You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
pytest paths filter omits data/traits/**, so corpus tests never run on data-only PRs — 13 records drifted off the round-trip invariant #554
tests/test_write_validated_round_trip.py::test_the_whole_corpus_round_tripsfails on main (verified at 3316e80, clean checkout):
AssertionError: 13 record(s) no longer round-trip through write_validated_trait; first few:
['data/traits/environment/facultatively_aerobic.yaml', 'data/traits/environment/microaerotolerant.yaml',
'data/traits/environment/nacl_delta_high.yaml', 'data/traits/environment/stenohaline.yaml',
'data/traits/environment/thermotolerant.yaml']
The 13 are records touched by today's canonical-example and graph-connectivity tranches (#544–#552). They were written in a form write_validated_trait would not emit — the difference is purely line wrapping, e.g. thermotolerant.yaml:
- note: Type strain CM3 (DSM 14292; LMG 21284) was described as thermotolerant,
- with an approximately 47 degrees C optimum and a 55 degrees C maximum. The source
+ note: Type strain CM3 (DSM 14292; LMG 21284) was described as thermotolerant, with
+ an approximately 47 degrees C optimum and a 55 degrees C maximum. The source published
No data differs. #322/#344 normalised the whole corpus precisely so this invariant could be enforced; it has now drifted back for these records.
There is no data/traits/**. Every canonical-example tranche is a data-only PR, so pytest never ran on any of them — confirmed on #551, whose only checks were canonical-example-taxonomy, history, pr-sanity, qc, validate-strict, vendored-sync. The corpus round-trip test reads data/traits/ directly, exactly like the research/** case already called out in that file's own comment (#250).
This is the same defect class as #184/#200/#250/#252 and as the DO_NOT_WORK.md gap fixed in #511: a filter narrower than what the job reads. audit-qc-paths guards the qc workflow this way but nothing guards pytest.
Fix
Add data/traits/** to both the pull_request and push filters in pytest.yaml, with a comment naming the test that reads it.
Activity
Address the second adversarial review
Adapt MicroGrowLink's review-open-issues rigor back into TraitMech (#559
Trigger dependency-installing jobs on dependency changes (#573)
Add a source catalogue, and register both investigated sources
Add a source catalogue, and register both investigated sources (#574)
Restore the workflow filters and guard that #574 reverted (#585)