Re: Creating remote branch called HEAD corrupts remote clones

On Mon, May 2, 2011 at 9:26 PM, Stephen Kelly [off-list ref] wrote:

On Wed, Apr 27, 2011 at 2:49 PM, Erik Faye-Lund [off-list ref] wrote:

quoted

On Wed, Apr 27, 2011 at 2:21 PM, Erik Faye-Lund [off-list ref] wrote:

quoted

On Wed, Apr 27, 2011 at 1:29 PM, Stephen Kelly [off-list ref] wrote:

quoted

On Wed, Apr 27, 2011 at 11:48 AM, Felipe Contreras
[off-list ref] wrote:

quoted

No problems here:

I had another go.

mkdir remote
cd remote/
git init --bare
cd ../
git clone remote/ alice
cd alice/
echo test >> file
git add file
git commit -am w
git push origin master
echo test >> file
git commit -am w
git branch HEAD

I'll stop you here. You reproduce the issue a lot simpler:

git init foo &&
cd foo &&
echo "foo" > bar &&
git add bar &&
git commit -m. &&
git branch HEAD &&
gitk

No need to involve remote branches. While remote branches makes the
issue worse, because you can get in a situation where gitk doesn't
when someone else made a nasty branch, and you fetched it.

The real problem is that "git rev-parse HEAD" outputs "warning:
refname 'HEAD' is ambiguous." to stderr (even if stderr is a non-tty),
and gitk does not like that.

This can be fixed by either doing "git -c core.warnambiguousrefs=0
rev-parse HEAD", which strikes me as ugly, or by making sure that we
don't issue this warning when not attached to a tty:

Of course, a third (and probably even better) option is to make gitk
warn about the ambiguous refname (like other commands will), but not
treat it as a fatal problem. But I'm not motivated enough to give that
solution a stab myself.

Not outputting that warning might be a regression for other users of
rev-parse (and/or the underlying mechanics).

Ok, if you can't see in the code why a branch called HEAD might
corrupt the remote and I can't demonstrate it with a testcase, maybe
it's not an issue anymore, I don't know.

No, it's still an issue, and I believe I pin-pointed it in my first
mail. You can try out the patch I sent, and see if that helps in your
case. If it does, I think it'd make sense to do something (preferably
a bit more robust) with it.