Inconsistence between out_edges and edge_index

[malbarbo]

Let's consider this example

julia> g = simple_graph(2, is_directed=false)
Undirected Graph (2 vertices, 0 edges)

julia> add_edge!(g, 1, 2)
edge [1]: 1 -- 2

julia> out_edges(1, g)[1] == out_edges(2, g)[1]
false

julia> edge_index(out_edges(1, g)[1], g) == edge_index(out_edges(2, g)[1], g)
true

Conceptually the out edges in the example should be equal (g is undirected). It's inconsistent and confusing the index of different edges to be equal, modifying a map using the index of different edges could change the same value.

Networkx gives the same result...

[sbromberger]

I think the existing behavior is reasonable, since the edges are technically different. That is, out_edges(1,g)[1] is the edge 1->2, and out_edges(2,g)[1] is the edge 2->1. The only way you could determine equality is if == was supplied enough information to determine whether the edges were part of a directed graph (in which case 1->2 != 2->1), or undirected (in which case an argument could be made that 1->2 == 2->1). Absent that directedness information, there's no way for == to tell.

I also note that out_edges() is undefined for undirected graphs in NetworkX.

[malbarbo]

A simple solution is to compare the index of the edges. This does not create a issue with directed graph, because the edges 1->2 and 2->1 will have different indexes.

Using edges method of undirect graphs in Networkx gives the same result. Interesting in graph-tool (https://graph-tool.skewed.de/) the edges are equals.

[malbarbo]

Boost got it right. See Undirect Graphs http://www.boost.org/doc/libs/1_59_0/libs/graph/doc/graph_concepts.html, out_edges http://www.boost.org/doc/libs/1_59_0/libs/graph/doc/IncidenceGraph.html, and in_edges http://www.boost.org/doc/libs/1_59_0/libs/graph/doc/BidirectionalGraph.html

[sbromberger]

A simple solution is to compare the index of the edges.

I've been struggling with this because in my implementation edges don't have indices (they're in a Set()). The other issue is that, in my case, edges are immutables and independent of graphs, so there's no way to define equality for undirected edges (that is, Edge(x,y) == Edge(y,x)) without relying on some information (directedness) contained within the graph type.

The concept of directedness of edges therefore is a new construct and introduces more complexity than benefit at this time in my case.

Interesting discussion though!