[Stackless] [PySide] PySide violates Python API - PyQt does not!

Christian Tismer tismer at stackless.com
Sat Dec 8 19:36:23 CET 2012


Correction:

this was too quick. Shiboken does not directly use the threadstate
but only a reference, which is probably ok.A typical mis-reading
that always hits me with C++.

Will investigate further, why Stackless Python causes crashes with PySide.

Update:
It is pretty likely that there is a problem with extra data fields behind
the type object structure, which is used by both pyside and stackless.

from basewrapper.h :

> /// PyTypeObject extended with C++ multiple inheritance information.
> struct LIBSHIBOKEN_API SbkObjectType
> {
>     PyHeapTypeObject super;
>     SbkObjectTypePrivate* d;
> };

This overlaps with the Stackless version of PyHealTypeObject,
which is compatible but larger than CPython's!
At the end of the stackless heaptype lives a bytemap:

slp_methodflagsslpflags;

which misinterprets the field

     SbkObjectTypePrivate* d;


cheers - chris


On 12/8/12 4:45 AM, Christian Tismer wrote:
> Resent with a clearer subject line
>
> This comes from the stackless mailing list.
>
> PySide directly uses the threadstate structure of Cpython, which
> is a protocol violation. See below...
>
> Am 08.12.2012 um 01:49 schrieb Christian Tismer <tismer at stackless.com>:
>
>> Hi friends,
>>
>> this evening, I looked a bit into shiboken, the module that is used
>> for creating python interfaces for PySide, and now I'm
>> pretty confident that I found a reason that is creating some issues.
>>
>> libshiboken has an include file called sbkpython.h, and this includes
>> python.h from the installed python.
>>
>> And now we come to the interesting stuff:
>>
>> There is threadstatesaver.h and threadstatesaver.cpp !
>> The latter has code like:
>>
>> void ThreadStateSaver::save()
>> {
>>      if (PyEval_ThreadsInitialized())
>>          m_threadState = PyEval_SaveThread();
>> }
>>
>> But that code uses the threadstate definition from the cpython threadstate
>> module, instead of the different (at least larger) structure in stackless
>> threadstate. ;-)
>>
>> IOW., things *might* work if pyside is built against stackless python
>> instead of CPython.
>>
>> Today I looked at a simple PySide script and a larger one - they both used
>> multiple threads for startup. No idea if this code is involved there,
>> python did not use threads, but I could imagine that is at least a time
>> bomb.
>>
>>
>> Now, the question is: how to resolve this?
>> ------------------------------------------
>>
>> One way that needs testing is if PySide works ok on stackless
>> when built from source. I could write an installer for this case.
>>
>> The other question is: can we change stackless in a way that works
>> with PySide compiled for CPython?
>>
>> That would mean: tstate must stay completely unchanged in size.
>> I guess the extra info that we use would neet to be put into a
>> different structure, and the linkage back to tstate is a little trick,
>> at least.
>>
>> Another way would be to add a patch to libshiboken, that checks the
>> real size of tstate dynamically. This would be an easier patch, although
>> I think that at the moment it is not so easy to get anything into pyside,
>> quickly.
>>
>> I would like to get your opinion.
>> And I definately want to make stackless work with pyside.
>>
>> The source code can be inspected here:
>>
>> http://qt.gitorious.org/pyside/shiboken/trees/master
>>
>> cheers - chris
>>
>> [cross-posting to the pyside list]
>>
>> -- 
>> Christian Tismer             :^)   <mailto:tismer at stackless.com>
>> Software Consulting          :     Have a break! Take a ride on Python's
>> Karl-Liebknecht-Str. 121     :    *Starship* http://starship.python.net/
>> 14482 Potsdam                :     PGP key -> http://pgp.uni-mainz.de
>> phone +49 173 24 18 776  fax +49 (30) 700143-0023
>> PGP 0x57F3BF04       9064 F4E1 D754 C2FF 1619  305B C09C 5A3B 57F3 BF04
>>        whom do you want to sponsor today?   http://www.stackless.com/
>>
>> _______________________________________________
>> PySide mailing list
>> PySide at qt-project.org
>> http://lists.qt-project.org/mailman/listinfo/pyside
> _______________________________________________
> PySide mailing list
> PySide at qt-project.org
> http://lists.qt-project.org/mailman/listinfo/pyside


-- 
Christian Tismer             :^)   <mailto:tismer at stackless.com>
Software Consulting          :     Have a break! Take a ride on Python's
Karl-Liebknecht-Str. 121     :    *Starship* http://starship.python.net/
14482 Potsdam                :     PGP key -> http://pgp.uni-mainz.de
phone +49 173 24 18 776  fax +49 (30) 700143-0023
PGP 0x57F3BF04       9064 F4E1 D754 C2FF 1619  305B C09C 5A3B 57F3 BF04
       whom do you want to sponsor today?   http://www.stackless.com/

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.stackless.com/pipermail/stackless/attachments/20121208/238452e5/attachment.html>


More information about the Stackless mailing list