On Thu, 9 Aug 2007, Joseph Gauterin wrote:
Would anyone be interested in having a couple of extra specialized iterator adaptors in Boost.Iterator?
I'm thinking of 'key_iterator' and 'value_iterator' adaptors (along with convenience 'make_key_iterator' and 'make_value_iterator' functions) that allow any associative container (that follows standard library idioms) to be accessed as either a range of keys or as a range of values.
They'd be used like this:
std::map<std::string,int> m; std::for_each(make_value_iterator(m.begin()),make_value_iterator(m.end()),func);
Where func() takes an int parameter rather than a std::pair<const std::string,int>.
Using these adaptors is considerably easier than using boost.lambda for this purpose and they can be implemented to incur no (or negligable) runtime overhead when compared to iterating normally through an associative container.
I find these useful - does anyone else?
Yes, I agree they are. But how about a generalization to tuples? Let's call it a get_iterator (any ideas for a better name?), i.e. template < typename BaseIterator , int NthElement > class get_iterator ... ; and define make_get_iterator accordingly and then, following your example: std::map< std::string , int > m ; std::for_each( make_get_iterator< 1 >( m.begin() ) , make_get_iterator< 1 >( m.end() ) , func ) ; That would require to either specialize boost::get for std::pair (easier; wasn't a similar specialization done with tr1::get?) or specialize that iterator for std::pairs (more code). And then, for convenience and/or readability, it could be possible to wrap make_get_iterator as make_key_iterator and make_value_iterator. I already have an implementation of this get_iterator (with all specializations for std::pair and boost::compressed_pair), if anyone is interested. -- François Duranleau LIGUM, Université de Montréal