transitiveclosure! silently unimplemented

[drbergman]

transitiveclosure!(::MetaGraph) will return without error (at least when the underlying graph is a SimpleDiGraph{Int64}) but the transitive closure is not actually computed.

MWE

using Graphs, MetaGraphsNext, GraphPlot
G = MetaGraph(SimpleDiGraph();
    label_type=Symbol,
    vertex_data_type=Nothing,
    edge_data_type=Nothing)

G[:a] = nothing
G[:b] = nothing
G[:c] = nothing

G[:a, :b] = nothing
G[:b, :c] = nothing

transitiveclosure!(G)
gplot(G) # shows original graph of G, i.e. `a -> b -> c`

# can get expected behavior this way:
transitiveclosure!(G.graph)
gplot(G) # shows transitive closure graph, i.e. `a -> b -> c <- a`

# however, the edge_data is not updated this way!
@assert !in((:a, :c), keys(G.edge_data))

Possible solutions

I think the simplest thing for the short term would be to not dispatch on MetaGraph types. Obviously, would be great to have this functionality for MetaGraphs, too

[gdalle]

Sounds good, want to open a PR with an explicit dispatch throwing an informative error?
The general issue is that the AbstractGraph interface doesn't care about metadata, so every single generic algorithm in Graphs.jl just disregards it. That means the problem is much broader than transitive closure

[drbergman]

bandaid PR opened