exp show: Use tag name for all_tags.

[daavoo]

Close #6383

• ❗ I have followed the Contributing to DVC checklist.

[pmrowla]

So the original issue isn't with tags in packed refs. The problem is with annotated (signed) tags. Normal/lightweight (unsigned) tags are already handled properly, even when they are in packed refs.

Lightweight tags are just regular git refs (where the ref points directly to a commit object's SHA).

Annotated tags are different in that the ref points to a (annotated) tag object SHA, and not a commit SHA. To go from the annotated tag name to the commit SHA, you have to actually read the tag object first to get the dereferenced commit SHA (this is generally referred to as peeling the tag object in git/dulwich/pygit/etc docs+code).

(see https://git-scm.com/docs/git-tag)

The underlying issue in DVC is that the current dulwich implementation for git.describe doesn't resolve annotated (signed) tags properly.
https://github.com/iterative/dvc/blob/cb941a836d77c845f612be29e27ae80f7a2a126c/dvc/scm/git/backend/dulwich/__init__.py#L599

I haven't investigated this issue deeply, but if I had to guess I would think that the actual bug is probably that dulwich get_ref(annotated_tag) is returning the SHA for the tag object itself, instead of returning the peeled commit SHA.

https://github.com/iterative/dvc/blob/cb941a836d77c845f612be29e27ae80f7a2a126c/dvc/scm/git/backend/dulwich/__init__.py#L613

(Note that follow is unrelated to tags - it's for whether or not to follow symbolic refs. For annotated tags get_ref should always return the peeled commit SHA.)

[pmrowla]

Also, if you are trying to test against packed refs, you can just make some tags and then force git to pack everything in your repo with

git pack-refs --all

Don't do this in non-testing repos as it will pack all of your active branch refs (which will probably end up affecting performance of normal git command usage)

https://git-scm.com/docs/git-pack-refs

[daavoo]

Also, if you are trying to test against packed refs, you can just make some tags and then force git to pack everything in your repo with

git pack-refs --all

Don't do this in non-testing repos as it will pack all of your active branch refs (which will probably end up affecting performance of normal git command usage)

https://git-scm.com/docs/git-pack-refs

Thanks! Indeed, I used that for testing on local repo but was asking more in the lines of: how to properly do this inside one of our tests? I didn't find a way to do it from Python (apart from subprocess call)

[daavoo]

Also, if you are trying to test against packed refs, you can just make some tags and then force git to pack everything in your repo with

git pack-refs --all

Don't do this in non-testing repos as it will pack all of your active branch refs (which will probably end up affecting performance of normal git command usage)

https://git-scm.com/docs/git-pack-refs

Also, if you are trying to test against packed refs, you can just make some tags and then force git to pack everything in your repo with

git pack-refs --all

Don't do this in non-testing repos as it will pack all of your active branch refs (which will probably end up affecting performance of normal git command usage)
https://git-scm.com/docs/git-pack-refs

Thanks! Indeed, I used that for testing on local repo but was asking more in the lines of: how to properly do this inside one of our tests? I didn't find a way to do it from Python (apart from subprocess call)

As the problem was really about annotated tags, not packed refs, there is not really need for that

[pmrowla]

Thanks! Indeed, I used that for testing on local repo but was asking more in the lines of: how to properly do this inside one of our tests? I didn't find a way to do it from Python (apart from subprocess call)

In tests you can use CLI git directly via the scm fixture and gitpython like

scm.gitpython.git.pack_refs(all=True)

(https://gitpython.readthedocs.io/en/stable/tutorial.html#using-git-directly)

[pmrowla]

This breaks the case where name is a symbolic reference and follow = False, like when HEAD points to refs/heads/master. (see the current test failures)

Basically we need to catch KeyError when we do the self.repo[ref] check, and pass as long as follow is also False

[skshetry]

@daavoo, please open this PR on the upstream. Really sorry for the inconvenience. 🙂