threads / discuss / 55773

CI: 'seen' minus a few passes tests, eh, almost

Subject: CI: 'seen' minus a few passes tests, eh, almost

## tl;dr

8 messages between May 25, 2021 and Jun 2, 2021.

replies: 7people: 5as markdown or json

Junio C Hamano· May 25, 2021, 22:25 UTC · lore

It is sad that the tip of 'seen' rarely passes all tests these days.

I excluded a few topics (chosen primarily by prejudice after looking at which tests failed earlier, without locating exactly where and why they fail) from 'seen' and the exclusion of them seems to have made 'seen' finally pass the tests, eh, almost [*1*].

82b1e47 is such a "reduced" 'seen' branch, and the CI run for it can be found here:

  https://github.com/git/git/actions/runs/876448385

Excluded are:

  ab/fsck-unexpected-type
  hn/reftable
  hn/refs-errno-cleanup
  zh/ref-filter-raw-data

Any of these may not be broken by themselves, and it could be that the tests may still pass even if one or some of them are merged on top of 82b1e47 and the problem may manifest only when two or more of these topics in certain combinations are merged to 'seen'.

The CI run for 'seen' with all these four topics merged back can be found here:

  https://github.com/git/git/actions/runs/876534959

[Footnote]

*1* Except for "git repack" segfaults only in one of the tests:

  https://github.com/git/git/runs/2669668815?check_suite_focus=true#step:5:2846

Junio C Hamano· May 26, 2021, 05:32 UTC · re: Junio C Hamano · lore

Re: CI: 'seen' minus a few passes tests, eh, almost

Junio C Hamano <gitster@pobox.com> writes:

Show 20 quoted lines

> It is sad that the tip of 'seen' rarely passes all tests these days.
>
> I excluded a few topics (chosen primarily by prejudice after looking
> at which tests failed earlier, without locating exactly where and
> why they fail) from 'seen' and the exclusion of them seems to have
> made 'seen' finally pass the tests, eh, almost [*1*].
> ...
> Excluded are:
>
>   ab/fsck-unexpected-type
>   hn/reftable
>   hn/refs-errno-cleanup
>   zh/ref-filter-raw-data
>
> ...
> [Footnote]
>
> *1* Except for "git repack" segfaults only in one of the tests:
>
>   https://github.com/git/git/runs/2669668815?check_suite_focus=true#step:5:2846

Just with a hunch, I further excluded tb/multi-pack-bitmaps from 'seen', in addition to these four topics. It seems to pass the whole test suite on both 32- and 64-bit build combinations.

    https://github.com/git/git/actions/runs/877225207

Except that vs-build sometimes fails to download the sdk, that is.

Junio C Hamano· May 26, 2021, 06:07 UTC · re: Junio C Hamano · lore

Re: CI: 'seen' minus a few passes tests, eh, almost

Junio C Hamano <gitster@pobox.com> writes:

Show 30 quoted lines

> Junio C Hamano <gitster@pobox.com> writes:
>
>> It is sad that the tip of 'seen' rarely passes all tests these days.
>>
>> I excluded a few topics (chosen primarily by prejudice after looking
>> at which tests failed earlier, without locating exactly where and
>> why they fail) from 'seen' and the exclusion of them seems to have
>> made 'seen' finally pass the tests, eh, almost [*1*].
>> ...
>> Excluded are:
>>
>>   ab/fsck-unexpected-type
>>   hn/reftable
>>   hn/refs-errno-cleanup
>>   zh/ref-filter-raw-data
>>
>> ...
>> [Footnote]
>>
>> *1* Except for "git repack" segfaults only in one of the tests:
>>
>>   https://github.com/git/git/runs/2669668815?check_suite_focus=true#step:5:2846
>
> Just with a hunch, I further excluded tb/multi-pack-bitmaps from
> 'seen', in addition to these four topics.  It seems to pass the
> whole test suite on both 32- and 64-bit build combinations.
>
>     https://github.com/git/git/actions/runs/877225207
>
> Except that vs-build sometimes fails to download the sdk, that is.

And https://github.com/git/git/actions/runs/877379968 tests the above with hn/refs-errno-cleanup added back. The usual tests all seem to pass, even though vs-build fails to download the sdk.

Taylor Blau· May 26, 2021, 14:39 UTC · re: Junio C Hamano · lore

Re: CI: 'seen' minus a few passes tests, eh, almost

On Wed, May 26, 2021 at 02:32:19PM +0900, Junio C Hamano wrote:

Show 7 quoted lines

> Just with a hunch, I further excluded tb/multi-pack-bitmaps from
> 'seen', in addition to these four topics.  It seems to pass the
> whole test suite on both 32- and 64-bit build combinations.
>
>     https://github.com/git/git/actions/runs/877225207
>
> Except that vs-build sometimes fails to download the sdk, that is.

Thanks for letting me know. I have accumulated a few local changes that I wanted to include in a reroll of that topic, but I was waiting on other reviewer eyes before sending it.

I should be able to spend some more time pushing that topic forward after the release. There are a few other patch series waiting on it from GitHub, too.

