[Python-Dev] Idea for a fast calling convention

[Phillip J. Eby]

At 04:49 PM 2/27/04 -0500, Fred L. Drake, Jr. wrote:
>On Friday 27 February 2004 03:20 am, Raymond Hettinger wrote:
>  > I've been working on ways to speedup parameter passing for function
>  > calls.  It's somewhat expensive to pull parameters off of the stack,
>  > assemble them into a new tuple, have the function disassemble the tuple,
>  > and ultimately free the tuple.
>..
>  > My idea is for a new flag, METH_STACK, that generalizes (and potentially
>  > replaces) METH_NOARGS AND METH_O.
>
>How would METH_STACK methods work when called via func(*args) or via the
>various functions of the C API?
>
>The nice thing about METH_NOARGS and METH_O is that they describe what the
>called function needs.  METH_STACK sounds like a core dump waiting to 
>happen.
>;-(

Maybe there should instead be a tp_call_stack slot.  Then the various CALL 
opcodes would call that slot instead of tp_call.  C API calls would still 
go through tp_call.

'object' would define tp_call_stack to call tp_call, using information from 
the stack.  Python function and method objects would then override 
tp_call_stack to implement the shortcut form.

In practice, though, I expect it would be faster to do as Jython and 
IronPython have done, and define a set of tp_call1, tp_call2, etc. slots 
that are optimized for specific calling situations, allowing C API calls to 
be sped up as well, provided you used things like PyObject_Call1(ob,arg), 
PyObject_Call2(ob,arg1,arg2), and so on.

Perhaps there is some information that can be gleaned from the Jython 
research as to what are the most common number of positional parameters for 
calls.