jlebar added a comment. Unlike https://reviews.llvm.org/D26998, there should be no performance question here, because ReversePostOrderTraversal itself builds a vector -- previously we were just copying that vector before iterating over it. https://reviews.llvm.org/D26997