Stop the release-script tests reading the file under them #384

Merged

robbertbos merged 1 commit from release-scripts-fixture into main 2026-09-20 08:11:07 +00:00

Conversation 0 Commits 1 Files changed 3 +26 -21

robbertbos commented 2026-09-20 07:54:40 +00:00

Owner

Copy link

release-scripts fails on main now and then, most recently on 5529ffc3 while a second run of the same workflow was in flight. #137 added a three-attempt retry for this; the retry does not help, because the other checkout outlives it.

Cause: the section tests ran changelog-section.sh against the checkout's own CHANGELOG.md. A concurrent job re-checking out the workspace empties that file for a moment, and the script then reports an empty section for a tag that has one - which is exactly what #137 observed without being able to name it.

The tests now read a snapshot taken once at the start, through the CHANGELOG_FILE variable the scripts already honour. The newer frozen-check tests already worked this way in their own temp repos. The retry in ci.yml goes with it: a flake that survives this has a different cause and should be visible rather than papered over.

Measured, not assumed: a harness that truncates CHANGELOG.md in a loop while the suite runs fails the two section tests on main's version and passes on this one. Locally: 29 tests, exit 0.

`release-scripts` fails on main now and then, most recently on 5529ffc3 while a second run of the same workflow was in flight. #137 added a three-attempt retry for this; the retry does not help, because the other checkout outlives it. Cause: the section tests ran `changelog-section.sh` against the checkout's own `CHANGELOG.md`. A concurrent job re-checking out the workspace empties that file for a moment, and the script then reports an empty section for a tag that has one - which is exactly what #137 observed without being able to name it. The tests now read a snapshot taken once at the start, through the `CHANGELOG_FILE` variable the scripts already honour. The newer frozen-check tests already worked this way in their own temp repos. The retry in `ci.yml` goes with it: a flake that survives this has a different cause and should be visible rather than papered over. Measured, not assumed: a harness that truncates `CHANGELOG.md` in a loop while the suite runs fails the two section tests on main's version and passes on this one. Locally: 29 tests, exit 0.

robbertbos added 1 commit 2026-09-20 07:54:40 +00:00

Stop the release-script tests reading the file under them ... 538737a275

The job failed whenever a second run checked out the workspace mid-test:
changelog-section.sh then read an empty CHANGELOG.md and reported an
empty section for a tag that has one. Three retries did not help, because
the other checkout outlives them.

The section tests and the frozen check now read a snapshot taken once at
the start, through the CHANGELOG_FILE the scripts already honour, so a
concurrent checkout cannot reach them. The retry in ci.yml goes with it:
a flake that survives this is a different cause and should be visible.

Measured both ways with a harness that truncates CHANGELOG.md in a loop:
the previous version fails the two section tests, this one passes.

robbertbos merged commit 538737a275 into main 2026-09-20 08:11:07 +00:00

robbertbos deleted branch release-scripts-fixture 2026-09-20 08:11:07 +00:00

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

bug

Something is not working

  

duplicate

This issue or pull request already exists

  

enhancement

New feature

  

help wanted

Need some help

  

invalid

Something is wrong

  

nldd ds

  

question

More information is needed

  

wontfix

This won't be fixed

No labels

bug

duplicate

enhancement

help wanted

invalid

nldd ds

question

wontfix

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

robbertbos/waggle!384

No description provided.