Re: v2.15.0-rc2 ref deletion bug

From: Jeff King <peff@peff.net>
Date: Oct 24, 2017, 19:04 UTC
Message-ID: <20171024190411.dvv7zsq6ax4nndti@sigill.intra.peff.net>
In-Reply-To: <ea715dae-b410-d4fb-77c7-2e61b764a343@alum.mit.edu>

On Tue, Oct 24, 2017 at 01:05:00PM +0200, Michael Haggerty wrote:

Show 16 quoted lines

> > I'd expect one of:
> > 
> >   1. We delete "foo" before updating "foo/bar", and we end up with a
> >      single ref.
> 
> I don't think that this is possible in the general case in a single
> transaction. The problem is that the code would need to take locks
> 
>     refs/tags/foo.lock
>     refs/tags/foo/bar.lock
> 
> But the latter couldn't coexist with the loose reference file
> 
>     refs/tags/foo
> 
> , which might already exist.

Yeah, you're right. I was thinking of the opposite case (where you could create "foo.lock" even though "foo/bar" exists), but this one is impossible with filesystem semantics.

Show 6 quoted lines

> It is only imaginable to do this in a single transaction if you pack and
> prune `refs/tags/foo` first, to get the loose reference out of the way,
> before executing the transaction. Even then, you would have to beware of
> a race where another process writes a loose version of `refs/tags/foo`
> between the time that `pack-refs` prunes it and the time that the
> transaction obtains the lock again.

Yeah, it's probably better to avoid playing games here. Moving to a non-filesystem storage backend would just make the problem go away.

-Peff