first release of PyPy

[Paul Rubin]

Jp Calderone <exarkun at divmod.com> writes:
>   Have you profiler data in support of this?  Suggesting
>optimizations, especially ones which require semantic changes to
>existing behavior, without actually knowing that they'll speed things
>up, or even that they are targetted at bottleneck code, is kind of a
>waste of time.

I don't have measurements for Python (no idea whether PyPy supports
profiling, and CPython interpreters aren't very interesting since
the effect matters mostly for compiled code), but Flavors faced
a very similar problem and caching was a big win.  I'm sure all
serious CLOS implementations do something similar.