2 use std::marker::PhantomData;
3 use std::ops::RangeBounds;
4 use std::ops::{Bound, Range};
7 use crate::vec::IndexVec;
8 use smallvec::SmallVec;
13 /// Stores a set of intervals on the indices.
15 /// The elements in `map` are sorted and non-adjacent, which means
16 /// the second value of the previous element is *greater* than the
17 /// first value of the following element.
18 #[derive(Debug, Clone)]
19 pub struct IntervalSet<I> {
21 map: SmallVec<[(u32, u32); 4]>,
23 _data: PhantomData<I>,
27 fn inclusive_start<T: Idx>(range: impl RangeBounds<T>) -> u32 {
28 match range.start_bound() {
29 Bound::Included(start) => start.index() as u32,
30 Bound::Excluded(start) => start.index() as u32 + 1,
31 Bound::Unbounded => 0,
36 fn inclusive_end<T: Idx>(domain: usize, range: impl RangeBounds<T>) -> Option<u32> {
37 let end = match range.end_bound() {
38 Bound::Included(end) => end.index() as u32,
39 Bound::Excluded(end) => end.index().checked_sub(1)? as u32,
40 Bound::Unbounded => domain.checked_sub(1)? as u32,
45 impl<I: Idx> IntervalSet<I> {
46 pub fn new(domain: usize) -> IntervalSet<I> {
47 IntervalSet { map: SmallVec::new(), domain, _data: PhantomData }
50 pub fn clear(&mut self) {
54 pub fn iter(&self) -> impl Iterator<Item = I> + '_
58 self.iter_intervals().flatten()
61 /// Iterates through intervals stored in the set, in order.
62 pub fn iter_intervals(&self) -> impl Iterator<Item = std::ops::Range<I>> + '_
66 self.map.iter().map(|&(start, end)| I::new(start as usize)..I::new(end as usize + 1))
69 /// Returns true if we increased the number of elements present.
70 pub fn insert(&mut self, point: I) -> bool {
71 self.insert_range(point..=point)
74 /// Returns true if we increased the number of elements present.
75 pub fn insert_range(&mut self, range: impl RangeBounds<I> + Clone) -> bool {
76 let start = inclusive_start(range.clone());
77 let Some(end) = inclusive_end(self.domain, range) else {
85 // This condition looks a bit weird, but actually makes sense.
87 // if r.0 == end + 1, then we're actually adjacent, so we want to
88 // continue to the next range. We're looking here for the first
89 // range which starts *non-adjacently* to our end.
90 let next = self.map.partition_point(|r| r.0 <= end + 1);
91 let result = if let Some(right) = next.checked_sub(1) {
92 let (prev_start, prev_end) = self.map[right];
93 if prev_end + 1 >= start {
94 // If the start for the inserted range is adjacent to the
95 // end of the previous, we can extend the previous range.
96 if start < prev_start {
97 // The first range which ends *non-adjacently* to our start.
98 // And we can ensure that left <= right.
99 let left = self.map.partition_point(|l| l.1 + 1 < start);
100 let min = std::cmp::min(self.map[left].0, start);
101 let max = std::cmp::max(prev_end, end);
102 self.map[right] = (min, max);
104 self.map.drain(left..right);
108 // We overlap with the previous range, increase it to
111 // Make sure we're actually going to *increase* it though --
112 // it may be that end is just inside the previously existing
115 self.map[right].1 = end;
122 // Otherwise, we don't overlap, so just insert
123 self.map.insert(right + 1, (start, end));
127 if self.map.is_empty() {
128 // Quite common in practice, and expensive to call memcpy
130 self.map.push((start, end));
132 self.map.insert(next, (start, end));
137 self.check_invariants(),
138 "wrong intervals after insert {start:?}..={end:?} to {self:?}"
143 pub fn contains(&self, needle: I) -> bool {
144 let needle = needle.index() as u32;
145 let Some(last) = self.map.partition_point(|r| r.0 <= needle).checked_sub(1) else {
146 // All ranges in the map start after the new range's end
149 let (_, prev_end) = &self.map[last];
153 pub fn superset(&self, other: &IntervalSet<I>) -> bool
157 let mut sup_iter = self.iter_intervals();
158 let mut current = None;
159 let contains = |sup: Range<I>, sub: Range<I>, current: &mut Option<Range<I>>| {
160 if sup.end < sub.start {
161 // if `sup.end == sub.start`, the next sup doesn't contain `sub.start`
162 None // continue to the next sup
163 } else if sup.end >= sub.end && sup.start <= sub.start {
164 *current = Some(sup); // save the current sup
170 other.iter_intervals().all(|sub| {
173 .and_then(|sup| contains(sup, sub.clone(), &mut current))
174 .or_else(|| sup_iter.find_map(|sup| contains(sup, sub.clone(), &mut current)))
179 pub fn is_empty(&self) -> bool {
183 /// Returns the maximum (last) element present in the set from `range`.
