The make-git-checkout-seed.sh script (https://infrastructure.fedoraproject.org/cgit/ansible.git/tree/roles/git/make_checkout_seed/files/make-git-checkout-seed.sh) runs daily at 0200 to create the git checkout seed and the specfile tarball (https://src.fedoraproject.org/repo/git-seed-latest.tar.xz and https://src.fedoraproject.org/repo/rpm-specs-latest.tar.xz).

Currently the git seed is nearly 10G compressed (> 24G exploded) and contains nearly six million files and directories.

Since this is big enough to put it out of the range of usefulness for many people, I was wondering if providing a shallow clone would be better. Just a depth 1 clone appears to be about a quarter of the size, but since it contains neither history nor any of the branches it might not be sufficiently useful.

So my basic question is whether it is reasonably possible to generate a checkout which:

• has at least a little history
• has all of the live branches
• is at least somewhat smaller
• isn't too much slower or more difficult to generate than the current checkout

The clone is generated at line 54 of the script. It should be pretty trivial to just pass --depth 5 --no-single-branch to that to get a somewhat smaller checkout.

Things I'm thinking about:

• --depth N --no-single-branch (the simple case)
• Using --shallow-since and asking for a year or so of history. If there is no history recent enough, current git will fetch all history. Git 2.19 (not yet released) this will error in that case, so it would need to be handled. (Just fall back to --depth)
• Using --depth N and manually fetching each live branch. But how can we get a list of live branch in a shell script?