Re: [boost] Re: Lambda/bind interface.

From: Peter Dimov (pdimov_at_[hidden])
Date: 2001-03-19 11:54:21

From: "Gary Powell" <Gary.Powell_at_[hidden]>

> > bind(f); // gives a library-compatible function object //
> > case 1
> >
> > bind(&Class::member); // ditto, works with Class &, Class *,
> > shared_ptr<Class>, variant<Class>, etc as a first argument.  // case 2
> >
> > bind(f, _2, _1); // lambda(x, y, ...) := f(y, x);   // case 3
> >
> > bind(f, 4, _, 5); // lambda(x, y, ...) := f(4, x, 5);  // case 4
> >
> > bind(_, 1, 2); // labmda(x, ...) := x(1, 2);  // case 5
> >
> >  bind.hpp, however, is doable
> > (in theory.)
> >
> > I think that we should propose a common interface, to give people a
chance
> > to implement it for the platform they are interested in.
> <<
> I think I can speak for Jaakko and I here, and that is yes a common
> interface would be fine. That's part of the reason we broke the LL library
> into the various sections.
>
> I think the other case you are missing is
> bind(&className::MemberFn, classReference/Pointer, arguments);
>
> as in bind(&data::foo, f, 1, 2) ==> f->*foo(1,2);  // case 6

Yes, I've thought of that as well, just didn't include it in the list.

> The only other case we don't handle this way is case #2. I'm not sure I
> understand what you intend the result of this call to bind to be.

std::for_each(v.begin(), v.end(), bind(&X::f));

bind(&X::f) is basically equivalent to bind(&X::f, _, _, ...) (insert the
number of arguments to X::f here) - it is a direct replacement to mem_fun
and mem_fun_ref.

> And that the naming convention for the "free" args that the two libraries
> use is different. (LL has a typedefs for the free args so this isn't much
of
> an issue for LL.)

_1 is my new favorite. :-) I think that it's much better than free1 or arg1.
The single _ (without a number) is very convenient, too.

--
Peter Dimov
Multi Media Ltd.