Re: [boost] Problem with random.hpp

From: Jens Maurer (Jens.Maurer_at_[hidden])
Date: 2001-02-25 15:52:31

That's my code, so I should probably say something.

Daryle Walker wrote:
> The problem is in a template parameter description in "random.hpp", lines
> 383 through 391:

[shortened to remove compiler-workaround]

> > // L'Ecuyer 1988
> > template<class MLCG1, class MLCG2, typename MLCG1::result_type val>
> > class additive_combine
> 
> The rest of the class template continues from there.  The problem is that my
> compiler chokes on the fourth line ("typename MLCG1::result_type") here,
> claiming something like "illegal name overloading."
> 
> I guess the problem is an ambiguity between the types of template
> parameters:
> 
> 1. A type
>     A. "class" [formal_parameter_name ["=" default_value]]
>     B. "typename" [formal_parameter_name ["=" default_value]]

> 3.  A value
>     A. type_name  [formal_parameter_name ["=" default_value]]
> 
> When using a type name that is within a class template, you have to use the
> "typename outer_type::inner_type" syntax.  That is what the writer did here.
> However, this case of 3A also looks like 1B!  The compiler is choosing 1B,
> then choking because the formal_parameter isn't a single identifier.

That's exactly the point: The formal_parameter is not a single identifier,
thus it cannot be case 1B.  If your compiler is unable to properly
determine this, use the work-around #define.  What compiler are you using,
btw.?

> The ironic part is that I'm not using this class, but another from the same
> header.  Is this an ambiguity that the original writer's compiler missed, or
> is my compiler wrong?  

I believe your compiler has enough information to handle this case, if
it chose to make full use of said information.

Jens Maurer