Repository navigation
ci: run the unit, uv-contract and e2e suites on every PR - #138
Merged
Merged
Conversation
Three jobs, split by what each needs from the runner: the unit suite wants only git-annex (so the two annex-proving tests run instead of skipping), the uv_build tests want a real uv, and the e2e wants the slurm-docker-ci image plus a ReproNim/containers clone holding the simbids SIF. Two things the e2e job has to work around: pull_request checks out a detached merge commit and run_in_podman.sh refuses a detached HEAD (the campaign pin is a branch or tag), so HEAD is put on a branch first; and the suite skips rather than fails when the SIF is missing, so the seed step asserts the file exists to turn a failed seed into a red job. Whether the image + SIF + campaign venv fit a hosted runner, and whether rootless podman there allows the nested apptainer namespace, is what the first run answers. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pfazng8PuPvbQNN25uzK2Y
First run of con#138 reddened all three jobs on two causes. Every suite commits — the unit tests build real datalad datasets, the e2e seeds a git-annex clone — and a hosted runner has no git identity, so git refused with "empty ident name" (3 unit failures in test_clean_check, the e2e seed dead in `git annex init`). A workflow-level GIT_AUTHOR_*/GIT_COMMITTER_* env covers all jobs in one place. The uv_build test pins this checkout at `rev-parse --abbrev-ref HEAD`, which on a detached pull_request checkout is the literal `HEAD`; uv then tried to fetch a branch or tag named HEAD. The uv-contract job now puts HEAD on a branch first, as the e2e job already did. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pfazng8PuPvbQNN25uzK2Y
uv (0.12.9) cannot resolve a branch out of a shallow repo: the fetch succeeds but neither the branch nor the tag ref lands, and it reports "failed to find branch or tag". The uv_build test pins this checkout as a git+file://…@<branch> source, so the default depth-1 checkout broke it. Reproduced locally against a `git clone --depth 1`; a full clone resolves fine, and the e2e job (already fetch-depth 0) ran the same kind of `uv lock` green. Also correct the timing comments to what the first green run measured: unit ~90 s, e2e ~6 min end to end. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pfazng8PuPvbQNN25uzK2Y
asmacdo
marked this pull request as ready for review
September 3, 2026 17:19
This was referenced Sep 4, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
tldr:
tests.ymlwith three jobs: unit (pytest, git-annex installed so nothing skips), uv-contract (pytest -m uv_buildwith a real uv), e2e (run_in_podman.shin rootless podman on the hosted runner).ReproNim/containersclone + the simbids SIF (cached) and pullsslurm-docker-ci:0.14as their own steps.pull_requestruns the workflow from the PR head, so the first run answers whether the image + SIF + campaign venv fit a hosted runner and whether rootless podman there allows the nested apptainer namespace.Why three jobs
They need different things from the runner. Splitting them keeps a 45-second unit failure from waiting on a 20-minute e2e, and makes the e2e's infrastructure dependencies (image, container dataset) visible as their own steps rather than buried in one log.
Two workarounds in the e2e job
pull_requestchecks out a detached merge commit, andrun_in_podman.shrefuses a detached HEAD because the campaign pin isURL@REFandgit clone --branchtakes a branch or tag. A step doesgit switch -c ci-under-testfirst.test -eon the SIF to make that red.Deferred
The suite could take the containers dataset by URL for CI, with the local clone as an env-var override; then the seed step goes away. Left as a
TODOcomment on the step.🤖 Generated with Claude Code
https://claude.ai/code/session_01Pfazng8PuPvbQNN25uzK2Y