Re: [ctypes-users] Dynamic structures..

[Bob I. <bo...@re...>]

On Wednesday, July 30, 2003, at 2:44PM, John Sutherland wrote:

> Ok, I think I have -some- of it figured out... But not all..
>
> Here's some example code:
>
> from ctypes import *
>
> class TEST(Structure):
>     _fields_ = []
>
> TEST._fields_.append( ( "test1", c_char * 4 ) )
> TEST._fields_.append( ( "test2", c_char * 4 ) )

If you don't know what the fields in a Structure subclass are, simply 
DON'T DEFINE IT YET :P

If you need to create a class on the fly, do this:

from ctypes import *
MyStructure = type('MyStructure', (Structure,), {'_fields_': [('test1', 
c_char*4), ('test2', c_char*4)]})

is equivalent to:
class MyStructure(Structure):
	_fields_ =  [('test1', c_char*4), ('test2', c_char*4)]

Thomas, you should probably make _fields_ a read-only property (via 
metaclass or whatever), or at least an n-tuple or 2-tuples.

When I've written code that did something VERY similar to what 
Structure does, I had a metaclass that mangled "_fields_", created 
property descriptors for each member, converted _fields_ to a readonly 
descriptor (which was a tuple, so you couldn't change it in-line 
either) that was only useful for introspection purposes anyway since 
all of the property descriptors had all the information they needed 
prebound.

Essentially, Structure and Union should probably be python classes that 
just have a superclass of whatever buffer... so long as creating 
c_char, etc that point to existing memory becomes easy.

-bob