#1028041 php-excimer: FTBFS on mipsel

Package:
src:php-excimer
Source:
src:php-excimer
Submitter:
Paul Gevers
Date:
2023-02-02 20:12:02 UTC
Severity:
normal
Tags:
#1028041#5
Date:
2023-01-06 07:59:08 UTC
From:
To:
Dear maintainer,

The php8.2 transition started recently, and while going over the
packages involved I noticed that your latest upload (currently in
unstable only) is failing to build from source on mipsel.

It fails its tests.
=====================================================================
FAILED TEST SUMMARY
---------------------------------------------------------------------
ExcimerProfiler CPU profile [tests/cpu.phpt]
=====================================================================

Please fix ASAP to not block the php8.2 transition.

Paul

https://buildd.debian.org/status/fetch.php?pkg=php-excimer&arch=mipsel&ver=1.0.4-2&stamp=1672548054&raw=0
#1028041#10
Date:
2023-01-06 18:47:49 UTC
From:
To:
It built after I gave it back to build on the right buildd,
so it's now rebuilt with PHP 8.2.

The issue might depend on buildd speed, or some weird difference
between buildds.

cu
Adrian
#1028041#15
Date:
2023-02-02 19:23:21 UTC
From:
To:
severity 1028041 normal
thanks

Hi,

Thanks.

Looking through my email, it's flaked before in the past (#1014801). The
test in question[1] has the following comment:

// Test aggregateByFunction
// Typically the parent functions foo() and bar() will have self=0 and
// inclusive ~= 30. The other 4 functions will have a count of about
30/4 = 7.5.
// The probability of C::member() or baz() having a count of zero is
about 1 in 5600.

Maybe the known flakiness is worse due to something on the mipsel build?
I'll ask the upstream author. Worst case we can just skip the test on
mipsel.

I'm lowering the severity because it no longer blocks the 8.2 transition
(please revert me if I'm wrong on that).

[1]
https://salsa.debian.org/mediawiki-team/php-excimer/-/blob/master/tests/cpu.phpt
#1028041#22
Date:
2023-02-02 19:49:15 UTC
From:
To:
Hi Kunal,

In my opinion flaky builds and flaky tests are bad because they cost
quite some time of people that are not involved in the package. E.g. in
this case Release Team, but imagine a security upload where the build
fails. But you have flaky and flaky. https://bugs.debian.org/844264 is
an interesting read in that regard.

As somebody that maintains infrastructure and does a lot of QA
(ci.debian.net) and is involved in transitions and migrations where I
see a lot of flaky tests, I'm biased. But really, a test that fails more
often than say 1/5 times is RC in my book. I didn't check the ratio of
php-excimer failures.

Paul
#1028041#27
Date:
2023-02-02 20:08:46 UTC
From:
To:
I am not convinced that this is flakiness, my guess (that could be wrong!)
would be tests failing on all except the fastest mipsel buildds.

It is problematic that the test doesn't output what exactly fails.

On the porterbox the test passes initially, but it fails after
  -$profiler->setPeriod(0.1);
  +$profiler->setPeriod(10);
I do not know whether or not this is the same failure as on the mipsel
buildds.

cu
Adrian