genBackEdgeMap fails on recursive subroutines about fortran-src

[mrd]

This is actually a much deeper issue but the root cause is that the recursive subroutine is not called by anything but itself. Therefore none of its basic blocks appear in a traversal of the basic block graph when starting from the beginning of the program, simply because they are not connected to the beginning of the program.

Because the basic blocks don't appear in a reverse postorder traversal then the dataflow analysis function dominators ignores them as well. This produces an error because the function genBackEdgeMap assumes that the result of dominators has an entry for every basic block (by definition, every basic block dominates itself).

This part can be 'fixed' by adding stub entries for every basic block but that doesn't really solve the problem: what to do with disconnected sections of the graph? There are dominance relations within the subgraph that can be extracted, but what is the meaning of 'reverse postorder traversal' of nodes that are not connected to the beginning of the program?

This only comes up in the context of supergraph construction, in which case the question becomes: what is the meaning of interprocedural analysis between procedures that are not connected?

[mrd]

fixed in 9de6d8e by refactoring BBGr to have a bbgrEntries field