Kernel never returns from np.prod() on quaternions

[noncom]

Performing the prod NumPy operation over an array of quaternions results in kernel becoming unresponsive and producing a 100% CPU load. Only restarting the kernel would bring the notebook back to life.

Example code:

np.prod([np.quaternion(1, 0, 0, 0), np.quaternion(0, 0, 0, 1)])

[moble]

I can verify that this happens with the newest versions of everything. I've tracked it back to the numpy.core.umath.multiply.reduce function. Here's a MWE:

import numpy as np
from numpy.core import umath as um
import quaternion

a = np.array([quaternion.x, quaternion.y])
um.multiply.reduce(a, axis=None, dtype=None, out=None, keepdims=False)

Unfortunately, I can't actually find where that function is defined, because I would hope to pass the buck to the numpy maintainers...

@eric-wieser Since you seem to know something about both repos, any thoughts?

[eric-wieser]

numpy\core\src\umath\ufunc_object.c:PyUFunc_Reduce is the place to start looking. That does indeed seem odd.

However, it works fine on np.core.test_rational.rational - so my hunch is that your multiply inner loop is broken

[eric-wieser]

np.multiply.reduce(a[:0]) shows the same problem, which is interesting

[eric-wieser]

The hang is within type_tuple_type_resolver.

[eric-wieser]

This is a numpy bug, in type_tuple_userloop_type_resolver, where there is an obvious inner loop.

Please file an issue upstream!

[eric-wieser]

*infinite loop

[moble]

Thanks! Issue filed. I'll keep an eye on that to test any fixes, and I'll leave this open until the fix makes it into a numpy release.

[moble]

The changes have filtered down to the conda package, so this issue is resolved. Thanks everyone!