For the records.
I've just committed the revised package to 10.5 and 10.4 unstable.
Begin forwarded message:
> Resent-From: Michael Wild <[EMAIL PROTECTED]>
> From: Michael Wild <[EMAIL PROTECTED]>
> Date: April 22, 2008 10:39:08 CEDT
> Resent-To: [EMAIL PROTECTED]
> To: [EMAIL PROTECTED]
> Subject: Patch for the Fink Numarray package
>
> Hi
>
> Using the Python Numarray package I came across a bug which
> apparently is fixed in CVS (see below conversation with the
> developers). I created a patchfile for the numarray-py package and
> adjusted the info file accordingly. You might want to adjust the
> revision number, I just set it to 1.patched and not 2 in order to
> get your update.
>
> Regards
>
>
> Michael
>
> Begin forwarded message:
>> From: Todd Miller <[EMAIL PROTECTED]>
>> Date: 21April, 2008 21:14:43 GMT+02:00
>> To: [EMAIL PROTECTED], [EMAIL PROTECTED]
>> Subject: Re: You have been assigned call CNSHD722278 regarding
>> PyArray_FromDims in Numarray leaking Py_None by chanley its status
>> is Assigned
>>
>> This is a known problem with no corresponding source release. The
>> patch against 1.5.2 patches Src/libnumarray.ch and is:
>>
>> [EMAIL PROTECTED] Src]$ cvs diff -r 1.33 -r 1.34 libnumarray.ch
>> Index: libnumarray.ch
>> ===================================================================
>> RCS file: /cvsroot/numpy/numarray/Src/libnumarray.ch,v
>> retrieving revision 1.33
>> retrieving revision 1.34
>> diff -c -r1.33 -r1.34
>> *** libnumarray.ch 24 Aug 2006 18:39:45 -0000 1.33
>> --- libnumarray.ch 20 Nov 2006 17:35:31 -0000 1.34
>> ***************
>> *** 877,882 ****
>> --- 877,883 ----
>> if (!buf) return NULL;
>> } else {
>> buf = Py_None;
>> + Py_INCREF(buf);
>> }
>> a = NA_NewAllFromBuffer( nd, dimensions, descr->type_num, buf,
>>
>> After patching, do:
>>
>> % python setup.py install --gencode --selftest
>>
>> Another alternative is to use the unofficial numarray-1.5.3 tarball
>> (attached) which should work on at least linux and Mac OS-X. It is
>> lightly tested with Python-2.5.2 but there are known self-test
>> problems on Windows (numarray floating point Python version
>> specific platform esoterica) or Mac OS-X (apparently in numpy). It
>> appeared to fix your problem though.
>>
>> Regards,
>> Todd
>>
>> [EMAIL PROTECTED] wrote:
>>> You have been assigned call CNSHD722278 regarding PyArray_FromDims
>>> in Numarray leaking Py_None by chanley its status is Assigned
>>>
>>> Request Log : 04/18/08 04:28:16 mcleskey
>>> Request from [EMAIL PROTECTED]
>>> To: [EMAIL PROTECTED]
>>> Cc: This is an OpenPGP/MIME signed message (RFC
>>> 2440 and 3156)
>>> --Apple-Mail-20-418780118
>>> Content-Type: multipart/mixed; boundary=Apple-Mail-19-418780066
>>> --Apple-Mail-19-418780066
>>> Content-Type: text/plain;
>>> charset=US-ASCII;
>>> format=flowed;
>>> delsp=yes
>>> Content-Transfer-Encoding: 7bit
>>> Hi
>>> I know that Numarray is being phased out and one should
>>> switch over to Numpy, but I encountered a problem with a
>>> rather large code (cantera) which makes use of Numarray, so
>>> I still direct this question to you.
>>> The cantera code I use had the strange behavior of
>>> crashing python with an error message stating "trying to
>>> deallocate None". I tracked it down to PyArray_FromDims
>>> being called very often, and found that after each call the
>>> reference count of the Py_None object is decreased by 1.
>>> This means, that after ~1500 calls to PyArray_FromDims,
>>> python tries to deallocate Py_None, resulting in the crash.
>>> First I thought it might be cantera "misusing" Numarray.
>>> But then I modified the "dimsdata" example provided with
>>> Numarray to use PyArray_FromDims, and the same behavior
>>> shows. Strange enough, PyArray_FromDimsAndData works just
>>> fine.
>>> Is this a known issue? Is there a patch fixing this? I
>>> hacked a quick solution by calling Py_INCREF(Py_None) after
>>> each call to PyArray_FromDims in cantera. This fixed the
>>> immediate symptoms, but I'm quite sure that this might make
>>> things even worse (and am well aware that this doesn't fix
>>> the actual cause of the problem).
>>> I attached the modified dimsdata example with a python
>>> driver script so you can try to reproduce the problem. To
>>> do so, save the attached files and in a shell execute:
>>> python setup.py build
>>> ln -s <path/to/dimsdata.so> .
>>> python test.py
>>> You should see that the reported reference count
>>> decreases by 1 after each call to PyArray_FromDims, but is
>>> unchanged after the call to PyArray_FromDimsAndData.
>>> Michael
>>> My specs are:
>>> MacBook Pro
>>> Intel Core 2 Duo 2.33 GHz
>>> 2GB of 667 MHz DDR2 SDRAM
>>> Mac OS X 10.5.2
>>> Python 2.5.2 installed with fink
>>> Numarray 1.5.2 installed with fink
>>> The same thing happens on a Linux box:
>>> AMD Athlon 64 X2 Dual Core Processor 3800+
>>> SuSE Linux 10.3, Kernel 2.6.16
>>> Python 2.4.2
>>> Numarray 1.5.2
>>> --
>>> Institute of Fluid Dynamics
>>> ETH-Zentrum, ML H59
>>> CH-8092 Zurich
>>> +41 44 632 8709
>>> --Apple-Mail-19-418780066
>>> Content-Disposition: attachment;
>>> filename=dimsdata.c
>>> Content-Type: application/octet-stream;
>>> x-unix-mode=0644;
>>> name="dimsdata.c"
>>> Content-Transfer-Encoding: 7bit
>>> /* This module shows how to turn a static C array into a
>>> NumArray. */
>>> .#include "Python.h"
>>> .#include "numarray/arrayobject.h"
>>> .#include <stdio.h>
>>> static double data[4][3] = {{1, 2, 3},
>>> {4, 5, 6},
>>> {7, 8, 9},
>>> {10, 11, 12}};
>>> static int shape[2] = { 4, 3 };
>>> static PyObject *
>>> _getarray(PyObject *module, PyObject *args)
>>> {
>>> if (!PyArg_ParseTuple(args,":getarray"))
>>> return NULL;
>>> int num;
>>> num = 100;
>>> printf("before PyArray_FromDims: %d\n",(int)Py_None-
>>> >ob_refcnt);
>>> PyObject* a =
>>> PyArray_FromDims(2, shape, PyArray_DOUBLE);
>>> printf("after PyArray_FromDims: %d\n",(int)Py_None->ob_refcnt);
>>> Py_DECREF(a);
>>> printf("before PyArray_FromDimsAndData: %d\n",(int)Py_None-
>>> >ob_refcnt);
>>> PyObject* res =
>>> PyArray_FromDimsAndData(2, shape, PyArray_DOUBLE, (char
>>> *)data);
>>> printf("after PyArray_FromDimsAndData: %d\n\n",(int)Py_None-
>>> >ob_refcnt);
>>> return res;
>>> }
>>> static PyMethodDef _dimsdata_functions[] = {
>>> {"getarray", _getarray, METH_VARARGS, "getarray()
>>> returns a static array cast as a numarray."},
>>> {NULL, NULL} /* Sentinel */
>>> };
>>> static char * _dimsdata_doc = "Example module for
>>> PyArray_DimsAndData.";
>>> void initdimsdata(void)
>>> {
>>> PyObject *m = Py_InitModule3("dimsdata",
>>> _dimsdata_functions,
>>> _dimsdata_doc);
>>> if (m == NULL) return;
>>> import_array();
>>> }
>>> --Apple-Mail-19-418780066
>>> Content-Disposition: attachment;
>>> filename=test.py
>>> Content-Type: text/x-python-script;
>>> x-unix-mode=0644;
>>> name="test.py"
>>> Content-Transfer-Encoding: 7bit
>>> from dimsdata import *
>>> for i in range(10):
>>> getarray()
>>> --Apple-Mail-19-418780066
>>> Content-Disposition: attachment;
>>> filename=setup.py
>>> Content-Type: text/x-python-script;
>>> x-unix-mode=0644;
>>> name="setup.py"
>>> Content-Transfer-Encoding: 7bit
>>> from distutils.core import Extension, setup
>>> e = Extension(name="dimsdata", sources=["dimsdata.c"])
>>> setup(name="dimsdata", ext_modules=[e])
>>> --Apple-Mail-19-418780066--
>>> --Apple-Mail-20-418780118
>>> content-type: application/pgp-signature; x-mac-type=70674453;
>>> name=PGP.sig
>>> content-description: This is a digitally signed message part
>>> content-disposition: inline; filename=PGP.sig
>>> content-transfer-encoding: 7bit
>>> -----BEGIN PGP SIGNATURE-----
>>> Version: GnuPG v1.4.9 (Darwin)
>>> iEYEARECAAYFAkgIWWoACgkQl7l0ufRsoXp0VACfc0lKwFDEKZrlINFQ/
>>> rMfiANz
>>> wGEAnj7Fa7ld4rYgvtN0ijduQVGrDMAo
>>> =dPKI
>>> -----END PGP SIGNATURE-----
>>> --Apple-Mail-20-418780118--
>>> 04/18/08 09:15:09 bshaw
>>> bshaw changed platform to STSDAS
>>> 04/21/08 09:40:08 chanley
>>> chanley changed status to Assigned
>>> chanley assigned call to jmiller
-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Fink-devel mailing list
[email protected]
http://news.gmane.org/gmane.os.apple.fink.devel
