1 // Copyright 2014-2015 The Rust Project Developers. See the COPYRIGHT
2 // file at the top-level directory of this distribution and at
3 // http://rust-lang.org/COPYRIGHT.
5 // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
6 // http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
7 // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
8 // option. This file may not be copied, modified, or distributed
9 // except according to those terms.
12 use self::SearchResult::*;
13 use self::VacantEntryState::*;
17 use cmp::{max, Eq, PartialEq};
19 use fmt::{self, Debug};
20 use hash::{Hash, SipHasher};
21 use iter::{self, Iterator, ExactSizeIterator, IntoIterator, FromIterator, Extend, Map};
23 use mem::{self, replace};
24 use ops::{Deref, FnMut, FnOnce, Index};
25 use option::Option::{self, Some, None};
26 use rand::{self, Rng};
38 use super::table::BucketState::{
42 use super::state::HashState;
44 const INITIAL_LOG2_CAP: usize = 5;
45 const INITIAL_CAPACITY: usize = 1 << INITIAL_LOG2_CAP; // 2^5
47 /// The default behavior of HashMap implements a load factor of 90.9%.
48 /// This behavior is characterized by the following condition:
50 /// - if size > 0.909 * capacity: grow the map
52 struct DefaultResizePolicy;
54 impl DefaultResizePolicy {
55 fn new() -> DefaultResizePolicy {
60 fn min_capacity(&self, usable_size: usize) -> usize {
61 // Here, we are rephrasing the logic by specifying the lower limit
64 // - if `cap < size * 1.1`: grow the map
68 /// An inverse of `min_capacity`, approximately.
70 fn usable_capacity(&self, cap: usize) -> usize {
71 // As the number of entries approaches usable capacity,
72 // min_capacity(size) must be smaller than the internal capacity,
73 // so that the map is not resized:
74 // `min_capacity(usable_capacity(x)) <= x`.
75 // The left-hand side can only be smaller due to flooring by integer
78 // This doesn't have to be checked for overflow since allocation size
79 // in bytes will overflow earlier than multiplication by 10.
85 fn test_resize_policy() {
86 let rp = DefaultResizePolicy;
88 assert!(rp.min_capacity(rp.usable_capacity(n)) <= n);
89 assert!(rp.usable_capacity(rp.min_capacity(n)) <= n);
93 // The main performance trick in this hashmap is called Robin Hood Hashing.
94 // It gains its excellent performance from one essential operation:
96 // If an insertion collides with an existing element, and that element's
97 // "probe distance" (how far away the element is from its ideal location)
98 // is higher than how far we've already probed, swap the elements.
100 // This massively lowers variance in probe distance, and allows us to get very
101 // high load factors with good performance. The 90% load factor I use is rather
104 // > Why a load factor of approximately 90%?
106 // In general, all the distances to initial buckets will converge on the mean.
107 // At a load factor of α, the odds of finding the target bucket after k
108 // probes is approximately 1-α^k. If we set this equal to 50% (since we converge
109 // on the mean) and set k=8 (64-byte cache line / 8-byte hash), α=0.92. I round
110 // this down to make the math easier on the CPU and avoid its FPU.
111 // Since on average we start the probing in the middle of a cache line, this
112 // strategy pulls in two cache lines of hashes on every lookup. I think that's
113 // pretty good, but if you want to trade off some space, it could go down to one
114 // cache line on average with an α of 0.84.
116 // > Wait, what? Where did you get 1-α^k from?
118 // On the first probe, your odds of a collision with an existing element is α.
119 // The odds of doing this twice in a row is approximately α^2. For three times,
120 // α^3, etc. Therefore, the odds of colliding k times is α^k. The odds of NOT
121 // colliding after k tries is 1-α^k.
123 // The paper from 1986 cited below mentions an implementation which keeps track
124 // of the distance-to-initial-bucket histogram. This approach is not suitable
125 // for modern architectures because it requires maintaining an internal data
126 // structure. This allows very good first guesses, but we are most concerned
127 // with guessing entire cache lines, not individual indexes. Furthermore, array
128 // accesses are no longer linear and in one direction, as we have now. There
129 // is also memory and cache pressure that this would entail that would be very
130 // difficult to properly see in a microbenchmark.
132 // ## Future Improvements (FIXME!)
134 // Allow the load factor to be changed dynamically and/or at initialization.
136 // Also, would it be possible for us to reuse storage when growing the
137 // underlying table? This is exactly the use case for 'realloc', and may
138 // be worth exploring.
140 // ## Future Optimizations (FIXME!)
142 // Another possible design choice that I made without any real reason is
143 // parameterizing the raw table over keys and values. Technically, all we need
144 // is the size and alignment of keys and values, and the code should be just as
145 // efficient (well, we might need one for power-of-two size and one for not...).
146 // This has the potential to reduce code bloat in rust executables, without
147 // really losing anything except 4 words (key size, key alignment, val size,
148 // val alignment) which can be passed in to every call of a `RawTable` function.
149 // This would definitely be an avenue worth exploring if people start complaining
150 // about the size of rust executables.
152 // Annotate exceedingly likely branches in `table::make_hash`
153 // and `search_hashed` to reduce instruction cache pressure
154 // and mispredictions once it becomes possible (blocked on issue #11092).
156 // Shrinking the table could simply reallocate in place after moving buckets
157 // to the first half.
159 // The growth algorithm (fragment of the Proof of Correctness)
160 // --------------------
162 // The growth algorithm is basically a fast path of the naive reinsertion-
163 // during-resize algorithm. Other paths should never be taken.
165 // Consider growing a robin hood hashtable of capacity n. Normally, we do this
166 // by allocating a new table of capacity `2n`, and then individually reinsert
167 // each element in the old table into the new one. This guarantees that the
168 // new table is a valid robin hood hashtable with all the desired statistical
169 // properties. Remark that the order we reinsert the elements in should not
170 // matter. For simplicity and efficiency, we will consider only linear
171 // reinsertions, which consist of reinserting all elements in the old table
172 // into the new one by increasing order of index. However we will not be
173 // starting our reinsertions from index 0 in general. If we start from index
174 // i, for the purpose of reinsertion we will consider all elements with real
175 // index j < i to have virtual index n + j.
177 // Our hash generation scheme consists of generating a 64-bit hash and
178 // truncating the most significant bits. When moving to the new table, we
179 // simply introduce a new bit to the front of the hash. Therefore, if an
180 // elements has ideal index i in the old table, it can have one of two ideal
181 // locations in the new table. If the new bit is 0, then the new ideal index
182 // is i. If the new bit is 1, then the new ideal index is n + i. Intuitively,
183 // we are producing two independent tables of size n, and for each element we
184 // independently choose which table to insert it into with equal probability.
185 // However the rather than wrapping around themselves on overflowing their
186 // indexes, the first table overflows into the first, and the first into the
187 // second. Visually, our new table will look something like:
189 // [yy_xxx_xxxx_xxx|xx_yyy_yyyy_yyy]
191 // Where x's are elements inserted into the first table, y's are elements
192 // inserted into the second, and _'s are empty sections. We now define a few
193 // key concepts that we will use later. Note that this is a very abstract
194 // perspective of the table. A real resized table would be at least half
197 // Theorem: A linear robin hood reinsertion from the first ideal element
198 // produces identical results to a linear naive reinsertion from the same
201 // FIXME(Gankro, pczarn): review the proof and put it all in a separate README.md
203 /// A hash map implementation which uses linear probing with Robin
204 /// Hood bucket stealing.
