Semantic problem in articulation_points

[hoehrmann]

Boost BGL reports vertices 4 and 7 as articulation points in this graph:

graph "" {
0 -- 1;
0 -- 2;
0 -- 3;
0 -- 7;
1 -- 1;
1 -- 2;
1 -- 3;
1 -- 7;
2 -- 2;
2 -- 3;
3 -- 4;
3 -- 5;
3 -- 6;
7 -- 10;
4 -- 4;
4 -- 5;
4 -- 6;
5 -- 6;
5 -- 8;
6 -- 9;
8 -- 9;
}

Python's networkx library reports 3 and 7:

>>> list(networkx.articulation_points(networkx.Graph([(0, 1),
(0, 2), (0, 3), (0, 7), (1, 1), (1, 2), (1, 3), (1, 7), (2, 2), (2, 3), (3, 4),
(3, 5), (3, 6), (7, 10), (4, 4), (4, 5), (4, 6), (5, 6), (5, 8), (6, 9), (8, 9)])))
[3, 7]

Looking at the graph it seems clear that removing vertex 4 does not increase the number of connected components, contrary to the definition of an articulation point (»Articulation points are vertices whose removal would increase the number of connected components in the graph.« in the Boost BGL documentation).

[hoehrmann]

My bad.