Especially when they are explicitly taken into account through escape hatches in the ownership system
Maybe that's an interesting way to think about the article too. The obvious way to think about a graph using ownership is to start with a root node, then to say that nodes own other nodes through their edges. But that model won't generalise to disjoint graphs (eg, social network with separate friend groups) and it isn't always clear which node is the "root" node.
struct Graph {
nodes: Vec<Node>,
edges: Vec<Edge>,
}
but the associations between nodes would have an arbitrary structure.The article suggests that handles are better than references anyway (I'd tend to agree in this case) so Edge would be a pair of handles. That seems totally fine, so maybe the "fatal flaw" is overstated...