206 /// The hashes are all keyed by the thread-local random number generator
207 /// on creation by default. This means that the ordering of the keys is
208 /// randomized, but makes the tables more resistant to
209 /// denial-of-service attacks (Hash DoS). This behavior can be
210 /// overridden with one of the constructors.
212 /// It is required that the keys implement the `Eq` and `Hash` traits, although
213 /// this can frequently be achieved by using `#[derive(PartialEq, Eq, Hash)]`.
214 /// If you implement these yourself, it is important that the following
218 /// k1 == k2 -> hash(k1) == hash(k2)
221 /// In other words, if two keys are equal, their hashes must be equal.
223 /// It is a logic error for a key to be modified in such a way that the key's
224 /// hash, as determined by the `Hash` trait, or its equality, as determined by
225 /// the `Eq` trait, changes while it is in the map. This is normally only
226 /// possible through `Cell`, `RefCell`, global state, I/O, or unsafe code.
228 /// Relevant papers/articles:
230 /// 1. Pedro Celis. ["Robin Hood Hashing"](https://cs.uwaterloo.ca/research/tr/1986/CS-86-14.pdf)
231 /// 2. Emmanuel Goossaert. ["Robin Hood
232 /// hashing"](http://codecapsule.com/2013/11/11/robin-hood-hashing/)
233 /// 3. Emmanuel Goossaert. ["Robin Hood hashing: backward shift
234 /// deletion"](http://codecapsule.com/2013/11/17/robin-hood-hashing-backward-shift-deletion/)
239 /// use std::collections::HashMap;
241 /// // type inference lets us omit an explicit type signature (which
242 /// // would be `HashMap<&str, &str>` in this example).
243 /// let mut book_reviews = HashMap::new();
245 /// // review some books.
246 /// book_reviews.insert("Adventures of Huckleberry Finn", "My favorite book.");
247 /// book_reviews.insert("Grimms' Fairy Tales", "Masterpiece.");
248 /// book_reviews.insert("Pride and Prejudice", "Very enjoyable.");
249 /// book_reviews.insert("The Adventures of Sherlock Holmes", "Eye lyked it alot.");
251 /// // check for a specific one.
252 /// if !book_reviews.contains_key("Les Misérables") {
253 /// println!("We've got {} reviews, but Les Misérables ain't one.",
254 /// book_reviews.len());
257 /// // oops, this review has a lot of spelling mistakes, let's delete it.
258 /// book_reviews.remove("The Adventures of Sherlock Holmes");
260 /// // look up the values associated with some keys.
261 /// let to_find = ["Pride and Prejudice", "Alice's Adventure in Wonderland"];
262 /// for book in &to_find {
263 /// match book_reviews.get(book) {
264 /// Some(review) => println!("{}: {}", book, review),
265 /// None => println!("{} is unreviewed.", book)
269 /// // iterate over everything.
270 /// for (book, review) in &book_reviews {
271 /// println!("{}: \"{}\"", book, review);
275 /// The easiest way to use `HashMap` with a custom type as key is to derive `Eq` and `Hash`.
276 /// We must also derive `PartialEq`.
279 /// use std::collections::HashMap;
281 /// #[derive(Hash, Eq, PartialEq, Debug)]
288 /// /// Create a new Viking.
289 /// fn new(name: &str, country: &str) -> Viking {
290 /// Viking { name: name.to_string(), country: country.to_string() }
294 /// // Use a HashMap to store the vikings' health points.
295 /// let mut vikings = HashMap::new();
297 /// vikings.insert(Viking::new("Einar", "Norway"), 25);
298 /// vikings.insert(Viking::new("Olaf", "Denmark"), 24);
299 /// vikings.insert(Viking::new("Harald", "Iceland"), 12);
301 /// // Use derived implementation to print the status of the vikings.
302 /// for (viking, health) in &vikings {
303 /// println!("{:?} has {} hp", viking, health);
307 #[stable(feature = "rust1", since = "1.0.0")]
308 pub struct HashMap<K, V, S = RandomState> {
309 // All hashes are keyed on these values, to prevent hash collision attacks.
312 table: RawTable<K, V>,
314 resize_policy: DefaultResizePolicy,
317 /// Search for a pre-hashed key.
318 fn search_hashed<K, V, M, F>(table: M,
321 -> SearchResult<K, V, M> where
322 M: Deref<Target=RawTable<K, V>>,
323 F: FnMut(&K) -> bool,
325 // This is the only function where capacity can be zero. To avoid
326 // undefined behavior when Bucket::new gets the raw bucket in this
327 // case, immediately return the appropriate search result.
328 if table.capacity() == 0 {
329 return TableRef(table);
332 let size = table.size();
333 let mut probe = Bucket::new(table, hash);
334 let ib = probe.index();
336 while probe.index() != ib + size {
337 let full = match probe.peek() {
338 Empty(b) => return TableRef(b.into_table()), // hit an empty bucket
342 if full.distance() + ib < full.index() {
343 // We can finish the search early if we hit any bucket
344 // with a lower distance to initial bucket than we've probed.
345 return TableRef(full.into_table());
348 // If the hash doesn't match, it can't be this one..
349 if hash == full.hash() {
350 // If the key doesn't match, it can't be this one..
351 if is_match(full.read().0) {
352 return FoundExisting(full);
359 TableRef(probe.into_table())
362 fn pop_internal<K, V>(starting_bucket: FullBucketMut<K, V>) -> (K, V) {
363 let (empty, retkey, retval) = starting_bucket.take();
364 let mut gap = match empty.gap_peek() {
366 None => return (retkey, retval)
369 while gap.full().distance() != 0 {
370 gap = match gap.shift() {
376 // Now we've done all our shifting. Return the value we grabbed earlier.
380 /// Perform robin hood bucket stealing at the given `bucket`. You must
381 /// also pass the position of that bucket's initial bucket so we don't have
382 /// to recalculate it.
384 /// `hash`, `k`, and `v` are the elements to "robin hood" into the hashtable.
385 fn robin_hood<'a, K: 'a, V: 'a>(mut bucket: FullBucketMut<'a, K, V>,
391 let starting_index = bucket.index();
393 let table = bucket.table(); // FIXME "lifetime too short".
396 // There can be at most `size - dib` buckets to displace, because
397 // in the worst case, there are `size` elements and we already are
398 // `distance` buckets away from the initial one.
399 let idx_end = starting_index + size - bucket.distance();
402 let (old_hash, old_key, old_val) = bucket.replace(hash, k, v);
404 let probe = bucket.next();
405 assert!(probe.index() != idx_end);
407 let full_bucket = match probe.peek() {
410 let b = bucket.put(old_hash, old_key, old_val);
411 // Now that it's stolen, just read the value's pointer
412 // right out of the table!
