1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380
//! Routine to compute the strongly connected components (SCCs) of a graph. //! //! Also computes as the resulting DAG if each SCC is replaced with a //! node in the graph. This uses [Tarjan's algorithm]( //! https://en.wikipedia.org/wiki/Tarjan%27s_strongly_connected_components_algorithm) //! that completes in *O(n)* time. use crate::fx::FxHashSet; use crate::graph::vec_graph::VecGraph; use crate::graph::{DirectedGraph, GraphSuccessors, WithNumEdges, WithNumNodes, WithSuccessors}; use rustc_index::vec::{Idx, IndexVec}; use std::ops::Range; #[cfg(test)] mod tests; /// Strongly connected components (SCC) of a graph. The type `N` is /// the index type for the graph nodes and `S` is the index type for /// the SCCs. We can map from each node to the SCC that it /// participates in, and we also have the successors of each SCC. pub struct Sccs<N: Idx, S: Idx> { /// For each node, what is the SCC index of the SCC to which it /// belongs. scc_indices: IndexVec<N, S>, /// Data about each SCC. scc_data: SccData<S>, } struct SccData<S: Idx> { /// For each SCC, the range of `all_successors` where its /// successors can be found. ranges: IndexVec<S, Range<usize>>, /// Contains the successors for all the Sccs, concatenated. The /// range of indices corresponding to a given SCC is found in its /// SccData. all_successors: Vec<S>, } impl<N: Idx, S: Idx> Sccs<N, S> { pub fn new(graph: &(impl DirectedGraph<Node = N> + WithNumNodes + WithSuccessors)) -> Self { SccsConstruction::construct(graph) } /// Returns the number of SCCs in the graph. pub fn num_sccs(&self) -> usize { self.scc_data.len() } /// Returns an iterator over the SCCs in the graph. /// /// The SCCs will be iterated in **dependency order** (or **post order**), /// meaning that if `S1 -> S2`, we will visit `S2` first and `S1` after. /// This is convenient when the edges represent dependencies: when you visit /// `S1`, the value for `S2` will already have been computed. pub fn all_sccs(&self) -> impl Iterator<Item = S> { (0..self.scc_data.len()).map(S::new) } /// Returns the SCC to which a node `r` belongs. pub fn scc(&self, r: N) -> S { self.scc_indices[r] } /// Returns the successors of the given SCC. pub fn successors(&self, scc: S) -> &[S] { self.scc_data.successors(scc) } /// Construct the reverse graph of the SCC graph. pub fn reverse(&self) -> VecGraph<S> { VecGraph::new( self.num_sccs(), self.all_sccs() .flat_map(|source| { self.successors(source).iter().map(move |&target| (target, source)) }) .collect(), ) } } impl<N: Idx, S: Idx> DirectedGraph for Sccs<N, S> { type Node = S; } impl<N: Idx, S: Idx> WithNumNodes for Sccs<N, S> { fn num_nodes(&self) -> usize { self.num_sccs() } } impl<N: Idx, S: Idx> WithNumEdges for Sccs<N, S> { fn num_edges(&self) -> usize { self.scc_data.all_successors.len() } } impl<N: Idx, S: Idx> GraphSuccessors<'graph> for Sccs<N, S> { type Item = S; type Iter = std::iter::Cloned<std::slice::Iter<'graph, S>>; } impl<N: Idx, S: Idx> WithSuccessors for Sccs<N, S> { fn successors(&self, node: S) -> <Self as GraphSuccessors<'_>>::Iter { self.successors(node).iter().cloned() } } impl<S: Idx> SccData<S> { /// Number of SCCs, fn len(&self) -> usize { self.ranges.len() } /// Returns the successors of the given SCC. fn successors(&self, scc: S) -> &[S] { // Annoyingly, `range` does not implement `Copy`, so we have // to do `range.start..range.end`: let range = &self.ranges[scc]; &self.all_successors[range.start..range.end] } /// Creates a new SCC with `successors` as its successors and /// returns the resulting index. fn create_scc(&mut self, successors: impl IntoIterator<Item = S>) -> S { // Store the successors on `scc_successors_vec`, remembering // the range of indices. let all_successors_start = self.all_successors.len(); self.all_successors.extend(successors); let all_successors_end = self.all_successors.len(); debug!( "create_scc({:?}) successors={:?