[Issue 1649] Variant coercion fails with delegates

[Christopher Wright]

Brad Roberts wrote:
> http://www.dsource.org/projects/phobos/browser/candidate/phobos/std/variant.d?rev=511
> 
> Why static assert(false) rather than return false? 

Coerce doesn't allow that. Besides which, it's better to fail early -- 
if Variant doesn't support that operation, it shouldn't compile, since 
it'll ALWAYS fail at runtime. And if the user wants something like that, 
they'll just put assert(false) in their code.

> Additionally, if 
> you're going to static assert(false), then a message ought to be included 
> since it's going to result in a nice meaningless compilation failure.
> 
> Additionally.. add a code block -> add a unit test!

We can actually do that now, using __traits(compiles).

unittest {
    Variant v = 5;
    assert (!__traits(compiles, v.coerce!(bool delegate())));
}

> Lastly, stop doing trivial bug fixes on the candidate branch.  They belong 
> directly on the trunk.