The Thermopylae excerpt of TDPL available online

[Robert Clipsham]

Don wrote:
>  I'm starting to think we need a
>> separate namespace for the CT stuff.
>>
>>   D.compiles(XXX)
>>   D.typeof(foo)
>>   D.stringof(T)
>>   D.allMembers(T)
> 
> That's not bad. Can't be 'D', though, has to look like a keyword. Maybe 
> something like 'traits' instead. In exchange, get rid of the '__traits' 
> and 'typeid' keywords. Not sure about typeof, though.
> 
> traits.compiles(XXX)
> traits.typeid(TTT)
> traits.stringof(T)
> traits.allMembers(T)
> traits.error("message");
> 
> IMHO this looks better than __traits(compiles, XXX) but it actually has 
> the same flexibility, and it's a straightforward transformation inside 
> the compiler.
> For bonus points, allow 'with(traits)':
> 
> with(traits) {
>   if (compiles(XXX)) return stringof(T);
>   else error("Can't use " ~ stringof(T) ~ " in a transmogrifier.");
> }
> 

This could even be implemented in phobos (or object.d to avoid an import 
when you want to use it), using something like:

pragma(traits) struct traits
{
   // Members automatically added by the compiler, or maybe just their
   // declarations eg:
   static T delegate()[] allMembers(T)(T type);
}

This way there's no need to force a certain name, as it can be chosen by 
the developer (by either using renaming imports or just giving the 
struct a different name eg pragma(traits) struct meta {}).