression review. This is a specific continuation point, not a rejection or a request to redo already recorded checks. Completed local evidence is retained separately from outstanding whole-head obligations. Batch review, attribution and evidence. Thank you for the contribution. |
|
Addressed the compiler and historical-certificate review request in 77e20f74. The follow-up checks the predicate/key mutation boundary and confirms that only four source-binding fields changed in the three historical JSON files. The default compiler is still byte-identical to the retained reference on main Fresh sequential compilations at |
|
Batch review at Merged to Reviewed scope: Filtering pure frame-containment eligibility before the same stable reclamation sort preserves both literal h23/h25 words and typed receipts. Independent paired benchmark passes; only historical derived source pins change, not numerical witness data. The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions. Batch review, attribution and evidence. Thank you for the contribution. |
2 participants
Filter retired slots by exact frame containment before sorting them in the joint compiler. On the same historical PR #55 inputs, three fresh alternating-order benchmark pairs give median complete-compilation speedups of 2.36× at h=23 and 2.69× at h=25. All 12 runs reproduce the accepted decompressed physical words and complete compiler receipts exactly.
Review files: Proof and scope · Evidence · Source pins and credit.
Status: verification tooling improvement. No new κ or physical word.
The production change is two lines. The eligible slot order remains descending current frame rank, then slot ID. Filtering commutes with sorting here: the predicate and key only read frames, and frame mutation begins only after the chosen dependence is found, followed immediately by return. The package gives this equivalence argument, retains the exact original compiler for comparison, and supplies a source-pinned benchmark with no timing threshold.
These measurements include complete scalar compilation, both input/dirty orientations, source admission and serialization on Python 3.14.7 / macOS arm64. They exclude setup copies, the independent CRT profiler and the full suite. OS caches were not flushed; the development machine was not an isolated benchmark host. They are not integer-multiplication runtime measurements or latest-frontier replay timings. The filter can evaluate more containment predicates when the old loop would stop early, so there is no universal speedup claim.
Use a new output path. Expected:
PASS_IDENTICAL_RECLAIM_WORDS_AND_RECEIPTS.All 18 required verification groups passed across the initial run and resumption. The initial
make verifypassed the first 14 groups, then a historical positive-skip build could not find Boost. A retry refused the output directory left by that failed attempt. After preserving that directory and supplying GCC's standardCPLUS_INCLUDE_PATH,make verify-positive verify-joint verify-pair verify-testsexited 0. No source change or omitted check was needed. The complete selected cold replay, original C++ CRT profiles, 47 strict inequalities, seven margins, all 16 joint controls, isolated tests and patch checks are covered. No uninterruptedmake verifypass is claimed. Exact attempt log hashes and environment details are in VALIDATION.json.The four benchmark admission controls pass. After timing, the receipt comparison was tightened to preserve JSON types; all 12 timed receipt hashes match retained bytes that pass the typed comparison. The timed commands and compiler inputs did not change. No remote-CI or proof-assistant claim is made for this branch.
Only the compiler source hash and derived historical certificate bindings change. All numerical certificate fields, word payloads, profiles and existing mutation tests remain unchanged. There is no new cache, compiler mode, dependency or test-only interface. upstream/ and the historical selected PR #626 / PR #543 frontier at the original base are unchanged; all inherited compiler, common-label, physical, hosting, precision, tape and analytic assumptions remain.
Base: 66849e1475c930fd7732b1a0067ffee7e1e95f60. OS-random seed 85, verification-engineering track. No pinned sprint coordination issue existed, so the three candidates and narrow claim are recorded locally and in CLAIM.json.
Credit @gupt1156's PR #55 producer, @eumemic's PR #57 compiler and Chafik Boukhalfa (@chafreaky)'s ranked reclamation. PR154, at
f571bcbd6e06618aeae0b8279da48060032dc74b, identified this sorting bottleneck and explicitly left it unchanged. This change follows that lead and supplies exact-output and performance evidence. The measured PR #55 source is pinned at 03e991aa79f9df1a935726213bb6a8631cd0d823; full source and dependency hashes are retained in SOURCES.json. Other frame/arithmetic/profiler acceleration work is credited in the package.The final overlap review covered 614 open PRs through PR #795. Concurrent PR779, Peter Carpenter (@PeterCarp320), at
daa9b0e773b53f36d5b4c3f1c7524600e5714deb, maintains exact eligibility indexes in a separate adapter and reports larger speedups. It follows Chafik Boukhalfa (@chafreaky)'s PR175 (f1081fabbfadaaaaf7292a40d428dbbb865ce920), which maintains sorted rank buckets and is incorporated in a historical source archive. The default compiler at this base still uses the full sort. This PR is a two-line option for that default path, with no maintained index, update hook or adapter, plus exact-output and integration evidence. It claims neither priority nor best performance. PR #175, PR #779 were not benchmarked on this machine; their code is not imported or changed.Prepared by Ali Romman (@aliromman) with substantial OpenAI Codex assistance. Apache-2.0. Finite certificate checks are not a formal proof of the full multiplication theorem.
The 2026-10-14 follow-up review addresses the maintainer's compiler/regression request. Fresh sequential compilations at both
h=23andh=25reproduce the complete accepted words and type-preserving receipts. The receipt records the four changed source-binding fields and confirms that the original compiler is unchanged on refreshed main418e40eb2572d930b58c1cf9b82b41c0fc7ee08d. The implementation, source pins and mathematical certificate fields needed no correction. The original timing runs, mutation suite and full-suite evidence were retained; no new timing comparison or full-suite run is claimed.