Comments on: Pushing Python Past the Present

By: Robert Zaremba

Robert Zaremba — Mon, 08 Oct 2012 20:35:56 +0000

If you looking for easy integration with legacy c code just check libffi ( http://cffi.readthedocs.org ) which PyPy guys created and is compatible both by CPython and PyPy

By: tehwalrus

tehwalrus — Sun, 07 Oct 2012 11:19:22 +0000

I work in Python, Cython, C (using the cython interface) and I write wrappers for other peoples’ FORTRAN (in cython, via direct C access). If all of that works in PyPy, then I’d be happy to try it.

Oh, I also require pygame, pyopengl, cjson (or a json library that doesn’t use 6GB RAM to open a 100MB document, like the default CPython module,) and so on.

Seriously, my entire PhD is (indirectly) trying to make python faster – Support this and I’ll switch.

By: Dmitri Lebedev

Dmitri Lebedev — Sun, 07 Oct 2012 09:09:27 +0000

Just to compare, see Jakob Kaplan-Moss’ keynote at DjangoCon 12, where he presented Meteor.js, a very threatening future competitor, said “this is too hard in Django”, yet the whole talk was very inspirational and made me personally to start using Sphinx finally.

By: Dmitri Lebedev

Dmitri Lebedev — Sun, 07 Oct 2012 07:46:13 +0000

Guido’s keynote was quite dissapointing for me, despite the content of the message. When you say “Python is everywhere”, saying “It’s not our fault that were not on mobile” sounds really bad. Then, I heard nothing about why Python is adopted this wide, like it just happened. So I get Guido’s message as “The adoption of Python is out of our control.”

By: Philip Neustrom

Philip Neustrom — Sat, 06 Oct 2012 22:30:34 +0000

RE: gevent – many people are working on this and it looks like it’s getting close. More details here: https://groups.google.com/forum/?fromgroups=#!topic/gevent/1JPOFqplTw4

By: Pushing Python Past the Present « thoughts…

Pushing Python Past the Present « thoughts… — Sat, 06 Oct 2012 21:49:41 +0000

[...] http://archlinux.me/dusty/2012/10/04/pushing-python-past-the-present/ Share this:TwitterFacebookLike this:LikeBe the first to like this. [...]

By: Volker Birk

Volker Birk — Sat, 06 Oct 2012 20:53:54 +0000

Why I stopped using PyPy? Well, with my tool chain, it was slower than CPython, so I saw no reason.

I think, PyPy is a cool project, though.

By: David Warde-Farley

David Warde-Farley — Sat, 06 Oct 2012 20:01:22 +0000

It's been said better elsewhere, but for the community of Python users with whom I most often interact (that is, scientific users), PyPy is currently a no-go as well. Part of what makes NumPy a bedrock is its extensive C API upon which mountains of packages are based; rewriting all of those packages will take an immense amount of time and effort, and even then, easy integration with legacy C and Fortran libraries is immensely important. Rewriting such libraries in Python is simply not a serious option for a lot of reasons. See Peter Wang's NumPy isn't about fast arrays and More thoughts on arrays in PyPy for details. Succinctly put, a NumPy C API is not an implementation detail, it is a necessity. I really do wish this were not the case. Interpreted CPython performance sucks, and PyPy and the STM patch seems like the best hope for real multi-threaded parallelism in Python. If I were writing a conventional Python web application or something like that, or doing non-numerical data crunching, I might consider PyPy. But array computation benchmarks alone are not going to rip me away from CPython for numerical work, simply because I'd be leaving behind the ecosystem and not gaining the tools necessary to rebuild it (setting aside, for the moment, the manpower problem).

By: Thomas Kluyver

Thomas Kluyver — Sat, 06 Oct 2012 18:51:30 +0000

Unfortunately, for a lot of the scientific community using Python, PyPy is still essentially an interesting research project. We have a lot of development invested in C extensions, which PyPy has no interest in supporting well. And since the critical parts are already in compiled code, there’s little payoff for the massive rewriting that would be necessary.

By: Justin Alan Ryan

Justin Alan Ryan — Sat, 06 Oct 2012 18:42:58 +0000

Last I checked, there’s some work to support traditional CPython apis, except for those related to refcounting, but the true golden path here is for C extensions to be developed using ctypes, which allows runtime dynamic linking. Pretty sure anything that uses ctypes should work on PyPy.