Bug 990668

Hmm, probably should have %{?_smp_flags} on the make check line as well.

These are the failures I'm seeing now in rawhide:

test 631                                                                                        
    Start 631: gmres_2                                                                          

631: Test command: /builddir/build/BUILD/eigen-eigen-ffa86ffb5570/test/runtest.sh "gmres_2"     
631: Test timeout computed to be: 1500                                                          
631: /builddir/build/BUILD/eigen-eigen-ffa86ffb5570/test/runtest.sh: line 20: 22113 Aborted                 ./$1 > /dev/null 2> .runtest.log                                                    
631:  Test gmres_2 failed: 
631:
631: Test check_sparse_square_solving(gmres_colmajor_diag) failed in /builddir/build/BUILD/eigen-eigen-ffa86ffb5570/unsupported/test/../../test/sparse_solver.h (39)
631:     x.isApprox(refX,test_precision<Scalar>())
631: Stack:
631:   - check_sparse_square_solving(gmres_colmajor_diag)
631:   - test_gmres_T<std::complex<double> >()
631:   - gmres
631:   - Seed: 1375306889
631:
631:
2/4 Test #631: gmres_2 ..........................***Failed    0.63 sec
test 632
    Start 632: minres_1

632: Test command: /builddir/build/BUILD/eigen-eigen-ffa86ffb5570/test/runtest.sh "minres_1"
632: Test timeout computed to be: 1500
632: /builddir/build/BUILD/eigen-eigen-ffa86ffb5570/test/runtest.sh: line 20: 22116 Aborted                 ./$1 > /dev/null 2> .runtest.log
632:  Test minres_1 failed:
632:
632: Test check_sparse_square_solving(minres_colmajor_diag) failed in /builddir/build/BUILD/eigen-eigen-ffa86ffb5570/unsupported/test/../../test/sparse_solver.h (39)
632:     x.isApprox(refX,test_precision<Scalar>())
632: Stack:
632:   - check_sparse_square_solving(minres_colmajor_diag)
632:   - test_minres_T<double>()
632:   - minres
632:   - Seed: 1375306890
632:
632:
3/4 Test #632: minres_1 .........................***Failed    0.01 sec
test 647
    Start 647: bdcsvd_2

647: Test command: /builddir/build/BUILD/eigen-eigen-ffa86ffb5570/test/runtest.sh "bdcsvd_2"
647: Test timeout computed to be: 1500
647: /builddir/build/BUILD/eigen-eigen-ffa86ffb5570/test/runtest.sh: line 20: 22119 Aborted                 ./$1 > /dev/null 2> .runtest.log
647:  Test bdcsvd_2 failed:
647:
647: Test ( bdcsvd(n, false) ) failed in /builddir/build/BUILD/eigen-eigen-ffa86ffb5570/unsupported/test/svd_common.h (46)
647:     test_isApprox(m, u * sigma * v.adjoint())
647: Stack:
647:   - ( bdcsvd(n, false) )
647:   - bdcsvd
647:   - Seed: 1375306890
647:
647:
4/4 Test #647: bdcsvd_2 .........................***Failed    0.01 sec

I'm also seeing intermittent errors with dontalign_3, and I've seen gmres_2 pass as well.  So definitely some flaky code that needs to be addressed.

Related upstream bug: http://eigen.tuxfamily.org/bz/show_bug.cgi?id=539

Yep I also see failures here and there (i.e. the metis backend test does not compile at all). I'll look further as time allows.

Thanks for your work on this. I'll look at updating the f19 and f18 branches with eigen 3.1.4 later today, and will carry over your spec enhancements there.

Another thing to look as is making eigen3 an arch specific package so that the tests always get run on all of the different architectures.  Right now we're building on arm:

http://koji.fedoraproject.org/koji/taskinfo?taskID=5693434

so we'll see how that goes...

Does a noarch -> arch change need special considerations concerning upgrade paths?

I'm really not sure.  Definitely worth asking/testing.

I found this [1] where I read

[12:27]  <rdieter> and it's not entirely clear (to me anyway) how yum handles/balks at arch -> noarch -> arch upgrade paths.
[...]
[12:29]  <skvidal> rdieter: yum handles those fine
[12:29]  <skvidal> arch->noarch->arch isn't a problem except for glibc and kernel

so should be straight forward (worth testing nevertheless though).

[1] https://fedoraproject.org/wiki/Packaging:Minutes20070403

So, after numerous attempts, I've managed to get a build to succeed (http://koji.fedoraproject.org/koji/buildinfo?buildID=447825). I'll close this bug, though I doubt that I've cornered out all flaky tests.