Evgeni Burovski

> T is possibly a quasi triangular array 2x2 blocks [[a, b], [c, a]] (notice the nonsymmetricity). Maybe I am missing it but how do we get eigenvalues from that?...

Yeah. How about keeping the current band-aid then? Specifically, I'd suggest we - keep the code as is ATM - keep this issue open, maybe rename it. - keep the...

There is also some Fortran code in stats, all probably rather low-hanging fruit. ``` $ ll scipy/stats/*f -rw-rw-r-- 1 br br 46387 мая 8 2022 scipy/stats/mvndst.f -rw-rw-r-- 1 br br...

This is great @ilayn ! Now that we have a laundry list, I think it would be helpful to add two more classification indicators: - whether a module causes problems...

cc https://github.com/scipy/scipy/issues/19079#issuecomment-1682108384 for a C++ port of AMOS, which seems to be license-compatible.

Possibly related: there is some large-v modified Bessel function expansion in scipy. special: https://github.com/scipy/scipy/blob/main/scipy/special/_hyp0f1.pxd

Can you add a reproducible example --- something we can run as a standalone script, which defines all input arrays etc.

Thanks. So sparse matrices here is only a means to construct (rather large) dense inputs to `griddata`: it only receives dense ndarrays and never sees any csr_matrices. Good -- because...

Going back to `_find_simplex` sounds right indeed. This would get us back to the `c` array being `double[NPY_MAXARGS]` instead of O(number_of_data_points times NPY_MAXARGS).

Off the cuff, if we're rolling it back, we should try being as close to the previous behavior without losing the benefits of https://github.com/scipy/scipy/pull/18376.