}", self.ranges.len(), &self.all_successors[all_successors_start..all_successors_end], ); self.ranges.push(all_successors_start..all_successors_end) } } struct SccsConstruction<'c, G: DirectedGraph + WithNumNodes + WithSuccessors, S: Idx> { graph: &'c G, /// The state of each node; used during walk to record the stack /// and after walk to record what cycle each node ended up being /// in. node_states: IndexVec<G::Node, NodeState<G::Node, S>>, /// The stack of nodes that we are visiting as part of the DFS. node_stack: Vec<G::Node>, /// The stack of successors: as we visit a node, we mark our /// position in this stack, and when we encounter a successor SCC, /// we push it on the stack. When we complete an SCC, we can pop /// everything off the stack that was found along the way. successors_stack: Vec<S>, /// A set used to strip duplicates. As we accumulate successors /// into the successors_stack, we sometimes get duplicate entries. /// We use this set to remove those -- we also keep its storage /// around between successors to amortize memory allocation costs. duplicate_set: FxHashSet<S>, scc_data: SccData<S>, } #[derive(Copy, Clone, Debug)] enum NodeState<N, S> { /// This node has not yet been visited as part of the DFS. /// /// After SCC construction is complete, this state ought to be /// impossible. NotVisited, /// This node is currently being walk as part of our DFS. It is on /// the stack at the depth `depth`. /// /// After SCC construction is complete, this state ought to be /// impossible. BeingVisited { depth: usize }, /// Indicates that this node is a member of the given cycle. InCycle { scc_index: S }, /// Indicates that this node is a member of whatever cycle /// `parent` is a member of. This state is transient: whenever we /// see it, we try to overwrite it with the current state of /// `parent` (this is the "path compression" step of a union-find /// algorithm). InCycleWith { parent: N }, } #[derive(Copy, Clone, Debug)] enum WalkReturn<S> { Cycle { min_depth: usize }, Complete { scc_index: S }, } impl<'c, G, S> SccsConstruction<'c, G, S> where G: DirectedGraph + WithNumNodes + WithSuccessors, S: Idx, { /// Identifies SCCs in the graph `G` and computes the resulting /// DAG. This uses a variant of [Tarjan's /// algorithm][wikipedia]. The high-level summary of the algorithm /// is that we do a depth-first search. Along the way, we keep a /// stack of each node whose successors are being visited. We /// track the depth of each node on this stack (there is no depth /// if the node is not on the stack). When we find that some node /// N with depth D can reach some other node N' with lower depth /// D' (i.e., D' < D), we know that N, N', and all nodes in /// between them on the stack are part of an SCC. /// /// [wikipedia]: https://bit.ly/2EZIx84 fn construct(graph: &'c G) -> Sccs<G::Node, S> { let num_nodes = graph.num_nodes(); let mut this = Self { graph, node_states: IndexVec::from_elem_n(NodeState::NotVisited, num_nodes), node_stack: Vec::with_capacity(num_nodes), successors_stack: Vec::new(), scc_data: SccData { ranges: IndexVec::new(), all_successors: Vec::new() }, duplicate_set: FxHashSet::default(), }; let scc_indices = (0..num_nodes) .map(G::Node::new) .map(|node| match this.walk_node(0, node) { WalkReturn::Complete { scc_index } => scc_index, WalkReturn::Cycle { min_depth } => { panic!("`walk_node(0, {:?})` returned cycle with depth {:?}", node, min_depth) } }) .collect(); Sccs { scc_indices, scc_data: this.scc_data } } /// Visits a node during the DFS. We first examine its current /// state -- if it is not yet visited (`NotVisited`), we can push /// it onto the stack and start walking its successors. /// /// If it is already on the DFS stack it will be in the state /// `BeingVisited`. In that case, we have found a cycle and we /// return the depth from the stack. /// /// Otherwise, we are looking at a node that has already been /// completely visited. We therefore return `WalkReturn::Complete` /// with its associated SCC index. fn walk_node(&mut self, depth: usize, node: G::Node) -> WalkReturn<S> { debug!