The accompanying patch file (stack-diffs.out) modifies the Python
interpreter to access the run-time stack more efficiently.  Fewer
temporary variables and fewer PUSH and POP operations are used.  By relying
less heavily on temporary variables, compiler optimizers have a better
chance of keeping critical variables in registers.  At least thirteen local
variables in eval_code2 are declared with the register attribute.  On a
register-starved architecture like the Intel x86, they can't all be placed
in registers.

Here is a brief summary of the changes:

    * Include/listobject.h - A new macro, PyList_SET_ITEM, was added.  It is 
      analogous to PyTuple_SET_ITEM and should only be used to initialize
      lists.  The Python interpreter uses it to speed up the BUILD_LIST
      opcode.

    * ceval.c - A new macro, FAST_STACK, is available.  When defined, these
      optimizations are enabled.  For example, no temporaries are used to
      duplicate the top of the stack:

	Py_INCREF(TOP());
	PUSH(TOP());
	
    * ceval.c - A few other miscellaneous optimizations are included:
      shortcutting error checking code when no error is possible and using
      macros to access tuples and lists where possible.

If you have any questions or problems with this patch, contact me at
skip@calendar.com.

Skip Montanaro