413 return Bucket::at_index(b.into_table(), starting_index)
419 Full(bucket) => bucket
422 let probe_ib = full_bucket.index() - full_bucket.distance();
424 bucket = full_bucket;
426 // Robin hood! Steal the spot.
438 /// A result that works like Option<FullBucket<..>> but preserves
439 /// the reference that grants us access to the table in any case.
440 enum SearchResult<K, V, M> {
441 // This is an entry that holds the given key:
442 FoundExisting(FullBucket<K, V, M>),
444 // There was no such entry. The reference is given back:
448 impl<K, V, M> SearchResult<K, V, M> {
449 fn into_option(self) -> Option<FullBucket<K, V, M>> {
451 FoundExisting(bucket) => Some(bucket),
457 impl<K, V, S> HashMap<K, V, S>
458 where K: Eq + Hash, S: HashState
460 fn make_hash<X: ?Sized>(&self, x: &X) -> SafeHash where X: Hash {
461 table::make_hash(&self.hash_state, x)
464 /// Search for a key, yielding the index if it's found in the hashtable.
465 /// If you already have the hash for the key lying around, use
467 fn search<'a, Q: ?Sized>(&'a self, q: &Q) -> Option<FullBucketImm<'a, K, V>>
468 where K: Borrow<Q>, Q: Eq + Hash
470 let hash = self.make_hash(q);
471 search_hashed(&self.table, hash, |k| q.eq(k.borrow()))
475 fn search_mut<'a, Q: ?Sized>(&'a mut self, q: &Q) -> Option<FullBucketMut<'a, K, V>>
476 where K: Borrow<Q>, Q: Eq + Hash
478 let hash = self.make_hash(q);
479 search_hashed(&mut self.table, hash, |k| q.eq(k.borrow()))
483 // The caller should ensure that invariants by Robin Hood Hashing hold.
484 fn insert_hashed_ordered(&mut self, hash: SafeHash, k: K, v: V) {
485 let cap = self.table.capacity();
486 let mut buckets = Bucket::new(&mut self.table, hash);
487 let ib = buckets.index();
489 while buckets.index() != ib + cap {
490 // We don't need to compare hashes for value swap.
491 // Not even DIBs for Robin Hood.
492 buckets = match buckets.peek() {
494 empty.put(hash, k, v);
497 Full(b) => b.into_bucket()
501 panic!("Internal HashMap error: Out of space.");
505 impl<K: Hash + Eq, V> HashMap<K, V, RandomState> {
506 /// Creates an empty HashMap.
511 /// use std::collections::HashMap;
512 /// let mut map: HashMap<&str, isize> = HashMap::new();
515 #[stable(feature = "rust1", since = "1.0.0")]
516 pub fn new() -> HashMap<K, V, RandomState> {
520 /// Creates an empty hash map with the given initial capacity.
525 /// use std::collections::HashMap;
526 /// let mut map: HashMap<&str, isize> = HashMap::with_capacity(10);
529 #[stable(feature = "rust1", since = "1.0.0")]
530 pub fn with_capacity(capacity: usize) -> HashMap<K, V, RandomState> {
531 HashMap::with_capacity_and_hash_state(capacity, Default::default())
535 impl<K, V, S> HashMap<K, V, S>
536 where K: Eq + Hash, S: HashState
538 /// Creates an empty hashmap which will use the given hasher to hash keys.
540 /// The created map has the default initial capacity.
545 /// #![feature(hashmap_hasher)]
547 /// use std::collections::HashMap;
548 /// use std::collections::hash_map::RandomState;
550 /// let s = RandomState::new();
551 /// let mut map = HashMap::with_hash_state(s);
552 /// map.insert(1, 2);
555 #[unstable(feature = "hashmap_hasher", reason = "hasher stuff is unclear",
557 pub fn with_hash_state(hash_state: S) -> HashMap<K, V, S> {
559 hash_state: hash_state,
560 resize_policy: DefaultResizePolicy::new(),
561 table: RawTable::new(0),
565 /// Creates an empty HashMap with space for at least `capacity`
566 /// elements, using `hasher` to hash the keys.
568 /// Warning: `hasher` is normally randomly generated, and
569 /// is designed to allow HashMaps to be resistant to attacks that
570 /// cause many collisions and very poor performance. Setting it
571 /// manually using this function can expose a DoS attack vector.
576 /// #![feature(hashmap_hasher)]
578 /// use std::collections::HashMap;
579 /// use std::collections::hash_map::RandomState;
581 /// let s = RandomState::new();
582 /// let mut map = HashMap::with_capacity_and_hash_state(10, s);
583 /// map.insert(1, 2);
586 #[unstable(feature = "hashmap_hasher", reason = "hasher stuff is unclear",
588 pub fn with_capacity_and_hash_state(capacity: usize, hash_state: S)
589 -> HashMap<K, V, S> {
590 let resize_policy = DefaultResizePolicy::new();
591 let min_cap = max(INITIAL_CAPACITY, resize_policy.min_capacity(capacity));
592 let internal_cap = min_cap.checked_next_power_of_two().expect("capacity overflow");
593 assert!(internal_cap >= capacity, "capacity overflow");
595 hash_state: hash_state,
596 resize_policy: resize_policy,
597 table: RawTable::new(internal_cap),
601 /// Returns the number of elements the map can hold without reallocating.
603 /// This number is a lower bound; the `HashMap<K, V>` might be able to hold
604 /// more, but is guaranteed to be able to hold at least this many.
609 /// use std::collections::HashMap;
610 /// let map: HashMap<isize, isize> = HashMap::with_capacity(100);
611 /// assert!(map.capacity() >= 100);
614 #[stable(feature = "rust1", since = "1.0.0")]
615 pub fn capacity(&self) -> usize {
616 self.resize_policy.usable_capacity(self.table.capacity())
619 /// Reserves capacity for at least `additional` more elements to be inserted
620 /// in the `HashMap`. The collection may reserve more space to avoid
621 /// frequent reallocations.
625 /// Panics if the new allocation size overflows `usize`.
630 /// use std::collections::HashMap;
631 /// let mut map: HashMap<&str, isize> = HashMap::new();
634 #[stable(feature = "rust1", since = "1.0.0")]
635 pub fn reserve(&mut self, additional: usize) {
636 let new_size = self.len().checked_add(additional).expect("capacity overflow");
637 let min_cap = self.resize_policy.min_capacity(new_size);
639 // An invalid value shouldn't make us run out of space. This includes
640 // an overflow check.
641 assert!(new_size <= min_cap);
643 if self.table.capacity() < min_cap {
644 let new_capacity = max(min_cap.next_power_of_two(), INITIAL_CAPACITY);
645 self.resize(new_capacity);
649 /// Resizes the internal vectors to a new capacity. It's your responsibility to:
650 /// 1) Make sure the new capacity is enough for all the elements, accounting
651 /// for the load factor.
652 /// 2) Ensure new_capacity is a power of two or zero.
