Evgeni Burovski

Fixed by gh-20195 ``` In [1]: import numpy as np ...: import scipy.linalg ...: ...: a = np.random.normal(size=(3, 3)) ...: np.vectorize(scipy.linalg.expm, signature="(m,m)->(m,m)")(a) Out[1]: array([[ 0.29571569, -0.0045622 , 0.48342085], [ 0.33301448,...

OK, musl CI failure is unrelated, all other failures are resolved, the tolerance for the gcvspl comparison is explained to be due to different minimization primitives: for gcvspl it's a...

CWT is deprecated in scipy 1.12, along with the rest of the wavelet functionality. So, this enhancement request cannot be acted upon. Thank you @jerabaul29 regardless.

No activity for ~10 months. The feature is borderline: easy to work around on the user side, easy to implement, not clear if it's worth implementing. If somebody feels like...

What about contract_path=True, should this be equivalent to "auto"?

That would be great @Kai-Striega ! I would just grep pyx files by subpackage, along the lines of `$ grep def scipy/optimize -rn` to list all cython function declarations. For...

Ouch. I don't know whether - cython does sensible things in `if lapack_t is float: ...` selections below https://github.com/scipy/scipy/blob/9c98ee762a95cf46d8f91f1c1a5547701538d286/scipy/linalg/_cythonized_array_utils.pyx#L29 - lapack bindings are prepared to handle `const float*` arguments. there...

> We really don't benefit from Cython which is best for mostly quick bursts of typed arithmetic and for loops but instead we are extensively using arrays and low-level C...

> I mean there is no significant speed benefit to do array operations in Cython. NumPy is already pretty performant, for many array operations. Yes, until you need a for...

> This is something Array API folks really need to make it crystal clear in my opinion. SciPy should not (and could not) be the only library providing anything that...