1 use crate::frozen::Frozen;
2 use crate::fx::FxIndexSet;
3 use rustc_index::bit_set::BitMatrix;
12 #[derive(Clone, Debug)]
13 pub struct TransitiveRelationBuilder<T> {
14 // List of elements. This is used to map from a T to a usize.
15 elements: FxIndexSet<T>,
17 // List of base edges in the graph. Require to compute transitive
23 pub struct TransitiveRelation<T> {
24 // Frozen transitive relation elements and edges.
25 builder: Frozen<TransitiveRelationBuilder<T>>,
27 // Cached transitive closure derived from the edges.
28 closure: Frozen<BitMatrix<usize, usize>>,
31 impl<T> Deref for TransitiveRelation<T> {
32 type Target = Frozen<TransitiveRelationBuilder<T>>;
34 fn deref(&self) -> &Self::Target {
39 impl<T: Clone> Clone for TransitiveRelation<T> {
40 fn clone(&self) -> Self {
42 builder: Frozen::freeze(self.builder.deref().clone()),
43 closure: Frozen::freeze(self.closure.deref().clone()),
48 // HACK(eddyb) manual impl avoids `Default` bound on `T`.
49 impl<T: Eq + Hash> Default for TransitiveRelationBuilder<T> {
50 fn default() -> Self {
51 TransitiveRelationBuilder { elements: Default::default(), edges: Default::default() }
55 #[derive(Copy, Clone, PartialEq, Eq, PartialOrd, Debug)]
58 #[derive(Clone, PartialEq, Eq, Debug)]
64 impl<T: Eq + Hash + Copy> TransitiveRelationBuilder<T> {
65 pub fn is_empty(&self) -> bool {
69 pub fn elements(&self) -> impl Iterator<Item = &T> {
73 fn index(&self, a: T) -> Option<Index> {
74 self.elements.get_index_of(&a).map(Index)
77 fn add_index(&mut self, a: T) -> Index {
78 let (index, _added) = self.elements.insert_full(a);
82 /// Applies the (partial) function to each edge and returns a new
83 /// relation builder. If `f` returns `None` for any end-point,
85 pub fn maybe_map<F, U>(&self, mut f: F) -> Option<TransitiveRelationBuilder<U>>
87 F: FnMut(T) -> Option<U>,
88 U: Clone + Debug + Eq + Hash + Copy,
90 let mut result = TransitiveRelationBuilder::default();
91 for edge in &self.edges {
92 result.add(f(self.elements[edge.source.0])?, f(self.elements[edge.target.0])?);
97 /// Indicate that `a < b` (where `<` is this relation)
98 pub fn add(&mut self, a: T, b: T) {
99 let a = self.add_index(a);
100 let b = self.add_index(b);
101 let edge = Edge { source: a, target: b };
102 if !self.edges.contains(&edge) {
103 self.edges.push(edge);
107 /// Compute the transitive closure derived from the edges, and converted to
108 /// the final result. After this, all elements will be immutable to maintain
109 /// the correctness of the result.
110 pub fn freeze(self) -> TransitiveRelation<T> {
111 let mut matrix = BitMatrix::new(self.elements.len(), self.elements.len());
112 let mut changed = true;
115 for edge in &self.edges {
116 // add an edge from S -> T
117 changed |= matrix.insert(edge.source.0, edge.target.0);
119 // add all outgoing edges from T into S
120 changed |= matrix.union_rows(edge.target.0, edge.source.0);
123 TransitiveRelation { builder: Frozen::freeze(self), closure: Frozen::freeze(matrix) }
127 impl<T: Eq + Hash + Copy> TransitiveRelation<T> {
128 /// Applies the (partial) function to each edge and returns a new
129 /// relation including transitive closures.
130 pub fn maybe_map<F, U>(&self, f: F) -> Option<TransitiveRelation<U>>
132 F: FnMut(T) -> Option<U>,
133 U: Clone + Debug + Eq + Hash + Copy,
135 Some(self.builder.maybe_map(f)?.freeze())
138 /// Checks whether `a < target` (transitively)
139 pub fn contains(&self, a: T, b: T) -> bool {
140 match (self.index(a), self.index(b)) {
141 (Some(a), Some(b)) => self.with_closure(|closure| closure.contains(a.0, b.0)),
142 (None, _) | (_, None) => false,
146 /// Thinking of `x R y` as an edge `x -> y` in a graph, this
147 /// returns all things reachable from `a`.
