[Python-Dev] Performance (non)optimization: 31-bit ints in pointers

Guido van Rossum guido@python.org
Mon, 12 Aug 2002 20:27:16 -0400

> > In Python I decided to stay away from cleverness except when
> > extensive performance analysis showed there was a real need to speed
> > something up.  That got us super-fast dicts, for example, and .pyc
> > files to cache the work of the (slow, but trick-free) parser.
> For small ints, it also got you the small int cache, which has nearly
> the same storage requirements as a pointer-as-int, and is probably as
> expensive (you can drop the tests for odd addresses, but need to add
> increfs and decrefs for ints).

But the increfs and decrefs for ints are goodness, because they
simplify the code.  You can incref/decref any object without having to
know its type.

--Guido van Rossum (home page: http://www.python.org/~guido/)