Repository navigation
Gate expensive CI fan-out on existing honesty checks - #413
Conversation
afflom
left a comment
There was a problem hiding this comment.
Maintainer source review, with an independent Sol review of the implementation and focused evidence, is complete. No blocking source finding at 1a5f87a; reviewed tree 7c3f96bb9299793f1d95f6201790bd629e01c25c matches the published head. The final delta after independent implementation review only clarifies the CI critical path and Cargo target selection in prose.
The existing honesty target is moved, not copied: actual harness listings prove 810 = 296 ordinary + 514 BDD tests, with the 23 BDD parts disjoint and complete. All three expensive roots require same-run honesty, and the existing check context is retained. Local full-suite and supported-host coverage remain. Both evidence consumers reject missing/failed/skipped honesty; large successful listings and late failures exercise the real shell paths. The eight-step plant fails the unchanged meta assertion, and the restored initial row, RP-10/13/14/18, library tests and Clippy passed. No identity input changed or artifact writer ran.
This is conditional source review, not merge approval or complete acceptance. Keep draft while #403 runs. After #403 lands, retarget to release/v0.4, review any integration delta, and require the complete final-head gate and merged-tree evidence before merging and explicitly closing #411. No aggregate runtime saving or post-change hosted measurement is claimed.
|
Corrected the hosted matrix naming defect in 80dab87 (tree 985dc31b33f557c16f66b3e4c7abc3c9fd22c392). The first draft run, 41639924023, passed the preflight but emitted RP-14 failed with the strengthened check and old static declaration, then passed after the correction; RP-18 also passes. Both existing evidence consumers additionally refuse the observed suffixed name, and their exact-match predicates remain unchanged. Focused Clippy, formatting and whitespace checks passed under the validation lock. The PR remains draft and stacked on #403's unchanged head 39399a2. Retargeting after that dependency lands and a complete final-head CI gate are still required. No full-gate rerun or merge was requested. |
afflom
left a comment
There was a problem hiding this comment.
Reviewed the hosted-name correction at 80dab87. The workflow now derives its name from matrix.part; the exact acceptance predicates remain unchanged. RP-14 rejects the old static declaration, and RP-14/RP-18 reject the observed suffixed job name. Focused red/green evidence and Clippy passed.
I independently checked hosted run 41643357309: job 125942946218 is now named exactly honesty and has concluded success. The draft rest job is running; hosts/Atlas are appropriately absent from draft acceptance. This resolves the naming defect missed by the earlier source-only review. Full final-head acceptance after #403 lands and retargeting to release/v0.4 is still required; this remains a COMMENT review, not merge approval.
|
The corrected head 80dab87 has now completed its intended draft run: CI 41643357309 passed plan, the exact This confirms the actual hosted check name and draft execution after the matrix-name fix. It is not full acceptance, a host-runtime comparison, or merge clearance. Keep this PR draft and stacked until #403's new-base integration lands, then retarget to release/v0.4, review the integration delta and run the complete final-head gate before closing #411. |
|
Prepared and independently reviewed the source integration with #403's now-published BS-20 head215a9d591. Only the VERIFICATION append conflicted; both records remain. The six CI implementation files are unchanged from80dab877. The combined plans retain BS-20's bootstrap-tooling-latex shard, so they now have24 parts:22 named ID groups, whole-target honesty, and dynamic rest. All186 explicitly assigned IDs are unique. The earlier810-test/23-part figures are historical measurements, not fixed assertions. This remains local preparation; the PR head, base and draft state are unchanged while #403's final gate runs. After #403 lands, retarget this existing PR, publish the reviewed integration, measure the actual inventory once with the focused model/meta/RP-13/14/18 checks, and require complete final-head CI. No new identity wave or duplicate full local gate is needed. |
afflom
left a comment
There was a problem hiding this comment.
Reviewed final release integration at tree5868cd66f765591c89583023be5247fb2fa07b48. It is byte-identical to the independently reviewed prepared integration; the new merge adds actual landed #403 ancestry only. All six CI implementation blobs retain the previously reviewed content. Static planning retains 24 parts and186 unique explicitly assigned IDs with a dynamic remainder, not a literal test-count gate.
The final-head hosted preflight and complete gate are the authoritative next validation, including RP-13/RP-14/RP-18 and dynamic inventory; no redundant local build/test phase is claimed while Host regeneration holds the shared lock. Required checks and same-run evidence binding are unchanged. This self-authored review is a COMMENT, not self-approval, and does not authorize merge before the complete final-head run passes.
afflom
left a comment
There was a problem hiding this comment.
Final-head evidence is complete: CI41696659763 passed all 44 jobs, including the same-run honesty preflight, supported hosts and full normative partitions; repro41696591207 passed. Rest126131363605 measured the disjoint complete 812-test inventory (296 ordinary + 516 BDD across 24 parts) and passed RP-13/RP-14/RP-18. No reduced checks or aggregate savings claim is involved.
Fresh metadata/comments preserve the reviewed non-draft head96d4bf0 and base5a44fe49. Head and synthetic merge cbe8fcf both have tree5868cd66f765591c89583023be5247fb2fa07b48; synthetic parents are exactly5a44+96d4. The source review and final integration review remain applicable. This self-authored COMMENT records merge clearance, not self-approval. Proceed with the expected-head merge, then verify the actual merge tree and explicitly close #411 on this non-default release branch.
An eight-step VR-11 scenario failed the existing honesty check, but expensive Intel and Atlas work still started independently. This change makes those existing checks a prerequisite of expensive CI fan-out.
The initial same-run
honestyjob preserves the check context and owns formatting, model/specification audits and the complete BDD meta target. The BDD remainder excludes that target by Cargo target ownership. Localjust vvand supported-host coverage remain complete. Parent-reuse and release evidence refuse missing, failed or skipped honesty. Both job-list consumers read complete listings, preserving late-failure refusal without producer SIGPIPE.The PR now targets
release/v0.4after #403 landed. Final head96d4bf0bb22070aa044a2c9e012b3166d2fad91dcontains actual release merge5a44fe49e5d73569c3bb377bb28b56e463ee0547; tree5868cd66f765591c89583023be5247fb2fa07b48is byte-identical to the independently reviewed prepared integration. The final ancestry merge changes no files, and the six CI implementation files retain their reviewed content.Validation and limits:
Original failure evidence: honesty 41605700268/job125814846351 versus CI41605700263. Later failures on other heads are not attributed to that eight-step scenario.
Closes #411.
Final acceptance: CI41696659763 passed all 44 jobs and repro41696591207 passed on head96d4bf0. Merged into release/v0.4 as5393f5528ec46a3fb0f2884a81c0c7087f1de95b; actual merge tree5868cd66f765591c89583023be5247fb2fa07b48 equals the tested/reviewed head and synthetic merge. Issue #411 explicitly closed with evidence in comment6616436387. Required forward integration into v0.5 remains tracked by #405/#404.