[Tutor] Interpreter level objects

[Danny Yoo]

On Wed, 23 Feb 2005, Jacob S. wrote:

> Say I put the code
>
> import psyco
> psyco.full()
>
> in sitecustomize.py and run a random file that I have already got an
> average execution time of. Then I run it again, with the above
> implemented. My execution time is has dropped. Which brings me to
> believe that psyco monitors all subprocesses and all subobjects.


Hi Jacob,

Yes, psyco.full() will do so, by patching itself in the Python runtime.
But normally, people will run psyco on specific functions that they know
are hot-spots in their code by using psyco.bind().


> So this being the case, why haven't the volunteers set up from
> __import__ division so that you can put it in sitecustomize.py and get
> it to run whenever you start python?

One thing is that psyco only works on the x86 platform.  It's also a bit
experimental; in the Drawbacks section in Psyco's introduction:

    http://psyco.sourceforge.net/introduction.html

the page mentions that the behavior of certain programs will change under
Psyco, and that's not ideal.  Finally, Psyco uses a lot of memory, and
that's probably the biggest drawback.

In short, Psyco is intrusive enough that it's probably not a good idea to
turn it on by default.


> Anyways, this all came up because several modules that used psyco.full()
> would greatly be hindered if they imported another module that did the
> same.  So I got the great idea, why not put it in sitecustomize.py?

I think the current thinking is that individual modules are responsible
for calling psyco on themselves --- that is, the authors of a module would
know best what functions would benefit most from Psyco's JIT compilation.

Calling Psyco on everything is probably not a good idea, just because it
does use a lot of memory, and because there is a time cost involved in
doing the JIT stuff itself.  If some code is going to be run just once
anyway, there's really not much gain in JIT-ing it first, since all that
analysis is probably going to cost more than running the code once.

Psyco can work wonders on things like inner loops, but it's not so
effective on top-level code.



> Then I got the idea, why not put from __import__ division in the
> sitecustomize.py, too?  psyco.full() worked, from __import__ division
> did not.

The statement:

###
from __future__ import division
###

is module specific; it won't automatically affect division throughout the
system, but limits itself to the particular module where it's been
declared.  So in your example, new-style division actually is being turned
on only for sitecustomize.py.

This is briefly mentioned in:

    http://www.python.org/peps/pep-0238.html

with:

"""
    - The future division statement, spelled "from __future__ import
      division", will change the / operator to mean true division
      throughout the module.
"""

If we want to make new-style division work throughout the system, we can
call Python with the '-Q new' command line argument.



Best of wishes to you!