Re: [PATCH 3/6] push: change `simple` to accommodate triangular workflows

[Ramkumar Ramachandra <artagnon@gmail.com>]

Johan Herland wrote:
>> An earlier round of this change by mistake broke the safety for
>> "simple" mode we have had since day 1 of that mode to make sure that
>> the branch in the repository we update is set to be the one we fetch
>> and integrate with, but it has been fixed.
>
> Shouldn't there be an acompanying test to demonstrate this mistake being fixed?

Read "earlier iteration": it didn't get merged.

>> +static void setup_push_current(struct remote *remote, struct branch *branch)
>> +{
>> +       if (!branch)
>> +               die(_(message_detached_head_die), remote->name);
>> +       add_refspec(branch->name);
>
> Here (and above) we add a refspec to tell Git exactly what to push
> from the local end, and into what on the remote end.

Nope, we add the refspec "foo", without the :destination part.  The
remote end is unspecified (and defaults to "foo", but that is in the
transport layer).

> Is it possible to
> end up with multiple simultaneous refspecs matching the same local
> ref, but mapping to different remote refs? If so, which will win, and
> does that make sense?

It is impossible.  We either:

- Get an explicit refspec from the user and never run
setup_default_push_refspecs() to begin with.

- Run setup_push_refspecs() and add *one* refspec depending on the
push.default value.