]>
Commit | Line | Data |
---|---|---|
f20569fa XL |
1 | //! A variant of `SortedMap` that preserves insertion order. |
2 | ||
3 | use std::borrow::Borrow; | |
4 | use std::hash::{Hash, Hasher}; | |
5 | use std::iter::FromIterator; | |
6 | ||
7 | use crate::stable_hasher::{HashStable, StableHasher}; | |
8 | use rustc_index::vec::{Idx, IndexVec}; | |
9 | ||
10 | /// An indexed multi-map that preserves insertion order while permitting both *O*(log *n*) lookup of | |
11 | /// an item by key and *O*(1) lookup by index. | |
12 | /// | |
13 | /// This data structure is a hybrid of an [`IndexVec`] and a [`SortedMap`]. Like `IndexVec`, | |
14 | /// `SortedIndexMultiMap` assigns a typed index to each item while preserving insertion order. | |
15 | /// Like `SortedMap`, `SortedIndexMultiMap` has efficient lookup of items by key. However, this | |
16 | /// is accomplished by sorting an array of item indices instead of the items themselves. | |
17 | /// | |
18 | /// Unlike `SortedMap`, this data structure can hold multiple equivalent items at once, so the | |
19 | /// `get_by_key` method and its variants return an iterator instead of an `Option`. Equivalent | |
20 | /// items will be yielded in insertion order. | |
21 | /// | |
22 | /// Unlike a general-purpose map like `BTreeSet` or `HashSet`, `SortedMap` and | |
23 | /// `SortedIndexMultiMap` require *O*(*n*) time to insert a single item. This is because we may need | |
24 | /// to insert into the middle of the sorted array. Users should avoid mutating this data structure | |
25 | /// in-place. | |
26 | /// | |
27 | /// [`SortedMap`]: super::SortedMap | |
28 | #[derive(Clone, Debug)] | |
29 | pub struct SortedIndexMultiMap<I: Idx, K, V> { | |
30 | /// The elements of the map in insertion order. | |
31 | items: IndexVec<I, (K, V)>, | |
32 | ||
33 | /// Indices of the items in the set, sorted by the item's key. | |
34 | idx_sorted_by_item_key: Vec<I>, | |
35 | } | |
36 | ||
37 | impl<I: Idx, K: Ord, V> SortedIndexMultiMap<I, K, V> { | |
38 | pub fn new() -> Self { | |
39 | SortedIndexMultiMap { items: IndexVec::new(), idx_sorted_by_item_key: Vec::new() } | |
40 | } | |
41 | ||
42 | pub fn len(&self) -> usize { | |
43 | self.items.len() | |
44 | } | |
45 | ||
46 | pub fn is_empty(&self) -> bool { | |
47 | self.items.is_empty() | |
48 | } | |
49 | ||
50 | /// Returns an iterator over the items in the map in insertion order. | |
51 | pub fn into_iter(self) -> impl DoubleEndedIterator<Item = (K, V)> { | |
52 | self.items.into_iter() | |
53 | } | |
54 | ||
55 | /// Returns an iterator over the items in the map in insertion order along with their indices. | |
56 | pub fn into_iter_enumerated(self) -> impl DoubleEndedIterator<Item = (I, (K, V))> { | |
57 | self.items.into_iter_enumerated() | |
58 | } | |
59 | ||
60 | /// Returns an iterator over the items in the map in insertion order. | |
61 | pub fn iter(&self) -> impl '_ + DoubleEndedIterator<Item = (&K, &V)> { | |
62 | self.items.iter().map(|(ref k, ref v)| (k, v)) | |
63 | } | |
64 | ||
65 | /// Returns an iterator over the items in the map in insertion order along with their indices. | |
66 | pub fn iter_enumerated(&self) -> impl '_ + DoubleEndedIterator<Item = (I, (&K, &V))> { | |
67 | self.items.iter_enumerated().map(|(i, (ref k, ref v))| (i, (k, v))) | |
68 | } | |
69 | ||
70 | /// Returns the item in the map with the given index. | |
71 | pub fn get(&self, idx: I) -> Option<&(K, V)> { | |
72 | self.items.get(idx) | |
73 | } | |
74 | ||
75 | /// Returns an iterator over the items in the map that are equal to `key`. | |
76 | /// | |
77 | /// If there are multiple items that are equivalent to `key`, they will be yielded in | |
78 | /// insertion order. | |
79 | pub fn get_by_key<Q: 'a>(&'a self, key: &Q) -> impl 'a + Iterator<Item = &'a V> | |
80 | where | |
81 | Q: Ord + ?Sized, | |
82 | K: Borrow<Q>, | |
83 | { | |
84 | self.get_by_key_enumerated(key).map(|(_, v)| v) | |
85 | } | |
86 | ||
87 | /// Returns an iterator over the items in the map that are equal to `key` along with their | |
88 | /// indices. | |
89 | /// | |
90 | /// If there are multiple items that are equivalent to `key`, they will be yielded in | |
91 | /// insertion order. | |
92 | pub fn get_by_key_enumerated<Q>(&self, key: &Q) -> impl '_ + Iterator<Item = (I, &V)> | |
93 | where | |
94 | Q: Ord + ?Sized, | |
95 | K: Borrow<Q>, | |
96 | { | |
97 | // FIXME: This should be in the standard library as `equal_range`. See rust-lang/rfcs#2184. | |
98 | match self.binary_search_idx(key) { | |
