[ann] Minimal Python project

Manuel M. Garcia mgarcia at cole-switches.com
Tue Jan 14 00:51:48 CET 2003


Please, please, please release a "first playable" as quickly as
possible, even if performance is below that of CPython.

Because imagine a crude version is quickly released that runs 50 times
slower than CPython (and *nobody* expects performance this bad).  It
will have fast context switching between threads (thanks to Christian
begin on the team ;-) ), so even if a "first playable" is this slow,
it will still be faster than CPython for an interesting set of
applications.  Maybe because it is feasable to customize the context
switching for the specific application (because it is just Python
code, after all), it will run at a pretty good clip!

Secondly, the pool of programmers willing to write finely-tuned code
will be pretty large, considering development will be in Python and
not it C.  I can imagine many Python programmers will love to work
man-years on this, even to the point of harming marriages!

Also, please allow "thread pickling" from day one.  (I remember
Christian saying Stackless 1.0 had thread pickling.)

Actually, thread pickling might give a performance boost.  I was
thinking about it during the weekend; you code a loop in a thread,
take it through its paces through test data, all the while <PyPy> is
figuring out how to make things run fast.  Then when you get the
performance you want, you just pickle the thread, and all the
performance "hacks" get pickled.  When you unpickle the thread in the
deliverable, it zips along immediately, without the "inspection" time
you would expect from the PSYCO technique.

It is hard *not* to be excited about this project!

Manuel




More information about the Python-list mailing list