Skip to content

ci: combine release efficiency fixes with Intel test partitions - #450

Merged
afflom merged 11 commits into
release/v0.4from
dot/449-intel-release-candidate
Nov 10, 2026
Merged

afflom merged 11 commits into
release/v0.4from
dot/449-intel-release-candidate

Conversation

@afflom

@afflom afflom commented Nov 9, 2026 •

Copy link
Copy Markdown
Member

Closes #449. Closes #447.

This is the combined release/v0.4 qualification candidate for #394, #444, #448 and the Intel host partition. It preserves all original contributor histories, including the exact #394 lexer source and generated products. Those PRs stay open until this exact combined candidate is qualified and merged.

Release changes

  1. Retain ci: release Atlas consumers from unrelated shard waits #444's Atlas dependency correction unchanged: cert4 replay no longer waits on unrelated census shards; closure waits only on its real prerequisites. Every shard, artifact and join obligation remains.
  2. Retain test: avoid unused BS-07 mutant preparation on other hosts #448's exact BS-07 source change: do not construct 150 + 1,500 mutants on hosts that never execute them. All native vectors/roots and normative Linux checks remain.
  3. Construct Intel workspace test products once, keep all ordinary/library/meta/future targets and doctests on that runner, and execute the conformance binary in the disjoint lanes of a committed plan. The initial plan has four lanes; correctness depends on complete coverage, not an arbitrary lane count.

The other three host jobs retain their original command and timeout. The exact four supported-host status names remain. The Intel aggregate runs after failure/cancellation and refuses missing, duplicate, foreign, changed, failed or incomplete results. SPEC §8.3/§9.4, RP-13/RP-14/RP-18 and release evidence checks move together.

Construction and evidence

  • Same repository/source tree/run/attempt only; bind native target, checkout path, pinned Rust, profile/features, lockfile, plans, executable bytes/mode, actual harness inventory, invocation and logs.
  • Transfer the conformance executable, measured Mach-O library closure and dependency data from a fresh crates.io Cargo home. No credentials, workspace build cache, cross-run receipt reuse, or foreign path/git/alternate-registry sources.
  • Preserve ignored/should-panic semantics and fail-closed inventory coverage, including future tests and targets.
  • Continue producer tests after a failure and still execute all runnable conformance lanes; the aggregate reports failure.
  • Keep the 240-minute execution bound. No skipped test, reduced corpus, higher timeout or larger runner.

Verification completed

  • Independent review of the integrated source and final macOS path/loader delta: no remaining source-level blocker.
  • 25 focused behavioral tests, including real fake-tool subprocess construction → every consumer → join, and producer-failure aggregation.
  • 60 complete-script fake-GitHub evidence fixtures across ci-plan/release-gate.
  • Additional archive traversal/link/mode/collision, dylib alias/transitive closure, inventory and source/run/environment mismatch probes.
  • YAML and shell syntax checks; parsed comparison confirms unchanged triggers, permissions, concurrency, approved Atlas jobs and other host commands.
  • All ten final remote blob identities read back and matched reviewed local files at head 25be1d8.

Qualification still required

The editing executor has no Rust compiler or Intel runtime. Draft CI is development evidence only. Rust formatting/compilation/conformance, real Intel Cargo/libtest/Mach-O execution, artifact transport, offline nested builds and measured timing remain pending. No runtime improvement or timeout resolution is claimed yet.

After draft checks and final-head review, run one complete acceptance gate on the combined candidate, then integrate into release/v0.4 and carry the landing forward through existing #405 ownership. Do not rerun #444/#448 independently. Preserve #442's count-writer design when reconciling the documented BS-07 assertion hunk.

Final intended release union

Commit fddf2e8017aa9ca2e3478143b690a31be21112af, tree 0c24b969dcc5c02d9afb7616f97736827b7b09e6, retains ordered parents a06facbff76e6625da500d1572ba500e7fb2fc19 (the reviewed efficiency candidate) and 1062e3951990a8ce9f9885085834fd898eef6c4c (the reviewed #394 lexer landing).

