Re: What's cooking in git.git (Dec 2021, #05; Thu, 23)

From: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Date: Dec 26, 2021, 17:15 UTC
Message-ID: <211226.8635mfgu7o.gmgdl@evledraar.gmail.com>
In-Reply-To: <CADMgQSSwM3SMysB5=UkXOYLOTUTTN3kF2aaaQwHc+o7Q5qWDcA@mail.gmail.com>

On Sat, Dec 25 2021, Teng Long wrote:

Show 23 quoted lines

> On Thu, 23 Dec 2021 15:42:33 -0800, Junio C Hamano wrote:
>
>
>  > "git ls-tree" learns "--oid-only" option, similar to "--name-only".
>  >
>  > Will merge to 'next'?
>  > source: <cover.1639721750.git.dyroneteng@gmail.com>
>
> Currently, the "source" is patch v6, and there are some test problems in
> [1]. They're not very hard to fix, but I'm considering whether to:
>
>      WAY-1: continue on the current implementation path;
>
>      WAY-2: Combine (steal :) the RFC patch from Ævar Arnfjörð Bjarmason
>      and some commits of mine, to the next patch, because some arguments
>      given from Ævar are compelling (of course the test problems will be
>      fixed too).
>
> So, I will work on it, I think I will send a new patch based on WAY1
> quickly, and send a further RFC patch on WAY2. If both are ok, let the
> community decide which one to use.
>
> Thanks.

Yes there's the test problem I mentioned in [1], but in addition to that your current set of patches have around a ~10% performance regression, as noted in [2]. My RFC series[2] side-steps that by leaving the current code in-place, and only introducing a new optional --format path for new output formats.

I really don't mind if you go for "WAY-1" first over my RFC --format "WAY-2", but I do think any such change should be prominently noting/selling that this new feature is worth the performance regression, or finding some alternate "WAY-1+" to avoid it.

1. https://lore.kernel.org/git/211217.86o85f8jey.gmgdl@evledraar.gmail.com/
2. https://lore.kernel.org/git/RFC-cover-0.7-00000000000-20211217T131635Z-avarab@gmail.com/