.gc within zarr_sync is not "in effect"?

[yarikoptic]

while assessing what consumes space I spotted that some zarrs have huge number of loose objects, which seems to be packed or cleaned up with git gc call. But it seems that we do call git gc after syncing zarr:

dandisets/tools/backups2datalad/zarr.py

Line 577 in 306e1ae

await ds.gc()

so it is not clear why we still have the case (may be failed syncs which never got git gc ran, only me doing git reset --hard?). I am still collecting stats across all zarrs under /tmp/dandizarrs-inodes.out but here is an example of one and effect of running git gc now:

dandi@drogon:/mnt/backup/dandi/dandizarrs/0cf3c734-4bb2-47d1-8e6a-528a3739af01$ du -scm --inodes .git/objects/; git gc; du -scm --inodes .git/objects/
232704  .git/objects/
232704  total
Enumerating objects: 541506, done.
Counting objects: 100% (541506/541506), done.
Delta compression using up to 4 threads
Compressing objects: 100% (469070/469070), done.
Writing objects: 100% (541506/541506), done.
Total 541506 (delta 224707), reused 156961 (delta 72419), pack-reused 0
Removing duplicate objects: 100% (256/256), done.
7       .git/objects/
7       total

so we went from 230k to 7 inodes in .git/objects.