Re: Boost algorithm library

From: helmut.zeisel_at_[hidden]
Date: 2001-07-26 07:00:50

--- In boost_at_y..., "Philip Nash" <philip.nash_at_c...> wrote:
> .. adding to the debate of container interface vs range iterator 
interface:

Just to change the terminology a little:
IMHO, we have no container interface vs. range interface debate.
We have just 2 different possible range interfaces.

Interface 2 is the classical STL interface that
needs 2 interators (begin, end) to specify a range.

Interface 1 is only slightly differently
and needs 1 object (offering a begin() and an end() method)
to specify a range.
By default, every STL container already offers Interface 1.

As demonstrated in my code example posted previously,
it is easy to write a make_range function
that creates a range object from a pair
of iterators (begin, end).

Interface 2 has the advantage that it is used
in the standard C++ algorithms.
It has the disadvantage that there is no possibility
(or at least that I do not know one)
to switch to interface 1 without writing
a second wrapper algorithm for every existing standard algorithm.

Interface 1 needs only class range and function make_range
to offer an interface 2 for every client that
needs the different interface.

The disadvantages are that
a) it is not the usual standard interface, and
b) it is not clear to me whether make_range
   can work without perfomrmance loss.

Using interface 1 is shorter for
the default range begin(), end() in a container,
and longer for every other range,
compared with interface 2. 

The decision which interface to prefer
might depend on the most common
applications of the algorithm(s).

Helmut