[BangPypers] The merge : Unladen Swallow into CPython
Anand Balachandran Pillai
abpillai at gmail.com
Thu Jan 21 07:40:56 CET 2010
On Thu, Jan 21, 2010 at 10:59 AM, Noufal Ibrahim <noufal at gmail.com> wrote:
> http://thread.gmane.org/gmane.comp.python.devel/109919
>
> Looks it's really lead laden rather than unladen. An interpreter binary is
> over 100 MB (as opposed to the original CPython one which is less than 10).
> More details in the PEP in the above email thread.
>
> I sense a disturbance in the force.
>
I read through the entire thread. The PEP is detailed, descriptive and
Collin has done a thorough job of discussing the pluses and minuses
of this approach. I don't see why the PEP should not be accepted.
What disappoints me is the amount of "performance improvement" they
have achieved. With most workloads he is testing on, what I see is an
average of 1.5x perf increase, that too on a super muscled box. The
minuses are more what with the average memory increase of
2.5x and mammoth executable size of 128 MB on 64 bit!
Basically if Collin is expecting a warm hug from pydev for this PEP
he is never going to get it, at best a lukewarm response or most likely
a *shrug*.
As I made a point earlier, they seemed to have assumed LLVM was
a silver bullet and it turned out to have its own problems which they
had to spend fixing. Also the approach of LLVM analysis of every code
path has to be optimized, and unboxing is required for most numerical
work-loads.
The addition of a C++ API to CPython is a rather bitter pill to swallow
IMHO :)
>
>
> --
> ~noufal
> http://nibrahim.net.in
> _______________________________________________
> BangPypers mailing list
> BangPypers at python.org
> http://mail.python.org/mailman/listinfo/bangpypers
>
--
--Anand
More information about the BangPypers
mailing list