ci(test.yml): fold coverage into ubuntu-24 lane, add scripts/ floor, single lint step

- Coverage gate (test:coverage:unit) now runs inside the ubuntu/24 full lane;
  the standalone coverage job duplicated that lane's entire unit run (~4 min
  of runner time per PR). required-tests gate updated accordingly.
- New second-tier floor: c8 check-coverage --lines 55 over scripts/**
  re-slices the same V8 data (measured 65.95%) — the CI/release/lint tooling
  was previously enforced at 0%.
- Coverage artifact now excludes coverage/tmp (>1 GB of raw V8 dumps).
- lint-tests runs npm run lint:ci — one orchestrated step, identical set
  locally and in CI; drops the no-op eslint --cache flag (CI never restored
  the cache directory).
- Delete unreferenced scripts/run-cross-platform-tests.cjs (+ its test);
  document the mutation UNMUTATED blind spot (~48% of lib lines) in
  stryker.config.mjs.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Colin
2026-06-09 23:03:14 -04:00
parent 622e4be6d8
commit a869df2acf
2 changed files with 46 additions and 64 deletions

View File

@@ -29,6 +29,15 @@
// force a full tsc rebuild per mutant — far too slow for the 30-min CI budget.)
// Large/low-coverage modules are excluded (the command's test set does not
// exercise them, so they would only ever produce survived mutants).
//
// KNOWN BLIND SPOT (2026-06 CI audit): this list excludes ~14.2k of ~29.8k
// lib lines (~48%), including the most central modules (state, core,
// commands, phase, verify). Mutation results therefore speak only for the
// well-tested half of the lib. Shrinking the list is deliberate tracked work:
// bring one module into scope per release by first giving it per-module
// *.unit.test.cjs / *.property.test.cjs coverage, then deleting its entry —
// never delete an entry without that coverage (it will only produce survived
// mutants and trip the break threshold).
const UNMUTATED = [
'!gsd-core/bin/lib/command-aliases.cjs',
'!gsd-core/bin/lib/commands.cjs',