"Marcin 'Qrczak' Kowalczyk" <qrczak@knm.org.pl> writes:

> > The problem is that C library is binary-only (Intel image processing
> > library), and CallBack type is exactly 
> > 
> > typedef int (*CallBack)(int);
> 
> Converting a function closure to a C function pointer can't be done
> portably, but it can be done with lots of ugly magic.
> 
> Glasgow Haskell does this (by generating a piece of assembler on
> the heap) and it's convenient to use from the level of Haskell.
> Such functions need to be explicitly freed of course.
> 
> GNU C does this for local functions, but only "downwards". If the
> function pointer doesn't need to live longer than the function which
> installs the callback, the GNU C extension can be used. It generates
> the piece of assembler on the stack. To use it - just define a function
> inside a function and take its address.

Hmm, this trick assumes that the stack is executable (IMHO, the serious
security hole). Don't know about Linux/x86 page/segment attributes policy,
but I am almost sure that Windows NT/2000 x86 (which is my platform) will
not allow to execute the code in data/stack pages.

> It would be nice if OCaml provided this functionality because not
> all C libraries provide the extra argument for simulating closures.
> Unfortunately it can't be implemented nicely.
> 
> I've once seen a C library which tries to provide it for several
> platforms but I forgot its name.

It would be nice if you recall it :-)

Hope to hear from you soon,
Dmitry


-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr