Ilhan Polat

> A recent array-api discussion [2], had a question if NumPy would consider changing this value-dependent behavior and always return a complex array for eigenvalues? I am assuming this is...

I still don't think this is the right move. Especially the argument to have it so that the return type is consistent is quite weak and unnecessarily making things complex...

Also the Schur complement needs to be written with `linalg.solve` and not `linalg.inv` s to improve the accuracy. In the degenerate cases, LAPACK will give eigenvalues with nonzero imaginary parts...

It is indeed better to clean up before we remove the output linting. I briefly checked a few of them and indeed, say `qtu` here is not used below and...

Probably best course of action is fix the linalg issues then turn off the build output linting and leave the ufuncs stuff alone as there is too much stuff happening...

Thanks @DVarela56 for this one. We have to do a few things before we do more work on this, one is that this probably should live in integrate and not...

Sounds good to me. Thank you for the persistence, I was busy with Work™ and arpack, so disappeared for a while. But now it seems like it is finally stabilizing...

See this discussion too https://discuss.scientific-python.org/t/add-lsrn-solver/1888/1

Based on the existing translations, It's typically on par, sometimes faster (due to avoiding the glue), or at most 2x-ish slower but removal of the maintenance burden, unnecessary compiling olympics...

Also it goes much faster if there is someone who knows what ODR code should do. So please let me know if you would like to take stab @markcampanelli. I'd...