Re: rename a remote does not update pushdefault (v1.9.5)

From: Jeff King <peff@peff.net>
Date: Mar 31, 2015, 03:52 UTC
Message-ID: <20150331035256.GA1408@peff.net>
In-Reply-To: <xmqq384mxoka.fsf@gitster.dls.corp.google.com>

On Mon, Mar 30, 2015 at 12:39:17PM -0700, Junio C Hamano wrote:

Show 11 quoted lines

> Whichever way you implement "remote rename", you will make half of
> the users happy while making other half unhappy.  One use case will
> be happier if remote.pushdefault is left intact; the other use case
> will be happier if remote.pushdefault is updated.  There are two
> sides to this coin.
> 
> I think the implementation took the most straight-forward path to
> say "we rename everything inside remote.C.* section and adjust the
> refspecs for remote-tracking branches because that is what appear in
> that section"; which allows the first use case and the second use
> case would be just a single "git config remote.pushDefault C" away.

I had a similar thought. But note that we do update "branch.*.remote" that points to the renamed remote. So it seems inconsistent that we do not similarly update "branch.*.pushremote". And if we update that, it seems inconsistent that we do not update "remote.pushdefault".

In other words, we should probably choose to update all references or none, but we are currently somewhere in between. Of course, the fact that the code has been in this limbo for so long makes it doubly awkward, as we do not know what people expect (and what they simply consider a bug).

So I don't know what the right answer is.

I did take a peek at the code. I don't think updating these variables would be too hard, but there needs to be some refactoring around remote.c's pushremote_name. After calling read_config(), we do not have a value that exactly corresponds to remote.pushdefault; we overwrite it with the branch-specific pushremote if there is one.

-Peff