Re: [boost] [Fit] formal review starts today

5 Mar 2016


      AMDG

On 03/05/2016 01:36 AM, Vicente J. Botet Escriba wrote:
...
Le 05/03/2016 01:11, Paul Fultz II a écrit :
...
On Friday, March 4, 2016 at 5:19:52 PM UTC-6, Steven Watanabe wrote:
...
On 03/04/2016 11:55 AM, Vicente J. Botet Escriba wrote:
...
#include 
struct identity {
     template<class T>
     T operator()(T t) { return t; }
};
int main() {
     boost::fit::conditional(identity(), identity())(0);
}
1> boost\fit\conditional.hpp(84): error C2500:
'boost::fit::detail::conditional_kernel': 'identity' is already a
direct base class
1>          with
1>          [
1>              F1=identity,
1>              F2=identity
1>          ]
1>
This is by design. I should document this. It makes absolutely no
sense to
pass the same function to conditional, and is most likely a mistake in
the
users code.
I disagree that this makes no sense.  It
would be true if we were talking about match,
but for conditional, this degenerate case has
a clear meaning, since conditional chooses
the first possible function.  Forbidding it
creates a special case, and special cases
are bad for generic code.  Of course, my
example is a bit silly, because I was
trying to create a minimal reproduction,
but it's quite possible to have a situation
where you don't know and don't care whether
the arguments are the same type.
(warning: untested code)

template
T my_accumulate(const S& s, T val, F f)
{
    fusion::for_each(s,
      phoenix::ref(val) += phoenix::bind(conditional(f, identity), _1));
    return val;
}

my_accumulate(fusion::make_vector(1, 2, 3), 0, identity);

In Christ,
Steven Watanabe