Hi Nilesh,
Thank you tons for thinking along.
It took me a bit but. Too long. The answer is that git does not
lose the sensation of a file being a git lfs reference even when
you download a tar.gz. For some reason I had expected that all
genomes were truly gzipped fasta files, but no, they were still
references. Maybe I had inadvertently transformed a few to what
they point to during a first build and that is why I then did not
find the issue a bit earlier.
The import of pristine-tar has worked after removing
.gitattributes, but then the git lfs references were still in the
tarball and the upstream branch. pristine-tar could be pushed, but
then the other branches would trigger git lfs errors when pushed
to salsa. Only after having all fasta.gz lfs files removed, the
upload went smoothly and you all now find this on
https://salsa.debian.org/med-team/broad-catch
Is that really intended?
We would not be able to run tests in this case since you essentially ended up repacking it , since several tests seem to be using that data.
BTW, I tried to do a little solution for the lfs thingy, for it to not store "references" and committed it to my personal repository[1]
Can you have a look and let me know if it looks sensible?
Also, you might as well want to have a look at pristine-lfs[2] which could be interesting to use. I've attempted to use this too, please consider taking a look
I admit, I'm not very used to the lfs workflow, so something could be wrong for sure.
Not sure why the CI fails though -- probably it does not work fine with lfs, but just thinking of it as a ground for more ideas here :)
I can clone it locally in a different space and re-produce the tarball, and I fixed a few tests (not all) -- please consider trying once and let me know
Nilesh