Re: [Jython-dev] CPyExt

[Tobias Ivarsson <thobes@gm...>]

Well, CPyExt is a pure reimplementation of the API needed by the Python C
extensions, and IronClad (last time I looked) reuses a lot of the actual
CPython implementation. My thinking is that a pure reimplementation would be
easier to port, reuse and maintain.

>From reading up on the pypy discussions about this code it doesn't seem like
it's ready to support NumPy yet, but they seem hopeful about it. And from a
quick glance at the code it is of course written in RPython. I think for
our purposes we would reuse the ideas but rewrite a lot in C, linking with
JNI.

The source code for CPyExt is in this pypy branch by the way:
http://codespeak.net/svn/pypy/branch/cpython-extension

/Tobias

On Sat, Apr 10, 2010 at 9:21 PM, Philip Jenvey <pjenvey@...>wrote:

>
> On Apr 10, 2010, at 3:52 AM, Tobias Ivarsson wrote:
>
> > Have you guys seen this:
> http://morepypy.blogspot.com/2010/04/using-cpython-extension-modules-with.html?
> >
> > I wonder if it could be integrated with Jython to enable Jython to use
> CPython extension modules such as PIL and NumPy.
> >
> > From an initial look, it seems to be an easier approach for us to
> integrate than IronClad would be, although I need to look deeper before I
> could tell for sure.
>
> It's very interesting but I haven't looked at it at all yet, why would it
> easier for us to adopt?
>
> --
> Philip Jenvey

[Jython-dev] CPyExt

[Tobias Ivarsson <thobes@gm...>]

Have you guys seen this:
http://morepypy.blogspot.com/2010/04/using-cpython-extension-modules-with.html
 ?

I wonder if it could be integrated with Jython to enable Jython to use
CPython extension modules such as PIL and NumPy.

>From an initial look, it seems to be an easier approach for us to integrate
than IronClad would be, although I need to look deeper before I could tell
for sure.

Cheers,
Tobias

Re: [Jython-dev] CPyExt

[Philip Jenvey <pjenvey@un...>]

On Apr 10, 2010, at 3:52 AM, Tobias Ivarsson wrote:

> Have you guys seen this: http://morepypy.blogspot.com/2010/04/using-cpython-extension-modules-with.html ?
> 
> I wonder if it could be integrated with Jython to enable Jython to use CPython extension modules such as PIL and NumPy.
> 
> From an initial look, it seems to be an easier approach for us to integrate than IronClad would be, although I need to look deeper before I could tell for sure.

It's very interesting but I haven't looked at it at all yet, why would it easier for us to adopt?

--
Philip Jenvey

Re: [Jython-dev] CPyExt

[Tobias Ivarsson <thobes@gm...>]

Well, CPyExt is a pure reimplementation of the API needed by the Python C
extensions, and IronClad (last time I looked) reuses a lot of the actual
CPython implementation. My thinking is that a pure reimplementation would be
easier to port, reuse and maintain.

>From reading up on the pypy discussions about this code it doesn't seem like
it's ready to support NumPy yet, but they seem hopeful about it. And from a
quick glance at the code it is of course written in RPython. I think for
our purposes we would reuse the ideas but rewrite a lot in C, linking with
JNI.

The source code for CPyExt is in this pypy branch by the way:
http://codespeak.net/svn/pypy/branch/cpython-extension

/Tobias

On Sat, Apr 10, 2010 at 9:21 PM, Philip Jenvey <pjenvey@...>wrote:

>
> On Apr 10, 2010, at 3:52 AM, Tobias Ivarsson wrote:
>
> > Have you guys seen this:
> http://morepypy.blogspot.com/2010/04/using-cpython-extension-modules-with.html?
> >
> > I wonder if it could be integrated with Jython to enable Jython to use
> CPython extension modules such as PIL and NumPy.
> >
> > From an initial look, it seems to be an easier approach for us to
> integrate than IronClad would be, although I need to look deeper before I
> could tell for sure.
>
> It's very interesting but I haven't looked at it at all yet, why would it
> easier for us to adopt?
>
> --
> Philip Jenvey

Re: [Jython-dev] CPyExt

[Wayne Meissner <wmeissner@gm...>]

If anyone is interested in working on this, I could see if there is
anything I can salvage from my jruby-cext project.  It includes a bit
of support code to take some pain out of the JNI-calling-java
experience.

I had a look at doing a cpython-extension layer for jython last year
(about the same time I resurrected the jruby-cext code), and I think
it won't be too horrendous.  Compared to ruby, the python C ext api
uses reference counting for object lifecycles, so it won't have to
emulate a totally different GC like jruby-cext has to.

There will still be some warts, which the pypy guys also had to deal
with (such as C exts which directly access the backing store of a
PyString).


On 11 April 2010 09:55, Tobias Ivarsson <thobes@...> wrote:
> Well, CPyExt is a pure reimplementation of the API needed by the Python C
> extensions, and IronClad (last time I looked) reuses a lot of the actual
> CPython implementation. My thinking is that a pure reimplementation would be
> easier to port, reuse and maintain.
> From reading up on the pypy discussions about this code it doesn't seem like
> it's ready to support NumPy yet, but they seem hopeful about it. And from a
> quick glance at the code it is of course written in RPython. I think for
> our purposes we would reuse the ideas but rewrite a lot in C, linking with
> JNI.
> The source code for CPyExt is in this pypy branch by the
> way: http://codespeak.net/svn/pypy/branch/cpython-extension
> /Tobias
>
> On Sat, Apr 10, 2010 at 9:21 PM, Philip Jenvey <pjenvey@...>
> wrote:
>>
>> On Apr 10, 2010, at 3:52 AM, Tobias Ivarsson wrote:
>>
>> > Have you guys seen this:
>> > http://morepypy.blogspot.com/2010/04/using-cpython-extension-modules-with.html
>> > ?
>> >
>> > I wonder if it could be integrated with Jython to enable Jython to use
>> > CPython extension modules such as PIL and NumPy.
>> >
>> > From an initial look, it seems to be an easier approach for us to
>> > integrate than IronClad would be, although I need to look deeper before I
>> > could tell for sure.
>>
>> It's very interesting but I haven't looked at it at all yet, why would it
>> easier for us to adopt?
>>
>> --
>> Philip Jenvey
>
> ------------------------------------------------------------------------------
> Download Intel&#174; Parallel Studio Eval
> Try the new software tools for yourself. Speed compiling, find bugs
> proactively, and fine-tune applications for parallel performance.
> See why Intel Parallel Studio got high marks during beta.
> http://p.sf.net/sfu/intel-sw-dev
> _______________________________________________
> Jython-dev mailing list
> Jython-dev@...
> https://lists.sourceforge.net/lists/listinfo/jython-dev
>
>

Re: [Jython-dev] CPyExt

[Stefan Behnel <stefan_ml@be...>]

Wayne Meissner, 11.04.2010 02:28:
> There will still be some warts, which the pypy guys also had to deal
> with (such as C exts which directly access the backing store of a
> PyString).

... and PyList/PyTuple, to name a few more. It's extremely common to access 
their content directly instead of going through the high-level C-API 
functions for each item. Cython does that a lot, for example. So that would 
mean that many data structures would basically have to get copied back and 
forth when using them through the JNI. May or may not be faster than 
separate item access, depending on their size and the C code that uses 
them, but I'd expect that you'd loose a lot of performance on average.