653 fn resize(&mut self, new_capacity: usize) {
654 assert!(self.table.size() <= new_capacity);
655 assert!(new_capacity.is_power_of_two() || new_capacity == 0);
657 let mut old_table = replace(&mut self.table, RawTable::new(new_capacity));
658 let old_size = old_table.size();
660 if old_table.capacity() == 0 || old_table.size() == 0 {
665 // Specialization of the other branch.
666 let mut bucket = Bucket::first(&mut old_table);
668 // "So a few of the first shall be last: for many be called,
671 // We'll most likely encounter a few buckets at the beginning that
672 // have their initial buckets near the end of the table. They were
673 // placed at the beginning as the probe wrapped around the table
674 // during insertion. We must skip forward to a bucket that won't
675 // get reinserted too early and won't unfairly steal others spot.
676 // This eliminates the need for robin hood.
678 bucket = match bucket.peek() {
680 if full.distance() == 0 {
681 // This bucket occupies its ideal spot.
682 // It indicates the start of another "cluster".
683 bucket = full.into_bucket();
686 // Leaving this bucket in the last cluster for later.
690 // Encountered a hole between clusters.
697 // This is how the buckets might be laid out in memory:
698 // ($ marks an initialized bucket)
700 // |$$$_$$$$$$_$$$$$|
702 // But we've skipped the entire initial cluster of buckets
703 // and will continue iteration in this order:
706 // ^ wrap around once end is reached
709 // ^ exit once table.size == 0
711 bucket = match bucket.peek() {
713 let h = bucket.hash();
714 let (b, k, v) = bucket.take();
715 self.insert_hashed_ordered(h, k, v);
717 let t = b.table(); // FIXME "lifetime too short".
718 if t.size() == 0 { break }
722 Empty(b) => b.into_bucket()
727 assert_eq!(self.table.size(), old_size);
730 /// Shrinks the capacity of the map as much as possible. It will drop
731 /// down as much as possible while maintaining the internal rules
732 /// and possibly leaving some space in accordance with the resize policy.
737 /// use std::collections::HashMap;
739 /// let mut map: HashMap<isize, isize> = HashMap::with_capacity(100);
740 /// map.insert(1, 2);
741 /// map.insert(3, 4);
742 /// assert!(map.capacity() >= 100);
743 /// map.shrink_to_fit();
744 /// assert!(map.capacity() >= 2);
746 #[stable(feature = "rust1", since = "1.0.0")]
747 pub fn shrink_to_fit(&mut self) {
748 let min_capacity = self.resize_policy.min_capacity(self.len());
749 let min_capacity = max(min_capacity.next_power_of_two(), INITIAL_CAPACITY);
751 // An invalid value shouldn't make us run out of space.
752 debug_assert!(self.len() <= min_capacity);
754 if self.table.capacity() != min_capacity {
755 let old_table = replace(&mut self.table, RawTable::new(min_capacity));
756 let old_size = old_table.size();
758 // Shrink the table. Naive algorithm for resizing:
759 for (h, k, v) in old_table.into_iter() {
760 self.insert_hashed_nocheck(h, k, v);
763 debug_assert_eq!(self.table.size(), old_size);
767 /// Insert a pre-hashed key-value pair, without first checking
768 /// that there's enough room in the buckets. Returns a reference to the
769 /// newly insert value.
771 /// If the key already exists, the hashtable will be returned untouched
772 /// and a reference to the existing element will be returned.
773 fn insert_hashed_nocheck(&mut self, hash: SafeHash, k: K, v: V) -> &mut V {
774 self.insert_or_replace_with(hash, k, v, |_, _, _, _| ())
777 fn insert_or_replace_with<'a, F>(&'a mut self,
781 mut found_existing: F)
783 F: FnMut(&mut K, &mut V, K, V),
785 // Worst case, we'll find one empty bucket among `size + 1` buckets.
786 let size = self.table.size();
787 let mut probe = Bucket::new(&mut self.table, hash);
788 let ib = probe.index();
791 let mut bucket = match probe.peek() {
794 return bucket.put(hash, k, v).into_mut_refs().1;
796 Full(bucket) => bucket
800 if bucket.hash() == hash {
802 if k == *bucket.read_mut().0 {
803 let (bucket_k, bucket_v) = bucket.into_mut_refs();
804 debug_assert!(k == *bucket_k);
805 // Key already exists. Get its reference.
806 found_existing(bucket_k, bucket_v, k, v);
811 let robin_ib = bucket.index() as isize - bucket.distance() as isize;
813 if (ib as isize) < robin_ib {
814 // Found a luckier bucket than me. Better steal his spot.
815 return robin_hood(bucket, robin_ib as usize, hash, k, v);
818 probe = bucket.next();
819 assert!(probe.index() != ib + size + 1);
823 /// An iterator visiting all keys in arbitrary order.
824 /// Iterator element type is `&'a K`.
829 /// use std::collections::HashMap;
831 /// let mut map = HashMap::new();
832 /// map.insert("a", 1);
833 /// map.insert("b", 2);
834 /// map.insert("c", 3);
836 /// for key in map.keys() {
837 /// println!("{}", key);
840 #[stable(feature = "rust1", since = "1.0.0")]
841 pub fn keys<'a>(&'a self) -> Keys<'a, K, V> {
842 fn first<A, B>((a, _): (A, B)) -> A { a }
843 let first: fn((&'a K,&'a V)) -> &'a K = first; // coerce to fn ptr
845 Keys { inner: self.iter().map(first) }
848 /// An iterator visiting all values in arbitrary order.
849 /// Iterator element type is `&'a V`.
854 /// use std::collections::HashMap;
856 /// let mut map = HashMap::new();
857 /// map.insert("a", 1);
858 /// map.insert("b", 2);
859 /// map.insert("c", 3);
861 /// for val in map.values() {
862 /// println!("{}", val);
865 #[stable(feature = "rust1", since = "1.0.0")]
866 pub fn values<'a>(&'a self) -> Values<'a, K, V> {
867 fn second<A, B>((_, b): (A, B)) -> B { b }
868 let second: fn((&'a K,&'a V)) -> &'a V = second; // coerce to fn ptr
870 Values { inner: self.iter().map(second) }
873 /// An iterator visiting all key-value pairs in arbitrary order.
874 /// Iterator element type is `(&'a K, &'a V)`.
879 /// use std::collections::HashMap;
881 /// let mut map = HashMap::new();
882 /// map.insert("a", 1);
883 /// map.insert("b", 2);
884 /// map.insert("c", 3);
886 /// for (key, val) in map.iter() {
887 /// println!("key: {} val: {}", key, val);
890 #[stable(feature = "rust1", since = "1.0.0")]
891 pub fn iter(&self) -> Iter<K, V> {
892 Iter { inner: self.table.iter() }
895 /// An iterator visiting all key-value pairs in arbitrary order,
896 /// with mutable references to the values.
897 /// Iterator element type is `(&'a K, &'a mut V)`.
902 /// use std::collections::HashMap;
904 /// let mut map = HashMap::new();
905 /// map.insert("a", 1);
906 /// map.insert("b", 2);
907 /// map.insert("c", 3);
909 /// // Update all values
910 /// for (_, val) in map.iter_mut() {
914 /// for (key, val) in &map {
915 /// println!("key: {} val: {}", key, val);
918 #[stable(feature = "rust1", since = "1.0.0")]
919 pub fn iter_mut(&mut self) -> IterMut<K, V> {
920 IterMut { inner: self.table.iter_mut() }
923 /// Gets the given key's corresponding entry in the map for in-place manipulation.
