RE: [boost] Submission: typelist

From: Aleksey Gurtovoy (alexy_at_[hidden])
Date: 2001-11-28 11:09:42

David Abrahams wrote:
> > template<typename SequenceTag>
> > struct at_algorithm_traits
> > {
> >     template<long N, typename Sequence> struct algorithm
> >     {
> >      private:
> >         typedef typename mpl::begin<Sequence>::iterator first_;
> >         typedef typename mpl::for_loop<
> >               mpl::int_t<0>
> >             , mpl::lt<N>
> >             , mpl::next<mpl::_1>
> >             , first_
> >             , mpl::compose_f_gxy<
> >                     mpl::next<mpl::placeholder>
> >                   , mpl::project1st<mpl::_1,mpl::_2>
> >                   >
> >             > loop_;
> >
> >      public:
> >         typedef typename loop_::executed::state type;
> >     };
> > };
> >
> > template<long N, typename Sequence>
> > struct at
> >     : at_algorithm_traits<
> >           typename mpl::sequence_traits<Sequence>::sequence_category
> >           >::template algorithm<N, Sequence>::type
> > {
> > };
> 
> To be fair, Andrei, you should compare a version of Loki 
> typelists that works without partial specialization.
> 
> I had hoped that when Aleksey and I "discovered" that 
> map/accumulate seemed to be /the/ fundamental iteration 
> algorithm, things like the above would have been rewritten 
> to use it. Maybe Aleksey can explain why he's still
> using for_loop<>. 

Just for a record - the for_loop above is accumulate-like, i.e. it separates
state from the behavior (function). But anyway, the above is just a
_default_ implementation of 'at' algorithm; it's supposed to work with _any_
compile-time sequence that provides begin<>/end<> interface. It's not
supposed to be a default implementation of 'at' algorithm for type_list, or
type_vector, or any other known for mpl types, for which one can write much
more efficient/simple code. The only reason MPL doesn't provide these
specializations is because I was lazy.

> It may be that it's just in an intermediate 
> state, with for_loops not yet replaced by an alternative.
> 
> In any case, I believe that primitive algorithms such as size/length
> probably make the most sense when written recursively, to 
> reduce the burden on the compiler's template instantiation 
> mechanism, if nothing else!

Sure, that was the intent from the beginning.

Aleksey