Re: [Jython-dev] speed obsession

[Samuele P. <pe...@in...>]

About the recurrent topic of speed obsession:

AFAIK
we probably can improve Jython speed, here and there.
(with some caching and maybe cutting some call-paths).

But Jython is a dynamic language implemented over the JVM, not hosted
by a CPU.

Yes, there is a series of "known" ways to improve speed for dynamic languages:
- tagged arithmetic
- inline (polymorphic) caches
- dynamic/adaptive compilation techniques
etc

(Actual JVMs use them, of course not tagged arithmetic for the obvious reason)

but for the moment I don't see any reasonable way to implement them for Jython 
over the JVM, which is a more rigid beast wrt a hardware CPU and have
a different execution time profile and don't execute everything and
always at full speed. It's also a matter of how much memory one can throw
at speed improvements.

The other possibilities:
- adding (optional) static typing (a remote and not very nice possibility, and a 
quite complicated problem and this has failed on Python side so far)
- add a type inferencer to jythonc, but probably we will loose too
much dynamism and in any case a production version of such a beast is far from
trivial and is a lot of work (!). I can post a list of papers on that.

So we can just wait for further improvements of JVMs implementetions,
expecially related to reflection.
Or wait Sun to change the JVM in order to better accomodate dynamic languages.

regards, Samuele Pedroni.