Re: binutils builder.sourceware.org status

[Jan Beulich <jbeulich@suse.com>]

On 03.07.2022 18:28, Mark Wielaard wrote:
> Hi,
> 
> A quick overview of the status of the buildbot binutils status (and
> enabling of libctf for those builders that use GCC12 now).
> 
> https://builder.sourceware.org/buildbot/#/builders?tags=binutils
> We have the following fully green CI builders:
> 
> binutils-debian-amd64
> binutils-debian-ppc64
> binutils-fedora-ppc64le
> binutils-fedora-x86_64
> binutils-opensuseleap-x86_64
> binutils-opensusetw-x86_64
> binutils-rawhide-x86_64
> 
> And the following red CI builders:
> 
> binutils-debian-arm64
> binutils-debian-armhf
> binutils-debian-i386
> binutils-fedora-s390x
> 
> All 4 fail some ld tests. For now I disabled the ld tests on arm64 and
> armhf to turn them green and also because they are both running
> debian-old stable and are somewhat slow.
> 
> binutils-debian-i386 fails:
> 
> FAIL: Run p_align-1b with PIE
> FAIL: Run p_align-1d with -Wl,-z,max-page-size=0x1000 with PIE

This presumably is, as H.J. would put it, because of a not-up-to-date
kernel. Personally I continue to think that binutils shouldn't test
kernel functionality, and hence the test should be skipped when run
on un-patched kernels. But as per my recent x86-64 side skipping
patch (to cover old gcc) I lack any good idea how to probe for this
without effectively using the testcase itself as a probe.

Jan

> FAIL: Run pr19031
> FAIL: Run got1
> FAIL: Undefined weak symbol (-fPIE -no-pie)
> FAIL: Undefined weak symbol (-fPIE -pie)
> 
> binutils-fedora-s390x fails:
> 
> FAIL: pr26580-3
> FAIL: Run pr19719 fun undefined
> 
> The are also two "full" binutils-gdb builders:
> 
> binutils-gdb-fedrawhide-x86_64
> binutils-gdb-clang-fedrawhide-x86_64
> 
> Which are currently in a somewhat limbo state since the gdb testsuite
> seems to crash the kernel... Test results (when the builder do
> succeed) are in bunsen though. Which is interesting to compare to see
> what difference a non-gcc compiler makes.
> https://builder.sourceware.org/testruns/?has_keyvalue_like_k=testrun.git_describe&has_keyvalue_like_v=%25binutils-gdb%25
> 
> There are try-builders for all of the above, minus the full
> binutils-gdb builder. The try builders aren't used that much yet. And
> one complaint was that since several CI builders have existing
> failures it isn't immediately clear whether a patch caused something
> or not. So it would be nice if we could resolve (or xfail) the above
> tests.
> 
> We are seeing some git update failures on s390x and ppc64le, it isn't
> entirely sure why this happens only on these arches, but it is
> annoying since it is seen as a new failure and so sents warning
> emails. Still investigating how to work around this.
> 
> The fedora-x86_64, fedora-s390x fedora-ppc64le, opensuse
> tumbleweed-x86_64 and rawhide-x86_64 have been upgraded and use GCC12
> now. For these builder (and the try builders) libctf is now explicitly
> enabled, compiled and tested. libctf logs are now also uploaded to
> bunsed. Please double check the master.cfg patch to see if it does the
> right thing now.
> 
> Thanks,
> 
> Mark