This has always bugged me: what reasons outside of convention do we have for preferring O(V + E) to O(E) on algorithms that only make sense in the context of a single connected component? I know this is slightly OT, but tangentially related.