Bug fix for map iterators

[Peter Kovacs]

The attached patch [e9c203fb003d] contains bug fixes for the map extenders. The reference members are replaced with pointers to handle the cases when they have to be initialized to NULL.

It is on the top of [c6acc34f98dc], it can be merged into all branches.

[Alpar Juttner]

When do they have to be initialized to NULL? (Wouldn't it be appropriate to add a test for it?)

[Peter Kovacs]

Replying to alpar:

When do they have to be initialized to NULL? (Wouldn't it be appropriate to add a test for it?)

The main reason for these fixes is to replace references with pointers. If we use pointers, then it is not so important to initialize them to NULL, however, the non-initialized references cause compiler errors.

A use case for this is mentioned in #320. Let's suppose you would like to apply STL algorithms (that operate on ranges) on standard graph maps using their MapIt or ConstMapIt iterators. E.g.

  typedef SmartDigraph::NodeMap<int>::ConstMapIt MapIt;
  Map map(g);
  for_each(MapIt(map), MapIt(INVALID), action_functor);

Currently, we have constructor for MapIt(INVALID), but it is faulty, so this code does not compile.

Similar tests are included in maps_test.cc in [8ddb7deabab9] (see #320), but they are disabled and waiting for these bug fixes.

[Alpar Juttner]

Replying to kpeter:

It is on the top of [c6acc34f98dc], it can be merged into all branches.

Done, see [2e9655402778], [994c7df296c9], [76689f2fc02d] and [1b89e29c9fc7].