Independent integration review and complete remote readback preserve all 10,903 entries, their modes and eight symlinks. The only shared path is VERIFICATION.md: the exact 7,343-byte #394 historical appendix follows #450's document. No compiler-semantics or project source-ID input changes, product rewriting, new performance measurement or repeated producer phase was introduced. All #394 products and measurement caveats retain their original scope.

This union makes the necessary complete qualification cover the intended release tree once. Earlier #394 and draft #450 results remain evidence for their original trees. The prior #450 draft run 42697157078 completed successfully (plan, honesty and rest); its reproducibility run 42697157402 also passed before this branch advanced. The final union still needs its own complete hosted gate and reproducibility, including real Intel construction, transport, offline nested Cargo, every conformance lane and the aggregate. No failed or skipped gate is accepted. Keep #394/#444/#448 open until the actual combined landing and tree equality are verified; #373 closes only after its combined proof/BS02 and full acceptance requirements pass. #405 remains the forward-integration lane.

humuhumu33 and others added 9 commits November 2, 2026 21:40
stepLexAt, stepLexToken, stepLexIdent and scanSegmentsStep bound every
candidate scan with a strict let before branching, so block comment, string,
raw string, line comment and quoted name scans ran at positions that never
used them. Each scan now sits in the branch that reads it. The packages,
records and PERFORMANCE.md row are regenerated in a later commit.

Part of #373.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The scans in TokenAudit.lex.tex moved into the branches that read them, so
the rendered lex and audit_tokens packages, bootstrap's build oracles and
coverage maps, and its normalized verification records change with them.
Every rewrite is under bootstrap/ and tests/production-export/bootstrap/;
no language identity moves and bootstrap/lexlean.lock is unchanged.

Regenerated on WSL2 with check-golden --write, check-production-export
--write and verify-examples --write (all projects but uor-atlas), then
check-golden and check-production-export pass without --write.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The bootstrap_perf rows for lex and audit_tokens at 343,104 B, measured
through the harness at d34c1fe with LEXLEAN_PERF_MAX=400000: lex takes
28.8 ms against 680,088 ms before and grows at 1.01 against 2.04;
audit_tokens takes 25.8 ms and grows at 0.94. The sections that called the
lexer quadratic, the stage 1 projection and the "faster first" list now
record it as fixed.

Closes #373

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Retain the contributor lexer changes and actual #393 landing. Refresh the
bootstrap products with the official export, golden and verification writers.
Record independently reviewed producer evidence; final combined CI is pending.
Preserve actual281 contributor/history and actual4dc release ancestry.
Reconcile only demonstrated golden source/build envelopes and retain
reviewed export/verification products with additive scoped evidence.
Integrate the reviewed Atlas prerequisite scheduling from #444 and the
BS-07 host-preparation hoist from #448 into one release/v0.4 candidate.
Preserve complete host coverage, exact evidence custody, and failure
aggregation without increasing the host timeout.

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Independent integrated source review found no remaining source-level blocker on this exact candidate; all ten remote blob identities matched the reviewed local files. The final helper SHA-256 is 35be55f14880175761ddcf9217195498c13f5c8565d2fd7820941015dd85ceb8. Validation includes 25 behavioral tests (full fake-toolchain construction/consumer/join and producer-failure paths), 60 complete-script evidence fixtures, archive/dylib/path negative probes, and unchanged Atlas/other-host YAML comparisons. This is a source-review comment, not acceptance or approval to merge: actual Rust/libtest, Intel Mach-O loading/transport/offline nested Cargo and full exact-head gate remain pending.

afflom commented Nov 9, 2026

Copy link
Copy Markdown
Member Author

Draft run42696143843 passed all 25 factory tests on GitHub, then stopped at cargo fmt --check before expensive jobs. Commit a06facb applies all ten exact pinned-rustfmt hunks in repository.rs, with string/comment and non-formatting token equivalence checked. It also disables Python bytecode writes only for the honesty test step to keep the checkout clean. No acceptance logic changed, and no full gate has been triggered. Separate reproducibility passed on the prior head; the new exact-head draft checks must establish their own result.

