TST: fix a set of test errors in master.

[rgommers]

These are the ones I ran into on 32-bit Linux with current master and Pathlib installed on Python 2.7:

======================================================================
ERROR: test_format.test_large_archive
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/rgommers/.local/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
    self.test(*self.arg)
  File "/home/rgommers/Code/numpy/numpy/lib/tests/test_format.py", line 841, in test_large_archive
    a = np.empty((2 ** 30, 2), dtype=np.uint8)
ValueError: array is too big; `arr.size * arr.dtype.itemsize` is larger than the maximum possible size.

======================================================================
ERROR: test_mafromtxt (test_io.TestPathUsage)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/rgommers/Code/numpy/numpy/testing/decorators.py", line 147, in skipper_func
    return f(*args, **kwargs)
  File "/home/rgommers/Code/numpy/numpy/lib/tests/test_io.py", line 1899, in test_mafromtxt
    f.write('1,2,3.0\n4,5,6.0\n')
TypeError: must be unicode, not str

======================================================================
ERROR: test_ndfromtxt (test_io.TestPathUsage)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/rgommers/Code/numpy/numpy/testing/decorators.py", line 147, in skipper_func
    return f(*args, **kwargs)
  File "/home/rgommers/Code/numpy/numpy/lib/tests/test_io.py", line 1888, in test_ndfromtxt
    f.write('1 2\n3 4')
TypeError: must be unicode, not str

======================================================================
ERROR: test_recfromcsv (test_io.TestPathUsage)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/rgommers/Code/numpy/numpy/testing/decorators.py", line 147, in skipper_func
    return f(*args, **kwargs)
  File "/home/rgommers/Code/numpy/numpy/lib/tests/test_io.py", line 1923, in test_recfromcsv
    f.write('A,B\n0,1\n2,3')
TypeError: must be unicode, not str

======================================================================
ERROR: test_recfromtxt (test_io.TestPathUsage)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/rgommers/Code/numpy/numpy/testing/decorators.py", line 147, in skipper_func
    return f(*args, **kwargs)
  File "/home/rgommers/Code/numpy/numpy/lib/tests/test_io.py", line 1910, in test_recfromtxt
    f.write('A,B\n0,1\n2,3')
TypeError: must be unicode, not str

======================================================================
ERROR: test_all_outliers (test_twodim_base.TestHistogram2d)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/rgommers/Code/numpy/numpy/lib/tests/test_twodim_base.py", line 257, in test_all_outliers
    r = rand(100) + 1. + 1e6  # histogramdd rounds by decimal=6
  File "/home/rgommers/Code/numpy/numpy/lib/utils.py", line 99, in newfunc
    warnings.warn(depdoc, DeprecationWarning)
DeprecationWarning: `rand` is deprecated!
numpy.testing.rand is deprecated in numpy 1.11. Use numpy.random.rand instead.

----------------------------------------------------------------------

[charris]

Hmm, is this big enough? The test is for

    BUG: Cast size to int64 when loading from archive

    Prevents overflow errors for large arrays on systems
    where the default int type is int32.

IIRC, the fix was for the product of dimensions that overflowed, so the tested array should probably be multidimensional with total size >= 2**31

[charris]

Thinking a bit more, this really only needs to be run on 64 bit windows systems.

[rgommers]

Ah, a comment on the test intent would have been helpful.....

Indeed, needs to be skipped on 32-bit systems then. For the original size I get:

In [1]: a = np.empty((2**30, 2), dtype=np.uint8)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-1-d8d46384be90> in <module>()
----> 1 a = np.empty((2**30, 2), dtype=np.uint8)

ValueError: array is too big.

Will add the skip, but keep the except MemoryError - that's always good when trying to allocate such large arrays.

[rgommers]

Updated. Also marked two tests as slow, for me np.lib.test() now runs in 8s instead of 45s.

[charris]

Thanks Ralf.