Re: [PATCH/RFC] rev-list: add --authorship-order alternative ordering

From: Junio C Hamano <gitster@pobox.com>
Date: Jun 6, 2013, 19:40 UTC
Message-ID: <7vobbjxc21.fsf@alter.siamese.dyndns.org>
In-Reply-To: <CAPZ477OFM6D4n_Wz-OozN=aYn5-LmNA2ggL+9GNrbGrRQh9pRQ@mail.gmail.com>

Elliott Cable <me@ell.io> writes:

Show 6 quoted lines

>> And update revs->lifo to use that same enum, without adding
>> use_author_date bit to rev_info.
>
> I'll look into replacing lifo with an enum as soon as I can sit back
> down to update this patch. For the moment, nothing more than
> committer_date_sort and author_date_sort, I suppose?

> I'll try and make this a separate patch. First commit, to replace lifo
> with an enum; second commit, to *actually implement* the code obeying
> that enum when it is set to author_date_order.

If you want to do this in a multi-step series (which may not be a bad idea), I would imagine that the enum starts as a choice between the two: traversal-order vs committer-date-order. The first patch would change nothing else.

And then you would add the third choice, author-date-order, and implement the logic to sort them using author instead of committer date in the same patch.

Show 16 quoted lines

>> Elliott: you can see the relevant changes to the topo-sort in commit
>> 96c4f4a (commit: allow associating auxiliary info on-demand,
>> 2013-04-09).
>>
>> -Peff
>
> Again, might be a little over my head. If you really think it's best
> that I look into that branch, I will try. :)
>
> Meantime, is there any other, more-immediate approach you can think
> of? I thought, for a moment, of only storing *either* the committer
> *or* the author date in the commit-struct at a given time, and
> flagging with a single bit ... but I'm not sure how widely-spread the
> nead for committer-date currently is. Maybe I can go back and
> parse-out the author date *when I need it*, instead, though that
> sounds slow …

You would parse all of them at the beginning of topo-sort function once and store these dates in the commit-info-slab (alongside with indegree). Once you are done sorting, you can discard the slab.

This could be done as a follow-up patch, but the tons of helper functions you added to compare by author date to revision.c will have to be removed in such a transition, because the whole point of using commit-info-slab is not to have commit->author_date field, which these new helpers work on.