How to handle self-loops?

[gdalle]

It is unclear how well self-loops are supported in Graphs.jl, but that should not prevent us from supporting them here. For now, I stated in the documentation that they are not properly handled, because of a few issues that were raised over the years and never closed in the original package:

• Doubling the weights of self-edges sbromberger/SimpleWeightedGraphs.jl#38
• Inconsistent adjacency matrix for weighted and unweighted graphs sbromberger/SimpleWeightedGraphs.jl#53

What do you think?

[pgrepds]

Good question. Is there any particular reason why self-loops are allowed for SimpleGraphs at all in Graphs.jl? It seems to violate the common definition.

[simonschoelly]

Adding self-loops to graphs it not very difficult from a technical perspective (unlike multi-edges), so while it does not strictly correspond to the mathematical definition of a simple graph, we still allow them. There are many cases where self-loops can be useful, for example in finite state diagrams.

In some cases, self-loops have given us some headaches over the last few years though, as one has to consider some extra cases when writing algorithms. And people often forget to write test cases with self-loops.