Re: What's cooking in git.git (Aug 2021, #09; Sun, 29)

From: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Date: Aug 30, 2021, 13:27 UTC
Message-ID: <87o89fxd5l.fsf@evledraar.gmail.com>
In-Reply-To: <CAFQ2z_PCUr3D0H-fzfHcFFayzdDJc64=JgVM5_2kuLJn-G9awA@mail.gmail.com>

On Mon, Aug 30 2021, Han-Wen Nienhuys wrote:

Show 13 quoted lines

> On Mon, Aug 30, 2021 at 2:03 PM Ævar Arnfjörð Bjarmason
> <avarab@gmail.com> wrote:
>
>> Han-Wen: It would be really nice to have this re-rolled sooner than
>> later, my and Junio's "fixup" commits at the end are a band-aid, but I
>> already ran into wanting to bisect something between master..seen that
>> was unnecessarily painful due to the series now not compiling without
>> the "fixup" commits at the end.
>
> I am folding in the __FUNCTION__ fix.
>
> However, I thought patches should be sent against the master branch
> and not against some intermediate  version of 'seen' ?

No, per "Decide what to base your work on" in Documentation/SubmittingPatches. See the part about "A new feature should be based on `master` in general[...]".

In this case the topic on Junio's side isn't based on "master", hence the bisect breakages noted above.

I don't know per the upthread what Junio's intended pace of merging down ab/refs-files-cleanup and hn/refs-errno-cleanup is, depending on the answer to that perhaps it would be best to wait and re-roll on a new master.

Also for any re-roll of hn/refable, see my <877dgch4rn.fsf@evledraar.gmail.com> for notes of other breakages in it. I also had the impression based on my <87h7jqz7k5.fsf@evledraar.gmail.com> and your reply in <CAFQ2z_P8vgY0RRT+XSH9K3VDQt39FLqXx6qfeZqaZPkwhq1w+A@mail.gmail.com> back in April/May that the intent was to re-roll this topic in a way that would be more split-up as described there.

I don't care much about the exact end-state there, other than:

 1. Trying to land reftable/ in some shape where we're not constantly
    re-reviewing the "add upstream library" part of it.

 2. Not have something like hn/reftable in-tree which per my above is in
    a demonstrably broken/experimental state, but our docs are in state
    of noting it like it's just another ref backend. Hence the
    suggestion of first landing the library with its own tests, getting
    the integration and GIT_TEST_REFABLE passing, and finally updating
    docs etc. to advertise the thing to users.