Incidence dictionary produces incorrect graph

[GearsAD]

If I create an undirected incidence dictionary with the following code, it produces incorrect edges.

using Graphs
const FGGdict = Graphs.GenericIncidenceList{Graphs.ExVertex,Graphs.Edge{Graphs.ExVertex},Dict{Int,Graphs.ExVertex},Dict{Int,Array{Graphs.Edge{Graphs.ExVertex},1}}}
fg = Graphs.incdict(Graphs.ExVertex,is_directed=false)
variables = [ExVertex(1, "x1"), ExVertex(2, "x2"), ExVertex(3, "x3"), ExVertex(4, "x4")]
add_vertex!(fg, variables[1])
add_vertex!(fg, variables[2])
add_vertex!(fg, variables[3])
add_vertex!(fg, variables[4])
# Factors
factors =[ExVertex(5, "x1x2f1"), ExVertex(6, "x2x3f1"), ExVertex(7, "x3x4f1")]
add_vertex!(fg, factors[1])
add_vertex!(fg, factors[2])
add_vertex!(fg, factors[3])
edge = Graphs.make_edge(fg, variables[1], factors[1]); Graphs.add_edge!(fg, edge)
edge = Graphs.make_edge(fg, factors[1], variables[2]); Graphs.add_edge!(fg, edge)
edge = Graphs.make_edge(fg, variables[2], factors[2]); Graphs.add_edge!(fg, edge)
edge = Graphs.make_edge(fg, factors[2], variables[3]); Graphs.add_edge!(fg, edge)
edge = Graphs.make_edge(fg, variables[3], factors[3]); Graphs.add_edge!(fg, edge)
edge = Graphs.make_edge(fg, factors[3], variables[4]); Graphs.add_edge!(fg, edge)

# Not great code but I wanted to actually write the data somewhere else, not to a file
m = PipeBuffer()
write(m,Graphs.to_dot(fg))
data = take!(m)
close(m)

open("/tmp/test.dot", "w") do fid
    write(fid,data)
end

This code produces:

If you change it to a directed graph:

Any ideas?

[dehann]

Hi thanks for the example. That's pretty weird. Maybe there is an internal assumption on the ordering of elements in the dict (which would clearly be incorrect). Should also double check weather this is a drawing issue or a graph lookup issue; likely a discrepancy between in and out neighbor searches.

I can go look at how inclist is built and see if something weird is there...

[dehann]

This doesn't look right in the case of dict:
https://github.com/JuliaAttic/Graphs.jl/blob/099ef8f30bd30d760ee5f4a13817928774c8bd20/src/graph.jl#L97

[GearsAD]

Wait one, actually it's producing the correct graph - I was just reading it wrong. If you look above, those are the vertex numbers and although the shape is different, the actual relationships are the same 1--5--2--6--3-7--4.

I thought that was a simplified version of the issue we were seeing earlier in the week, but actually it seems correct. I need to dig a bit more. I'm going to close this as this specific example is performing correctly.

[GearsAD]

So I wasn't using ExVertex, I had created a new type of vertex, and it seems the reason I was getting strange graphs is because I needed to provide an implementation of vertex_index(v), e.g.:

# This is insanely important - if we don't provide a valid index, the edges don't work correctly.
vertex_index(v::GraphsNode) = v.index

Apologies, the example code above actually produces the correct graphs with ExVertex.