We will give an update on the state of PyPy, the Python
implementation in Python.

Slides:

1. http://codespeak.net/svn/pypy/extradoc/talk/ep2009/status/pypy-status.pdf
2. http://codespeak.net/svn/pypy/extradoc/talk/ep2009/jit/pypy-jit.pdf


Part 1 - becoming complete - 30 minutes of the projected 1 hour
===============================================================

PyPy has come a long way.  No longer merely a research project and
proof of concept, PyPy now is
able to run sophisticated python applications such as Django and
Twisted.  We will focus on the improvements that happened during
the last year in PyPy, which made this possible.

Thus -- even before the JIT work -- we could assert that PyPy
now provides a complete, flexible and reasonably fast Python
interpreter, itself implemented in Python.  ("Complete" means
it is fully compliant but misses a lot of third-party
extension modules.)

This section of the talk is mostly intended for people who are already
familiar with the PyPy project, and want an update as to how we are doing.
We will also explain the motivations driving current PyPy development
and its targets for the future.


Part 2 - becoming fast - 30 minutes of the projected 1 hour
===========================================================

As we write this abstract we are finally integrating our Just-in-Time 
compiler generator into PyPy.  Preliminary results look very good
(more during the talk!).

The PyPy JIT compiler, in addition to being a *specialising* compiler,
is now a *tracing* compiler.  Another example of a tracing compiler is
JavaScript's TraceMonkey, soon to be included in
FireFox.  We will give the status of the implementation of
our JIT, with speed measures.

We will then describe how it all works.  No familiarity with tracing
or specialising compilers required.

Due to our architecture we can generate code for C but also for the
JVM and .NET; it is possible to generate a JIT for these platforms too,
running on top of the platform's native JIT.

Also, as usual in PyPy, one can generate such a tracing compiler for 
*any* computer language, only by writing an interpreter for it.  We already
have ones for Squeak, Prolog and GameBoy, and experimental ones for
JavaScript, Scheme, etc. so people wishing
to speed up other languages now have a clear and simple path to follow.


Authors: antocuni, cfbolz, pedronis, arigo