One Python 2.1 idea

[Neelakantan Krishnaswami]

On Tue, 26 Dec 2000 19:43:52 +0100, Alex Martelli <aleaxit at yahoo.com> wrote:
><rturpin at my-deja.com> wrote in message news:92ab9u$dou$1 at nnrp1.deja.com...
>>
>> Most work in a Python program involves object methods
>> or attributes. (At least, this is the case in my
>> programs. ;-) Increasing the speed of reference to
>> these should have appreciable benefit, across the
>> board. Yeah, it would make the code a bit more
>> complex to maintain. But this isn't a special case --
>> it is a general mechanism central to how Python works.
>
> It does seem likely that caching lookups is where most
> across-the-board performance might be gained, yes.  The trick is
> finding out a way to invalidate the cache every time some dynamic
> manipulation may need lookups to happen afresh.

I'm not sure this would help very much. Tim posted a link to an
attempt to add method caches, and reported inconclusive results.  As
of late last week, I'm not surprised; the Python interpreterseems to
be spending most of its time in bytecode instruction dispatch. Logic
suggests that that's where performance enhancements would have the
most impact.

Fortunately, I don't have to put my money where my mouth is, since
someone has already done it for me. :) Vladimir Marangozov has
modified the Python interpreter (as of 1.5) to use the GNU computed
labels extension and implemented direct threading.

  http://starship.python.net/crew/vlad/archive/threaded_code/

Bringing that up to date with 2.0 would be an interesting experiment. 


Neel