Numpy 1.16 linspace fails for very large start stop values

[jenshnielsen]

Reproducing code example:

The following corner case works correctly with numpy 1.15.4 but raises an error with numpy 1.16.2

import numpy as np
np.linspace(start = -9223372036854775809, stop = -9223372036854775809, num =1)

This was detected in a test with auto generated input using hypothesis to generate the input.

Error message:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-3-aa8cbb862cf0> in <module>
----> 1 np.linspace(start = -9223372036854775809, stop = -9223372036854775809, num =1)

~\AppData\Local\Continuum\miniconda3\envs\qcodes\lib\site-packages\numpy\core\function_base.py in linspace(start, stop, num, endpoint, retstep, dtype, axis)
    141 
    142     delta = stop - start
--> 143     y = _nx.arange(0, num, dtype=dt).reshape((-1,) + (1,) * delta.ndim)
    144     # In-place multiplication y *= delta/div is faster, but prevents the multiplicant
    145     # from overriding what class is produced, and thus prevents, e.g. use of Quantities,

AttributeError: 'float' object has no attribute 'ndim'

Numpy/Python version information:

1.16.2 3.6.8 |Anaconda, Inc.| (default, Feb 21 2019, 18:30:04) [MSC v.1916 64 bit (AMD64)]

Details

This seems to happen when the number used is sufficiently large end up being represented
as an object dtype. This in turn means that delta will be a regular floating point number and not a numpy float. Therefore it does not have a ndim method. The 1.15 code looks different and does not require the ndim method to be defined.

type(np.asanyarray(-9223372036854775809))
dtype('O')
type(np.asanyarray(-9223372036854775809) * 1.0 - np.asanyarray(-9223372036854775809) *1.0)
float

[mhvk]

Oops, that's an unexpected one. But fortunately easy to solve...

[mhvk]

See #13092 for a possible fix.

[jenshnielsen]

@mhvk thanks looks good