PyPy plans to support Python 3

Moritz Lenz moritz at faui2k3.org
Tue Sep 13 08:32:17 UTC 2011


Am 13.09.2011 02:49, schrieb Jochen Plumeyer:
> What is really stunning to me, the performance of PyPy, is now better
> than the C implementation (really?).

(Dangerous half knowledge alert: the following comes from reading random 
blog posts, reddit and HN; I haven't personally tested anything).

My understanding is that PyPy is faster than the C implementation for 
some (few) benchmarks, but I haven't read any reports that somebody 
switched his web application to pypy, and got a huge overal performance 
boost.

 > Are there saint cows to slaughter, perhaps it is better to drop the
 > "all dynamic languages VM" approach, and first implement a Perl6 JIT
 > compiler or something made in Perl5 and bootstrap that engine later
 > to Perl6 itself? And do that JIT compiler a bit more generic, to
 > support other languages?

All the very fast JIT compilers I've heard of (PyPy, V8 for JS, LuaJIT 
or whatever it's called) hugely benefit from assumptions about the 
language they implement, like the structure of their objects.

Perl 6 OTOH is so flexible that the structure of the objects is 
determined by the meta object protocol, and can be changed at will. 
Likewise are other parts of the language sufficiently flexible that you 
can't build a JIT compiler which special-cases all the common Perl 6 
idioms, and thus becomes fast.

So I personally wouldn't know how to write a "Perl 6 JIT compiler", 
except by writing a general JIT compiler on the PIR (or M0 or whatever) 
level, and then doing some performance tweaks.

Cheers,
Moritz


More information about the parrot-dev mailing list