Bug 2045186

Created attachment 1853611 [details]
build.log

file build.log too big, will only attach last 32768 bytes

Created attachment 1853612 [details]
root.log

file root.log too big, will only attach last 32768 bytes

Created attachment 1853613 [details]
state.log

Currently there is one test failure on architectures other than s390x. It looks like it could be a Google Test issue.

Scratch build: https://koji.fedoraproject.org/koji/taskinfo?taskID=82195530

(In reply to Ben Beasley from comment #4)
> Currently there is one test failure on architectures other than s390x. It
> looks like it could be a Google Test issue.
> 
> Scratch build: https://koji.fedoraproject.org/koji/taskinfo?taskID=82195530

Sorry, that last comment was with LTO disabled. Running another scratch build to see if there are still LTO problems, as in RHBZ#2042434: https://koji.fedoraproject.org/koji/taskinfo?taskID=82197683

*** Bug 2042434 has been marked as a duplicate of this bug. ***

Looks like LTO is OK now, so there’s just the test failure to deal with.

It looks like we now have a version of googletest that understands GTEST_ALLOW_UNINSTANTIATED_PARAMETERIZED_TEST. If I try removing this from %prep:

> # Remove macro only defined in googletest git master
> sed -i 's|GTEST_ALLOW_UNINSTANTIATED_PARAMETERIZED_TEST|//|' absl/container/internal/unordered_map_modifiers_test.h

then I get a build that is successful on everything but ppc64le: https://koji.fedoraproject.org/koji/taskinfo?taskID=82204878

The following tests FAILED:
	 81 - absl_random_beta_distribution_test (Failed)
	 87 - absl_random_exponential_distribution_test (Failed)
	 88 - absl_random_gaussian_distribution_test (Failed)
	 90 - absl_random_uniform_real_distribution_test (Subprocess aborted)
	111 - absl_random_internal_iostream_state_saver_test (Failed)

It looks like most of the failures are related to __ieee128 types:

[  FAILED  ] ExponentialDistributionTypedTest/2.SerializeTest, where TypeParam = __ieee128
[  FAILED  ] GaussianDistributionInterfaceTest/2.SerializeTest, where TypeParam = __ieee128 (282 ms)
[  FAILED  ] GaussianDistributionInterfaceTest/2.SerializeTest, where TypeParam = __ieee128
[  FAILED  ] IOStreamStateSaver.RoundTripLongDoubles
[  FAILED  ] BetaDistributionInterfaceTest/2.SerializeTest, where TypeParam = __ieee128


The subprocess aborted was also related to long double:

Expected equality of these values:
  before
    Which is: 6.953283485615657412940250901535356313e-310 6.953283485615657412940250901535356313e-310
  after
    Which is: 6.953283485615657412940250901535356313e-310 6.953283485615657412940250901535356313e-310
absl_random_uniform_real_distribution_test: /builddir/build/BUILD/abseil-cpp-20210324.2/absl/random/uniform_real_distribution.h:75: absl::lts_20210324::uniform_real_distribution<RealType>::param_type::param_type(absl::lts_20210324::uniform_real_distribution<RealType>::result_type, absl::lts_20210324::uniform_real_distribution<RealType>::result_type) [with RealType = long double; absl::lts_20210324::uniform_real_distribution<RealType>::result_type = long double]: Assertion `lo <= hi' failed.

This might be related to the ppc64 long double changes, I'm blocking the ppc tracker

Yes, it will be related. Does abseil depend on some external libs? Perhaps it's just a build order issue ...

After looking at the code then at least the beta distribution contains some expectation about the behaviour of the "old" IBM long double type for PPC ...

So there is something wrong somewhere :-) I have reduced the absl_random_beta_distribution_test source substantially, so it should be easier to debug.

Created attachment 1858360 [details]
reduced test

Do you have preprocessed version of the reduced test?

Looks to me there is something wrong with the stringstream reader/writer, I am slowly working on it. IMO it's not clear it's GCC's fault, could be libstdc++ ...

Created attachment 1859579 [details]
preprocessed test case source

and the command line:
/usr/bin/g++  -I/home/sharkcz/abseil-cpp/abseil-cpp-20210324.2 -O2 -flto=auto -ffat-lto-objects -fexceptions -g -grecord-gcc-switches -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -fstack-protector-strong -specs=/usr/lib/rpm/redhat/redhat-annobin-cc1  -m64 -mcpu=power8 -mtune=power8 -fasynchronous-unwind-tables -fstack-clash-protection -Wall -Wextra -Wcast-qual -Wconversion-null -Wformat-security -Wmissing-declarations -Woverlength-strings -Wpointer-arith -Wundef -Wunused-local-typedefs -Wunused-result -Wvarargs -Wvla -Wwrite-strings -DNOMINMAX -Wno-conversion-null -Wno-deprecated-declarations -Wno-missing-declarations -Wno-sign-compare -Wno-unused-function -Wno-unused-parameter -Wno-unused-private-field -std=gnu++17 -MD -MT absl/random/CMakeFiles/absl_random_beta_distribution_test.dir/beta_distribution_test.cc.o -MF CMakeFiles/absl_random_beta_distribution_test.dir/beta_distribution_test.cc.o.d -o CMakeFiles/absl_random_beta_distribution_test.dir/beta_distribution_test.cc.o -c /home/sharkcz/abseil-cpp/abseil-cpp-20210324.2/absl/random/beta_distribution_test.cc

I can reproduce it with just:
#include <sstream>

int
main ()
{
  std::stringstream ss;
  long double a = 9.99999999999999945153271454209571652e-21L;
  long double b;
  ss << a;
  ss >> b;
  __builtin_printf ("%Lg %Lg\n", a, b);
  return 0;
}

which with -mabi=ibmlongdouble prints
1e-20 1e-20
but with -mabi=ieeelongdouble prints
1e-20 6.95326e-310

So looks like libstdc++ issue.

This bug appears to have been reported against 'rawhide' during the Fedora 36 development cycle.
Changing version to 36.

Fixed in upstream GCC now.

Thanks! I opened a PR with the Google Test fix. Hopefully once GCC is updated in Rawhide it will be possible merge and build it without further changes.

https://src.fedoraproject.org/rpms/abseil-cpp/pull-request/2

Fix is present in gcc-12.0.1-0.8, which is currently building in koji:

https://src.fedoraproject.org/rpms/gcc/c/9743a6a9b014b9a1127626dfae25753feab77ae4?branch=rawhide
https://koji.fedoraproject.org/koji/buildinfo?buildID=1918054
https://koji.fedoraproject.org/koji/buildinfo?buildID=1918053

FEDORA-2022-6c93a8a110 has been submitted as an update to Fedora 36. https://bodhi.fedoraproject.org/updates/FEDORA-2022-6c93a8a110

FEDORA-2022-6e05296377 has been submitted as an update to Fedora 37. https://bodhi.fedoraproject.org/updates/FEDORA-2022-6e05296377

FEDORA-2022-6c93a8a110 has been pushed to the Fedora 36 stable repository.
If problem still persists, please make note of it in this bug report.

FEDORA-2022-6e05296377 has been pushed to the Fedora 37 stable repository.
If problem still persists, please make note of it in this bug report.