Re: [boost] [Function] Function pointer-like syntax introduced on CVStrunk

From: Douglas Gregor (gregod_at_[hidden])
Date: 2002-10-08 09:11:29

On Tuesday 08 October 2002 07:55 am, Peter Dimov wrote:
> From: "Douglas Gregor" <gregod_at_[hidden]>
> >   function<int (int x, int y)> f;
> >   f = std::plus<int>();
> >   assert(f != 0); // same as assert(!f.empty())
> >   f = 0; // same as f.clear()
> >   assert(0 == f); // same as assert(f.empty());
>
> What is the rationale for the change?

It gives boost::function nearly identical syntax to function pointers, making 
the two interchangeable and smoothing out the learning curve a bit. The 
OO-style '.clear()' and '.empty()' always felt like syntactic warts in a 
class that has such a trivial set of operator overloads. I had originally 
used nil_t/null_t instead of clear(), but that met with a bit more opposition 
than I wanted to deal with :)

> Actually I have a regression to report. Since function::operator= takes a
> non-const operand, this code now fails:
>
> #include <boost/function.hpp>
>
> typedef boost::function< void * (void * reader) > reader_type;
> typedef std::pair<int, reader_type> mapped_type;
>
> int main()
> {
>     mapped_type m;
>     m = mapped_type();
> }
>
> as std::pair's implicit assignment operator takes a non-const reference,
> too.

Yeah, that's a problem. Thanks for the testcase. The really annoying part is 
that if one writes out the trivial copy assignment operator

pair& operator=(const pair& other)
{
  this->first = other.first;
  this->second = other.second;
  return *this;
}

everything compiles and works properly. Perhaps 12.8/5 is just too strict, and 
should be considered a defect?

In any case, back to the whiteboard...

        Doug