184 pub fn last_set_in(&self, range: impl RangeBounds<I> + Clone) -> Option<I> {
185 let start = inclusive_start(range.clone());
186 let Some(end) = inclusive_end(self.domain, range) else {
193 let Some(last) = self.map.partition_point(|r| r.0 <= end).checked_sub(1) else {
194 // All ranges in the map start after the new range's end
197 let (_, prev_end) = &self.map[last];
198 if start <= *prev_end { Some(I::new(std::cmp::min(*prev_end, end) as usize)) } else { None }
201 pub fn insert_all(&mut self) {
203 if let Some(end) = self.domain.checked_sub(1) {
204 self.map.push((0, end.try_into().unwrap()));
206 debug_assert!(self.check_invariants());
209 pub fn union(&mut self, other: &IntervalSet<I>) -> bool
213 assert_eq!(self.domain, other.domain);
214 let mut did_insert = false;
215 for range in other.iter_intervals() {
216 did_insert |= self.insert_range(range);
218 debug_assert!(self.check_invariants());
222 // Check the intervals are valid, sorted and non-adjacent
223 fn check_invariants(&self) -> bool {
224 let mut current: Option<u32> = None;
225 for (start, end) in &self.map {
226 if start > end || current.map_or(false, |x| x + 1 >= *start) {
229 current = Some(*end);
231 current.map_or(true, |x| x < self.domain as u32)
235 /// This data structure optimizes for cases where the stored bits in each row
236 /// are expected to be highly contiguous (long ranges of 1s or 0s), in contrast
237 /// to BitMatrix and SparseBitMatrix which are optimized for
238 /// "random"/non-contiguous bits and cheap(er) point queries at the expense of
241 pub struct SparseIntervalMatrix<R, C>
246 rows: IndexVec<R, IntervalSet<C>>,
250 impl<R: Idx, C: Step + Idx> SparseIntervalMatrix<R, C> {
251 pub fn new(column_size: usize) -> SparseIntervalMatrix<R, C> {
252 SparseIntervalMatrix { rows: IndexVec::new(), column_size }
255 pub fn rows(&self) -> impl Iterator<Item = R> {
259 pub fn row(&self, row: R) -> Option<&IntervalSet<C>> {
263 fn ensure_row(&mut self, row: R) -> &mut IntervalSet<C> {
264 self.rows.ensure_contains_elem(row, || IntervalSet::new(self.column_size));
268 pub fn union_row(&mut self, row: R, from: &IntervalSet<C>) -> bool
272 self.ensure_row(row).union(from)
275 pub fn union_rows(&mut self, read: R, write: R) -> bool
279 if read == write || self.rows.get(read).is_none() {
282 self.ensure_row(write);
283 let (read_row, write_row) = self.rows.pick2_mut(read, write);
284 write_row.union(read_row)
287 pub fn insert_all_into_row(&mut self, row: R) {
288 self.ensure_row(row).insert_all();
291 pub fn insert_range(&mut self, row: R, range: impl RangeBounds<C> + Clone) {
292 self.ensure_row(row).insert_range(range);
295 pub fn insert(&mut self, row: R, point: C) -> bool {
296 self.ensure_row(row).insert(point)
299 pub fn contains(&self, row: R, point: C) -> bool {
300 self.row(row).map_or(false, |r| r.contains(point))