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Independent bounded source/acceptance-contract review of exact a06facb found no concrete source blocker across all ten changed paths. The actual conformance inventory must form complete disjoint lanes, including future tests and ignored semantics; other harnesses and doctests remain on the producer. Construction/consumer/join receipts bind source, same run/attempt, target/toolchain, plans, executable and dependency bytes, invocation and results. Failure aggregation, all four exact supported-host contexts, the other three host commands/timeouts, and every Atlas shard/join obligation remain intact. The BS-07 change preserves all-host native vectors and the complete normative mutant/plant path; #442's assertion/count-writer ownership stays separate.

The exact-head draft run42697157078 has passed plan and honesty (including factory tests and pinned formatting); rest is still active. Repro42697157402 passed. These are development signals: real Intel construction, Mach-O transport/loading, offline nested Cargo behavior, complete test execution and the final full gate remain unqualified. No runtime saving or timeout fix is established by source inspection or fake-tool fixtures.

#450 is the declared combined carrier for #444/#448/#449; keep the original PR histories open and do not rerun them independently. Before promoting the final carrier, refresh the live release/base and account for the concurrently qualifying #394 tree, so the one complete qualification covers the actual intended landing. Preserve #405's forward integration lane. This COMMENT is source review only; no readiness transition, new gate or merge is performed here.

@afflom afflom mentioned this pull request Nov 9, 2026

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Exact-head source-review continuation: compared a06facb against independently reviewed25be1d8. Only the ten pinned-rustfmt corrections (verified strings/comments and non-formatting token equivalence) and the honesty Python step's PYTHONDONTWRITEBYTECODE=1 environment changed. No acceptance behavior, artifact custody, test selection or helper bytes changed. Honesty and separate reproducibility now pass on this head; rest remains in progress. No source-level blocker found. Full actual Intel qualification and complete acceptance are still required before merging.

Preserve the exact #394 source and generated products together with #450's
Atlas scheduling, BS-07 host preparation and Intel construction changes.
Retain both contributor histories and historical verification scopes.

No producer regeneration is needed: the only shared path is the additive
VERIFICATION record, and the CI changes do not alter compiler or project
identity inputs. Complete combined-head hosted acceptance remains required.
@afflom
afflom marked this pull request as ready for review November 9, 2026 22:14

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The intended release union is independently reviewed at fddf2e8. Actual remote tree 0c24b969dcc5c02d9afb7616f97736827b7b09e6 and all 10,903 path/mode/type/blob entries match the reviewed inventory; eight symlinks and both exact parent histories are retained. Relative to #394, the nine other #450 paths remain exact and VERIFICATION.md appends #394's exact 7,343-byte historical record to #450's version. Neither compiler-semantics nor project source-ID inputs change, so existing products remain exact without repeated producer work.

The prior a06facb draft run 42697157078 and reproducibility 42697157402 finished successfully before publication. They do not accept this union. Marking this final candidate ready requests one complete exact-head gate, including genuine Intel construction, transport/offline execution, every conformance lane and the aggregate, plus all original recipe/Atlas/host obligations. No complete gate or runtime improvement is claimed yet. #394's existing run is untouched; its prior-tree outcomes cannot replace final-union acceptance. No merge or issue closure is performed.

@afflom
afflom marked this pull request as draft November 9, 2026 22:19
@afflom
afflom marked this pull request as ready for review November 9, 2026 22:19

afflom commented Nov 9, 2026

Copy link
Copy Markdown
Member Author

The complete qualification is now run 42710839907 on unchanged fddf2e8. Its completed plan records DRAFT=false and checks out synthetic merge bf0ed1a, whose tree equals the reviewed candidate 0c24b969dcc5c02d9afb7616f97736827b7b09e6.

GitHub delivered the synchronize event after the first readiness event: native concurrency cancelled run 42710137070, and the delayed run 42710413564 captured DRAFT=true. A fresh readiness transition corrected that ordering before the full fanout. Both earlier runs are cancelled and supply no acceptance. Final-head reproducibility 42710413660 remains intact. No source changed, completed producer repeated, or old-head evidence reused.

afflom commented Nov 9, 2026

Copy link
Copy Markdown
Member Author

Genuine Intel qualification passed in run 42710839907, attempt 1, on fddf2e8 / tree 0c24b969. Construction executed 349 tests across 13 harnesses plus workspace doctests. The four native lanes passed all 478 inventoried conformance cases exactly once: lower 1, print 11, validate 85 and moves 381. Aggregate job 129520804964 passed.

