Re: What's cooking in git.git (Apr 2013, #05; Mon, 15)

From: Felipe Contreras <felipe.contreras@gmail.com>
Date: Apr 16, 2013, 05:32 UTC
Message-ID: <CAMP44s3shDEx9dfWw_jNu0zQrReabgAMW-mti86wyoxhk+N2Ow@mail.gmail.com>
In-Reply-To: <7v1uabp109.fsf@alter.siamese.dyndns.org>

On Mon, Apr 15, 2013 at 11:12 PM, Junio C Hamano <gitster@pobox.com> wrote:

Show 27 quoted lines

> Felipe Contreras <felipe.contreras@gmail.com> writes:
>
>> So it should be clear now: the remote namespace refs/origin/master is
>> updated, but not the remote helper's namespace
>> refs/testgit/origin/master, which is what I already said. I don't know
>> what more do you expect. When you push 'refs/heads/master' to origin,
>> you expect 'refs/remotes/origin/master' to point to the same commit,
>> same with 'refs/testgit/origin/master', why would you expect to point
>> somewhere else?
>
> Let me play somebody who comes later and wonders about this exchange
> three months down the road...
>
> You mention three refs/ here.  Do they live in the same repository?
> Any Git person is expected to know refs/heads/master, which is "my
> local branch I have worked on and I am pushing".  It also is easy to
> guess what "refs/remotes/origin/master" is, even though we are not
> talking about a usual Git remote.  It is to keep track of the remote
> behind the helper we are pushing into, and is updated to pretend as
> if we fetched immediately from the place we just pushed.  The latter
> being in sync with what we pushed is something that can naturally be
> expected.
>
> Now, what is this third "refs/testgit/origin/master" thing?  Is it
> expected to always be the same as "refs/remotes/origin/master"?  If
> that is the case, why do we even need such a redundant information
> in the first place?

Answering that question is beyond the scope of the commit message. The purpose of the commit message is not to educate people about the current design of transport-helpers, we have Documentation/gitremote-helpers.txt for that. We also have discussions in the mailing list.

The untold answer is 'if you have to ask, you don't understand this code', but if you must, the short answer is: because it doesn't work otherwise.

Yes, in theory not using a remote helper ref namespace should work, but it doesn't, and I sent patches to try to fix this behavior, but those, along with this particular patch, where ignored. So I gave up on those patches that tried to fix the behavior for more corner cases and tried to push the simplest one I could find, still finding trouble. I tried to document that in t/t5801-remote-helpers.sh, although to be honest, the tests that pass can be hardly be considered proper behavior. This is of course, for import/export, which are the only operations I'm familiar with (maybe the namespace makes sense for push/fetch operations.

So basically I'm just tired of explaining the same things over and over again, and if you try to explain every single detail, I think any reasonable person would reach the conclusion that the design doesn't really make sense. But the only person that seems to be trying to explain it and improve it seems to be me.

The commit message is no place to explain all these subtleties, it would be huge and would need to refer to plenty of mails in the mailing list.

I could send the whole patch series I have, and then it might become clearer why not having the refspec doesn't work, but then ,the chances of this patch not getting through would be higher, as the last time I sent such series, it didn't go through.

Cheers.

-- 
Felipe Contreras