Note: what's required for direct betweenness is cutoff versions of igraph_get_all_shortest_paths(_dijkstra)()

This brings up an interesting question:

Should the comparison with the cutoff be done with tolerances, even in the current functions? Do we get paths which are < cutoff or <= cutoff? It this documented for existing functions? Are there cases where they return an unexpected result due to roundoff errors?

This brings up an interesting question:

Should the comparison with the cutoff be done with tolerances, even in the current functions? Do we get paths which are < cutoff or <= cutoff? It this documented for existing functions? Are there cases where they return an unexpected result due to roundoff errors?

😬 There's probably going to be issues without tolerances, but I haven't seen them yet.

I see IGRAPH_SHORTEST_PATH_EPSILON is used for comparisons.

I think you want inclusive? Like if you say the cutoff is 3.0, then you want to include paths with length 3.0? To me that seems more usual than wanting everything below 3.0.

Please skip the tolerances for now. I'll write later about why (on my phone).

I think elsewhere we used <= cutoff, so let's use that here too for consistency