This also includes proper test strings.

Now, if 'm' is larger than the seed graph, it is set to the size of the
seed graph, and is increased progressively until it matches the desired
value. This removes the need to start with a random seed graph of the
appropriate size.

This also includes modifications which make the code more robust
against parameter choices which don't make much sense, e.g. lead to
negative probabilites.

This makes the calculation coherent with the definition given in the
documentation.

This removes reverse_graph.hpp, reverse_graph.hpp and copy.hpp, and
introduces some modifications in graph_adaptor.hh. This modification
fixes compilation problems across different boost versions.

I'm looking at you, macports.

This removes a strict requirement on matplotlib in the "draw" module.

This moves the documentation out of the constructor, and into the class
definition.

This also changes the semantics of of bin list of size 2, and also
simplifies the histogram code slightly.

This allows easy creation of unfiltered copies of filtered graphs.

Now if a GraphView object is deleted, any property map created for it
reverts its ownership to the base Graph object. This eliminates the
problem of property maps created with transient GraphView objects.

This also fully honors the PYTHON_VERSION environment variable.

This removes some 'undirected' parameters, in favour of graph views.

The default vertex shape is now a circle (not an ellipse), and the
default vertex/edge colors correspond to the Tango palette.

This correctly maintains the edge index metadata.

This fixes a bug where the edge index metadata was not properly copied.

This fixes a leftover bug from the removal of the 'core' submodule.