[Python-3000] Ctypes as cross-interpreter C calling interface

[Guido van Rossum]

On 8/10/06, Paul Prescod <paul at prescod.net> wrote:
> > I don't know enough about ctypes, but assuming I have a reason to
> > write an extension in C (e.g. Tkinter, which uses the Tcl/Tk API), how
> > to I use ctypes to call things like PyDict_GetItem() or
> > PyErr_SetString()?
>
> There are two answers to your question. The simplest is that if you have a
> dict object called "foo" you just call 'foo["abc"]'. It's just Python. Same
> for the other one: you'd just call 'raise'.

That doesn't make sense if you want to write your extension in C.
Surely you don't propose to rewrite all of tkinter.c in Python? That
would be insane. Or Numeric? That would kill performance.

> Ctypes is the opposite model of the standard extension stuff. You're writing
> in Python so Python stuff is straightforward (just Python) and C stuff is a
> bit weird. So if you had to populate a Python dictionary from a C struct
> then it is the reading from the C struct that takes a bit of doing. The
> writing the Python dictionary is straightforward.
>
> If there was a reason to call PyDict_GetItem directly (performance maybe???)
> then that's possible. You need to set up the function prototype (which you
> would probably do in a helper library) and then you just call
> PyDict_GetItem. CTypes would coerce the types. py_object is a native data
> type.
>
> So I think it ends up looking like
>
> from PythonConvenienceFunctions import PyDict_GetItem
>
> obj = {}
> key = "Guido"
>
> rc = PyDict_GetItem(obj, key)
>
> I'm sure an expert will correct me if I'm wrong...

I guess I object against the idea that we have to write all extensions
in Python using ctypes for all C calls. This is okay if there's
relatively little interaction with C code. It's insane if you're doing
serious C code. And what about C++ extensions?

-- 
--Guido van Rossum (home page: http://www.python.org/~guido/)