miandmhoennig
87775359cf
## The Problem The first run of the build `mutation` (PR#93) on mih09, right after the merge on 2026-10-03, failed twice: - PIT stopped in its coverage run: `BuildExecutorIntegrationTest` did not pass without mutation, although the normal test build on mih09 and PIT locally pass it. - The deploy of v1.28.0, queued behind it, found no runtime bundle: `mutation` runs in `main`'s worktree, and its `cleanCommand: rm -rf build` deleted the bundle the test build had left there. The run started at once instead of at 02:00, because the definition was new and the day's slot was due. ## The Solution - `mutation` cleans only `build/reports/pitest` and `build/reports/codecharta`; PIT recompiles what changed, and the runtime bundle stays for the deploy. - `mutation` runs at 00:00 UTC instead of 02:00: 01:00 or 02:00 in Germany, depending on daylight saving time, which UTC does not have; later nightly builds of other repositories follow it on mih09. - `BuildExecutorIntegrationTest` no longer runs under PIT. Two probe builds on this branch, with PIT's verbose output, could not reproduce the failure: the spec passed alone and with every other spec. It is most likely a timing-dependent case; the best guess is Scenario#67.04, where `true` must use no measurable processor time, which one clock tick under PIT's instrumentation breaks. One failing spec in the coverage run aborts the whole night, and the spec costs 24-29 s per covering mutant. `BuildExecutor` shows mostly uncovered until its logic moves into unit tests. ## Follow-up PRs - `BuildExecutor`: its decisions into small classes with unit tests (already planned in PR#93). - Make the timing-dependent cases of `BuildExecutorIntegrationTest` robust, starting with Scenario#67.04. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: mhoennig <michael@hoennig.de> Reviewed-on: #94
1.9 KiB
1.9 KiB
PR#94: Fixes for the Nightly Mutation Build
This document describes the change of this pull request at the time it was made. Later pull requests may change the behaviour without updating it.
The Problem
The first run of the build mutation (PR#93) on mih09, right after the merge on 2026-10-03, failed twice:
- PIT stopped in its coverage run:
BuildExecutorIntegrationTestdid not pass without mutation, although the normal test build on mih09 and PIT locally pass it. - The deploy of v1.28.0, queued behind it, found no runtime bundle:
mutationruns inmain's worktree, and itscleanCommand: rm -rf builddeleted the bundle the test build had left there.
The run started at once instead of at 02:00, because the definition was new and the day's slot was due.
The Solution
mutationcleans onlybuild/reports/pitestandbuild/reports/codecharta; PIT recompiles what changed, and the runtime bundle stays for the deploy.mutationruns at 00:00 UTC instead of 02:00: 01:00 or 02:00 in Germany, depending on daylight saving time, which UTC does not have; later nightly builds of other repositories follow it on mih09.BuildExecutorIntegrationTestno longer runs under PIT. Two probe builds on this branch, with PIT's verbose output, could not reproduce the failure: the spec passed alone and with every other spec. It is most likely a timing-dependent case; the best guess is Scenario#67.04, wheretruemust use no measurable processor time, which one clock tick under PIT's instrumentation breaks. One failing spec in the coverage run aborts the whole night, and the spec costs 24-29 s per covering mutant.BuildExecutorshows mostly uncovered until its logic moves into unit tests.
Follow-up PRs
BuildExecutor: its decisions into small classes with unit tests (already planned in PR#93).- Make the timing-dependent cases of
BuildExecutorIntegrationTestrobust, starting with Scenario#67.04.