BUG: Matrix multiplication of a coo_matrix with an invalid type value does not raise a TypeError

[asi1024]

Describe your issue.

In SciPy 1.15, matrix multiplication of a coo_matrix with an invalid type value does not raise a TypeError. This is a minor bug that will not cause major confusion, but it is likely not the intended behavior.

Reproducing Code Example

In SciPy 1.15.1,

>>> import scipy
>>> x = scipy.sparse.coo_matrix((3, 4))
>>> x.dot(None)
NotImplemented

In SciPy 1.14.1 (Expected),

>>> import scipy
>>> x = scipy.sparse.coo_matrix((3, 4))
>>> x.dot(None)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/.../.pyenv/versions/3.11.11/lib/python3.11/site-packages/scipy/sparse/_base.py", line 450, in dot
    return self @ other
           ~~~~~^~~~~~~
TypeError: unsupported operand type(s) for @: 'coo_matrix' and 'NoneType'

SciPy/NumPy/Python version and system information

OS             : Linux-5.4.0-193-generic-x86_64-with-glibc2.31
Python Version : 3.11.11
NumPy Version  : 2.2.1
SciPy Version  : 1.15.1

[j-bowhay]

cc @dschult

[dschult]

Yes indeed. Thanks for bringing this up. Returning NotImplemented was added in #21435 and is supposed to be a feature when used for matmul (@). But I think it shouldn't be used for dot.

By returning NotImplemented matmul gives the "other" object a chance to complete the operation if they implement __rmatmul__. This make sense for matmul. But since there is no Python operator associated with the method dot, we probably should not return NotImplemented from the dot method. There will be no automatic attempt to run other.dot(self). Of course, we could try that before raising an exception too. Not sure if that would be helpful though.

[dschult]

Thanks @asi1024 for reporting this! I think it is fixed in #22373 and that should get backported to the next bug-fix release (not sure on timing for that).

Very helpful to know when something is not working as it should! Thanks...