why is this cast necessary?

[Steven Schveighoffer]

On Mon, 07 Jun 2010 23:02:48 -0400, Graham Fawcett <fawcett at uwindsor.ca>  
wrote:

> Hi folks,
>
> This program works as expected in D2:
>
>     import std.stdio;
>     import std.algorithm;
>
>     T largestSubelement(T)(T[][] lol) {
>       alias reduce!"a>b?a:b" max;
>       return cast(T) max(map!max(lol));   // the cast matters...
>     }
>
>     void main() {
>       auto a = [[1,2,3],[4,5,6],[8,9,7]];
>       assert (largestSubelement(a) == 9);
>
>       auto b = ["howdy", "pardner"];
>       assert (largestSubelement(b) == 'y');
>
>       auto c = [[1u, 3u, 45u, 2u], [29u, 1u]];
>       assert (largestSubelement(c) == 45u);
>     }
>
> But if I leave out the 'cast(T)' in line 7, then this program will not
> compile:
>
> lse.d(6): Error: cannot implicitly convert expression
> 		  (reduce(map(lol))) of type dchar to immutable(char)
> lse.d(14): Error: template instance
> 		   lse.largestSubelement!(immutable(char)) error
> instantiating
>
> Where did the 'dchar' came from? And why does the cast resolve the issue?

In a recent update, Andrei changed char[] and wchar[] to bi-directional  
ranges of dchar instead of straight arrays (at least, I think that was the  
change) in the eyes of the range types.  I think this is where the dchar  
comes from.

If you had a char[], and the 'max' element was a sequence of 2 code  
points, how do you return a single char for that result?

-Steve