NumPy 1.12.0 has failing tests

[nils-werner]

Somewhere there is auto-densification happening, as I have many failing tests on my setup:

sparse/tests/test_core.py::test_elemwise[func0] FAILED
sparse/tests/test_core.py::test_elemwise[func1] FAILED
sparse/tests/test_core.py::test_elemwise[func2] FAILED
sparse/tests/test_core.py::test_elemwise[func3] FAILED
sparse/tests/test_core.py::test_elemwise[func4] FAILED
sparse/tests/test_core.py::test_elemwise[func5] FAILED
sparse/tests/test_core.py::test_elemwise[func6] FAILED
sparse/tests/test_core.py::test_elemwise[func7] FAILED
sparse/tests/test_core.py::test_elemwise[func8] FAILED
sparse/tests/test_core.py::test_elemwise[func9] FAILED
sparse/tests/test_core.py::test_elemwise[func10] PASSED
sparse/tests/test_core.py::test_elemwise[func11] FAILED
sparse/tests/test_core.py::test_elemwise[func12] PASSED
sparse/tests/test_core.py::test_elemwise[func13] FAILED
sparse/tests/test_core.py::test_elemwise[func14] FAILED
sparse/tests/test_core.py::test_elemwise[func15] PASSED
sparse/tests/test_core.py::test_elemwise[func16] PASSED

with

func = <ufunc 'floor'>

    @pytest.mark.parametrize('func', [np.expm1, np.log1p, np.sin, np.tan,
                                      np.sinh, np.tanh, np.floor, np.ceil,
                                      np.sqrt, np.conj, np.round, np.rint,
                                      lambda x: x.astype('int32'), np.conjugate,
                                      np.conj, lambda x: x.round(decimals=2), abs])
    def test_elemwise(func):
        s = sparse.random((2, 3, 4), density=0.5)
        x = s.todense()

        fs = func(s)

>       assert isinstance(fs, COO)
E       assert False
E        +  where False = isinstance(array([[[ 0.,  0.,  0.,  0.],\n        [ 0.,  0.,  0.,  0.],\n        [ 0.,  0.,  0.,  0.]],\n\n       [[ 0.,  0.,  0.,  0.],\n        [ 0.,  0.,  0.,  0.],\n        [ 0.,  0.,  0.,  0.]]]), COO)

Additionally, the two tests

sparse/tests/test_core.py::test_addition_not_ok_when_large_and_sparse
sparse/tests/test_core.py::test_raise_dense

basically never finish (I aborted them using Ctrl+C).

To reproduce this behaviour you can use the following tox.ini file

[tox]
envlist = {py27,py36}-{numpy12,numpylatest}
[testenv]
deps =
    numpy12: numpy==1.12.0
basepython =
    py27: python2.7
    py36: python3.6
commands=
    pip list
    py.test {posargs}
extras=
    docs
    tests

and run the specific tests using

tox -e py36-numpy12 -- -k test_elemwise
tox -e py36-numpy12 -- -k test_addition_not_ok_when_large_and_sparse
tox -e py36-numpy12 -- -k test_raise_dense

[hameerabbasi]

Numpy support is being dropped, see #84.