Re: What's cooking in git.git (Sep 2021, #06; Mon, 20)

["Ævar Arnfjörð Bjarmason" <avarab@gmail.com>]

On Thu, Sep 23 2021, Junio C Hamano wrote:

> Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes:
>
>>> * ab/sanitize-leak-ci (2021-09-20) 2 commits
>>>  - tests: add a test mode for SANITIZE=leak, run it in CI
>>>  - Makefile: add SANITIZE=leak flag to GIT-BUILD-OPTIONS
>>>
>>>  CI learns to run the leak sanitizer builds.
>>>
>>>  Will merge to 'next'?
>>
>> Yes please, as noted in
>> https://lore.kernel.org/git/87r1dh8r62.fsf@evledraar.gmail.com/ more
>> leak fixes are waiting on this.
>
> Isn't the direction of dependency the other way?  Once fixes go in,
> checks will no longer complain, but until fixes are reviewed and
> applied, checking at CI will break the testing and this would need
> working around by marking various tests as "not ready to be tested".

The fixes are structured as fixing the code, and then for both
self-documentation & testing turning on:

    TEST_PASSES_SANITIZE_LEAK=true

In the same commit, I could fix the leaks and have a second series later
for turning on the regression test, or just turning on the test and
having it kick in once it's merged with ab/sanitize-leak-ci, but waiting
until ab/sanitize-leak-ci hit master first seemed less confusing for
everyone.

But if you'd like to have 'em now with either of those caveats...

>> It seems due to the size of this series we might be better off splitting
>> up this already split-up series.
>>
>> I was trying to convince you to merge down the much more trivial changes
>> up to the <mark0> above, which are purely build system prep work. Any
>> chance you'd reconsider?
>>
>> I think a plan that might be better would be:
>>
>>  1. Eject both ab/config-based-hooks-base & es/config-based-hooks
>>  2. I'd submit a series up to the <mark0>, hopefully that can land fairly
>>     soon thereafter.
>>  3. Once that's in, another one for <mark0>..<mark1>, i.e. the "git hook
>>     run" command, but only the more basic bits, and migrating fairly
>>     simple hooks.
>>  4. Then <mark1>..<mark2>, followed by <mark2>..ab/config-based-hooks-base
>>  5. Then Emily's es/config-based-hooks.
>>
>> That's converting a 2-step process (ab/config-based-hooks-base +
>> es/config-based-hooks) into 5 steps, but I suspect it would go faster,
>> right now it seems there's no takers for a review of this rather large
>> series. What do you & Emily think?

[Just for anyone reading along, I've since pulled the trigger on that
proposed plan at
https://lore.kernel.org/git/cover-0.8-00000000000-20210923T095326Z-avarab@gmail.com/]

> Today I learned a phrase "six of one and half a dozen of another" ;-)
>
> I have been wondering if something like a book reading club is
> doable on the list.  A chairperson nominates (and perhaps resends) a
> series of patches that is of manageable size to be reviewed, perhaps
> even prime the discussion with some comments, likeminded folks chime
> in and we end up with a reviewed series?

Some of your colleagues at Google have been running such a review for
what I understand to have been a while, and have recently started
inviting external people. I attended one & applaud the effort.

I think we could run a similar thing here on-list if there's interest
(I'd participate). Perhaps just:

 1. Include a list of un-loved topics in What's Cooking (or someone
    could do it as a follow-up)
 2. Pick N, and distill it down to one, either asa BDFL decision or via
    some easy voting mechanism, like https://doodle.com or E-Mail replies.
 3. Send out a call for reviews with a re-send some time (a week?) later.

If you're not up for organizing it with everything you've got on your
plate I'd be up for doing it. I'd basically just:

 1. Read What's Cooking, subjectively note topics that seem to need love.
 2. Pick one for next week's on-list review club. Via the BFDL-decreed
    methof of "shuf | head -n 1" or similar.
 3. Send out weekly REVIEW CLUB E-Mail with a link to the relevant
    series (if active), or have it be a --cover-letter with a full
    re-send if it's been dead on-list for long enough for nobody to be
    bothered by two parallel threads as a result.

What do you & others think?