Re: [boost] Overloading a function taking a boost::function

From: Douglas Gregor (gregod_at_[hidden])
Date: 2002-05-11 13:54:53

On Saturday 11 May 2002 12:22 pm, you wrote:
> Wouldn't make_function work for std::unary_function, std::binary_function
> and
> boost::function too? But yes, handling arbitrary function objects would be
> difficult.

Yes, it could be made to work with unary_function, binary_function, and 
boost::function.

> Well I guess I'll write make_function myself then. It won't work for any
> function
> object but I can live with that.

That's probably the best option. Sorry :(

> Just one thing though, I'd like it to work for the result of a boost::bind
> as well.
> But the result of a boost::bind is implementation-defined according to the
> documentation. So there is no way to portably create boost::bind overloads
> for
> make_function, right?

It wouldn't work anyway, because boost::bind function objects don't have 
exact argument types to deduce. For instance, how would the following work?

template<typename T>
struct add_to {
  typedef T result_type;
  template<typename U>
  T& operator()(T& t, const U& u) const 
  { 
    return t += u; 
  }
};

std::string& foo;
make_function(bind<std::string&>(add_to<std::string>(), ref(foo), _1));

The boost::bind function object takes at least 1 argument (any more would be 
ignored), but the type of that first argument isn't specified. Boost.Function 
doesn't (and can't) do partial bindings.

> > If we _did_ have a working set of function traits, it would be possible
> > to make Boost.Function objects implicitly constructed only from function
>
> objects
>
> > that are callable by the Boost.Function object.
>
> This seems to me as an awesome addition to Boost.Function! Too bad there
> are compilers out there that don't support the needed functionality. :(
>
> Dirk Gerrits

There are workarounds for most compiler deficiencies, but of course we won't 
know until someone tries. I'll work on it sometime.

        Doug