Tags: cortexkit/aft
Tags
fix(test): deflake watcher pending-ignore replay on loaded macOS runners search_pending_replay_skips_file_that_becomes_ignored panicked when the watcher never delivered/settled the .aftignore refresh within 15s on a loaded macOS release runner (FSEvents is throttled/async). The actual regression under test — an ignored pending replay must not re-index the secret — is asserted by the total_matches==0 check that only runs once the index settles; non-delivery proves nothing, so skip it (mirroring the trigger-not-observed skip earlier in the same test). Also commits the Cargo.lock 0.38->0.39 workspace bump release.sh regenerated but did not stage, so the retag has a consistent lockfile. Co-authored-by: Alfonso <[email protected]>
fix(aft-tokenizer): restore claude_data.rs truncated by a concurrent … …build.rs write A killed-but-detached cargo test's build.rs regenerator was mid-write on claude_data.rs when the release chain's git add -A snapshotted it, committing a file cut mid-tuple. Restored content is byte-identical to the vendored original (c279ee3). Atomic-write hardening for the regenerator follows separately. Co-authored-by: Alfonso <[email protected]>
test(pty): barrier on REPL prompt before writing in python round-trip Cold Windows CI runners need >5s for shell+python startup; writing before the REPL exists flaked two release runs today. Wait for the >>> prompt (30s) before the write, then read with a 10s budget. Co-authored-by: Alfonso <[email protected]>
test(callgraph): deadline-based retry for async build test — fix Wind… …ows CI flake 50x100ms fixed window was too tight for Windows debug runners under full-suite load; 90s deadline keeps regression detection. Co-authored-by: Alfonso <[email protected]>
PreviousNext