Please document if it is possible to build extensions interfacing with fortran libraries using real128

[callegar]

Because there are fortran libraries that use real128, it would be interesting to know if via numpy-quadtype numpy it can be possible to take advantage of them. Being able to use cython to build python extensions acting as bridges to them would be a nice option.

[ngoldbaum]

I think that would require numpy_quaddtype to expose a C API, which introduces a lot of headaches for a Python package. NumPy does it but it's a big pain.

Ideally it would all work using Cython's support for typed memoryviews, but that can only work with types supported by the buffer protocol, which is limited to a set of enumerated types that doesn't include float128.

I have long-term plans to improve the buffer protocol and enable support for custom data types, but that's something that needs a PEP and is not a short-term fix.

All that to say: yes this is possible and desirable, but it will require a lot of work and some design decisions. If numpy-quaddtype starts exposing a C API that adds a lot of complexity since we need to be careful to maintain both forward and backward compatibility. I'm not sure we want to take that on.