add_edge!(g, u, v) deletes existing edges if called twice with the same arguments

[degregat]

add_edge!(g, u, v) from here, when called with the same u, v twice, adds the edge the first time, deletes it the second time (and does not add it in subsequent calls), since the length of g.graph does not change.

Example

using Graphs
using MetaGraphsNext

dg = MetaGraph(
    DiGraph();
    label_type = String,
    vertex_data_type = String,
)

add_vertex!(dg, "u", "foo")
add_vertex!(dg, "v", "bar")
add_edge!(dg, "u", "v")
println(length(dg.edge_data)) # == 1
add_edge!(dg, "u", "v")
println(length(dg.edge_data)) # == 0

g = MetaGraph(
    Graph();
    label_type = String,
    vertex_data_type = String,
)

add_vertex!(g, "u", "foo")
add_vertex!(g, "v", "bar")
add_edge!(g, "u", "v")
println(length(dg.edge_data)) # == 1 
add_edge!(g, "u", "v")
println(length(dg.edge_data)) # == 0

Workaround

# For directed graphs
if !haskey(g.edge_data, (u, v))
  add_edge!(g, u, v)
end

# For undirected graphs
if !haskey(g.edge_data, (u, v)) && !haskey(g.edge_data, (v, u))
  add_edge!(g, u, v)
end

Proposed Fix

Remove this call of delete! and only return false. If that does not lead to the expected behavior, add checks as described in the workaround inside add_edge!, s.t. users don't have to do that.

[gdalle]

Thanks for the issue! It seems like a problem indeed, do you want to open a PR?

[degregat]

I can do that 😄

Do you think removing delete! will have unintended consequences, e.g. what's the worst that can happen if add_edge!() (from Graphs.jl) is called repeatedly with the same codes it should resolve to from the labels?

[bramtayl]

I think the only reason the delete! was there was in case adding the edge failed for another reason, e.g. the graph ran out of room or something. In which case, we should only delete if the edge doesn't exist. So we could change this line to !has_edge(meta_graph.graph, code_1, code_2)?

[cecileane]

I discovered that too, and will add a fix to a PR very soon.