Re: [PATCH 2/2] transport-helper: update remote helper namespace

From: Felipe Contreras <felipe.contreras@gmail.com>
Date: Apr 11, 2013, 04:53 UTC
Message-ID: <CAMP44s0FkiwPMJVhVBNa32J3rgghRZy6xDTN-YnHKcQ4Fj0BMQ@mail.gmail.com>
In-Reply-To: <20130411043346.GE14551@sigill.intra.peff.net>

On Wed, Apr 10, 2013 at 11:33 PM, Jeff King <peff@peff.net> wrote:

Show 22 quoted lines

> On Wed, Apr 10, 2013 at 07:07:12PM -0500, Felipe Contreras wrote:
>
>> When pushing, the remote namespace is updated correctly
>> (e.g. refs/origin/master), but not the remote helper's
>> (e.g. refs/testgit/origin/master).
>>
>> Let's update it correctly.
>
> I would have thought it was the helper's responsibility to update these.
> Obviously remote-testgit can handle this fine, but will any other
> helpers be using these refs as a marker to know the last point they
> imported, and get confused if we update the refs behind their back?
>
> For example, during the import, a helper might know that it has imported
> up to X on a foreign vcs, and that resulted in commit Y in git, which it
> stored in refs/$helper/heads/master during the last import. When we
> fetch from it again, it picks up from X to the tip of the foreign vcs,
> and then imports that history on top of commit Y.
>
> But if we push some commits to the helper, moving Y up to Z, then it
> would build the new commit (which contains the foreign-vcs's equivalent of
> Y..Z) on top of Z, not Y.

Why would it do that? If X points to say revision 100, presumably it was stored somewhere while doing a fetch. Similarly, if foreign version of Z is 150, it can update that number while doing a push. The next fetch it would start from 151.

All this is hypothetical of course, because...

> I do not offhand know of any helpers that are implemented this way,
> though. vcs-svn does not seem to use the refspec feature at all, and I
> assume that your hg/bzr helpers do not have this problem. So perhaps it
> is not worth worrying about.

They cannot be implemented this way, because as I have already argued and shown[1], remote helpers must be using marks, they don't work otherwise (transport helper is broken for those cases).

Cheers.

[1] http://article.gmane.org/gmane.comp.version-control.git/210306

-- Felipe Contreras