[Python-Dev] Py_TPFLAGS_HEAPTYPE, what's its real meaning ?

[Boris Boutillier]

No answers on this ? I posted the question two times on c.l.py and got 
no answers., help would be appreciated.

Boris


Boris Boutillier wrote:

> Hi all,
>
> I've posted this question to the main python list, but got no answers, 
> and I didn't see the issue arose on Python-dev (but I subscribed only 
> two weeks ago).
> It concerns problems with the Py_TPFLAGS_HEAPTYPE and the new 
> 'hackcheck' in python 2.3.
>
> I'm writing a C-extension module for python 2.3.
> I need to declare a new class, MyClass.
> For this class I want two things :
> 1) redefine the setattr function on objects of this class
>   (ie setting a new tp_setattro)
> 2) I want that the python user can change attributes on MyClass (the
> class itself).
>
> Now I have a conflict on the Py_TPFLAGS_HEAPTYPE with new Python 2.3.
> If I have Py_TPFLAGS_HEAPTYPE set on MyClass, I'll have problem with the
> new hackcheck (Object/typeobject.c:3631), as I am a HEAPTYPE but I also
> redefine tp_setattro.
> If I don't have Py_TPFLAGS_HEAPTYPE, the user can't set new attributes on
> my class because of a check in type_setattro (Object/typeobject.c:2047).
>
> The only solution I've got without modifying python source is to 
> create a specific Metaclass for Myclass, and write the tp_setattr.
> But I don't like the idea of making a copy-paste of the type_setattr 
> source code, just to remove a check, this is not great for future 
> compatibility with python (at each revision of Python I have to check 
> if type_setattr has not change to copy-paste the changes).
> In fact I'm really wondering what's the real meaning of this flags, 
> but I think there is some history behind it.
>
> If you think this is not the right place for this question, just 
> ignore it, and sorry for disturbance.
>
> Boris
>
>
> _______________________________________________
> Python-Dev mailing list
> Python-Dev at python.org
> http://mail.python.org/mailman/listinfo/python-dev
> Unsubscribe: 
> http://mail.python.org/mailman/options/python-dev/boris.boutillier%40arteris.net 
>