[Pyrex] access to c data class members

[Daniele Varrazzo]

ahalda at cs.mcgill.ca wrote:
> Hi,
> 
> I am having trouble accessing c-data in a pyrex class. My class has a pointer
> to some c data (in a struct). It looks like some people have recently
> mentioned similar problems on the list, but I wasn't clear on the answer.
> 
> The goal is to have a non-member function modify the data and return a
> (new) python object. The tricky thing is that the data does not map to a
> python type. (It works otherwise)
> 
> EG:
> ------------------------------------
> cdef class MyClass:
> 	cdef OBJSTRUCT *data      #pointer to a struct
> 
> def load_File():
> 	newObj = MyClass()               #alternately, 'cdef Module newMod'
> 	newObj.data = c_mod.getData() #returns pointer to the new object
> 	return newObj
> ------------------------------------

You can declare newObj being of the exact class you expect:

     def load_File():
         cdef MyClass newObj
         newObj = MyClass()
         newObj.data = getData()
         return newObj

so newObj is C-typed and you can access its C members.

Alternatively you can cast the python instance on the fly when you need 
to access C members:

     def load_File():
         newObj = MyClass()
         (<MyClass>newObj).data = getData()
         return newObj

The first method involves a __Pyx_TypeTest call: if you try to set to 
newObj anything different from a MyClass instance, a TypeError is 
elegantly raised.

The second methods doesn't involve such check: it is faster when you are 
_really_ sure your object is the instance you expect (as in your 
example, unless MyClass has some funky metaclass). If you are not sure 
(e.g. you are getting a Python object from outside) it is basically 
asking for a core dump.

Disclaimer: this is what i learned reading the C code generated by 
Pyrex: if there is a serious reason to avoid such code, i'd be glad to 
know it.

Daniele