Add mapping dict as graph attribute in condensation.

[jtorrents]

I've found myself several times needing a mapping of original nodes to condensation nodes (ie strongly connected components). It is easy to compute that myself (especially since condensation function accepts precomputed SCC as a parameter). But then the function itself recalculates the mapping for building the condensation graph. Adding the mapping as graph attribute is simple, backwards compatible, and avoids redundant computations.

I've also updated condensation docstrings and added a test.

[ysitu]

Maybe some code maintenance can be piggybacked.

for i, component in enumerate(scc):
    mapping.update((n, i) for n in component)
number_of_components = i + 1
C.add_nodes_from(range(number_of_components))
C.add_edges_from((u, v) for u, v in G.edges_iter()
                 if mapping[u] != mapping[v])

[ysitu]

LGTM. But exposing the computed SCCs directly is also a convenience to the user:

for i, component in enumerate(scc):
    C.node[i]['members'] = list(scc)

[jtorrents]

@ysitu, I've added the node attribute members with the list of nodes in the SCC. And also improved the condensation function as you suggested. Maybe having both mapping as a graph attribute and members as node attribute is a bit redundant. I don't think is a big problem, but if we have to keep only one, I'd prefer to keep the node attribute because, as you said, it's a nice way to expose directly the computed SCCs.