Recomputing the DFS will cause m_dfs and m_loops to return nonsensical results (i.e. the results aren't just outdated). We should validate in post-phase checks that the DFS tree we're storing matches the flow graph, and explicitly null out the relevant information when we invalidate it.
Recomputing the DFS will cause
m_dfsandm_loopsto return nonsensical results (i.e. the results aren't just outdated). We should validate in post-phase checks that the DFS tree we're storing matches the flow graph, and explicitly null out the relevant information when we invalidate it.