fetch: tag following too ambitious?

Michael Haggerty· Oct 28, 2013, 12:42 UTC · lore

When investigating the exact semantics of tag-following, I discovered that the tag auto-following behavior of "git fetch" is more ambitious than I would have expected: it fetches any tag that references an object that is known to the local repository, *even if that object is not currently reachable* (i.e., neither reachable before the fetch or after the fetch of non-auto-followed references). This makes it hard to renounce interest in a branch.

Suppose there is a remote repo with

    o---o---o        <- master
     \
      o---A---B      <- pu

When I clone this repo, of course I get all of the commits and both branches.

Now suppose I decide I'm not interested in "branch" anymore, so I delete its remote-tracking branch from my repository and change the config to only fetch "master":

    git config remote.origin.fetch \
            '+refs/heads/master:refs/remotes/origin/master'
    git update-ref -d refs/remotes/origin/pu

It looks like I'm free of the "pu" branch, right?

But if a week later somebody pushes a tag "t" to origin that points at commit A, and then I do

    git fetch origin

then Git (un)helpfully fetches tag "t" into my repo, because even though commit "A" isn't reachable in my repo, it hasn't been pruned yet from the object database.

I admit this is not likely to be a serious problem in practice, but I found it surprising and strangely disturbing. I would call it a bug.

Thoughts?

Michael

-- 
Michael Haggerty
mhagger@alum.mit.edu
http://softwareswirl.blogspot.com/

Junio C Hamano· Oct 28, 2013, 16:34 UTC · re: Michael Haggerty · lore

Re: fetch: tag following too ambitious?

Michael Haggerty <mhagger@alum.mit.edu> writes:

Show 38 quoted lines

> When investigating the exact semantics of tag-following, I discovered
> that the tag auto-following behavior of "git fetch" is more ambitious
> than I would have expected: it fetches any tag that references an object
> that is known to the local repository, *even if that object is not
> currently reachable* (i.e., neither reachable before the fetch or after
> the fetch of non-auto-followed references).  This makes it hard to
> renounce interest in a branch.
>
> Suppose there is a remote repo with
>
>     o---o---o        <- master
>      \
>       o---A---B      <- pu
>
> When I clone this repo, of course I get all of the commits and both
> branches.
>
> Now suppose I decide I'm not interested in "branch" anymore, so I delete
> its remote-tracking branch from my repository and change the config to
> only fetch "master":
>
>     git config remote.origin.fetch \
>             '+refs/heads/master:refs/remotes/origin/master'
>     git update-ref -d refs/remotes/origin/pu
>
> It looks like I'm free of the "pu" branch, right?
>
> But if a week later somebody pushes a tag "t" to origin that points at
> commit A, and then I do
>
>     git fetch origin
>
> then Git (un)helpfully fetches tag "t" into my repo, because even though
> commit "A" isn't reachable in my repo, it hasn't been pruned yet from
> the object database.
>
> I admit this is not likely to be a serious problem in practice, but I
> found it surprising and strangely disturbing.  I would call it a bug.

Sounds like a bug to me. Does upload-pack to pack-object codepath actually pack the tag object and give it to you, or is it done all by reconnecting an existing and dangling tag back to your ref namespace?