Skip to content

ci: reuse verified partitions for cold PR mutation runs - #56

Merged
choskeli merged 1 commit into
mainfrom
ci/parallel-cold-pr-mutations
Oct 28, 2026
Merged

choskeli merged 1 commit into
mainfrom
ci/parallel-cold-pr-mutations

Conversation

@choskeli

Copy link
Copy Markdown
Contributor

An ordinary PR with no compatible mutation baseline currently executes every parser mutant on one runner. This checks the existing cache admission before test discovery or execution: compatible PRs continue native incremental testing on that runner; cold PRs hand off to the existing verified partition matrix and collector. A failed warm run never starts a second campaign.

  • Preserve the exact input fingerprint, baseline authentication, source-bound equivalence proofs, complete mutant inventory, raw report provenance and 100% adjusted gate. Distinguish missing, incompatible, corrupt/unauditable and unreadable baselines in diagnostics.
  • Require an explicit route, keep failures blocking, and honor workflow cancellation without losing explicit-full scheduling after skipped ancestors. Reuse the existing planner, shards and collector unchanged.
  • Keep ordinary main execution native so successful cold main runs can publish the trusted whole-report baseline. Cold PR partitions publish neither native nor own-PR baseline envelopes: reports are not merged or rewritten, and repeated cold PR runs can remain cold until a compatible main baseline exists. Non-full manual and merge-group invocations retain their existing native command.

Production footprint is four files, +89/−18 lines. The nine-file diff also includes documentation and regression fixtures; no new dependency, timing store, cache engine or parser refactor.

Validation

  • 25 runner-routing tests and 14 workflow tests pass, including early cold handoff, warm reuse, malformed baseline diagnostics, unknown-input rejection via existing admission, failure without retry, explicit-full skipped ancestors, missing outputs and cancellation. Existing transport/CI contract tests pass. Workflow fixtures model documented GitHub conditions; hosted scheduling remains to be verified.
  • Actual CLI cold handoff with a simulated hosted PR environment completed in 321 ms: no discovery, tests, mutation execution, audit or baseline promotion; lock released.
  • Real Stryker nine-case cache fixture passed in 27.6 s: unchanged/passive-doc runs reused 5/5 mutants; cold/forced/source/helper/fixture/config/lock changes ran 5/5 fresh.
  • Existing native two-partition fixture passed in 13.1 s: 10/10 fresh mutants collected, with stale/partial evidence and ordinary-test failure rejected.
  • Actionlint 1.7.12, formatting, lint, both TypeScript compilers, 162 mutation-policy tests and build passed. npm run check stopped at six Windows HTTP cleanup failures (ECONNRESET/ECONNABORTED), with 2,124 tests passing and 12 skipped. The cache unit suite also retains its pre-existing Windows executable-bit assertion failure. No assertions were weakened.
  • Independent source review is clear. No local full repository mutation campaign or workflow cancellation/restart was launched; required hosted exact-head checks remain mandatory.

The previously accepted 8m32 partition campaign belongs to PR #53's tested implementation, not a new timing measurement for this PR. This draft makes that existing path available to cold PRs; it adds no timing weighting.

@choskeli
choskeli marked this pull request as ready for review October 28, 2026 01:28
@choskeli
choskeli merged commit 3a9ab07 into main Oct 28, 2026
39 checks passed
@choskeli
choskeli deleted the ci/parallel-cold-pr-mutations branch October 28, 2026 01:29
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant