[ann] Minimal Python project

John Roth johnroth at ameritech.net
Sun Jan 12 02:20:21 CET 2003


"Manuel M. Garcia" <mgarcia at cole-switches.com> wrote in message
news:dgb12von92hgnkrqmibja28mmp602coi0j at 4ax.com...
> On Sat, 11 Jan 2003 20:09:36 +0100, holger krekel
> <pyth at devel.trillke.net> wrote:
> (edit)
> >We probably should have said 'hope to' instead of
> >'are confident' regarding speed prospects.
>
> Yeah, I got nervous when I read "we are confident that MinimalPython
> will eventually run faster than today's CPython".
>
> It reminded me of the "microthread OS" and "Perl 6 Parrot registers VM
> versus stack VM".  Before coding even begins, performance claims are
> made, mainly based on the "coolness" of the technique.  When the
> performance does not materialize, or the coding is stalled, the
> technique is still treated as above reproach.  And when the old way is
> still faster, pleading that will more time, and with more coders, the
> old way will be left in the dust.
>
> I can see how MinimalPython could be faster than CPython.  No
> programmer would write C code tuned to run optimally on every machine
> under every different multi-tasking work load with every different
> internal state.  Too many machines, too many ways of trading space for
> speed, too many ways of striving for something small enough to fit in
> the cache and avoid paging, too many different tricks.  Of course, a
> sane C programmer would write a single reasonably efficient
> implementation, and in the program's hot spots, maybe specialized code
> to take advantage of a few common situations.
>
> A JIT compliler would supply code tuned for the specific machine in
> the specific situation.  It might be different low level code each
> time, but the JIT compliler does not care.
>
> The part of the project I am most excited about is playing with a
> reasonalbly efficient Python implementation written mainly in Python!
> If the speed never materializes, we still have a test bed for many
> Python programmers to try out new ideas for the traditional Python
> implementations.  And if the speed claim comes true, that is a whole
> another level of goodness!

What about writing it *entirely* in Python?

I'm currently playing around with Squeak, which is a Smalltalk
variant whose interpreter is written entirely in a subset of Smalltalk
that can be machine translated to portable C!

John Roth

>
> Manuel






More information about the Python-list mailing list