`linalg.inv` fails for arrays of shape (0, 0)

My issue is about the behavior of linalg.inv for arrays of shape (0, 0). I would like to move from numpy.linalg to scipy.linalg, but with scipy inverting this kind of matrices is not possible. Normally my functions runs with (M, M) matrices with M > 0 but I would like to handle the edge case as well where M = 0. I can work around this issue by checking the shape beforehand, but it seams to me like a bug with scipy’s linalg.inv that it can’t handle this case.

Reproducing code example:

import numpy
import scipy
z = numpy.zeros((0, 0))
numpy.linalg.inv(z) # --> array([], shape=(0, 0), dtype=float64)
scipy.linalg.inv(z) # --> ValueError

Error message:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
~/test.py in <module>
      3 z = numpy.zeros((0, 0))
      4 numpy.linalg.inv(z) # --> array([], shape=(0, 0), dtype=float64)
----> 5 scipy.linalg.inv(z) # --> ValueError

~/anaconda3/lib/python3.8/site-packages/scipy/linalg/basic.py in inv(a, overwrite_a, check_finite)
    963         raise LinAlgError("singular matrix")
    964     if info < 0:
--> 965         raise ValueError('illegal value in %d-th argument of internal '
    966                          'getrf|getri' % -info)
    967     return inv_a

ValueError: illegal value in 4-th argument of internal getrf|getri

Scipy/Numpy/Python version information:

1.6.2 1.20.2 sys.version_info(major=3, minor=8, micro=5, releaselevel='final', serial=0)

[ilayn]

You can just return a1.copy() after the squareness check.

[ev-br]

Thanks @ilayn, clicking though… found it —It’s de Boor, not Higham, it turns out: https://github.com/scipy/scipy/pull/4672#issuecomment-124718692