[Stackless] PySide problem, take #2: typeobject clash

Christian Tismer tismer at stackless.com
Sat Dec 8 22:44:56 CET 2012


Hi again,

after some more investigation, I think I found it.
See thread [PySide violates Python API - PyQt does not!]
The first message was not correct, but the second one very likely is.

As said there, the PyHeapTypeObject is extended like so:

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

This additional pointer gets aliased to the first 4 bytes (win 32 bit)
of

typedef struct _slp_methodflags {
     signed char tp_itemsize;
     signed char tp_weaklistoffset;
     signed char tp_dictoffset;
     signed char nb_add;
     signed char nb_subtract;
     signed char nb_multiply;
     signed char nb_divide;
     signed char nb_remainder;
...


The first three seem irrelevant. But the nb_add might influence the way
how __add__ is interpreted. Due to the pointer value, stackless tries
to call __add__ using the soft-switching protocol, which then fails
miserably.

The macro that tries this is:

#define STACKLESS_PROMOTE_METHOD(obj, meth) \
     (obj->ob_type->tp_flags & Py_TPFLAGS_HAVE_STACKLESS_EXTENSION ? \
     slp_try_stackless = stackless & obj->ob_type->slpflags.meth : 0)

and it is used by

#define STACKLESS_PROPOSE_METHOD(obj, meth)                         \
{                                                               \
     int stackless = STACKLESS_POSSIBLE();                       \
     STACKLESS_PROMOTE_METHOD(obj, meth);                        \
     }

Unfortunately, I could not find out if there is a way for Stackless
to fall into this trap. But a good thing would be a little test:

Lars, Richard:
Can you please try your crashing examples, by putting
> import stackless
> stackless.enable_softswitch(False)

right after the program start?

That makes sure that stackless support of the interpreter is completely
disabled, so the misinterpreted frags are of no influence.

If that leads to significantly less crashes, then there is a quick fix
for PySide:
Insert a dummy void* into object.h before slpflags:

>     PyObject *ht_name, *ht_slots;
>     slp_methodflags slpflags;
> #endif
> } PyTypeObject;

Hope this helps, but I'm not confident -- Chris

-- 
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/




More information about the Stackless mailing list