> we could have only **one** initialization for the whole lifespan of the program. If
> I get it right, their proposal would, prior to a search, to "re-set" the
> values from distances vector and predecessors vector only for the reached
> vertices of the last search. I think it worth a try because, from what I
> timed the python side of initialization could cost around 18ms (5ms for
> distance map, 13ms for predecessor map) see timeit snipet below.
In order for this to work, the function would need to keep and return a list
of reached vertices. This is is not difficult to do, but I wonder how muchit would actually improve...