Re: [boost] generic script language binding

From: Sascha Krissler (boost-dev_at_[hidden])
Date: 2006-12-20 08:13:34

On Tue, Dec 19, 2006 at 03:04:38PM -0500, David Abrahams wrote:
> Sascha Krissler <boost-dev_at_[hidden]> writes:
> 
> > On Mon, Dec 18, 2006 at 04:45:18PM -0500, David Abrahams wrote:
> >> > boost library: a generic dispatcher for foreign languages and C++ itself.
> >> > Right now i am finding out whether one can write an awkward-syntax-predicate-
> >> > dispatcher in c++.
> >> 
> >> Huh?  What's that?
> >> 
> >
> > in the same sense not perfect like boost::lambda is not becuase of the syntactic
> > restrictions of c++. so somewhere in the middle between unusable and
> > without syntactic clutter
> 
> I think I'm wondering less about the awkward-syntax than about the
> "predicate-dispatcher" part.
> 
when you regsiter a candidate method to the multimethod you can optionally
pass a function returning bool and the candidate method is only called
when the function returns true.
This is a generalization of normal multiple dispatch and emulates the latter
with:
bool fun(argn, ...) { return dynamic_cast<T1>(arg1) && dynamic_cast<T2>(arg2); }

the nontrivial part is the ordering of the predicates so that
arg1->property1() && arg2->property2()
is tested before
arg1->property1()
because the former is more specific.

i was thinking about passing the predicate as a lambda expression and then
work on it with a compile time algorithm that operates on the parse tree
of the expression template.