feat(runner): report a scratch directory that disappears mid-run by Chemaclass · Pull Request #1168 · TypedDevs/bashunit

Chemaclass · 2026-08-13T20:27:26Z

🤔 Background

Related #1167

#1164 made a run survive its scratch directory going missing, so a valid test
file is no longer reported as unsourceable. But it recovers silently, and
#1137 is open precisely because that directory goes missing on CI and nobody
can say what removes it. No trace at all is worse for diagnosis than the
misleading error it replaced.

💡 Changes

Warn once per run, on stderr, naming the directory and pointing at The suites acceptance test is flaky on Bash 3.0 under load #1137
Once per run rather than per file — verified across four files: one warning, all four still run

Surviving the disappearance silently leaves no trace, which is worse for diagnosis than the misleading 'failed to source' it replaced: #1137 is open precisely because that directory goes missing on CI and nobody can say what removes it. Warn once per run on stderr, naming the directory. Once per run rather than per file: a run that loses it early would otherwise repeat the warning for every remaining file. Closes#1167

Chemaclass · 2026-08-13T20:27:26Z

🤔 Background

Related #1167

#1164 made a run survive its scratch directory going missing, so a valid test
file is no longer reported as unsourceable. But it recovers silently, and
#1137 is open precisely because that directory goes missing on CI and nobody
can say what removes it. No trace at all is worse for diagnosis than the
misleading error it replaced.

💡 Changes

Warn once per run, on stderr, naming the directory and pointing at The suites acceptance test is flaky on Bash 3.0 under load #1137
Once per run rather than per file — verified across four files: one warning, all four still run

Surviving the disappearance silently leaves no trace, which is worse for diagnosis than the misleading 'failed to source' it replaced: #1137 is open precisely because that directory goes missing on CI and nobody can say what removes it. Warn once per run on stderr, naming the directory. Once per run rather than per file: a run that loses it early would otherwise repeat the warning for every remaining file. Closes#1167

Chemaclass · 2026-08-13T20:27:26Z

🤔 Background

Related #1167

#1164 made a run survive its scratch directory going missing, so a valid test
file is no longer reported as unsourceable. But it recovers silently, and
#1137 is open precisely because that directory goes missing on CI and nobody
can say what removes it. No trace at all is worse for diagnosis than the
misleading error it replaced.

💡 Changes

Warn once per run, on stderr, naming the directory and pointing at The suites acceptance test is flaky on Bash 3.0 under load #1137
Once per run rather than per file — verified across four files: one warning, all four still run

Surviving the disappearance silently leaves no trace, which is worse for diagnosis than the misleading 'failed to source' it replaced: #1137 is open precisely because that directory goes missing on CI and nobody can say what removes it. Warn once per run on stderr, naming the directory. Once per run rather than per file: a run that loses it early would otherwise repeat the warning for every remaining file. Closes#1167

Chemaclass · 2026-08-13T20:27:26Z

🤔 Background

Related #1167

#1164 made a run survive its scratch directory going missing, so a valid test
file is no longer reported as unsourceable. But it recovers silently, and
#1137 is open precisely because that directory goes missing on CI and nobody
can say what removes it. No trace at all is worse for diagnosis than the
misleading error it replaced.

💡 Changes

Warn once per run, on stderr, naming the directory and pointing at The suites acceptance test is flaky on Bash 3.0 under load #1137
Once per run rather than per file — verified across four files: one warning, all four still run

Surviving the disappearance silently leaves no trace, which is worse for diagnosis than the misleading 'failed to source' it replaced: #1137 is open precisely because that directory goes missing on CI and nobody can say what removes it. Warn once per run on stderr, naming the directory. Once per run rather than per file: a run that loses it early would otherwise repeat the warning for every remaining file. Closes#1167

Chemaclass · 2026-08-13T20:27:26Z

🤔 Background

Related #1167

#1164 made a run survive its scratch directory going missing, so a valid test
file is no longer reported as unsourceable. But it recovers silently, and
#1137 is open precisely because that directory goes missing on CI and nobody
can say what removes it. No trace at all is worse for diagnosis than the
misleading error it replaced.

💡 Changes

Warn once per run, on stderr, naming the directory and pointing at The suites acceptance test is flaky on Bash 3.0 under load #1137
Once per run rather than per file — verified across four files: one warning, all four still run

Surviving the disappearance silently leaves no trace, which is worse for diagnosis than the misleading 'failed to source' it replaced: #1137 is open precisely because that directory goes missing on CI and nobody can say what removes it. Warn once per run on stderr, naming the directory. Once per run rather than per file: a run that loses it early would otherwise repeat the warning for every remaining file. Closes#1167

Chemaclass · 2026-08-13T20:27:26Z

🤔 Background

Related #1167

#1164 made a run survive its scratch directory going missing, so a valid test
file is no longer reported as unsourceable. But it recovers silently, and
#1137 is open precisely because that directory goes missing on CI and nobody
can say what removes it. No trace at all is worse for diagnosis than the
misleading error it replaced.

💡 Changes

Warn once per run, on stderr, naming the directory and pointing at The suites acceptance test is flaky on Bash 3.0 under load #1137
Once per run rather than per file — verified across four files: one warning, all four still run

Surviving the disappearance silently leaves no trace, which is worse for diagnosis than the misleading 'failed to source' it replaced: #1137 is open precisely because that directory goes missing on CI and nobody can say what removes it. Warn once per run on stderr, naming the directory. Once per run rather than per file: a run that loses it early would otherwise repeat the warning for every remaining file. Closes#1167

Chemaclass · 2026-08-13T20:27:26Z

🤔 Background

Related #1167

#1164 made a run survive its scratch directory going missing, so a valid test
file is no longer reported as unsourceable. But it recovers silently, and
#1137 is open precisely because that directory goes missing on CI and nobody
can say what removes it. No trace at all is worse for diagnosis than the
misleading error it replaced.

💡 Changes

Warn once per run, on stderr, naming the directory and pointing at The suites acceptance test is flaky on Bash 3.0 under load #1137
Once per run rather than per file — verified across four files: one warning, all four still run

Surviving the disappearance silently leaves no trace, which is worse for diagnosis than the misleading 'failed to source' it replaced: #1137 is open precisely because that directory goes missing on CI and nobody can say what removes it. Warn once per run on stderr, naming the directory. Once per run rather than per file: a run that loses it early would otherwise repeat the warning for every remaining file. Closes#1167

Chemaclass · 2026-08-13T20:27:26Z

🤔 Background

Related #1167

#1164 made a run survive its scratch directory going missing, so a valid test
file is no longer reported as unsourceable. But it recovers silently, and
#1137 is open precisely because that directory goes missing on CI and nobody
can say what removes it. No trace at all is worse for diagnosis than the
misleading error it replaced.

💡 Changes

Warn once per run, on stderr, naming the directory and pointing at The suites acceptance test is flaky on Bash 3.0 under load #1137
Once per run rather than per file — verified across four files: one warning, all four still run

Surviving the disappearance silently leaves no trace, which is worse for diagnosis than the misleading 'failed to source' it replaced: #1137 is open precisely because that directory goes missing on CI and nobody can say what removes it. Warn once per run on stderr, naming the directory. Once per run rather than per file: a run that loses it early would otherwise repeat the warning for every remaining file. Closes#1167

Chemaclass added the enhancement New feature or request label Aug 13, 2026

Chemaclass self-assigned this Aug 13, 2026

Chemaclass merged commit bc68ebe into mainAug 13, 2026
37 checks passed

Chemaclass deleted the feat/1167-warn-scratch-dir-vanished branch August 13, 2026 20:34