Re: What's cooking in git.git (Sep 2021, #01; Thu, 2)

From: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Date: Sep 6, 2021, 22:21 UTC
Message-ID: <87h7exnxzh.fsf@evledraar.gmail.com>
In-Reply-To: <xmqq4kaxe5dt.fsf@gitster.g>

On Mon, Sep 06 2021, Junio C Hamano wrote:

Show 29 quoted lines

> Han-Wen Nienhuys <hanwen@google.com> writes:
>
>> On Fri, Sep 3, 2021 at 12:48 AM Junio C Hamano <gitster@pobox.com> wrote:
>>> * hn/reftable (2021-08-26) 29 commits
>>>  - SQUASH??? https://github.com/git/git/runs/3439941236?check_suite_focus=true#step:5:700
>>>  - reftable: fixup for new base topic 3/3
>>>  - reftable: fixup for new base topic 2/3
>>>  - reftable: fixup for new base topic 1/3
>> ..
>>>  The "reftable" backend for the refs API.
>>
>> I posted a subset of these patches as
>> https://lore.kernel.org/git/pull.1081.git.git.1630335476.gitgitgadget@gmail.com/
>
> Thanks for a ping.  I saw it but haven't read it.
>
>> As discussed with AEvar, it will probably speed things up if we can
>> focus on getting the base library submitted without hooking it up to
>> Git. This would avoid cross-interactions with other pending topics,
>> and reduce the size of the more controversial topic (hooking it up to
>> Git).
>
> My worry is that a "base library" that is not hooked up to anything
> that works in the system would not be properly reviewed at all.  Of
> course, without review, it would speed things up, but it is unclear
> if that the kind of speed we want.
>
> Anyway, I'll eject the old topic and replace them with the latest
> one soonish.

The proposal is to still have it hooked up to t/helper/test-reftable.c, the build system, and t0032-reftable-unittest.sh.

So we'd build and test it on all platforms, getting that working in some stable fashion would already be a big step forward.

What we wouldn't get right away is the series as of things like 0a0d5fe74d4 (refs: RFC: Reftable support for git-core, 2021-08-17), i.e. hooking up refs/reftable-backend.c, changes to refs.[ch], running reftable with some mode of "git init".

As discussed elsewhere that step would currently fail with some tests failing, but most/all of those failures are due to the integration of the reftable library into git, not failings of the library itself or the reftable format.