Very interesting stuff. PyPy/rPython is doing the same thing only it taks out the OCAML step. This backs up the premis with a more direct example.
I think not many people in the python comminuty have yet to react because:
1) we didn't know about it.
2) been there, doing that. ^_^
Wasn't that a PyCon talk? If not, the question to the schedulers is, "Why not?" Especially since some of the accepted talks seem to be re-runs of talks from previous conferences: a tendency which I hope is eliminated from EuroPython this year.
hm, this "we have implemented about half of Python and it is 2 times faster than CPython" sounds so completely familiar. Didn't IronPython claim the same originally?
Doug: RPython can only compile a 'restricted subset' of Python of course. (For example it lacks most of the built-in functions and you can't change the type of variables).
Hmm. I see the video is from a Google Tech Talk (with Guido apparently in the front row), but I noticed that an abstract had also been posted to the PyCon Wiki, so I guess the presenter wanted to talk about it there, too.
Anyway, shouldn't we be encouraging this kind of thing? That people want to implement Python for (or in) other environments (OCaml, Lisp, C++, CLR) says a lot about the desirability of the language.
Implementing half of the language is easy. It's getting the edge cases that things like parrotbench massage that is difficult. If the guy can pull this off more power to him, but the language is not as simple as some people might think when it comes to the complete semantics.
As for encouraging, no one is discouraging him. But he has never come forward to python-dev with questions or let us know this was happening. This video was the first I ever heard about it. So it is not so much lack of encouragement but more that people just don't let it be known they are working on this. Besides having his degree hinge on this will probably be a good motivator. =)
fuzzyman: Yes, its only a subset. The same goes for the OCAML converter. It does not support the full python language, so I don't really see the difference, except an added complexity and not as good type inferencing. Did I miss something?
Years ago Max Skaller was working on a similar project: Vyper. If you Google for it there are still some references...
Doug: I assumed that the aim of the Ocaml translation project was to implement *all* of Python, rather than just a restricted subset - setting it apart from RPython.
Fuzzyman: It seems I did miss something: the extend of support the Ocaml converter has for the python language. There are indeed language features which the Ocaml converter supports which PyPy does not yet support. That said, the goal of PyPy and rPython is to implement all of the cPython standard as well. It is a goal they have been working toward from the beginning. There was alot of discussion at PyCon on wether PyPy+rPython would ever replace cPython. Brett Cannon has a blog post about this (answer: most likely not).
PyPy will certainly cover the whole language spec - but RPython doesn't, and never will. It is a restricted subset of Python.
Commenting by HaloScan
