Repository navigation
Conversation
qian-harvard
left a comment
There was a problem hiding this comment.
Thanks, the coordinator/executor/auditor study workflow is a good addition, and all the holistic-run fixes from #135 are here too. Two things need fixing before this can merge:
-
The
powermcp confighook is gone again. This branch was cut from main before #135 landed, so it doesn't havepowermcp_adapter.py, and the syncedcore.pydrops main's_hope_setting(). Supposehope.repo_root,hope.julia_binandhope.julia_depot_pathare set throughpowermcp configand the environment variables aren't. Then this branch resolvesrepo_rootto the package's parent directory, runsjuliafromPATH, and sets no depot. That is the regression #135 fixed.I'm merging #135 first, so please rebase onto main after that. Six HOPE files will conflict. One resolution works: take this branch's copies of the synced files and keep the
apply_powermcp_hope_config()call inserver.py. I tried it locally, and the config paths were honoured again. -
test_registered_validation_profile_materializes_pinned_casesfails on macOS. The study manifest stores resolved paths, but the test compares them againstself.studies, which is built from an unresolvedTemporaryDirectory(). On macOS/varis a symlink to/private/var, sois_relative_toreturns false. CI runs on Linux and doesn't catch it. Settingself.root = Path(self.temporary.name).resolve()insetUpfixes it, and with that change the HOPE suite passes on macOS. The test is synced from HOPE.jl, so the fix belongs upstream.
Summary
dc4d1fcVerification
uv run python -m unittest discover -s tests -v: 45 passed, 3 opt-in solver tests skippeduv run --with pytest pytest tests -q -rs: 45 passed, 3 opt-in solver tests skipped