Adam Williamson <adamwill(a)fedoraproject.org> writes:
Neal Gompa wrote:
> Adam Williamson <adamwill(a)fedoraproject.org> wrote:
>> Robbie Harwood wrote:
>>> Elliott Sales de Andrade <quantum.analyst(a)gmail.com> writes:
>>>
>>>> Now that ppc64 is gone, s390x is the only big-endian architecture
>>>> left. Bugs around endianness are not usually difficult to fix,
>>>> _if_ I can debug it and see where exactly the problem is. However,
>>>> this requires a tedious guess-a-patch, try a scratch build, check
>>>> the result, rinse and repeat.
>>>>
>>>> Mock (with --forcearch) is completely useless for this. The
>>>> programs just crash during the build in such a way that I can't
>>>> even use `catchsegv`, and gdb is unusable in the container. And
>>>> besides, the programs don't actually crash on real s390x anyway..
>>>>
>>>> Just like we have test machines for other less used architectures
>>>> [1], I am wondering if there is some way we can spin up a test
>>>> machine for s390x?
>>>>
>>>> [1]
https://fedoraproject.org/wiki/Test_Machine_Resources_For_Package_Maintai...
>>>
>>> It's very strange to me that having test hardware available isn't a
>>> requirement for being a Primary architecture, or for that
>>> architecture being present in koji. IMO we should change that
>>> going forward.
>>
>> s390x isn't a primary arch. It's an alternative arch.
That's true, which is why I had the "or". I'd like it to be a
requirement for either/both.
>>
https://fedoraproject.org/wiki/Architectures
>
> That page is out of date. All architectures are effectively primary
> now, since failures for any arch block builds from releasing in Koji.
We still draw a distinction between the two, it just doesn't have that
dimension to it any more. The page even explains this in its definition
at the top. The distinction is rather smaller now, but still there.
In the release criteria we've mostly switched to using the term
"release-blocking arches", and s390x isn't one of those either. :)
You're right, I'm being loose with language. Neal's point is what I'm
trying to articulate: whatever the formal position is, we as packagers
have to care about making this architecture work, since our builds won't
go through if it doesn't.
Thanks,
--Robbie