RE: [boost] Named template parameters

From: Aleksey Gurtovoy (alexy_at_[hidden])
Date: 2001-03-05 02:42:49

Jeremy Siek wrote:
> So I'm in favor of going with the following
> interface style (suggested by someone in an earlier email):
> 
>   typedef 
>     boost::iterator_adaptor_generator<int*, 
> boost::default_iterator_policies,
>       boost::iterator_adaptor_types
>       ::reference<int>
>       ::iterator_category<std::input_iterator_tag>::type
>     >::type Iter;
> 

IMO this is very nice!

One small concern, though: I would like to see
'boost::iterator_adaptor_types' to be named along the lines of
(non-existent) 'boost::iterator_traits_builder', for two reasons:

1) 'builder' suffix reminds one to write '::type' at the end of the '::'
chain; it's true that if you forget the compiler will let you know :), but
it's still a nice convention that worked well for me;
2) IMO 'iterator_traits' part gives you more explicit hint about naming of
the "parameters" than 'iterator_adaptor_types' does.

Disadvantage of the particular 'iterator_traits_builder' name is that it is
so generic that one can think that

typedef std::iterator<random_access_iterator_tag, int> Iter;

and

typedef boost::iterator_traits_builder
  ::iterator_category<random_access_iterator_tag>
  ::value_type<int>
  ::type
  Iter;

should give the same result (in terms of defined nested typedefs), which is,
of course, not true. However, on second thought, I don't think that such
assumption is valid at all; it would be if the class was named
'boost::std_iterator_builder' or something alike, but
'iterator_traits_builder' seems to be not bad after all..

> The trick with getting this approach right, as Dave pointed 
> out, is that the creation of defaults can only happen for 
> types that are suppose to be defaulted. I've attached a file 
> that does this. I've done a quick check to make sure it
> works for VC++ and g++. The file creates a 
> iterator_adaptor_generator class... it should be 
> straightforward to apply the same technique to 
> iterator_adaptor itself.

A small issue with the code - we don't need a semicolon after
BOOST_NAMED_TEMPLATE_PARAMETER macro invocation, and CW6.1 issues the
"illegal empty declaration" warning for it:

BOOST_NAMED_TEMPLATE_PARAMETER(value_type,
boost::detail::default_value_type); // not needed here

Besides this, it all looks great!

Aleksey