928 /// use std::collections::HashMap;
930 /// let mut letters = HashMap::new();
932 /// for ch in "a short treatise on fungi".chars() {
933 /// let counter = letters.entry(ch).or_insert(0);
937 /// assert_eq!(letters[&'s'], 2);
938 /// assert_eq!(letters[&'t'], 3);
939 /// assert_eq!(letters[&'u'], 1);
940 /// assert_eq!(letters.get(&'y'), None);
942 #[stable(feature = "rust1", since = "1.0.0")]
943 pub fn entry(&mut self, key: K) -> Entry<K, V> {
947 let hash = self.make_hash(&key);
948 search_entry_hashed(&mut self.table, hash, key)
951 /// Returns the number of elements in the map.
956 /// use std::collections::HashMap;
958 /// let mut a = HashMap::new();
959 /// assert_eq!(a.len(), 0);
960 /// a.insert(1, "a");
961 /// assert_eq!(a.len(), 1);
963 #[stable(feature = "rust1", since = "1.0.0")]
964 pub fn len(&self) -> usize { self.table.size() }
966 /// Returns true if the map contains no elements.
971 /// use std::collections::HashMap;
973 /// let mut a = HashMap::new();
974 /// assert!(a.is_empty());
975 /// a.insert(1, "a");
976 /// assert!(!a.is_empty());
979 #[stable(feature = "rust1", since = "1.0.0")]
980 pub fn is_empty(&self) -> bool { self.len() == 0 }
982 /// Clears the map, returning all key-value pairs as an iterator. Keeps the
983 /// allocated memory for reuse.
988 /// use std::collections::HashMap;
990 /// let mut a = HashMap::new();
991 /// a.insert(1, "a");
992 /// a.insert(2, "b");
994 /// for (k, v) in a.drain().take(1) {
995 /// assert!(k == 1 || k == 2);
996 /// assert!(v == "a" || v == "b");
999 /// assert!(a.is_empty());
1002 #[stable(feature = "drain", since = "1.6.0")]
1003 pub fn drain(&mut self) -> Drain<K, V> {
1004 fn last_two<A, B, C>((_, b, c): (A, B, C)) -> (B, C) { (b, c) }
1005 let last_two: fn((SafeHash, K, V)) -> (K, V) = last_two; // coerce to fn pointer
1008 inner: self.table.drain().map(last_two),
1012 /// Clears the map, removing all key-value pairs. Keeps the allocated memory
1018 /// use std::collections::HashMap;
1020 /// let mut a = HashMap::new();
1021 /// a.insert(1, "a");
1023 /// assert!(a.is_empty());
1025 #[stable(feature = "rust1", since = "1.0.0")]
1027 pub fn clear(&mut self) {
1031 /// Returns a reference to the value corresponding to the key.
1033 /// The key may be any borrowed form of the map's key type, but
1034 /// `Hash` and `Eq` on the borrowed form *must* match those for
1040 /// use std::collections::HashMap;
1042 /// let mut map = HashMap::new();
1043 /// map.insert(1, "a");
1044 /// assert_eq!(map.get(&1), Some(&"a"));
1045 /// assert_eq!(map.get(&2), None);
1047 #[stable(feature = "rust1", since = "1.0.0")]
1048 pub fn get<Q: ?Sized>(&self, k: &Q) -> Option<&V>
1049 where K: Borrow<Q>, Q: Hash + Eq
1051 self.search(k).map(|bucket| bucket.into_refs().1)
1054 /// Returns true if the map contains a value for the specified key.
1056 /// The key may be any borrowed form of the map's key type, but
1057 /// `Hash` and `Eq` on the borrowed form *must* match those for
1063 /// use std::collections::HashMap;
1065 /// let mut map = HashMap::new();
1066 /// map.insert(1, "a");
1067 /// assert_eq!(map.contains_key(&1), true);
1068 /// assert_eq!(map.contains_key(&2), false);
1070 #[stable(feature = "rust1", since = "1.0.0")]
1071 pub fn contains_key<Q: ?Sized>(&self, k: &Q) -> bool
1072 where K: Borrow<Q>, Q: Hash + Eq
1074 self.search(k).is_some()
1077 /// Returns a mutable reference to the value corresponding to the key.
1079 /// The key may be any borrowed form of the map's key type, but
1080 /// `Hash` and `Eq` on the borrowed form *must* match those for
1086 /// use std::collections::HashMap;
1088 /// let mut map = HashMap::new();
1089 /// map.insert(1, "a");
1090 /// if let Some(x) = map.get_mut(&1) {
1093 /// assert_eq!(map[&1], "b");
1095 #[stable(feature = "rust1", since = "1.0.0")]
1096 pub fn get_mut<Q: ?Sized>(&mut self, k: &Q) -> Option<&mut V>
1097 where K: Borrow<Q>, Q: Hash + Eq
1099 self.search_mut(k).map(|bucket| bucket.into_mut_refs().1)
1102 /// Inserts a key-value pair into the map.
1104 /// If the map did not have this key present, `None` is returned.
1106 /// If the map did have this key present, the key is not updated, the
1107 /// value is updated and the old value is returned.
1108 /// See the [module-level documentation] for more.
1110 /// [module-level documentation]: index.html#insert-and-complex-keys
1115 /// use std::collections::HashMap;
1117 /// let mut map = HashMap::new();
1118 /// assert_eq!(map.insert(37, "a"), None);
1119 /// assert_eq!(map.is_empty(), false);
1121 /// map.insert(37, "b");
1122 /// assert_eq!(map.insert(37, "c"), Some("b"));
1123 /// assert_eq!(map[&37], "c");
1125 #[stable(feature = "rust1", since = "1.0.0")]
1126 pub fn insert(&mut self, k: K, v: V) -> Option<V> {
1127 let hash = self.make_hash(&k);
1130 let mut retval = None;
1131 self.insert_or_replace_with(hash, k, v, |_, val_ref, _, val| {
1132 retval = Some(replace(val_ref, val));
1137 /// Removes a key from the map, returning the value at the key if the key
1138 /// was previously in the map.
1140 /// The key may be any borrowed form of the map's key type, but
1141 /// `Hash` and `Eq` on the borrowed form *must* match those for
1147 /// use std::collections::HashMap;
1149 /// let mut map = HashMap::new();
1150 /// map.insert(1, "a");
1151 /// assert_eq!(map.remove(&1), Some("a"));
1152 /// assert_eq!(map.remove(&1), None);
1154 #[stable(feature = "rust1", since = "1.0.0")]
1155 pub fn remove<Q: ?Sized>(&mut self, k: &Q) -> Option<V>
1156 where K: Borrow<Q>, Q: Hash + Eq
1158 if self.table.size() == 0 {
1162 self.search_mut(k).map(|bucket| pop_internal(bucket).1)
1166 fn search_entry_hashed<'a, K: Eq, V>(table: &'a mut RawTable<K,V>, hash: SafeHash, k: K)
1169 // Worst case, we'll find one empty bucket among `size + 1` buckets.
