Re: [PATCH 2/6] setup: move creation of "refs/" into the files backend

From: Patrick Steinhardt <ps@pks.im>
Date: Jan 3, 2024, 08:33 UTC
Message-ID: <ZZUb0rj7be6wLQz5@tanuki>
In-Reply-To: <CAOLa=ZTR6aW5aoxcMOS3U3TL1VxSfmyVno9fu7B5201pJTqyyg@mail.gmail.com>

On Tue, Jan 02, 2024 at 05:23:18AM -0800, Karthik Nayak wrote:

Show 11 quoted lines

> Patrick Steinhardt <ps@pks.im> writes:
> 
> > Move the code to create the directory into the files backend itself to
> > make it so. This means that future ref backends will also need to have
> > equivalent logic around to ensure that the directory exists, but it
> > seems a lot more sensible to have it this way round than to require
> > callers to create the directory themselves.
> >
> 
> Why not move it to refs.c:refs_init_db() instead? this way each
> implementation doesn't have to do it?

True, that would be another possibility. But I think it is conceptually unclean to split up creation of the refdb into multiple locations. The "files" backend already has to create "refs/heads/" and "refs/tags/", and the "reftable" backend will set up "refs/heads" as a file so that it's impossible to create branches as loose files by accident. So both do have specific knowledge around how exactly this directory hierarchy should look like, and thus I think it is sensible to make the code self contained inside of the backends.

My opinion on this would be different if we expected a proliferation of backends to happen, but that's quite unlikely. Furthermore, we may at some point decide that repos don't need "refs/" and/or "HEAD" at all anymore, at which point it is easier to drop the creation of those files and dirs from the reftable backend.

I'll update the commit message to include these considerations.

Patrick