From: Jeff King <peff@peff.net>
Date: Fri, 15 Jan 2016 22:28:00 GMT
Subject: Re: [issue?]'git branch -D' case sensitive issue
Message-ID: <20160115222800.GA32081@sigill.intra.peff.net>
In-Reply-To: <xmqq4megkrf9.fsf@gitster.mtv.corp.google.com>

On Thu, Jan 14, 2016 at 09:21:46AM -0800, Junio C Hamano wrote:

> Torsten Bögershausen <tboegi@web.de> writes:
> 
> > No, this is typically what we observe on a case insensitive FS.
> 
> There is an interesting bit in delete_branches::builtin/branch.c
> 
> 	if (kinds == FILTER_REFS_BRANCHES && !strcmp(head, bname.buf)) {
> 		error(_("Cannot delete the branch '%s' "
> 		      "which you are currently on."), bname.buf);
> 		ret = 1;
> 		continue;
> 	}
> 
> The strcmp() does not protect you from attempting to remove "FrOtZ"
> branch when you are on (and more importantly, Git thinks you are on)
> "frotz" branch.  A case-insensitive system then can tell Git that
> the request to remove "refs/heads/FrOtZ" succeeded, even though in
> reality it didn't have such file, and silently "refs/heads/frotz"
> gets removed instead.  You can end up removing your current branch
> that way.

This seemed eerily familiar. And indeed, there's some discussion in:

  http://article.gmane.org/gmane.comp.version-control.git/276506

Changing that strcmp to strcmp_icase() is still wrong, though I suspect
it's less wrong in practice (and errs on the side of caution). But I
think there are probably a lot of similar problems lurking.

-Peff