Re: Creating remote branch called HEAD corrupts remote clones

From: Jeff King <peff@peff.net>
Date: Jan 20, 2011, 21:54 UTC
Message-ID: <20110120215456.GB11468@sigill.intra.peff.net>
In-Reply-To: <7vbp3bqmiy.fsf@alter.siamese.dyndns.org>

On Thu, Jan 20, 2011 at 01:43:17PM -0800, Junio C Hamano wrote:

Show 10 quoted lines

> Jeff King <peff@peff.net> writes:
> 
> > Hmm. It seems like the symbolic ref is the culprit, not just HEAD. The
> > HEAD thing is the most likely, of course, but I could do something like:
> >
> >   git symbolic-ref refs/remotes/origin/convenient-alias \
> >                    refs/remotes/origin/some-name-you-dont-like
> 
> Isn't it already wrong to do the above locally, in the sense that it is
> equally wrong to do this?

Probably. My argument in favor would be "that is what we are doing with remotes/*/HEAD", but I think the logic is somewhat circular there.

Thinking on it more, yeah, if you want "convenient-alias", you are probably better off to do:

  git symbolic-ref refs/convenient-alias refs/remotes/origin/whatever

>     git update-ref refs/remotes/origin/no-such-thing-exists-over-there 
>     	refs/heads/master

Actually, don't we end up with that in the case of upstream deleting a ref? Which isn't to say it isn't a stupid thing to do, but that we can and do get into that state.

Thinking on it even more, I don't think we can cover all of the weird "you have a ref named $foo and they suddenly created a ref named $foo" push corner cases. There is no substitution for actual communication and organization of ref names if you are going to be pushing along with other people into a central repo. Because fundamentally the ref name is the unique identifier, and matching refs during push tries to reconcile matches based on those identifiers.

Show 6 quoted lines

> I personally like this line of thought, especially as a thought experiment
> to see what corner cases we could find, but I doubt I will be able to say
> we covered all the corner cases with confidence without thinking long and
> very hard.  For now, I do not find this issue worth spending that kind of
> deep thinking, especially when a lot simpler and easier to explain
> workaround is available, but others may disagree and perfect your idea.

Yeah, after reading your response and considering a bit, I think the simple "don't make HEAD" thing (or at least "don't pull or push HEAD") is a sane workaround.

-Peff