Speaking about Unlanden Swallow,

you should probably watch this:
http://pycon.blip.tv/#1956861
There is guy from Google explaining why they are doing that: at Google they are 
stucked with a lot of Python 2.5 code linked to C++ code by SWIG interfaces. 
They would have looked at Jython, but for the moment they have to make that 
CPython faster in a manner that will stay compatible with their adhoc C++ 
wrappers.

They plan to add some trivial JIT support inside the Python bytecode 
interpreter and expect a x5 in perf. They are planning to backport any 
improvement in CPython trunk.

Meanwhile, this video also compares the other Python VM's. I would rather go 
for Jython in the long run (and I hope Google will too), but it might take some 
time. Unladen might be the best short term option to win speed. In the long run 
things are different, I doubt artisanal engineering can come close to say the 
JVM. The thing is that Jython focused more on compatibility those years, now 
they are going to unleash the perf side. 

Notice that OpenERP is not Google code, the situation is better I think. 
Indeed, OpenERP uses relatively few native libs and among those ones, some will 
be driven by Python "standard" CTypes. Meaning that we could possible use a 
better VM (jython) as long as the support CTypes (Jython will soon).

In any case, OpenERP is luckily probably fast enough for lots of uses already.

Raphaël Valyi

------------------------
http://www.smile.fr




-------------------- m2f --------------------

--
http://www.openobject.com/forum/viewtopic.php?p=34324#34324

-------------------- m2f --------------------


_______________________________________________
Tinyerp-users mailing list
http://tiny.be/mailman/listinfo/tinyerp-users

Reply via email to