Evgeni Burovski

@TimoRoth any luck further chasing down this issue?

> in a fresh venv with everything fully updated, I am now unable to reproduce this on python 3.12 and 3.13. So to double-check: once you use a fresh venv,...

Here's a numpy issue to discuss this change: https://github.com/numpy/numpy/issues/29000. Also https://github.com/cupy/cupy/pull/8980#issuecomment-2890821763 is an attempt to convince CuPy devs to not follow numpy and have a dtype-stable API. Overall, I believe...

CuPy merged the support for `eig` and `eigvals` in https://github.com/cupy/cupy/pull/8980/ for v14.0.a2. IIRC, the CuPy interface follows pytorch and jax.numpy and always return eigenvalues as a complex array (with zero...

> And the documentation is geared towards people who already know what they are doing. This is probably a good topic for a tutorial and/or scipy cookbook.

> At first glance, this is a compromise I'm not particularly keen on. I'm definitely in the camp of error handling being a measure of correctness for which performance shouldn't...

> It would be easy to factor out the test changes into a PR that introduces Array API compatible tests. +1. A strategy of converting tests in a separate PR...

Tried looking into it and I've to admit to not understanding what's going on here. First, where do the cases come from? Second, dropping a breakpoint, ``` (Pdb) p case_str...

Deprecating it works too --- either an update or a deprecation would be an improvement IMO.

Great catch Tyler! gh-21713 has a fix: https://github.com/scipy/scipy/pull/21713/files#diff-351836adc98d076c1552d17a57c52e6aa8ca43760bae44bea9190e55b4769b7fR250 (not that the fix was deliberate---the whole `_reject_objects` call needed a bit of care for non-numpy backends)