1170 let size = table.size();
1171 let mut probe = Bucket::new(table, hash);
1172 let ib = probe.index();
1175 let bucket = match probe.peek() {
1178 return Vacant(VacantEntry {
1181 elem: NoElem(bucket),
1184 Full(bucket) => bucket
1188 if bucket.hash() == hash {
1190 if k == *bucket.read().0 {
1191 return Occupied(OccupiedEntry{
1197 let robin_ib = bucket.index() as isize - bucket.distance() as isize;
1199 if (ib as isize) < robin_ib {
1200 // Found a luckier bucket than me. Better steal his spot.
1201 return Vacant(VacantEntry {
1204 elem: NeqElem(bucket, robin_ib as usize),
1208 probe = bucket.next();
1209 assert!(probe.index() != ib + size + 1);
1213 #[stable(feature = "rust1", since = "1.0.0")]
1214 impl<K, V, S> PartialEq for HashMap<K, V, S>
1215 where K: Eq + Hash, V: PartialEq, S: HashState
1217 fn eq(&self, other: &HashMap<K, V, S>) -> bool {
1218 if self.len() != other.len() { return false; }
1220 self.iter().all(|(key, value)|
1221 other.get(key).map_or(false, |v| *value == *v)
1226 #[stable(feature = "rust1", since = "1.0.0")]
1227 impl<K, V, S> Eq for HashMap<K, V, S>
1228 where K: Eq + Hash, V: Eq, S: HashState
1231 #[stable(feature = "rust1", since = "1.0.0")]
1232 impl<K, V, S> Debug for HashMap<K, V, S>
1233 where K: Eq + Hash + Debug, V: Debug, S: HashState
1235 fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
1236 f.debug_map().entries(self.iter()).finish()
1240 #[stable(feature = "rust1", since = "1.0.0")]
1241 impl<K, V, S> Default for HashMap<K, V, S>
1243 S: HashState + Default,
1245 fn default() -> HashMap<K, V, S> {
1246 HashMap::with_hash_state(Default::default())
1250 #[stable(feature = "rust1", since = "1.0.0")]
1251 impl<'a, K, Q: ?Sized, V, S> Index<&'a Q> for HashMap<K, V, S>
1252 where K: Eq + Hash + Borrow<Q>,
1259 fn index(&self, index: &Q) -> &V {
1260 self.get(index).expect("no entry found for key")
1264 /// HashMap iterator.
1265 #[stable(feature = "rust1", since = "1.0.0")]
1266 pub struct Iter<'a, K: 'a, V: 'a> {
1267 inner: table::Iter<'a, K, V>
1270 // FIXME(#19839) Remove in favor of `#[derive(Clone)]`
1271 #[stable(feature = "rust1", since = "1.0.0")]
1272 impl<'a, K, V> Clone for Iter<'a, K, V> {
1273 fn clone(&self) -> Iter<'a, K, V> {
1275 inner: self.inner.clone()
1280 /// HashMap mutable values iterator.
1281 #[stable(feature = "rust1", since = "1.0.0")]
1282 pub struct IterMut<'a, K: 'a, V: 'a> {
1283 inner: table::IterMut<'a, K, V>
1286 /// HashMap move iterator.
1287 #[stable(feature = "rust1", since = "1.0.0")]
1288 pub struct IntoIter<K, V> {
1289 inner: iter::Map<table::IntoIter<K, V>, fn((SafeHash, K, V)) -> (K, V)>
1292 /// HashMap keys iterator.
1293 #[stable(feature = "rust1", since = "1.0.0")]
1294 pub struct Keys<'a, K: 'a, V: 'a> {
1295 inner: Map<Iter<'a, K, V>, fn((&'a K, &'a V)) -> &'a K>
1298 // FIXME(#19839) Remove in favor of `#[derive(Clone)]`
1299 #[stable(feature = "rust1", since = "1.0.0")]
1300 impl<'a, K, V> Clone for Keys<'a, K, V> {
1301 fn clone(&self) -> Keys<'a, K, V> {
1303 inner: self.inner.clone()
1308 /// HashMap values iterator.
1309 #[stable(feature = "rust1", since = "1.0.0")]
1310 pub struct Values<'a, K: 'a, V: 'a> {
1311 inner: Map<Iter<'a, K, V>, fn((&'a K, &'a V)) -> &'a V>
1314 // FIXME(#19839) Remove in favor of `#[derive(Clone)]`
1315 #[stable(feature = "rust1", since = "1.0.0")]
1316 impl<'a, K, V> Clone for Values<'a, K, V> {
1317 fn clone(&self) -> Values<'a, K, V> {
1319 inner: self.inner.clone()
1324 /// HashMap drain iterator.
1325 #[stable(feature = "drain", since = "1.6.0")]
1326 pub struct Drain<'a, K: 'a, V: 'a> {
1327 inner: iter::Map<table::Drain<'a, K, V>, fn((SafeHash, K, V)) -> (K, V)>
1330 /// A view into a single occupied location in a HashMap.
1331 #[stable(feature = "rust1", since = "1.0.0")]
1332 pub struct OccupiedEntry<'a, K: 'a, V: 'a> {
1333 elem: FullBucket<K, V, &'a mut RawTable<K, V>>,
1336 /// A view into a single empty location in a HashMap.
1337 #[stable(feature = "rust1", since = "1.0.0")]
1338 pub struct VacantEntry<'a, K: 'a, V: 'a> {
1341 elem: VacantEntryState<K, V, &'a mut RawTable<K, V>>,
1344 /// A view into a single location in a map, which may be vacant or occupied.
1345 #[stable(feature = "rust1", since = "1.0.0")]
1346 pub enum Entry<'a, K: 'a, V: 'a> {
1347 /// An occupied Entry.
1348 #[stable(feature = "rust1", since = "1.0.0")]
1350 #[cfg_attr(not(stage0), stable(feature = "rust1", since = "1.0.0"))] OccupiedEntry<'a, K, V>
1354 #[stable(feature = "rust1", since = "1.0.0")]
1356 #[cfg_attr(not(stage0), stable(feature = "rust1", since = "1.0.0"))] VacantEntry<'a, K, V>
1360 /// Possible states of a VacantEntry.
1361 enum VacantEntryState<K, V, M> {
1362 /// The index is occupied, but the key to insert has precedence,
1363 /// and will kick the current one out on insertion.
1364 NeqElem(FullBucket<K, V, M>, usize),
1365 /// The index is genuinely vacant.
