Library hard crashing when computing BBFG permanent with a 0x0 matrix

[GregoryMorse]

Before posting a bug report

• I have searched exisisting GitHub issues to make sure the issue does not already exist.

Expected behavior

Returning one (1+0j)

Actual behavior

Native code hard crashing Python in Windows/Linux with "overflow to inf" output

Reproduces how often

Always

System information

Python 3.6-3.8, Windows/Linux version 0.16.1

Source code

import numpy as np
from thewalrus.libwalrus import perm_BBFG_complex
perm_BBFG_complex(np.zeros((0, 0))+np.zeros((0, 0)) * 1j)

Tracebacks

overflow to inf

Additional information

Probably the real number case has the same bug.
perm_complex returns 0 instead of 1+0j so it is also bugged.  None of the 0x0 dimension permanents are working correctly.

[nquesada]

Hi @GregoryMorse --- Could you tell us which version of thewalrus are you using?

[GregoryMorse]

Hi @nquesada, version 0.16.1, the one in PyPi

[thisac]

Thanks for bringing this to our attention @GregoryMorse. As of the latest release (v0.18.0), The Walrus no longer uses libwalrus for any calculations. The function corresponding to thewalrus.libwalrus.perm_BBFG_complex is now thewalrus.perm.

The equivalent code to your example above would be:

import numpy as np
import thewalrus as tw

mat = np.zeros((0, 0))+np.zeros((0, 0)) * 1j
tw.perm(mat, method="bbfg")

With that said, it seems like this function has the same issue, so a fix would still be needed to make sure that the permanent function returns 1 for an empty matrix. We will look into fixing this in an upcoming release!

[nquesada]

or, alternatively, @GregoryMorse , you are most welcome to do a PR fixing it and become a contributor! You can use the example of how this is solved in the hafnian function here

thewalrus/thewalrus/_hafnian.py

Line 770 in 4452c64

if matshape == (0, 0):