Thanks, Taylor

Jeff King· May 26, 2021, 15:53 UTC · re: Taylor Blau · lore

Re: CI: 'seen' minus a few passes tests, eh, almost

On Wed, May 26, 2021 at 10:39:53AM -0400, Taylor Blau wrote:

Show 12 quoted lines

> On Wed, May 26, 2021 at 02:32:19PM +0900, Junio C Hamano wrote:
> > Just with a hunch, I further excluded tb/multi-pack-bitmaps from
> > 'seen', in addition to these four topics.  It seems to pass the
> > whole test suite on both 32- and 64-bit build combinations.
> >
> >     https://github.com/git/git/actions/runs/877225207
> >
> > Except that vs-build sometimes fails to download the sdk, that is.
> 
> Thanks for letting me know. I have accumulated a few local changes that
> I wanted to include in a reroll of that topic, but I was waiting on
> other reviewer eyes before sending it.

I think the best thing in a situation like this (where reviewers like me have been very slow to respond) is to repost it explaining the situation. That makes it clear what the current state is, and brings it to the forefront of the list's attention (both reviewers you'd expect to look at it, but people in general). Doing that too often can be obnoxious, of course, but after 6-8 weeks saying "hey, this didn't get attention; can anybody take a look at it?" is perfectly reasonable.

> I should be able to spend some more time pushing that topic forward
> after the release. There are a few other patch series waiting on it from
> GitHub, too.

But yeah, waiting until after the release is good, to keep attention on regression fixes in the meantime.

-Peff

Taylor Blau· May 26, 2021, 16:48 UTC · re: Jeff King · lore

Re: CI: 'seen' minus a few passes tests, eh, almost

On Wed, May 26, 2021 at 11:53:34AM -0400, Jeff King wrote:

Show 22 quoted lines

> On Wed, May 26, 2021 at 10:39:53AM -0400, Taylor Blau wrote:
>
> > On Wed, May 26, 2021 at 02:32:19PM +0900, Junio C Hamano wrote:
> > > Just with a hunch, I further excluded tb/multi-pack-bitmaps from
> > > 'seen', in addition to these four topics.  It seems to pass the
> > > whole test suite on both 32- and 64-bit build combinations.
> > >
> > >     https://github.com/git/git/actions/runs/877225207
> > >
> > > Except that vs-build sometimes fails to download the sdk, that is.
> >
> > Thanks for letting me know. I have accumulated a few local changes that
> > I wanted to include in a reroll of that topic, but I was waiting on
> > other reviewer eyes before sending it.
>
> I think the best thing in a situation like this (where reviewers like me
> have been very slow to respond) is to repost it explaining the
> situation. That makes it clear what the current state is, and brings it
> to the forefront of the list's attention (both reviewers you'd expect to
> look at it, but people in general). Doing that too often can be
> obnoxious, of course, but after 6-8 weeks saying "hey, this didn't get
> attention; can anybody take a look at it?" is perfectly reasonable.

Indeed, but I had been putting even doing that off. I've had some sporadic time off in the last few months (as you know) and so usually by the time that I'm back enough other stuff has accumulated that I don't feel like adding more to the list by encouraging further review.

So the topic (and my participation on the list in general) has kind of languished, but I'm hoping that since things have mostly cooled down for me outside of work that I'll be able to participate and submit patches more consistently.

Thanks, Taylor

Johannes Schindelin· Jun 2, 2021, 11:59 UTC · re: Junio C Hamano · lore

Re: CI: 'seen' minus a few passes tests, eh, almost

Hi Junio,

On Wed, 26 May 2021, Junio C Hamano wrote:

> [CI failed, and was fixed...]
>
> Except that vs-build sometimes fails to download the sdk, that is.

I have a patch series to help with that (at least a little bit), but I am holding it until after v2.32.0.

Ciao, Dscho

ZheNing Hu· May 26, 2021, 15:02 UTC · re: Junio C Hamano · lore

Re: CI: 'seen' minus a few passes tests, eh, almost

Junio C Hamano <gitster@pobox.com> 于2021年5月26日周三 上午6:25写道：

Show 20 quoted lines

>
> It is sad that the tip of 'seen' rarely passes all tests these days.
>
> I excluded a few topics (chosen primarily by prejudice after looking
> at which tests failed earlier, without locating exactly where and
> why they fail) from 'seen' and the exclusion of them seems to have
> made 'seen' finally pass the tests, eh, almost [*1*].
>
> 82b1e47 is such a "reduced" 'seen' branch, and the CI run for it can
> be found here:
>
>   https://github.com/git/git/actions/runs/876448385
>
> Excluded are:
>
>   ab/fsck-unexpected-type
>   hn/reftable
>   hn/refs-errno-cleanup
>   zh/ref-filter-raw-data
>

Hi, Junio, I find my patch may cause some bug: I ignored those blobs with no content (objectsize == 0) before, Some things about "--sort=contents:raw" need to be changed. so please remove my patch from `seen`.

-- ZheNing Hu