1366 NoElem(EmptyBucket<K, V, M>),
1369 #[stable(feature = "rust1", since = "1.0.0")]
1370 impl<'a, K, V, S> IntoIterator for &'a HashMap<K, V, S>
1371 where K: Eq + Hash, S: HashState
1373 type Item = (&'a K, &'a V);
1374 type IntoIter = Iter<'a, K, V>;
1376 fn into_iter(self) -> Iter<'a, K, V> {
1381 #[stable(feature = "rust1", since = "1.0.0")]
1382 impl<'a, K, V, S> IntoIterator for &'a mut HashMap<K, V, S>
1383 where K: Eq + Hash, S: HashState
1385 type Item = (&'a K, &'a mut V);
1386 type IntoIter = IterMut<'a, K, V>;
1388 fn into_iter(mut self) -> IterMut<'a, K, V> {
1393 #[stable(feature = "rust1", since = "1.0.0")]
1394 impl<K, V, S> IntoIterator for HashMap<K, V, S>
1395 where K: Eq + Hash, S: HashState
1398 type IntoIter = IntoIter<K, V>;
1400 /// Creates a consuming iterator, that is, one that moves each key-value
1401 /// pair out of the map in arbitrary order. The map cannot be used after
1407 /// use std::collections::HashMap;
1409 /// let mut map = HashMap::new();
1410 /// map.insert("a", 1);
1411 /// map.insert("b", 2);
1412 /// map.insert("c", 3);
1414 /// // Not possible with .iter()
1415 /// let vec: Vec<(&str, isize)> = map.into_iter().collect();
1417 fn into_iter(self) -> IntoIter<K, V> {
1418 fn last_two<A, B, C>((_, b, c): (A, B, C)) -> (B, C) { (b, c) }
1419 let last_two: fn((SafeHash, K, V)) -> (K, V) = last_two;
1422 inner: self.table.into_iter().map(last_two)
1427 #[stable(feature = "rust1", since = "1.0.0")]
1428 impl<'a, K, V> Iterator for Iter<'a, K, V> {
1429 type Item = (&'a K, &'a V);
1431 #[inline] fn next(&mut self) -> Option<(&'a K, &'a V)> { self.inner.next() }
1432 #[inline] fn size_hint(&self) -> (usize, Option<usize>) { self.inner.size_hint() }
1434 #[stable(feature = "rust1", since = "1.0.0")]
1435 impl<'a, K, V> ExactSizeIterator for Iter<'a, K, V> {
1436 #[inline] fn len(&self) -> usize { self.inner.len() }
1439 #[stable(feature = "rust1", since = "1.0.0")]
1440 impl<'a, K, V> Iterator for IterMut<'a, K, V> {
1441 type Item = (&'a K, &'a mut V);
1443 #[inline] fn next(&mut self) -> Option<(&'a K, &'a mut V)> { self.inner.next() }
1444 #[inline] fn size_hint(&self) -> (usize, Option<usize>) { self.inner.size_hint() }
1446 #[stable(feature = "rust1", since = "1.0.0")]
1447 impl<'a, K, V> ExactSizeIterator for IterMut<'a, K, V> {
1448 #[inline] fn len(&self) -> usize { self.inner.len() }
1451 #[stable(feature = "rust1", since = "1.0.0")]
1452 impl<K, V> Iterator for IntoIter<K, V> {
1455 #[inline] fn next(&mut self) -> Option<(K, V)> { self.inner.next() }
1456 #[inline] fn size_hint(&self) -> (usize, Option<usize>) { self.inner.size_hint() }
1458 #[stable(feature = "rust1", since = "1.0.0")]
1459 impl<K, V> ExactSizeIterator for IntoIter<K, V> {
1460 #[inline] fn len(&self) -> usize { self.inner.len() }
1463 #[stable(feature = "rust1", since = "1.0.0")]
1464 impl<'a, K, V> Iterator for Keys<'a, K, V> {
1467 #[inline] fn next(&mut self) -> Option<(&'a K)> { self.inner.next() }
1468 #[inline] fn size_hint(&self) -> (usize, Option<usize>) { self.inner.size_hint() }
1470 #[stable(feature = "rust1", since = "1.0.0")]
1471 impl<'a, K, V> ExactSizeIterator for Keys<'a, K, V> {
1472 #[inline] fn len(&self) -> usize { self.inner.len() }
1475 #[stable(feature = "rust1", since = "1.0.0")]
1476 impl<'a, K, V> Iterator for Values<'a, K, V> {
1479 #[inline] fn next(&mut self) -> Option<(&'a V)> { self.inner.next() }
1480 #[inline] fn size_hint(&self) -> (usize, Option<usize>) { self.inner.size_hint() }
1482 #[stable(feature = "rust1", since = "1.0.0")]
1483 impl<'a, K, V> ExactSizeIterator for Values<'a, K, V> {
1484 #[inline] fn len(&self) -> usize { self.inner.len() }
1487 #[stable(feature = "rust1", since = "1.0.0")]
1488 impl<'a, K, V> Iterator for Drain<'a, K, V> {
1491 #[inline] fn next(&mut self) -> Option<(K, V)> { self.inner.next() }
1492 #[inline] fn size_hint(&self) -> (usize, Option<usize>) { self.inner.size_hint() }
1494 #[stable(feature = "rust1", since = "1.0.0")]
1495 impl<'a, K, V> ExactSizeIterator for Drain<'a, K, V> {
1496 #[inline] fn len(&self) -> usize { self.inner.len() }
1499 impl<'a, K, V> Entry<'a, K, V> {
1500 #[stable(feature = "rust1", since = "1.0.0")]
1501 /// Ensures a value is in the entry by inserting the default if empty, and returns
1502 /// a mutable reference to the value in the entry.
1503 pub fn or_insert(self, default: V) -> &'a mut V {
1505 Occupied(entry) => entry.into_mut(),
1506 Vacant(entry) => entry.insert(default),
1510 #[stable(feature = "rust1", since = "1.0.0")]
1511 /// Ensures a value is in the entry by inserting the result of the default function if empty,
1512 /// and returns a mutable reference to the value in the entry.
1513 pub fn or_insert_with<F: FnOnce() -> V>(self, default: F) -> &'a mut V {
1515 Occupied(entry) => entry.into_mut(),
1516 Vacant(entry) => entry.insert(default()),
1521 impl<'a, K, V> OccupiedEntry<'a, K, V> {
1522 /// Gets a reference to the value in the entry.
1523 #[stable(feature = "rust1", since = "1.0.0")]
1524 pub fn get(&self) -> &V {
1528 /// Gets a mutable reference to the value in the entry.
