[boost] Re: Policy or Trait?

From: Gennadiy Rozental (gennadiy.rozental_at_[hidden])
Date: 2004-04-01 20:00:38

"Mark Rodgers" <mark.rodgers_at_[hidden]> wrote in message
news:002601c41836$33148560$0100a8c0_at_cadenza.co.nz...
> Well I think in terms of code, and it really is quite simple IMHO:
>
> Traits look like this:
>
> template< typename T >
> class SomeTraits
> {
>     // Stuff that tells us about T
> };
>
> and Policies look like this
>
> template< typename Policy >
> class SomeThing
> {
>     // Delegates some behaviour to Policy
> };
>
> IOW, traits are class templates and policies are template parameters so
> there really can't be any confusion between the two.  So in the case
> std::basic_string
>
>    template<class charT, class traits = char_traits<charT>,
>             class Allocator = allocator<charT> >
>    class basic_string {
>
> "char_traits" is a traits class template but "traits" is actually a policy
> template parameter.

That's is namely the problem I have with basic_string design. Trais class is
used as policy. Actually it's even worse - char_traits contain partially
functionality that may belong to some kind of policy (for example comparison
method). IMO it should be separated.

> It is quite reasonable for specialisations of traits
> templates to be used as arguments to policy parameters, which is exactly
> what happens when we write
>
>    std::basic_string< char,std::char_traits<char> >
>
> But I don't think there is anything in basic_string that requires the
> template argument to be a specialisation of a class template.  I think it
> would be quite legitimate to write
>
>    class MyFunkyCharBehaviour { /*...*/ };
>    typedef std::basic_string<char,MyFunkyCharBehaviour> MyFunkyString;

Exactly, so we could use complete different class that will define different
trait values. And this is not good IMO.

> so the "traits" parameter is indeed misnamed.
>
> However std::char_traits is most definitely a traits template and doesn't
> *have* to be used as a policy.  In fact std::basic_string could have been
> written to exclusively use char_traits *instead* of a policy.

IMO it should've been.

> Mark
>

Gennadiy.