Hi,
I'm studying Boost on Power systems.
On Fedora31/PPC64LE, I've noticed that the tests are not automatically run for Boost
and for some of its dependencies, and that there are several issues:
- Boost :
- boost-1.69.0 releases 9 & 11
- by default, tests of Boost are not run by means of the .spec file:
- empty %check part in the .spec file of boost-1.69.0-11.fc32 and
boost-1.69.0-9.fc31
- tests are not even run as part of the %build or %install phases
- I've run the tests, and before it was complete (my disk space went full), I
got:
- failures are detected : 20,099
- Passed : 102,720
- Failed : 10,383
- numpy :
- by default, no tests are run on ppc64le:
numpy-1.17.2-1.fc31.spec :
%check
%if %{_arch} != s390x && %{_arch} != ppc64le
python3 runtests.py
%endif
- however, results are good:
1 failed, 10146 passed, 69 skipped, 173 deselected, 19 xfailed, 2 xpassed in
376.67 seconds
- lapack :
- lapack-3.8.0-13.fc31
- by default, no tests done since no %check phase
- trying to run the tests by hand, it failed saying: "missing a rule for
building librefblas.a" .
- atlas:
- atlas-3.10.3-9.fc31
- the build crashes while trying to use some libptlapack.a library which was not
built.
Maybe I made some mistake. However, for other big packages, like golang, MariaDB, or
MySQL, that worked fine.
I've tried to get some information from the Boost community, talking to RedHat people
and to the developers email list, but I got no answer yet.
And PPC64LE does not appear in the BuildFarm of the Boost community.
Who could provide me with some information and help ?
I'd like to have official tests results for these 4 packages: Boost, numpy, lapack,
and atlas, as I'll have to port them on AIX/PPC64BE.
I need some reference for further tests results comparison between Fedora/PPC64LE and
AIX.
And I'm not sure that these 4 packages are fully tested on Fedora/PPC64LE.
Thanks/Regards
Tony
Grenoble - France
IBM AIX OpenSource project
Show replies by thread
As I said in my recent mail, the place to address this is upstream.
And PPC64LE does not appear in the BuildFarm of the Boost community.
I can see an obvious solution to that: somebody with access to Power systems should set up
an automated regression tester and contribute test results to the upstream project. That
way the Boost devs will know there are problems. Fedora is not the place to fix it if the
code is in such poor state for Power.
https://www.boost.org/development/running_regression_tests.html
Hi Jonathan,
I talked in this place since someone in IBM recommended me to do so.
I do not master yet all the places where I should talk.
About Fedora, whatever the HW, at least one thing should be done: add tests (cd status;
../b2) in the %check section of the .spec file.
Boost tests take a lot of time and consume about 65GB of disk space. Anyway, this is the
best solution for building an idea of the quality of a software. Customers are expecting
us to test the software we deliver and to not rely on tests that were done previously and
in different environments.
I'm now running Boost tests on Fedora31/x86_64. There are also failures. However, MUCH
MUCH less than on PPC64LE.
However, I may have made some mistake when building/testing Boost on PPC64LE. However, I
followed exactly what is explained on Boost test page you provide: cd status ; ../b2 .
About adding a Fedora/PPC64LE machine to Boost buildfarm, that is the best solution. I
agree. However, for now, I see only Intel HW and only Linux & Windows Operating
Systems on Boost buildfarm. Adding a different HW and some OS like BSD would help
improving the overall quality and stability of Boost.
Thanks !
Tony