Re: [OE-core] [PATCH v2 2/4] package_pkgdata: fix typo to stop calling undefined function

["Mathieu Dubois-Briand" <mathieu.dubois-briand@bootlin.com>]

On Sat Apr 4, 2026 at 7:14 PM CEST, Adam Blank via lists.openembedded.org wrote:
> On Fri, 3 Apr 2026 at 13:31, Mathieu Dubois-Briand <
> mathieu.dubois-briand@bootlin.com> wrote:
>
>> On Fri Apr 3, 2026 at 12:32 PM CEST, Adam Blank wrote:
>> > Hmmm, can't see any connection between those tests and this particular
>> > patch...
>> > The thing is, that since the patch fixes a call to an undefined function,
>> > there could not have been a test covering this case in the first place
>> ;-)
>> > In addition, the function call takes place only in the 'bitbake -b'
>> case...
>> >
>> > Where can it be found, which other changes took place in those builds?
>> >
>> > On Fri, 3 Apr 2026 at 10:00, Mathieu Dubois-Briand <
>> > mathieu.dubois-briand@bootlin.com> wrote:
>> >
>>
>> The tested branch can be found here:
>>
>> https://git.yoctoproject.org/poky-ci-archive/log/?h=oecore/autobuilder.yoctoproject.org/valkyrie/a-full-3579
>>
>> But I confirm git bisect points to this commit, and the error is gone
>> after a revert. You can easily reproduce the issue on your side:
>> oe-selftest -r sstatetests.SStateHashSameSigs2.test_sstate_allarch_samesigs
>>
>>
> I've run this test with my change cherry-picked onto the master, and have
> got some inconclusive, partially random results... Here's what I did:
> - the test started FAILING (as in the autobuilder)
> - I replaced the corrected function invocation with an exception throw and
> nothing changed (this proves, that this branch does not take part in this
> test)
> - I limited the packages built during the test to only some of the ones
> failing - the test started PASSING
> - I started restoring the original package list, and finally restored its
> original state - the test kept PASSING
> - I restored the original state of the workspace - master + my commit (the
> exact same configuration that I started with) and the test kept PASSING
> - I rerun the test a few times, each time in a new, fresh build, and after
> about 3 FAILING runs it suddenly PASSED 2 times, before it started FAILING
> again - all the time in the same state of the workspace
>
> I did confirm, that the signatures differed indeed in the failing runs. I
> chased it for a while, looked here and there (e.g. found it suspicious that
> native, nativesdk, cross-canadian, etc. clear
> out do_packagedata[stamp-extra-info] and allarch does not) but I cannot
> even dream of the type of free time I'd need to familiarise myself
> sufficiently with BitBake's hashing machinery.
>
> I think the way to go now is to file a bug - what do you think?
>
> Thanks,
> Adam
>

Thanks for taking the time to make these tests.

I am not familiar either with these bitbake parts, so yes, except if
someone else comes with an explanation, opening a bug entry is probably
the best move here.

Thanks,
Mathieu

-- 
Mathieu Dubois-Briand, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com