("walk_node(depth = {:?}, node = {:?})", depth, node); match self.find_state(node) { NodeState::InCycle { scc_index } => WalkReturn::Complete { scc_index }, NodeState::BeingVisited { depth: min_depth } => WalkReturn::Cycle { min_depth }, NodeState::NotVisited => self.walk_unvisited_node(depth, node), NodeState::InCycleWith { parent } => panic!( "`find_state` returned `InCycleWith({:?})`, which ought to be impossible", parent ), } } /// Fetches the state of the node `r`. If `r` is recorded as being /// in a cycle with some other node `r2`, then fetches the state /// of `r2` (and updates `r` to reflect current result). This is /// basically the "find" part of a standard union-find algorithm /// (with path compression). fn find_state(&mut self, r: G::Node) -> NodeState<G::Node, S> { debug!("find_state(r = {:?} in state {:?})", r, self.node_states[r]); match self.node_states[r] { NodeState::InCycle { scc_index } => NodeState::InCycle { scc_index }, NodeState::BeingVisited { depth } => NodeState::BeingVisited { depth }, NodeState::NotVisited => NodeState::NotVisited, NodeState::InCycleWith { parent } => { let parent_state = self.find_state(parent); debug!("find_state: parent_state = {:?}", parent_state); match parent_state { NodeState::InCycle { .. } => { self.node_states[r] = parent_state; parent_state } NodeState::BeingVisited { depth } => { self.node_states[r] = NodeState::InCycleWith { parent: self.node_stack[depth] }; parent_state } NodeState::NotVisited | NodeState::InCycleWith { .. } => { panic!("invalid parent state: {:?}", parent_state) } } } } } /// Walks a node that has never been visited before. fn walk_unvisited_node(&mut self, depth: usize, node: G::Node) -> WalkReturn<S> { debug!("walk_unvisited_node(depth = {:?}, node = {:?})", depth, node); debug_assert!(match self.node_states[node] { NodeState::NotVisited => true, _ => false, }); // Push `node` onto the stack. self.node_states[node] = NodeState::BeingVisited { depth }; self.node_stack.push(node); // Walk each successor of the node, looking to see if any of // them can reach a node that is presently on the stack. If // so, that means they can also reach us. let mut min_depth = depth; let mut min_cycle_root = node; let successors_len = self.successors_stack.len(); for successor_node in self.graph.successors(node) { debug!("walk_unvisited_node: node = {:?} successor_ode = {:?}", node, successor_node); match self.walk_node(depth + 1, successor_node) { WalkReturn::Cycle { min_depth: successor_min_depth } => { // Track the minimum depth we can reach. assert!(successor_min_depth <= depth); if successor_min_depth < min_depth { debug!( "walk_unvisited_node: node = {:?} successor_min_depth = {:?}", node, successor_min_depth ); min_depth = successor_min_depth; min_cycle_root = successor_node; } } WalkReturn::Complete { scc_index: successor_scc_index } => { // Push the completed SCC indices onto // the `successors_stack` for later. debug!( "walk_unvisited_node: node = {:?} successor_scc_index = {:?}", node, successor_scc_index ); self.successors_stack.push(successor_scc_index); } } } // Completed walk, remove `node` from the stack. let r = self.node_stack.pop(); debug_assert_eq!(r, Some(node)); // If `min_depth == depth`, then we are the root of the // cycle: we can't reach anyone further down the stack. if min_depth == depth { // Note that successor stack may have duplicates, so we // want to remove those: let deduplicated_successors = { let duplicate_set = &mut self.duplicate_set; duplicate_set.clear(); self.successors_stack .drain(successors_len..) .filter(move |&i| duplicate_set.insert(i)) }; let scc_index = self.scc_data.create_scc(deduplicated_successors); self.node_states[node] = NodeState::InCycle { scc_index }; WalkReturn::Complete { scc_index } } else { // We are not the head of the cycle. Return back to our // caller. They will take ownership of the // `self.successors` data that we pushed. self.node_states[node] = NodeState::InCycleWith { parent: min_cycle_root }; WalkReturn::Cycle { min_depth } } } }