Re: [Jython-dev] skeleton benchmarks

[<bc...@wo...>]

[John Coppola]

>Was the results of this experiment uninteresting? No
>comments?  There seemed to be a lot of enthusiasm at
>one point but no longer.  Hmmm... what are you fellas
>up to?

I'm reviewing a book, implementing the _weakref module and fixing some
random jython bugs. Thanks for asking <wink>.

Your finding of a 8x speedup matches my own experience when I
implemented cPickle and cStringIO. Hardening a non-trivial python module
into a java class can typically give 5x to 9x faster code. 

The fact that cPickle.java looks so much faster then pickle.py is due to
the extremely low performing (on jython) way pickle.py reads strings
from the pickle. Each string read from a file cause the compilation,
creation and loading of a new java class. cPickle.java avoids this by
reaching deep into the internal jython implementation to reuse the
string parsing from the parser.

I'll look closer at your example over the weekend.

regards,
finn