The scalar methods still don't use METH_FASTCALL for their forwarding. Changing this would improve performance at least if kwargs are involved (maybe just moderately, these are slow anyway due to creating an array).

It may also just simplify the code a bit more. E.g. NumPy array methods (methods.c) already forward via fastcall.

Related to this, it is nice to call methods with the Python string, rather than the C char * for speed reasons.

(This is fully in C code and requires acquiring understanding how the Python C-API works for method calls.)