Re: [RFC PATCH 0/8] Get rid of "git --super-prefix"

From: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Date: Nov 11, 2022, 21:17 UTC
Message-ID: <221111.86fsepmbhe.gmgdl@evledraar.gmail.com>
In-Reply-To: <kl6lv8nl1h19.fsf@chooglen-macbookpro.roam.corp.google.com>

On Fri, Nov 11 2022, Glen Choo wrote:

Show 17 quoted lines

> Glen Choo <chooglen@google.com> writes:
>
>> Rereading this series and thinking about this some more, let's go with
>> your approach, primarily because it avoids global state.
>>
>> From this series, it seems that it's not that hard to make this change
>> and support whatever use cases we currently have.
>>
>> This does make it more tedious to add more "--super-prefix" in the
>> future, but that's a good push for us to do more things in-process
>> and/or be more principled about passing context objects through the call
>> stack instead of relying on globals.
>>
>> Let me know what I can add to this effort besides reviewing :)
>
> Specifically, if you have other things on your plate, I'd be happy to
> pick up where where this RFC has left off.

I was going to get around to re-rolling this in the next few days, but I'd also be happy to have you beat me to it.

My plan was basically:

 * Steal the test from your series, put them at the beginning, and for
   those that fail make them "test_expect_failure", then
   "test_expect_success" later when they pass.

 * Pretty much my RFC as-is. If you're re-rolling it I'll leave to you
   whether it makes sense to do it with the "read-tree" included (I
   think probably yes, but "just the submodule--helper" is smaller).

   Rewording the commit message referring to "the other approach"
   (i.e. your series) probably makes sense in light of later discussion
   (probably just dropping it).

 * Right now I can't remember if that one test failed until the
   "read-tree" patch, or if the "submodule--helper" was sufficient, so
   maybe we need the "read-tree" one to flip the
   "test_expect_failure"...

 * The 8/8 has a wart where I just removed "SUPPORT_SUPER_PREFIX" from
   git.c, but didn't adjust the rest of the bitfields, i.e. it should be
   1<<0..6, not 1<<0..3,5..9 at the end (having removed "1<<4". You got
   that right in your version...

If you can do that and address any other nits/issues you find that would be great. I don't think I'd get to it before next week otherwise, but it's earlier in the -0800 TZ :)