numpy failures

[hmaarrfk]

maybe it is because i'm building with openblas 0.3.5

Here are some builds that I'm going to let go to completion.
https://app.shippable.com/github/Archiconda/numpy-feedstock/runs/6/summary/console

python 2.7 failures

self = <numpy.core.tests.test_umath.TestComplexFunctions object at 0xffffad4c2690>

    @pytest.mark.skipif(is_longdouble_finfo_bogus(),
                        reason="Bogus long double finfo")
    def test_loss_of_precision_longcomplex(self):
>       self.check_loss_of_precision(np.longcomplex)

self       = <numpy.core.tests.test_umath.TestComplexFunctions object at 0xffffad4c2690>

../_test_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeho/lib/python2.7/site-packages/numpy/core/tests/test_umath.py:2597: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../_test_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeho/lib/python2.7/site-packages/numpy/core/tests/test_umath.py:2547: in check_loss_of_precision
Tests failed for numpy-1.15.4-py27_blas_openblashfae12ec_1000.tar.bz2 - moving package to /root/archiconda3/conda-bld/broken
WARNING:conda_build.build:Tests failed for numpy-1.15.4-py27_blas_openblashfae12ec_1000.tar.bz2 - moving package to /root/archiconda3/conda-bld/broken
    check(x_series, 50*eps)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

x = array([1.00000000e-20, 1.21736864e-20, 1.48198641e-20, 1.80412378e-20,
       ...871e-04, 6.73218092e-04, 8.19554595e-04, 9.97700064e-04],
      dtype=float128)
rtol = 9.629649721936179265279889712924637e-33

    def check(x, rtol):
        x = x.astype(real_dtype)
    
        z = x.astype(dtype)
        d = np.absolute(np.arcsinh(x)/np.arcsinh(z).real - 1)
        assert_(np.all(d < rtol), (np.argmax(d), x[np.argmax(d)], d.max(),
>                                 'arcsinh'))
E       AssertionError: (120, 1.7809563130653488765574048895753284e-10, 5.286342315078822882670006871837315e-21, 'arcsinh')

d          = array([0.00000000e+00, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00,
       ...000e+00, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00],
      dtype=float128)
dtype      = <type 'numpy.complex256'>
real_dtype = dtype('float128')
rtol       = 9.629649721936179265279889712924637e-33
x          = array([1.00000000e-20, 1.21736864e-20, 1.48198641e-20, 1.80412378e-20,
       ...871e-04, 6.73218092e-04, 8.19554595e-04, 9.97700064e-04],
      dtype=float128)
z          = array([1.00000000e-20+0.j, 1.21736864e-20+0.j, 1.48198641e-20+0.j,
       1.80...8092e-04+0.j,
       8.19554595e-04+0.j, 9.97700064e-04+0.j], dtype=complex256)

../_test_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeho/lib/python2.7/site-packages/numpy/core/tests/test_umath.py:2520: AssertionError

[hmaarrfk]

@tylerjereddy

numpy 1.11 python 2.7 failure:

======================================================================
FAIL: test_longdouble.test_repr_roundtrip
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/root/archiconda3/conda-bld/numpy_1546900037232/_test_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeho/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
    self.test(*self.arg)
  File "/root/archiconda3/conda-bld/numpy_1546900037232/_test_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeho/lib/python2.7/site-packages/numpy/core/tests/test_longdouble.py", line 35, in test_repr_roundtrip
    "repr was %s" % repr(o))
  File "/root/archiconda3/conda-bld/numpy_1546900037232/_test_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeho/lib/python2.7/site-packages/numpy/testing/utils.py", line 381, in assert_equal
    raise AssertionError(msg)
AssertionError: 
Items are not equal: repr was 1.0
 ACTUAL: 1.0
 DESIRED: 1.0

in addition to the one above
I got too lazy to let the 3.6 and 3.7 builds even start since I assumed it had to do with the floating point unit in the arm processor.

https://app.shippable.com/github/Archiconda/numpy-feedstock/runs/12/1/console

[tylerjereddy]

Hmm 1.11 is pretty old! Pretty sure we're not using openblas yet on NumPy master branch ARMv8 builds, though I suppose we should in order to match wheel builds on other platforms.

We're just using some apt-get commands for the linalg backends at the moment. I guess you could try that if you really want, but probably not really consistent with the conda approach.

Hopefully we can resolve the single (?) failure for 1.15 or at least for master branch if that fails any tests with openblas on ARM

[hmaarrfk]

I know numpy 1.11 is pretty old, but I think it is now considered the defacto LTS. Especially since you have such strong forward compatibility, people are compiling against it ;).

So even for this super new distribution, I had to compile it. I think @jakirkham wanted to drop the in between ones for conda-forge, but couldn't.....

I'm not really sure it is openblas related, but it didn't seem that anybody else was raising that issue, and I feel like that might be the only difference between my and your builds.

Oh, and gcc 7.3, so it could be that too!!!! Ubuntu 18.04

[hmaarrfk]

.... they built as far back as 1.9... that isn't going to happen here

[hmaarrfk]

Here are some fresh builds that should reproduce the failures on python 3.6 (you know, when they run and finish that is)

https://app.shippable.com/github/Archiconda/numpy-feedstock/runs/17/summary/console