149 /// Really this probably ought to be `impl Iterator<Item = &T>`, but
150 /// I'm too lazy to make that work, and -- given the caching
151 /// strategy -- it'd be a touch tricky anyhow.
152 pub fn reachable_from(&self, a: T) -> Vec<T> {
153 match self.index(a) {
155 self.with_closure(|closure| closure.iter(a.0).map(|i| self.elements[i]).collect())
161 /// Picks what I am referring to as the "postdominating"
162 /// upper-bound for `a` and `b`. This is usually the least upper
163 /// bound, but in cases where there is no single least upper
164 /// bound, it is the "mutual immediate postdominator", if you
165 /// imagine a graph where `a < b` means `a -> b`.
167 /// This function is needed because region inference currently
168 /// requires that we produce a single "UB", and there is no best
169 /// choice for the LUB. Rather than pick arbitrarily, I pick a
170 /// less good, but predictable choice. This should help ensure
171 /// that region inference yields predictable results (though it
172 /// itself is not fully sufficient).
174 /// Examples are probably clearer than any prose I could write
175 /// (there are corresponding tests below, btw). In each case,
176 /// the query is `postdom_upper_bound(a, b)`:
179 /// // Returns Some(x), which is also LUB.
185 /// // Returns `Some(x)`, which is not LUB (there is none)
186 /// // diagonal edges run left-to-right.
192 /// // Returns `None`.
196 pub fn postdom_upper_bound(&self, a: T, b: T) -> Option<T> {
197 let mubs = self.minimal_upper_bounds(a, b);
198 self.mutual_immediate_postdominator(mubs)
201 /// Viewing the relation as a graph, computes the "mutual
202 /// immediate postdominator" of a set of points (if one
203 /// exists). See `postdom_upper_bound` for details.
204 pub fn mutual_immediate_postdominator<'a>(&'a self, mut mubs: Vec<T>) -> Option<T> {
208 1 => return Some(mubs[0]),
210 let m = mubs.pop().unwrap();
211 let n = mubs.pop().unwrap();
212 mubs.extend(self.minimal_upper_bounds(n, m));
218 /// Returns the set of bounds `X` such that:
220 /// - `a < X` and `b < X`
221 /// - there is no `Y != X` such that `a < Y` and `Y < X`
222 /// - except for the case where `X < a` (i.e., a strongly connected
223 /// component in the graph). In that case, the smallest
224 /// representative of the SCC is returned (as determined by the
225 /// internal indices).
227 /// Note that this set can, in principle, have any size.
228 pub fn minimal_upper_bounds(&self, a: T, b: T) -> Vec<T> {
229 let (Some(mut a), Some(mut b)) = (self.index(a), self.index(b)) else {
233 // in some cases, there are some arbitrary choices to be made;
234 // it doesn't really matter what we pick, as long as we pick
235 // the same thing consistently when queried, so ensure that
236 // (a, b) are in a consistent relative order
238 mem::swap(&mut a, &mut b);
241 let lub_indices = self.with_closure(|closure| {
242 // Easy case is when either a < b or b < a:
243 if closure.contains(a.0, b.0) {
246 if closure.contains(b.0, a.0) {
250 // Otherwise, the tricky part is that there may be some c
251 // where a < c and b < c. In fact, there may be many such
252 // values. So here is what we do:
254 // 1. Find the vector `[X | a < X && b < X]` of all values
255 // `X` where `a < X` and `b < X`. In terms of the
256 // graph, this means all values reachable from both `a`
257 // and `b`. Note that this vector is also a set, but we
258 // use the term vector because the order matters
259 // to the steps below.
260 // - This vector contains upper bounds, but they are
261 // not minimal upper bounds. So you may have e.g.
262 // `[x, y, tcx, z]` where `x < tcx` and `y < tcx` and
263 // `z < x` and `z < y`:
265 // z --+---> x ----+----> tcx
270 // In this case, we really want to return just `[z]`.
