Re: What's cooking in git.git (Feb 2014, #06; Wed, 19)

From: Junio C Hamano <gitster@pobox.com>
Date: Feb 24, 2014, 16:57 UTC
Message-ID: <xmqqtxbobg98.fsf@gitster.dls.corp.google.com>
In-Reply-To: <8732A8C8-145E-47F5-BD9A-ECD6E9DE07EF@quendi.de>

Max Horn <max@quendi.de> writes:

Show 15 quoted lines

> On 21.02.2014, at 19:04, Junio C Hamano <gitster@pobox.com> wrote:
>
>>  Isn't it possible for some helpers to _do_ want to
>> tell us that it did not have to force after all by _not_ saying
>> "forced update" and overwrite ->forced_update with zero?
>
> Yes to the first part, no to the last bit: Yes, a transport helper
> can (and frequently does) tell us that no force happened -- by not
> saying "forced update".
> ...
>>  How do we tell helpers that do want to do so apart from other
>> helpers that say "forced update" only when they noticed they are
>> indeed forcing?
>
> I am not completely sure I even understand that bit?

I think I phrased it too imprecisely.

If nobody even knew about the "forced update" before hg helper, then they by definition do not wish to overwrite, of course. But I was worried if we are closing the door for this possible scenario:

 * the calling side sets ref->forced_update to true before invoking
   the helper, knowing that this update is not fast-forward; and

 * the helper does a "magic" (after all, we are talking with an
   external mechanism, which may be a different SCM like darcs) to
   rebase our change on top of the history that the other side
   already have, and makes it a fast-forward, non-forced push.

Such a helper would want a way to say "You may have thought that this does not fast-forward, but the push result ended up to be a fast-forward update", and if we wanted to support that, one thing we may need to allow it to do is to reset ref->forced_update to zero.

But I think I was worried too much into the future---I agree that the code can stay as you proposed until such a remote-helper needs more support, because "overwrite with zero" is necessary but is probably not sufficient---it also may need to be able to tell us what the final resulting commit of the push is, for example.