[Pyrex] Access to nested structures and arrays

[Pierre GM]

On Friday 16 March 2007 16:19:29 Robert Bradshaw wrote:
> > First question:
> > When writing extension types, am I limited to use only Python
> > objects as
> > arguments of __init__ ?
>
> Yes, as __init__ cannot be a cdef function (otherwise it wouldn't be
> callable via python).

OK, duely noted. 

> > Third question:
> > how can I modify the underlying C array from Python ?
>
> You can't. You have to write a python function that when called
> modifies the array.

Well, I realized that I don't really need a numpy array for that, a simple 
list should do the trick as I have a limited number of elements. So I wrote a 
small extension:

cdef class modelflags:
    cdef int *c_list[8]
    cdef object str_list
    def __new__(self):
        cdef int i
        self.str_list = [False] * 8
    def __getitem__(self, idx):
        return self.str_list[i]
    def __setitem__(self, idx, val):
        cdef int tmpval
        tmpval = val
        self.c_list[i] = &tmpval
        self.str_list[i] = bool(val)
    def __str__(self):
        return self.str_list

The idea was to overload the __setitem__, so that I could modify the 
underlying C array directly.

Elsewhere in the code, I initalize the pyrex extension type wrapped around one 
of the substructures, with a tailor-made setup function

cdef setup(self, c_struct  *base, long npar):
	...
        self.parametric_flags = modelflags()
        self.parametric_flags.c_list[0] = base.parametric[0]

where c_struct is  the underlying C structure declared in a .pxd

That compiles OK, but the Python code sends me a 
AttributeError: 'modelflags' object has no attribute c_list. 
when the setup function gets called.

My understanding was that c_list, even if private, could still be accessed by 
the pyrex code. Obviously I'm missing something here, but could you tell me 
what ?

> Perhaps if you think of things in terms of providing a python
> interface to manipulating the underlying c structure, rather than
> trying to expose the c structure itself, things would become easier
> to do.

Er, yes, but I have no idea how. The rest of the external code uses this 
nested structure as the key object, using some substructures for inputs, some 
others for outputs, and calling the structure at once for some 
initialization. I tried to get rid of the initialization part and cinstruct 
the structure from my extnsion types, but that doesn't work either... At 
least, with the current version I tried to describe you,  I'm able to convert 
the C outputs back to python, converts most of the inputs from python to C, 
but for this particular array that needs to be read/write.

Thanks again for your time