You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
changed the title [-]several test errors (numpy.core.tests.test_print.test_locale_longdouble, numpy.core.tests.test_scalarprint.TestRealScalars.test_dragon4_interface, numpy.core.tests.test_scalarprint.TestRealScalars.test_str, numpy.f2py.tests.test_kind.TestKind.test_all, numpy.ma.tests.test_core.TestMaskedArrayMethods.test_sort_flexible, numpy.core.tests.test_arrayprint.TestComplexArray.test_str, numpy.core.tests.test_longdouble.test_repr_roundtrip)[/-][+]BUG: several test errors on SPARC[/+]on Jan 1, 2019
Sounds a bit like the bug might be the equality function for structured types with none native byteorder fields and not the specific tests. The arrays all print identical.
Hmm, couple of endianess problems. I'm guessing your SPARC is big endian? Not sure about the other record failures. @ahaldane Thoughts? The Fortran kind problem is a perpetual hassle. SPARC has quad precision long doubles (16) and it looks like we are returning extended precision (10).
Hmm, Lets try to track that down then. Are you testing with runtests.py? I'm going to guess that these lines in runtests.py are responsible.
if isinstance(result, bool):
sys.exit(0 if result else 1)
elif result.wasSuccessful():
sys.exit(0)
else:
sys.exit(1)
I've never seen that problem here, so what versions of Python 2.7 an pytest are you using? Did the error message say more? It looks like the isinstance bit may be failing even though the error message says the result is bool.
Activity
[-]several test errors (numpy.core.tests.test_print.test_locale_longdouble, numpy.core.tests.test_scalarprint.TestRealScalars.test_dragon4_interface, numpy.core.tests.test_scalarprint.TestRealScalars.test_str, numpy.f2py.tests.test_kind.TestKind.test_all, numpy.ma.tests.test_core.TestMaskedArrayMethods.test_sort_flexible, numpy.core.tests.test_arrayprint.TestComplexArray.test_str, numpy.core.tests.test_longdouble.test_repr_roundtrip)[/-][+]BUG: several test errors on SPARC[/+]mattip commentedon Jan 1, 2019
Can you test against 1.16.rc1?
DerDakon commentedon Jan 1, 2019
It died earlier with this:
::TestAlignment::test_various_alignments <- ../work/numpy-1.16.0_rc1-python2_7/test/lib/numpy/core/tests/test_multiarray.py /var/tmp/portage/dev-python/numpy-1.16.0_rc1/temp/environment: line 3960: 37258 Bus error ${EPYTHON} -c "
charris commentedon Jan 1, 2019
@DerDakon Hmm, that is not unexpected.
charris commentedon Jan 1, 2019
Reference #12626.
charris commentedon Jan 3, 2019
@DerDakon I think the bus errors may be fixed. I'll put out an rc2 soon that you can check.
charris commentedon Jan 5, 2019
OK, NumPy 1.16.0rc2 is out. It would be much appreciated if you could test it.
DerDakon commentedon Jan 5, 2019
Failing test:
-TestKind.test_all
-TestRecFunctions.test_structured_to_unstructured
-TestArrayEqual.test_recarrays
-TestEqual.test_recarrays
charris commentedon Jan 5, 2019
So no bus errors and four failing tests. I assume the TestKind failure is for Fortran. Could you post more information on the failures?
DerDakon commentedon Jan 5, 2019
seberg commentedon Jan 5, 2019
Sounds a bit like the bug might be the equality function for structured types with none native byteorder fields and not the specific tests. The arrays all print identical.
charris commentedon Jan 5, 2019
Hmm, couple of endianess problems. I'm guessing your SPARC is big endian? Not sure about the other record failures. @ahaldane Thoughts? The Fortran kind problem is a perpetual hassle. SPARC has quad precision long doubles (16) and it looks like we are returning extended precision (10).
DerDakon commentedon Jan 5, 2019
Yes, sparc is big endian, as well as ppc which had similar problems as in the initial report.
9 remaining items
DerDakon commentedon Jan 5, 2019
If I apply the patches from #12671 and #12672 I see no failing tests anymore. The only bad thing I see is this (with Python 2.7):
charris commentedon Jan 5, 2019
That was for nose, you should be using pytest. Maybe need to clean up the environment?
EDIT: Before building.
DerDakon commentedon Jan 5, 2019
I fixed my ebuild, test is restarted.
DerDakon commentedon Jan 6, 2019
I have no nose installed anymore, but the message stays the same.
charris commentedon Jan 6, 2019
Hmm, Lets try to track that down then. Are you testing with
runtests.py? I'm going to guess that these lines inruntests.pyare responsible.I've never seen that problem here, so what versions of Python 2.7 an pytest are you using? Did the error message say more? It looks like the
isinstancebit may be failing even though the error message says the result isbool.charris commentedon Jan 6, 2019
It also occurs to me that that code snippet should just be
DerDakon commentedon Jan 6, 2019
I honestly have no clue, I just run the Gentoo ebuild.
charris commentedon Jan 6, 2019
Ah, makes me suspect Gentoo. So you don't actually run the tests yourself?
DerDakon commentedon Jan 6, 2019
It's just me being stupid, the ebuild contains:
charris commentedon Jan 6, 2019
Thanks for tracking that down.
DerDakon commentedon Jan 7, 2019
New error log from 3.6:
DerDakon commentedon Jul 4, 2019
Here is the updated report for 1.16.1. All tests with python2 pass.
Edit: (by mattip) to put the log into triple-backquotes for formatting
TST: f2py: extend the close-regression pack