Re: [boost] [bimap][multi_index][range] sub_range vs std::pair<iter, iter>

From: JOAQUIN LOPEZ MU?Z (joaquin_at_[hidden])
Date: 2007-05-23 17:54:20

----- Mensaje original -----
De: Matias Capeletto <matias.capeletto_at_[hidden]>
Fecha: Miércoles, Mayo 23, 2007 11:20 pm
Asunto: [boost] [bimap][multi_index][range] sub_range vs 
std::pair<iter, iter>
Para: "boost_at_[hidden]" <boost_at_[hidden]>

> Hi,
> 
> Boost.MultiIndex and Boost.Bimap implements a function named 
> range. It is designed to integrate well with the
> new Boost.Range framework.
> 
> The actual signature is:
> 
> template< class LowerBound, class UpperBound >
> std::pair<iterator,iterator> range(
>   LowerBound lower, UpperBound upper);
> template< class LowerBound, class UpperBound >
> std::pair<const_iterator, const_iterator> range(LowerBound lower,
>   UpperBound upper) const;
> 
> In bimap docs it is stated that the way to use it is:
> 
> typedef bimap<string,int> bm_type;
> bm_type bm;
> //...
> std::pair<bm_type::left_iterator, bm_type::left_iterator>r =
> bm.left.range(...,...);
> 
> I want to know if this is the best approach.
> Boost.Range defines two range classes (iterator_range and sub_range)
> http://www.boost.org/libs/range/doc/utility_class.html
> It is said that they are better abstraction for ranges that 
> std::pair<iter,iter>
>
> 1) Do you think that the range functions should return a 
> sub_range<map_type>instead of a std::pair<iter,iter>?
[...]

IMHO, no, so as to avoid unasked for dependencies between
libraries --both at a conceptual and #include level.
Moreover, one can always convert the std::pair to a range
very easily, although this conversion would be much
simpler if iterator_range and sub_range provided ctors
taking std::pairs of iterators --Thorsten, would you
consider adding this a good idea?

> 2) In any case do you think that including typedef for the returned
> range type will make code more readable?
> The idea is to define:
> 
> typedef -range_type- sub_range; // the name can be different
> typedef -const_range_type- const_sub_range;
> 
> So the user code will look like this:
> 
> typedef bimap<string,int> bm_type;
> bm_type bm;
> //...
> bm_type::left_sub_range r = bm.left.range(...,...);

I am not sure. On one hand, the syntax above is obviously
more concise, but on the other hand a casual reader would
have to look up what sub_range is to fully understand the
code, whereas the std::pair is self explanatory and moreover
it's honored by the tradition of equal_range().

Just my 2c,

Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo