Nuitka Python Compiler

[Lucian Branescu]

It seems to be that nuitka could possibly become what Unladen Swallow
couldn't: a blessed CPython speedup, as a stopgap until everyone moves
to PyPy.

Interesting either way. There have been several other similar efforts,
one of which targeted x86 assembly directly.

On 13 October 2011 17:00, Allison Randal <allison at parrot.org> wrote:
> -------- Original Message --------
> Subject: Re: Request for packaging - Nuitka the Python Compiler
> Resent-Date: Wed, 12 Oct 2011 22:33:55 +0000 (UTC)
> Resent-From: debian-python at lists.debian.org
> Date: Thu, 13 Oct 2011 00:33:33 +0200
> From: Kay Hayen <kayhayen at gmx.de>
> To: debian-python at lists.debian.org
>
> Hello Jakub,
>
>>> http://www.nuitka.net/blog/nuitka-a-python-compiler/what-is-nuitka/
>>
>> How is it different/better than pure-Python[0] mode of Cython?
>
> It's more compatible to CPython than anything that exists. Nuitka passes
> practically 100% of the test suite. Currently I don't have support for
> threading, and that's it. Frame stack works perfect in branch already.
>
> Nuitka is a project not about a hybrid language, and not about C types,
> and it's intended for whole programs acceleration. That's probably
> differences that currently matter.
>
> It's design is cleaner, it uses the CPython parser to parse Python, it
> uses Scons to build the generated code, etc.
>
> Cython is trying to be more Python compatibility recently, but Nuitka
> already had full language coverage, before Cython started with
> generators. Now they have it. But there is still "unimportant" things,
> not supported.
>
> I believe Cython is currently the best choice available for something
> productive, but Nuitka has a cleaner plan (only Python semantics matter)
> and I believe a quicker road to success, and has already uses.
>
> Ultimately, I agree with Dr. Stefan Behnel, one of the lead developers
> of Cython, that the projects are coming from different ends, but reach
> out to similar goals.
>
> I gave my reasons on recent PyCON DE. But it boils down to willingness
> to move and different goals. Stefan agrees with me that Nuitka has
> different enough goals, or so I understood.
>
> To give an example, parameter errors. The error messages of Nuitka are
> identical to CPython and that's the test. The error message of Cython
> are not identical and arguably not better. The generated code may be or
> or less faster.
>
> To me, the only correct solution is the one that 100% imitates CPython
> and even avoids improvements to CPython. To Stefan the faster solution
> is an acceptable compromise.
>
> With this approach, a 100% compatibility cannot be achieved, which also
> means that you have to have your own tests. Cython needs to have a lot
> of efforts, because it has data driven testing that describes the
> non-CPython behaviour of Cython. I can just compare CPython and Nuitka
> and every difference is a bug.
>
> Yours,
> Kay
>
> _______________________________________________
> http://lists.parrot.org/mailman/listinfo/parrot-dev
>