You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Avoid unused BS-07 mutant preparation on non-normative hosts #447
In crates/conformance/src/cases/bootstrap/rust_lower.rs at release/v0.4 4dc5b803, BS-07 generates and oracle-lowers 150 evaluator mutants and 1,500 native-Rust mutants before its existing non-normative-host return. Those hosts execute neither mutant corpus. Each generated vector computes typing sites, lowering, fallibility and four oracle outcomes.
Full CI run 42647185072 completed 44 of 45 jobs successfully, including all Atlas shards and their join. The Intel job was cancelled at the 240-minute limit while tests were still passing. BS-07 spent at least 159 minutes in this contended run. That measurement does not attribute the whole duration to mutant preparation.
Bounded fix
Move the two independently seeded mutant generators and their corpus-count checks below the existing normative-host guard. Preserve:
22 rule programs, all 350 base vectors and the 292-vector evaluator-size classification;
the four-root native differential over all base vectors on every host;
both exact seeds, the 150/1,500 corpus sizes, all evaluator/native mutant checks, planted defects and Lean-backed certification on normative Linux.
No timeout increase, cache, identity regeneration or reduction in executed differential vectors is part of this fix. A broader host construction/partition design is separate work.
Verification and coordination
Review the control flow and count ownership; check source compilation/formatting and the unchanged native/normative execution paths. Quantify preparation separately before claiming a runtime saving, and retain full final-tree acceptance before merge.
The assertion block also changes in #442's count-report tooling. Keep that PR's ownership and integrate the small assertion split explicitly; do not create a competing count writer. This is a new observed preparation defect, not a reopening of the completed #235/#400/#403 changes.
Completed in #450, merged into release/v0.4 as cf76090. The actual landed tree is 0c24b969dcc5c02d9afb7616f97736827b7b09e6, exactly the reviewed and fully tested tree. Full CI https://github.com/UOR-Foundation/LexLean/actions/runs/42710839907 passed all 50 expected jobs; exact-head reproducibility https://github.com/UOR-Foundation/LexLean/actions/runs/42710413660 passed. Independent Intel receipt review verified actual native execution and complete coverage of all 478 conformance cases. No gate, test coverage, timeout or protection was weakened. Forward integration remains in #405.
Activity
afflom commentedon Nov 10, 2026
Completed in #450, merged into release/v0.4 as cf76090. The actual landed tree is 0c24b969dcc5c02d9afb7616f97736827b7b09e6, exactly the reviewed and fully tested tree. Full CI https://github.com/UOR-Foundation/LexLean/actions/runs/42710839907 passed all 50 expected jobs; exact-head reproducibility https://github.com/UOR-Foundation/LexLean/actions/runs/42710413660 passed. Independent Intel receipt review verified actual native execution and complete coverage of all 478 conformance cases. No gate, test coverage, timeout or protection was weakened. Forward integration remains in #405.