Re: [Bug] Porcelain allows creation of '@' branch
- From
Jeff King <peff@peff.net>
- Date
- Sep 2, 2026, 07:37 UTC
- Message-ID
- <20260902073753.GC70165@coredump.intra.peff.net>
- In-Reply-To
- <CALnO6CCph_xC394v_BetLPyoriYc9dLZY42LsXhjVNdvt2e-cQ@mail.gmail.com>
On Mon, Aug 31, 2026 at 03:59:56PM -0400, D. Ben Knoble wrote:
Show 11 quoted lines
> But (with git refs list) we should see this creates a ref "refs/heads/@". > > I happen to think that's extremely confusing given that "@" is the > shorthand for HEAD, but… it's not against the current documented > rules, I think. (e.g., "git switch @" will fail, since it sees "git > switch HEAD"; using "refs/heads/@" will also fail.) git-checkout > doesn't fail but also doesn't change to the "@" branch. Futzing with > .git/HEAD and restoring the working tree works, but… yikes. > > Of course, --branch mode is allowed to be stricter; maybe we should > reject this case?
Yeah, sadly bare "@" is allowed in a component (but not "@{") because we've maintained strict backwards compatibility. So I don't think it's a _bug_ exactly, but in the past we have restricted the porcelain interfaces (so git-branch, but not git-update-ref) to avoid such footguns. The most notable one is HEAD:
$ git check-ref-format refs/heads/HEAD; echo $? 0
$ git check-ref-format --branch HEAD; echo $? fatal: 'HEAD' is not a valid branch name 128
which also shows off that "--branch" is a convenience option for using those stricter porcelain rules.
I think it would make sense to treat bare "@" the same as HEAD here. We probably _don't_ want to forbid any "@", though. Branches and tags like foo@bar are legal (if IMHO somewhat gross), and less likely to be used accidentally.
So I think there is no bug, in the sense that the system is working as designed, but it sure would be a nice feature to block this footgun.
> (I don't recall offhand where we describe valid branch names, if at all.)
I don't think we document the magic porcelain-tightening rules anywhere. That's probably OK in practice; the point of "check-ref-format --branch" is to check against them programatically.
-Peff