TST: disable overflow exception test of numpy.power on AIX #29740

kkwli · 2025-09-11T17:05:59Z

numpy.power() does not raise the overflow exception with numpy.float64 type.
It can be illustrated in the following code.

import numpy as np
#fi = np.finfo(np.float64)
dtyp = np.dtype(np.float64).type
with np.errstate(all='raise'):
  try:
    x = np.power(dtyp(2), dtyp(2048))   # 2**fi.nexp == 2048
    print(f'No fpe error ({x})')
  except FloatingPointError as exc:
    print(f'FloatingPointError')

Output on AIX:

No fpe error (inf)

Output on Linux/PowerPC and Linux/x86:

FloatingPointError

This patch is to guard the self.assert_raises_fpe(overflow, np.power, ftype(2), ftype(2**fi.nexp)) test to skip it on AIX.

ngoldbaum · 2025-09-11T17:56:33Z

numpy/_core/tests/test_numeric.py

+            # it returns inf. Long double is the same as double.
+            if sys.platform != 'aix' or typecode not in 'dDgG':
+                self.assert_raises_fpe(overflow,
+                                       np.power, ftype(2), ftype(2**fi.nexp))


Is this papering over a bug in our floating point exception detection or is this just a fundamental limitation of the AIX math library?

TST: disable overflow exception test of numpy.power on AIX

58270e6

github-actions bot added the 05 - Testing label Sep 11, 2025

ngoldbaum reviewed Sep 11, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

TST: disable overflow exception test of numpy.power on AIX #29740

TST: disable overflow exception test of numpy.power on AIX #29740

kkwli commented Sep 11, 2025

Uh oh!

ngoldbaum Sep 11, 2025

Uh oh!

Uh oh!

Uh oh!

TST: disable overflow exception test of numpy.power on AIX #29740

Are you sure you want to change the base?

TST: disable overflow exception test of numpy.power on AIX #29740

Conversation

kkwli commented Sep 11, 2025

Uh oh!

ngoldbaum Sep 11, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!