py27-numpy fresh installation on Sierra fails test

[shigemi.ohta@…]

numpy.test('full') quickly fails:

FAIL: test_refcount_vdot (test_regression.TestRegression)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/opt/local/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/numpy/core/tests/test_regression.py", line 659, in test_refcount_vdot
    _assert_valid_refcount(np.vdot)
  File "/opt/local/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/numpy/testing/utils.py", line 1336, in _assert_valid_refcount
    assert_(sys.getrefcount(i) >= rc)
  File "/opt/local/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/numpy/testing/utils.py", line 75, in assert_
    raise AssertionError(smsg)
AssertionError

----------------------------------------------------------------------
Ran 5940 tests in 1441.503s

FAILED (KNOWNFAIL=6, SKIP=30, failures=1)

[larryv (Lawrence Velázquez)]

Thanks for the report. Please Cc port maintainers in the future:

$ port info --maintainers py27-numpy

[shigemi.ohta@…]

Replying to larryv@…:

Thanks for the report. Please Cc port maintainers in the future:

$ port info --maintainers py27-numpy

Thank you. The original failure was on my MacBook Pro newly running on Sierra. Now I found the same failure occurs with my office iMac, also newly on Sierra. In addition, on the latter, scipy.test('full') fails rather quickly too. Shall I open another ticket for that?

[michaelld (Michael Dickens)]

Please just keep everything on this ticket since they are directly related.

[michaelld (Michael Dickens)]

a few thoughts / personal opinions:

• I'm glad you reported this issue, and please don't take my comments here personally or as flippant.

• if NumPy otherwise works for your needs, then that's what matters. If NumPy is failing in some critical way, then please report the issue both here and to the NumPy folks themselves. The issue might be with MacPorts, or it might be with NumPy.

• Given that this is a runtime issue, I'd advise you -- if you're adamant about this part of NumPy working -- to take the issue upstream to the NumPy folks themselves. Quite honestly, these sorts of issues are a bit beyond us port maintainers' "pay grade". Most of us port maintainers are doing MacPorts work pro-bono (for free), so we get to what we can when we can.

• We welcome patches! If you feel compelled to hack on NumPy to fix these or other issues, please do so & attach a patch to this ticket & a comment about the fix. Again, nothing personal; we do the best we can.

[shigemi.ohta@…]

Thank you. No, I am not adamant about f2py, or fortran, but am annoyed because this prevents other tests. I would be willing to bring this to upstream, but am not sure how to: I would appreciate if you could tell me how. Also, can you try the newer 1.11.2?

Replying to michaelld@…:

a few thoughts / personal opinions:

• I'm glad you reported this issue, and please don't take my comments here personally or as flippant.

• if NumPy otherwise works for your needs, then that's what matters. If NumPy is failing in some critical way, then please report the issue both here and to the NumPy folks themselves. The issue might be with MacPorts, or it might be with NumPy.

• Given that this is a runtime issue, I'd advise you -- if you're adamant about this part of NumPy working -- to take the issue upstream to the NumPy folks themselves. Quite honestly, these sorts of issues are a bit beyond us port maintainers' "pay grade". Most of us port maintainers are doing MacPorts work pro-bono (for free), so we get to what we can when we can.

• We welcome patches! If you feel compelled to hack on NumPy to fix these or other issues, please do so & attach a patch to this ticket & a comment about the fix. Again, nothing personal; we do the best we can.

[michaelld (Michael Dickens)]

I just pushed 1.11.2 in r153815. It won't fix the f2py issue; it might address the other issue.

[shigemi.ohta@…]

I joined numpy-discussion@… and reported the issue there at Wed, 12 Oct 2016 23:08:44 +0900. No direct response, but today I ran "port upgrade outdated" on my office iMac that resulted in updating dbus, gcc6, glib2, hdf5, sqlite3, ImageMagick, jasper, netcdf, py27-setuptools, poppler, py27-numpy, snappy, and texlive-bin. After this upgrade "numpy.test()," "scipy.test()," and "scipy.test('full')" all pass without an unknown failure, while numpy.test('full') still fails the same way with f2py. The situation appears the same on my own MacBook Pro. So I am satisfied now. Thank you very much.

Replying to michaelld@…:

I just pushed 1.11.2 in r153815. It won't fix the f2py issue; it might address the other issue.