Re: Fwd: Errors when diff'ing arbitrary upstream remotes are not intuitive if git pull --all hasn't been done

From: Jeff King <peff@peff.net>
Date: Oct 25, 2013, 07:10 UTC
Message-ID: <20131025071048.GA16856@sigill.intra.peff.net>
In-Reply-To: <CALKQrgc0u8TA4mXc+2Hv+Fyo8EYNuiickF_janeCxkzFn+xAvA@mail.gmail.com>

On Fri, Oct 25, 2013 at 09:03:41AM +0200, Johan Herland wrote:

Show 12 quoted lines

> Actually, I don't think there's much refspec stuff to be done here.
> When running "git diff $remote/$branch", there are 3 possible
> outcomes:
> 
>  - $remote is not a valid remote name, the user probably meant
> something different (like "nested/branch"). The current error message
> is fine.
> 
>  - $remote is a valid remote name, but $branch has not (yet) been
> fetched from there. Suggest the user run "git fetch $remote"
> 
>  - $remote/$branch is a valid remote-tracking branch. The diff works! No errors.

Right, I think it is the second case we are talking about.

Show 9 quoted lines

> So, AFAICS, the patch should simply:
> 
>  1. Split the input on the first '/' into $remote/$branch, and use the
> preceding part ($remote) as a potential remote name, and the following
> part ($branch) as a potential branch name. (Although it is
> theoretically possible to have remote names containing slashes, I
> don't think anybody uses them, and we have considered disallowing
> them, mainly because of this very issue: it makes "$remote/$branch"
> parsing (even more) ambiguous)

What I specifically meant is that this breaks with a remote like:

  [remote "foo"]
    fetch = +refs/heads/*:refs/remotes/bar/*

The correct advice for "bar/branch" is to recommend "git fetch foo", and the correct advice for "foo/branch" is nothing at all.

I know such config is unusual, but I thought there was a recent push for us to be accurate about finding the local side of remote tracking branches, rather than just assuming they start with "$remote". Maybe I am misremembering, though; I thought it was related to potentially shifting the default refspecs.

The procedure along those lines would be:

  for each remote
    for each fetch-refspec in remote
      if refspec.rhs contains "refs/remotes/$failed_branch"
        recommend "git fetch $remote"

I was just wondering if we had something to make that "does this refspec contain this ref" part easier.

-Peff