I'm worried the previous algorithmically-naive implementation may
behave badly on larger-scale projects. We still keep a list around to
return results in the exact same order as previously.
git-svn-id: http://caml.inria.fr/svn/ocaml/trunk@15694 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02
When reporting a circular dependency, refine the printed filenames to
those that are really part of the cycle -- instead of those that
happened to be traversed during the DFS that found a cycle. This gives
much more readable error messages.
git-svn-id: http://caml.inria.fr/svn/ocaml/trunk@15693 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02