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
Found while publishing iEEG datasets with annexed NIfTI files.
What happens
run-bids-validation fails with NIFTI_HEADER_UNREADABLE (and INVALID_GZIP for annexed .tsv.gz) on files that are git-annex pointers, although the workflow is meant to skip those. Every summary ends with "(skipped 0 annex-pointer error(s))":
nm000372: 1 error, sub-01/anat/sub-01_T1w.nii.gz, run 37598060353
nm000378: 36 errors, one per sub-*_T1w.nii, run 37595917824
GNU find rounds a file's size up to whole units before comparing, so -size -1k means "0 KiB after rounding up", i.e. only empty files. The pointers on main are unlocked pointer files (mode 100644, ~106 bytes; nemar-cli pushes the adjusted/unlocked branch), so they are never collected, and the symlink branch of the step does not match them either.
Reproduction (GNU findutils 4.7.0):
printf '/annex/objects/SHA256E-s1--x.nii.gz\n' > p
find . -type f -size -1k # no output
find . -type f -size -1024c # ./p
Proposed fix
Replace -size -1k with -size -1024c (or -size -2k). The existing location filter should then skip these errors as intended.
Changing the datasets instead does not look right: nemar-cli's annex policy annexes NIfTI on purpose and strips inherited annex.largefiles attributes.
It does not appear to block publication (nm000254 is published), but it puts a red BIDS check on the affected datasets.
Activity