aleaxit at yahoo.com
Wed Oct 20 11:55:54 CEST 2004
Maurice LING <mauriceling at acm.org> wrote:
> Mark Asbach wrote:
> > If you need an interesting topic: why not write a just-in-time compiler
> > for Python?
> > There's already some interesting stuff in module "weave" and the pythonc
> > from Jython, but having a python interpreter use (say) gcc at runtime to
> > convert and probably optimize Python code to natively executable
> > binaries, you could speed up Python to the max.
> I think another issue here is that Python bytecode is not stable, as
> everyone had said it. So unless a specification on that is present,
> implementing JIT compiler for Python is futile... in my opinions.
Nah - just call it "a JIT compiler for Python 2.4". If there are minor
bytecode changes in 2.5, it will be correspondingly minor to adjust the
JIT. And you can start with the body of work (previous art) already
present in psyco and in the pypy project -- indeed since they're open
source you can build on top of either or both; psyco's dormant, but
pypy's quite active and you may end up fruitfully cooperating with pypy
people (we _ARE_ extremely open to community participation!!!) among
which is Armin Rigo, psyco's author. pypy minimizes issues due to
unfamiliarity with C -- it uses pyrex as a backend.
If you're keen to remain in the Java/JVM world for whatever reasons,
there may be interesting work in helping complete Jython _AND_ ensuring
it cooperates smoothly with JVM JIT's (I hear it doesn't, so far, but
this IS just hearsay -- I know nothing of the issue). Not sure this is
easy to package up as a thesis topic. Another possibility would be to
work on Python/Eclipse integration -- there's quite a lot of previous
art and some projects may be ongoing, but that is one extremely useful
wide set of tasks which look like they'd be suitable for packaging up a
thesis topic from.
More information about the Python-list