99 | Err(_) => self.idxs_to_items_enumerated(&[]), | |
100 | ||
101 | Ok(idx) => { | |
102 | let start = self.find_lower_bound(key, idx); | |
103 | let end = self.find_upper_bound(key, idx); | |
104 | self.idxs_to_items_enumerated(&self.idx_sorted_by_item_key[start..end]) | |
105 | } | |
106 | } | |
107 | } | |
108 | ||
109 | fn binary_search_idx<Q>(&self, key: &Q) -> Result<usize, usize> | |
110 | where | |
111 | Q: Ord + ?Sized, | |
112 | K: Borrow<Q>, | |
113 | { | |
114 | self.idx_sorted_by_item_key.binary_search_by(|&idx| self.items[idx].0.borrow().cmp(key)) | |
115 | } | |
116 | ||
117 | /// Returns the index into the `idx_sorted_by_item_key` array of the first item equal to | |
118 | /// `key`. | |
119 | /// | |
120 | /// `initial` must be an index into that same array for an item that is equal to `key`. | |
121 | fn find_lower_bound<Q>(&self, key: &Q, initial: usize) -> usize | |
122 | where | |
123 | Q: Ord + ?Sized, | |
124 | K: Borrow<Q>, | |
125 | { | |
126 | debug_assert!(self.items[self.idx_sorted_by_item_key[initial]].0.borrow() == key); | |
127 | ||
128 | // FIXME: At present, this uses linear search, meaning lookup is only `O(log n)` if duplicate | |
129 | // entries are rare. It would be better to start with a linear search for the common case but | |
130 | // fall back to an exponential search if many duplicates are found. This applies to | |
131 | // `upper_bound` as well. | |
132 | let mut start = initial; | |
133 | while start != 0 && self.items[self.idx_sorted_by_item_key[start - 1]].0.borrow() == key { | |
134 | start -= 1; | |
135 | } | |
136 | ||
137 | start | |
138 | } | |
139 | ||
140 | /// Returns the index into the `idx_sorted_by_item_key` array of the first item greater than | |
141 | /// `key`, or `self.len()` if no such item exists. | |
142 | /// | |
143 | /// `initial` must be an index into that same array for an item that is equal to `key`. | |
144 | fn find_upper_bound<Q>(&self, key: &Q, initial: usize) -> usize | |
145 | where | |
146 | Q: Ord + ?Sized, | |
147 | K: Borrow<Q>, | |
148 | { | |
149 | debug_assert!(self.items[self.idx_sorted_by_item_key[initial]].0.borrow() == key); | |
150 | ||
151 | // See the FIXME for `find_lower_bound`. | |
152 | let mut end = initial + 1; | |
153 | let len = self.items.len(); | |
154 | while end < len && self.items[self.idx_sorted_by_item_key[end]].0.borrow() == key { | |
155 | end += 1; | |
156 | } | |
157 | ||
158 | end | |
159 | } | |
160 | ||
161 | fn idxs_to_items_enumerated(&'a self, idxs: &'a [I]) -> impl 'a + Iterator<Item = (I, &'a V)> { | |
162 | idxs.iter().map(move |&idx| (idx, &self.items[idx].1)) | |
163 | } | |
164 | } | |
165 | ||
166 | impl<I: Idx, K: Eq, V: Eq> Eq for SortedIndexMultiMap<I, K, V> {} | |
167 | impl<I: Idx, K: PartialEq, V: PartialEq> PartialEq for SortedIndexMultiMap<I, K, V> { | |
168 | fn eq(&self, other: &Self) -> bool { | |
169 | // No need to compare the sorted index. If the items are the same, the index will be too. | |
170 | self.items == other.items | |
171 | } | |
172 | } | |
173 | ||
174 | impl<I: Idx, K, V> Hash for SortedIndexMultiMap<I, K, V> | |
175 | where | |
176 | K: Hash, | |
177 | V: Hash, | |
178 | { | |
179 | fn hash<H: Hasher>(&self, hasher: &mut H) { | |
180 | self.items.hash(hasher) | |
181 | } | |
182 | } | |
183 | impl<I: Idx, K, V, C> HashStable<C> for SortedIndexMultiMap<I, K, V> | |
184 | where | |
185 | K: HashStable<C>, | |
186 | V: HashStable<C>, | |
187 | { | |
188 | fn hash_stable(&self, ctx: &mut C, hasher: &mut StableHasher) { | |
189 | self.items.hash_stable(ctx, hasher) | |
190 | } | |
191 | } | |
192 | ||
193 | impl<I: Idx, K: Ord, V> FromIterator<(K, V)> for SortedIndexMultiMap<I, K, V> { | |
194 | fn from_iter<J>(iter: J) -> Self | |
195 | where | |
196 | J: IntoIterator<Item = (K, V)>, | |
197 | { | |
198 | let items = IndexVec::from_iter(iter); | |
199 | let mut idx_sorted_by_item_key: Vec<_> = items.indices().collect(); | |
200 | ||
201 | // `sort_by_key` is stable, so insertion order is preserved for duplicate items. | |
202 | idx_sorted_by_item_key.sort_by_key(|&idx| &items[idx].0); | |
203 | ||
204 | SortedIndexMultiMap { items, idx_sorted_by_item_key } | |
205 | } | |
206 | } | |
207 | ||
208 | impl<I: Idx, K, V> std::ops::Index<I> for SortedIndexMultiMap<I, K, V> { | |
209 | type Output = V; | |
210 | ||
211 | fn index(&self, idx: I) -> &Self::Output { | |
212 | &self.items[idx].1 | |
213 | } | |
214 | } | |
215 | ||
216 | #[cfg(tests)] | |
217 | mod tests; |