Re: [PATCH v2 4/4] log: --author-date-order

From: Junio C Hamano <gitster@pobox.com>
Date: Jun 20, 2013, 19:36 UTC
Message-ID: <7v61x8tw0a.fsf@alter.siamese.dyndns.org>
In-Reply-To: <20130610184918.GC2084@sigill.intra.peff.net>

Jeff King <peff@peff.net> writes:

Show 18 quoted lines

>> Or we could extend parse_commit() API to take an optional commit
>> info slab to store not just author date but other non-essential
>> stuff like people's names, and we arrange that extended API to be
>> triggered when we know --author-date-order is in effect?
>
> I like the latter option. It takes a non-trivial amount of time to load
> the commits from disk, and now we are potentially doing it 2 or 3 times
> for a run (once to parse, once to get the author info for topo-sort, and
> possibly later to show it if --pretty is given; though I did not check
> and maybe we turn off save_commit_buffer with --pretty). It would be
> nice to have an extended parse_object that handled that. I'm not sure of
> the interface. Maybe variadic with pairs of type/slab, like:
>
>   parse_commit_extended(commit,
>                         PARSE_COMMIT_AUTHORDATE, &authordate_slab,
>                         PARSE_COMMIT_DONE);
>
> ?

What I had in mind actually was a custom slab tailored for each caller that is an array of struct. If the caller is interested in authordate and authorname, instead of populating two separate authordate_slab and authorname_slab, the caller declares a

	struct {
        	unsigned long date;
                char name[FLEX_ARRAY];
	} author_info;

prepares author_info_slab, and use your commit_parser API to fill them.