271 // The following steps below achieve this by gradually
272 // reducing the list.
273 // 2. Pare down the vector using `pare_down`. This will
274 // remove elements from the vector that can be reached
275 // by an earlier element.
276 // - In the example above, this would convert `[x, y,
277 // tcx, z]` to `[x, y, z]`. Note that `x` and `y` are
278 // still in the vector; this is because while `z < x`
279 // (and `z < y`) holds, `z` comes after them in the
281 // 3. Reverse the vector and repeat the pare down process.
282 // - In the example above, we would reverse to
283 // `[z, y, x]` and then pare down to `[z]`.
284 // 4. Reverse once more just so that we yield a vector in
285 // increasing order of index. Not necessary, but why not.
287 // I believe this algorithm yields a minimal set. The
288 // argument is that, after step 2, we know that no element
289 // can reach its successors (in the vector, not the graph).
290 // After step 3, we know that no element can reach any of
291 // its predecessors (because of step 2) nor successors
292 // (because we just called `pare_down`)
294 // This same algorithm is used in `parents` below.
296 let mut candidates = closure.intersect_rows(a.0, b.0); // (1)
297 pare_down(&mut candidates, closure); // (2)
298 candidates.reverse(); // (3a)
299 pare_down(&mut candidates, closure); // (3b)
306 .map(|i| self.elements[i])
310 /// Given an element A, returns the maximal set {B} of elements B
315 /// - for each i, j: `B[i]` R `B[j]` does not hold
317 /// The intuition is that this moves "one step up" through a lattice
318 /// (where the relation is encoding the `<=` relation for the lattice).
319 /// So e.g., if the relation is `->` and we have
327 /// then `parents(a)` returns `[b, c]`. The `postdom_parent` function
328 /// would further reduce this to just `f`.
329 pub fn parents(&self, a: T) -> Vec<T> {
330 let Some(a) = self.index(a) else {
334 // Steal the algorithm for `minimal_upper_bounds` above, but
335 // with a slight tweak. In the case where `a R a`, we remove
336 // that from the set of candidates.
337 let ancestors = self.with_closure(|closure| {
338 let mut ancestors = closure.intersect_rows(a.0, a.0);
340 // Remove anything that can reach `a`. If this is a
341 // reflexive relation, this will include `a` itself.
342 ancestors.retain(|&e| !closure.contains(e, a.0));
344 pare_down(&mut ancestors, closure); // (2)
345 ancestors.reverse(); // (3a)
346 pare_down(&mut ancestors, closure); // (3b)
353 .map(|i| self.elements[i])
357 fn with_closure<OP, R>(&self, op: OP) -> R
359 OP: FnOnce(&BitMatrix<usize, usize>) -> R,
364 /// Lists all the base edges in the graph: the initial _non-transitive_ set of element
365 /// relations, which will be later used as the basis for the transitive closure computation.
366 pub fn base_edges(&self) -> impl Iterator<Item = (T, T)> + '_ {
369 .map(move |edge| (self.elements[edge.source.0], self.elements[edge.target.0]))
373 /// Pare down is used as a step in the LUB computation. It edits the
374 /// candidates array in place by removing any element j for which
375 /// there exists an earlier element i<j such that i -> j. That is,
376 /// after you run `pare_down`, you know that for all elements that
377 /// remain in candidates, they cannot reach any of the elements that
380 /// Examples follow. Assume that a -> b -> c and x -> y -> z.
382 /// - Input: `[a, b, x]`. Output: `[a, x]`.
383 /// - Input: `[b, a, x]`. Output: `[b, a, x]`.
384 /// - Input: `[a, x, b, y]`. Output: `[a, x]`.
385 fn pare_down(candidates: &mut Vec<usize>, closure: &BitMatrix<usize, usize>) {
387 while let Some(&candidate_i) = candidates.get(i) {
392 while let Some(&candidate_j) = candidates.get(j) {
393 if closure.contains(candidate_i, candidate_j) {
394 // If `i` can reach `j`, then we can remove `j`. So just
395 // mark it as dead and move on; subsequent indices will be
396 // shifted into its place.
399 candidates[j - dead] = candidate_j;
403 candidates.truncate(j - dead);