Dataflow cluster analysis

[Ellpeck]

We want to calculate clusters on the dataflow graph that determine which parts of code are highly dependent on each other (ie "belong together"). Open question: when can we ignore opposite-facing directed edges, and when do we have to traverse them?

Output should be a set of flowR node IDs that form a cluster, to be evaluated further later on.

Step 1: Simple/naive cluster calculation using reachability analysis.

Step 2: The result will likely be one large cluster because there are shared dependencies on setup steps, reused functions etc. We can implement a "bottleneck" node calculation that splits clusters on these sorts of nodes and creates separate clusters that all individually contain the "bottleneck" node. Open question: what constitutes a "bottleneck" node, ie when is it reused enough, and when is the cluster around it small enough, to be splittable?

Implementation as a separate "post analysis" module rather than a pipeline step.