Repository navigation
Dockerfile missing runtime deps for newly-merged modules (astroquery, numba, fitsio) #712
Description
Symptom
Several modules on develop import packages that aren't installed in the ghcr.io/cosmostat/shapepipe:develop image:
| Module | Missing package |
|---|---|
shapepipe.modules.mask_package.mask + mask_runner |
astroquery |
shapepipe.modules.ngmix_package.ngmix + ngmix_runner |
numba |
shapepipe.modules.uncompress_fits_package.uncompress_fits + uncompress_fits_runner |
fitsio |
Each raises ModuleNotFoundError at import time.
Root cause
Likely from a recent merge (PR #702?) that landed module changes requiring new deps without updating the Dockerfile's pip install list. The v2.0 PR (#706) does include these in its Dockerfile rewrite, but develop's Dockerfile lags.
Fix
Add the missing deps to the pip install block in Dockerfile on develop. Quick patch; doesn't need to wait for the v2.0 Dockerfile overhaul in #706.
Caught by
tests/unit/test_imports.py::test_module_imports[...] for the 6 affected modules (xfailed in #708).
Metadata
Metadata
Assignees
Labels
No labels
Activity
martinkilbinger commentedon May 18, 2026
Some packages are still (newly?) missing:
cailmdaley commentedon May 28, 2026
We dug into this empirically against current
ghcr.io/cosmostat/shapepipe:develop(dev target, post-#728), testing each listed package both in read-only mode and in a writable sandbox. All import cleanly:All three originally-flagged runner modules (
shapepipe.modules.{mask,ngmix,uncompress_fits}_package.*) also import cleanly. Read-only and writable sandbox give identical results.Pyproject confirms
astroquery/numba/joblib/tqdmare core deps;scipy/pyyaml/astropy_iers_data/pyerfa/LSSTDESC.Coord/importlib_metadatacome in as transitives ofastropy/mccd/modopt/galsim;fitsiois gated in both targets (runtime:--extra jupyter --extra fitsio;dev:--extra dev).@martinkilbinger — could the May 18 report have been against a cached or older image? If you can repro on a fresh
apptainer pull docker://ghcr.io/cosmostat/shapepipe:develop, drop the trace and we'll dig back in. Otherwise we'd like to close.— Claude on behalf of Cail
cailmdaley commentedon May 29, 2026
Resolved on
develop. The image's dependencies are now declared inpyproject.tomland frozen inuv.lock:astroqueryandnumbaare core dependencies, andfitsioships via thefitsioextra (the runtime image installs--extra fitsio; the dev image--extra dev, which includes it). All three of the flagged modules (mask,ngmix,uncompress_fits) import cleanly in the published image, and the test suite now runs inside that image in CI, so this class of missing-dep regression is caught automatically going forward.— Claude on behalf of Cail