Ilhan Polat

Thanks @ivirshup this is definitely something I would love to see. We had such discussions over the years and everytime it gets a bit more closer. I have the following...

Thanks sounds like there is more momentum this time, > The underlying collection of arrays should not change at all. It's just the python API. Was this your concern? If...

> To build on what @ivirshup said, numpy.array_api is fully compliant with the Array API. NumPy implemented many things and deprecated many things, if they are not used out in...

Don't get me wrong I'm not trying to torpedo the attempt. But I also got burned enough times adopting a standard prematurely and spending unnecessary amount of work trying to...

Thanks for all the clarifications, makes more sense now. If I am understanding correctly this will have indirect effect on the `sparse.linalg` module. Because most functions are using `LinearOperator`'s `(r)matvec`...

Pinging also @zaikunzhang

This is due to how Fortran works because everything is pass by reference. Hence it only consumes what it needs from that "pointer" and does not bother with the rest....

Did you try to link directly to our vendored OpenBLAS? It should be a static library in my limited understanding. Meson has some magic and probably would be possible to...

I think I am missing some details. Once we have the dll or the a file, why are we still dealing with fortran headers? Symbol names and name mangling hell...

> No, the only reason we are vendoring OpenBLAS in the wheels on PyPI is because we have to to make wheels work. It's an anti-pattern basically that is specific...