Performance difference of C calling conventions

[nineteendo]

Documentation

The performance difference of these calling conventions is undocumented:

• METH_VARARGS
• METH_FASTCALL
• METH_O
• METH_NOARGS

You would expect that METH_FASTCALL is faster than METH_VARARGS, but for METH_O & METH_NOARGS it isn't so clear.

[encukou]

We don't guarantee the relative performance of these calling conventions. Additionally, the performance also depends on the caller, not just the called function.

If this matters to you, you should experiment and measure the variants for your own use case, and for a particular build of Python.

I'll close the issue in a week if there are no objections.