Repository navigation
Reduce repeated host conformance work and bound scaling to Linux (#235) - #400
Conversation
|
Post-change Intel evidence is now complete at head Independent source review confirms the retained host corpus/limit checks and unchanged normative Linux stress coverage; the tested merge |
afflom
left a comment
There was a problem hiding this comment.
Final review cleared after independent Sol 6.1 source review and measured Intel evidence. CI 41592896657 now passes all 42 jobs, including the final bootstrap-package leg; honesty and reproducibility also pass. The tested tree is 354438f4d6f7e48268ae3e50885c5f6a96581ac4. A clean merge against current v0.4 b37bd79 produces tree e02d1637615872ccf1eff1588cb57b1b9fda3ce6; the complete difference from that tested tree is the already reviewed and fully green #397 documentation (README, SPEC §21.5, docs/LEAN-CONSUMER.md). No code, claim, fixture or CI conflict resolution is required. Intel actual job time fell from 144m02s to 117m01s; conformance time fell 19.41%, with all 468 cases passing. Normative stress and supported-host corpus/limit checks remain intact. Cleared to merge and close #235.
The Intel macOS host gate took about 144 minutes in successful run 41478333299 at
afe1a002; its 468 conformance tests took 7,470.84 seconds, with SP-02, TC-03 and TC-05 forming the tail.This keeps the full SP-02 scaling/seeded stress families and TC-03 seeded argument/fuel sweep on normative Linux x86-64. Every supported host retains the committed corpus/fixture assertions, negative certificate-limit cases and fixed small certificate vectors. TC-03 and TC-05 now render and parse the shared fixture modules once into a declaration-name set, preserving every name-membership and template assertion. The register, scenarios and SPEC state the host split; CONFORMANCE.md was regenerated. No acceptance recipe, build cache, additional macOS shard or kernel replay is introduced or removed.
Implements the measured CI-cost reduction for #235. The remaining acceptance item—the new measurement in the workflow timeout comment—is carried by #406, which will close #235 after its final gate passes.
Measured Intel result: job 125773640497 in run 41592896657 passed in 117m01s, versus 144m02s for baseline job 125390659100/run 41478333299 (27m01s less, 18.76%). All 468 conformance cases passed in 6021.07s versus 7470.84s (19.41% less). The 240-minute timeout is unchanged.
Validation on the final source:
Based on release/v0.4 at
90fb7f152. Published tree354438f4d6f7e48268ae3e50885c5f6a96581ac4matches the locally reviewed and validated commit tree.Merged as 477a508. The actual merge tree e02d1637615872ccf1eff1588cb57b1b9fda3ce6 matches the reviewed integration: the complete difference from the tested source tree is the independently reviewed and fully passing #397 documentation. Forward propagation is tracked by #404/#405.