Evgeni Burovski

A couple of quick superficial comments: 1. For the build, you'll need to use meson, see `ndimage/meson.build` file. 2. It may (or may not be) easier to use cython for...

> Can we re-run the tests? Done. (ain't this GH feature a tad annoying)

Looks good as a means to unblock development locally. I wonder though what it takes to fix the underlying issue. Relying on a

Seems to work on main for me locally: ``` In [3]: st.logser(0.25739239425774363).ppf(0.9999999999999999) Out[3]: 99.0 In [4]: import scipy In [5]: scipy.__version__ Out[5]: '1.13.0.dev0+1282.c87635c' ```

cross-ref https://github.com/scipy/scipy/issues/3890

Bumpted the milestone: the fix is unlikely to materialize in time for https://github.com/scipy/scipy/pull/19880#issuecomment-1992010395

Linking to various LAPACK variants is IIUC abstracted away with `+ dependencies: [lapack, np_dep],` in `meson.build`. What's left is ``` $ cat scipy/linalg/fortran_defs.h /* * Handle different Fortran conventions. */...

> I think the correct thing to do is to use scipy/_build_utils/src/npy_cblas.h for Fortran mangling handling in a central place. So if this header needs to be used in constructing...

> There are a limited number of LAPACK functions we use from C/C++, so it seems very much feasible to add hand-written wrappers a la the special/lapack_defs.h as we need...

> the correct thing to do is to use [scipy/_build_utils/src/npy_cblas.h](https://github.com/scipy/scipy/blob/main/scipy/_build_utils/src/npy_cblas.h) for Fortran mangling handling in a central place. Gave it a try. With this diff on top of https://github.com/scipy/scipy/pull/19970 ```diff...