Independent artifact review verified the archived x86_64 Mach-O executable, native Rust/Cargo 1.97.1, construction/archive/member hashes, every recorded producer/lane log hash and exact test-name coverage. Producer, lanes and join retain one construction digest and the same run/attempt/source tree. Each lane records its isolated bundled Cargo dependencies and CARGO_NET_OFFLINE=true; actual cases passed there, while logs do not separately trace every nested subprocess.

This discharges actual Intel execution, not just the earlier fake-tool fixtures. At this check, the full gate is 48/50 jobs green; Windows and bootstrap-package remain active. Reproducibility 42710413660 has passed. No merge is performed until the remaining full-gate results, final head/base/review checks and actual landing-tree verification.

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Final acceptance passed on fddf2e8. Full CI https://github.com/UOR-Foundation/LexLean/actions/runs/42710839907 completed successfully with exactly all 50 expected jobs and all 552 listed steps successful. Exact-head reproducibility https://github.com/UOR-Foundation/LexLean/actions/runs/42710413660 also passed. The independently reviewed genuine Intel evidence covers all 478 conformance cases, the producer's 349 tests and doctests, native Mach-O transport, offline dependency execution and the complete aggregate.

Fresh release/v0.4 is 4318b17. Current synthetic merge ec6a555 retains that base and this head, with tree 0c24b969dcc5c02d9afb7616f97736827b7b09e6 exactly equal to the reviewed and fully tested tree at bf0ed1a. No unresolved review threads or blocking reviews were found. No timeout, coverage, protection or security setting was weakened.

This is the final acceptance record, not a claim that all CI overhead is solved. The complete run took about 2h57m, dominated by BS-15 at 10,304.13 seconds; further optimization must preserve that contract. Merge by the normal PR flow, then verify the actual landed tree before closing #447/#449 and forwarding through #405.

@afflom
afflom merged commit cf76090 into release/v0.4 Nov 10, 2026
53 of 81 checks passed

afflom commented Nov 10, 2026

Copy link
Copy Markdown
Member Author

Merged and verified: release/v0.4 now points to cf76090, with parents 4318b17 and fddf2e8. Its actual tree is exactly the fully tested 0c24b969dcc5c02d9afb7616f97736827b7b09e6. #447/#449 are closed completed; original #444/#448 histories are retained and GitHub reports both merged. The accepted input and remaining integration obligations have been recorded in #405.

afflom commented Nov 10, 2026

Copy link
Copy Markdown
Member Author

Release-plan alignment: this delivered CI change and its original #444/#448 carriers and #447/#449 issues are now assigned to milestone 3, LexLean v0.4.1, the next planned patch after already-published v0.4.0. The accepted code is already in that milestone’s release/v0.4 line at cf76090. This is not a claim that v0.4.1 is tagged or published. #394/#373 retain their existing v0.4.2 milestone; unrelated release scope is not relabeled. Forward inclusion in v0.5 remains required through #405 and has not yet been verified.

humuhumu33 added a commit that referenced this pull request Nov 10, 2026
No bootstrap source changes, so nothing regenerates.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
humuhumu33 added a commit that referenced this pull request Nov 10, 2026
#450 maps every BDD part to one Intel lane; bootstrap-structural-grammar
joins the print lane with the other light bootstrap parts.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
humuhumu33 added a commit that referenced this pull request Nov 10, 2026
…rint lane

#450 maps every BDD part to one Intel lane; bootstrap-lock-text joins the
print lane with the other light bootstrap parts.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
humuhumu33 added a commit that referenced this pull request Nov 10, 2026
The conflicts are generated bootstrap files only, taken from the line and
regenerated in the commits that follow. The three surface parts join the
Intel print lane that #450 requires of every BDD part.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
humuhumu33 added a commit that referenced this pull request Nov 11, 2026
The conflicts are generated bootstrap files only, taken from the line and
regenerated in the commits that follow. The surface tooling part joins the
Intel print lane that #450 requires of every BDD part.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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