Re: cherry-pick and 'log --no-walk' and ordering

From: Junio C Hamano <gitster@pobox.com>
Date: Aug 11, 2012, 06:28 UTC
Message-ID: <7vpq6ygcy1.fsf@alter.siamese.dyndns.org>
In-Reply-To: <CAOeW2eHz5un9cNoy-=7Y8=F_G6u-n8kk7kXGHQ+dKrHD8wW6BA@mail.gmail.com>

Martin von Zweigbergk <martin.von.zweigbergk@gmail.com> writes:

Show 18 quoted lines

> On Fri, Aug 10, 2012 at 2:38 PM, Junio C Hamano <gitster@pobox.com> wrote:
>> Martin von Zweigbergk <martin.von.zweigbergk@gmail.com> writes:
>>
>>> There is also cherry-pick/revert, which I _think_ does not really want
>>> the revisions sorted.
>>
>> Yes, I think sequencer.c::prepare_revs() is wrong to unconditoinally
>> call prepare_revision_walk().
>>
>> It instead should first check the revs->pending.objects list to see
>> if what was given by the caller is a mere collection of individual
>> objects or a range expression (i.e. check if any of them is marked
>> with UNINTERESTING), and refrain from going into the body of the
>> preparation steps, which has to involve sorting.
>
> Do you mean "has to involve sorting" as in "has to involve sorting in
> order not to break current users of e.g. 'git log --no-walk
> --branches'"  or "revision walking inherently involves sorting"?

Range limited revision walking, e.g. "git cherry-pick A..B D~4..D", fundamentally implies sorting and you cannot assume B would appear before D only because B comes before D on the command line (B may even be inside D~4..D range in which case it would not even appear in the final output).

Any caller that wants to retrieve the objects given from the command line in the order the user gave it, e.g. "git cherry-pick A B C", using setup_revisions() and without walking the history, must look at revs->pending.objects without calling prepare_revision_walk().