[Python-Dev] RE: [Python-checkins] python/dist/src/Objects typeobject.c, 2.244, 2.245

Fri Oct 24 11:59:04 EDT 2003

> > Py_INCREF(Py_True);
> > return Py_True;
> > 
> > takes less time than
> > 
> > return PyBool_FromLong(1);
> > 
> > Maybe a pair of macros Py_return_True and Py_return_False would make
> > sense?
>> Sorry if this was already suggested and hastily rejected, but why do
> we care at all about the reference counter of the few heavily-used
> immortal objects of CPython?

It was never discussed; I don't recall that it has ever occurred to
me.
> I guess allowing their counter not to be carefully maintained
> ventures to the slippery slopes of bad code. Anyway, my two cents
> for a (very) slightly faster and shorter code would be to be allowed
> never to do Py_INCREF or Py_DECREF when we know that the object is
> Py_None, Py_False or Py_True. These three would have a dummy
> tp_dealloc that just resets the reference counter to some large
> value if it ever reaches zero.

I think there are debugging modes where this would upset some counters
that maintain a balance of the total number of references in the
world.
I also don't think that the performance ggain would be measurable.
Maybe the slight code size decrease would have some benefits.
I'm worried that there would be a negative effect in terms of people
copying the pattern for other objects.
--Guido van Rossum (home page: http://www.python.org/~guido/)