float128 should have __floor__, __ceil__ and __trunc__

math.floor() and math.ceil() can produce incorrect result for float128 due to double rounding. math.trunc() doe not work with float128 (but int() works).

Reproducing code example:

>>> import numpy as np
>>> import math
>>> x = np.float128('0.99999999999999999995')
>>> x < 1
True
>>> np.floor(x)
0.0
>>> math.floor(x)
1
>>> np.trunc(x)
0.0
>>> y = np.float128('1.0000000000000000001')
>>> y > 1
True
>>> np.ceil(y)
2.0
>>> math.ceil(y)
1
>>> z = np.float128()
>>> np.trunc(z)
0.0
>>> math.trunc(z)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: type numpy.float128 doesn't define __trunc__ method
>>> int(z)
0

This can be solved by implementing __floor__, __ceil__ and __trunc__ in float128.

Numpy/Python version information:

1.13.3 3.6.8 (default, Oct 7 2019, 12:59:55) [GCC 8.3.0]

[eric-wieser]

The latter one surprised me.

Your version of numpy is (very) old, that’s fixed in master

[rossbar]

Closing as duplicate to keep the discussion centralized in #13375. Please add any additional comments (including a potentially a summary of the discussion here) to that thread.