Re: [PATCH 1/4] Build in git-repack

From: Duy Nguyen <pclouds@gmail.com>
To: Junio C Hamano <gitster@pobox.com>,
	Stefan Beller <stefanbeller@googlemail.com>
Cc: Matthieu Moy <Matthieu.Moy@grenoble-inp.fr>,
	Fredrik Gustafsson <iveqy@iveqy.com>,
	Git Mailing List <git@vger.kernel.org>
Subject: Re: [PATCH 1/4] Build in git-repack
Date: Thu, 8 Aug 2013 09:44:50 +0700	[thread overview]
Message-ID: <CACsJy8D3FL0rG9YHuSOtobNffiqffU6J7_gfc85CiGuwUdRNnQ@mail.gmail.com> (raw)
In-Reply-To: <7v38qlec2w.fsf@alter.siamese.dyndns.org>

On Wed, Aug 7, 2013 at 10:48 PM, Junio C Hamano <gitster@pobox.com> wrote:
> Matthieu Moy <Matthieu.Moy@grenoble-inp.fr> writes:
>
>> [ It's cool you're working on this, I'd really like a git-repack in C.
>>   That would fix this
>>   http://thread.gmane.org/gmane.comp.version-control.git/226458 ]
>>
>> Stefan Beller <stefanbeller@googlemail.com> writes:
>>
>>> From: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
>>>
>>> pack-objects learns a few more options to take over what's been done
>>> by git-repack.sh. cmd_repack() becomes a wrapper around
>>> cmd_pack_objects().
>>
>> I think the patch would read easier if these were split into two
>> patches: one doing the real stuff in pack-objects, and then getting rid
>> of git-repack.sh to replace it with a trivial built-in.
>>
>> Actually, I'm wondering why pack-objects requires so much changes.
>> git-repack.sh was already a relatively small wrapper around
>> pack-objects, and did not need the new options you add, so why are they
>> needed? In particular adding the new --update-info option that just does
>>
>>> +    if (repack_flags & REPACK_UPDATE_INFO)
>>> +            update_server_info(0);
>>
>> seems overkill to me: why don't you just let cmd_repack call
>> update_server_info(0)?
>
> My feeling exactly.  I would rather see a patch that does not touch
> pack-objects at all, and use run_command() interface to spawn it.
> Once we do have to pack, the necessary processing cycle will dwarf
> the fork/exec latency anyway, no?

I'm not opposed to run_command(). I think the reason I wanted to move
repack functionality to pack-objects is to avoid reading sha-1 from
pack-objects and reconstruct the paths again from the sha-1. But for
simplicity, perhaps we should not touch pack-objects at all. Then we
could have builtin/repack.c instead of stuffing cmd_repack in
builtin/pack-objects.c

@Stefan, if you want to push this work, feel free to take it as _your_
patch, rewrite as will. You don't need to retain my name.
-- 
Duy