1529 #[stable(feature = "rust1", since = "1.0.0")]
1530 pub fn get_mut(&mut self) -> &mut V {
1531 self.elem.read_mut().1
1534 /// Converts the OccupiedEntry into a mutable reference to the value in the entry
1535 /// with a lifetime bound to the map itself
1536 #[stable(feature = "rust1", since = "1.0.0")]
1537 pub fn into_mut(self) -> &'a mut V {
1538 self.elem.into_mut_refs().1
1541 /// Sets the value of the entry, and returns the entry's old value
1542 #[stable(feature = "rust1", since = "1.0.0")]
1543 pub fn insert(&mut self, mut value: V) -> V {
1544 let old_value = self.get_mut();
1545 mem::swap(&mut value, old_value);
1549 /// Takes the value out of the entry, and returns it
1550 #[stable(feature = "rust1", since = "1.0.0")]
1551 pub fn remove(self) -> V {
1552 pop_internal(self.elem).1
1556 impl<'a, K: 'a, V: 'a> VacantEntry<'a, K, V> {
1557 /// Sets the value of the entry with the VacantEntry's key,
1558 /// and returns a mutable reference to it
1559 #[stable(feature = "rust1", since = "1.0.0")]
1560 pub fn insert(self, value: V) -> &'a mut V {
1562 NeqElem(bucket, ib) => {
1563 robin_hood(bucket, ib, self.hash, self.key, value)
1566 bucket.put(self.hash, self.key, value).into_mut_refs().1
1572 #[stable(feature = "rust1", since = "1.0.0")]
1573 impl<K, V, S> FromIterator<(K, V)> for HashMap<K, V, S>
1574 where K: Eq + Hash, S: HashState + Default
1576 fn from_iter<T: IntoIterator<Item=(K, V)>>(iterable: T) -> HashMap<K, V, S> {
1577 let iter = iterable.into_iter();
1578 let lower = iter.size_hint().0;
1579 let mut map = HashMap::with_capacity_and_hash_state(lower,
1580 Default::default());
1586 #[stable(feature = "rust1", since = "1.0.0")]
1587 impl<K, V, S> Extend<(K, V)> for HashMap<K, V, S>
1588 where K: Eq + Hash, S: HashState
1590 fn extend<T: IntoIterator<Item=(K, V)>>(&mut self, iter: T) {
1591 for (k, v) in iter {
1597 #[stable(feature = "hash_extend_copy", since = "1.4.0")]
1598 impl<'a, K, V, S> Extend<(&'a K, &'a V)> for HashMap<K, V, S>
1599 where K: Eq + Hash + Copy, V: Copy, S: HashState
1601 fn extend<T: IntoIterator<Item=(&'a K, &'a V)>>(&mut self, iter: T) {
1602 self.extend(iter.into_iter().map(|(&key, &value)| (key, value)));
1606 /// `RandomState` is the default state for `HashMap` types.
1608 /// A particular instance `RandomState` will create the same instances of
1609 /// `Hasher`, but the hashers created by two different `RandomState`
1610 /// instances are unlikely to produce the same result for the same values.
1612 #[unstable(feature = "hashmap_hasher",
1613 reason = "hashing an hash maps may be altered",
1615 pub struct RandomState {
1620 #[unstable(feature = "hashmap_hasher",
1621 reason = "hashing an hash maps may be altered",
1624 /// Constructs a new `RandomState` that is initialized with random keys.
1626 #[allow(deprecated)] // rand
1627 pub fn new() -> RandomState {
1628 let mut r = rand::thread_rng();
1629 RandomState { k0: r.gen(), k1: r.gen() }
1633 #[unstable(feature = "hashmap_hasher",
1634 reason = "hashing an hash maps may be altered",
1636 impl HashState for RandomState {
1637 type Hasher = SipHasher;
1639 fn hasher(&self) -> SipHasher {
1640 SipHasher::new_with_keys(self.k0, self.k1)
1644 #[stable(feature = "rust1", since = "1.0.0")]
1645 impl Default for RandomState {
1647 fn default() -> RandomState {
1652 impl<K, S, Q: ?Sized> super::Recover<Q> for HashMap<K, (), S>
1653 where K: Eq + Hash + Borrow<Q>, S: HashState, Q: Eq + Hash
1657 fn get(&self, key: &Q) -> Option<&K> {
1658 self.search(key).map(|bucket| bucket.into_refs().0)
1661 fn take(&mut self, key: &Q) -> Option<K> {
1662 if self.table.size() == 0 {
1666 self.search_mut(key).map(|bucket| pop_internal(bucket).0)
1669 fn replace(&mut self, key: K) -> Option<K> {
1670 let hash = self.make_hash(&key);
1673 let mut retkey = None;
1674 self.insert_or_replace_with(hash, key, (), |key_ref, _, key, _| {
1675 retkey = Some(replace(key_ref, key));
1686 use super::Entry::{Occupied, Vacant};
1688 use rand::{thread_rng, Rng};
1691 fn test_create_capacity_zero() {
1692 let mut m = HashMap::with_capacity(0);
1694 assert!(m.insert(1, 1).is_none());
1696 assert!(m.contains_key(&1));
1697 assert!(!m.contains_key(&0));
1702 let mut m = HashMap::new();
1703 assert_eq!(m.len(), 0);
1704 assert!(m.insert(1, 2).is_none());
1705 assert_eq!(m.len(), 1);
1706 assert!(m.insert(2, 4).is_none());
1707 assert_eq!(m.len(), 2);
1708 assert_eq!(*m.get(&1).unwrap(), 2);
1709 assert_eq!(*m.get(&2).unwrap(), 4);
1712 thread_local! { static DROP_VECTOR: RefCell<Vec<isize>> = RefCell::new(Vec::new()) }
1714 #[derive(Hash, PartialEq, Eq)]
1720 fn new(k: usize) -> Dropable {
1721 DROP_VECTOR.with(|slot| {
1722 slot.borrow_mut()[k] += 1;
1729 impl Drop for Dropable {
1730 fn drop(&mut self) {
1731 DROP_VECTOR.with(|slot| {
1732 slot.borrow_mut()[self.k] -= 1;
1737 impl Clone for Dropable {
1738 fn clone(&self) -> Dropable {
1739 Dropable::new(self.k)
1745 DROP_VECTOR.with(|slot| {
1746 *slot.borrow_mut() = vec![0; 200];
1750 let mut m = HashMap::new();
1752 DROP_VECTOR.with(|v| {
1754 assert_eq!(v.borrow()[i], 0);
1759 let d1 = Dropable::new(i);
1760 let d2 = Dropable::new(i+100);
1764 DROP_VECTOR.with(|v| {
1766 assert_eq!(v.borrow()[i], 1);
1771 let k = Dropable::new(i);
1772 let v = m.remove(&k);
1774 assert!(v.is_some());
1776 DROP_VECTOR.with(|v| {
1777 assert_eq!(v.borrow()[i], 1);
1778 assert_eq!(v.borrow()[i+100], 1);
1782 DROP_VECTOR.with(|v| {
1784 assert_eq!(v.borrow()[i], 0);
1785 assert_eq!(v.borrow()[i+100], 0);
1789 assert_eq!(v.borrow()[i], 1);
1790 assert_eq!(v.borrow()[i+100], 1);
1795 DROP_VECTOR.with(|v| {
1797 assert_eq!(v.borrow()[i], 0);
1803 fn test_move_iter_drops() {
1804 DROP_VECTOR.with(|v| {
1805 *v.borrow_mut() = vec![0; 200];
1809 let mut hm = HashMap::new();
1811 DROP_VECTOR.with(|v| {
1813 assert_eq!(v.borrow()[i], 0);
1818 let d1 = Dropable::new(i);
1819 let d2 = Dropable::new(i+100);
1823 DROP_VECTOR.with(|v| {
1825 assert_eq!(v.borrow()[i], 1);
1832 // By the way, ensure that cloning doesn't screw up the dropping.
1836 let mut half = hm.into_iter().take(50);
1838 DROP_VECTOR.with(|v| {
1840 assert_eq!(v.borrow()[i], 1);
1844 for _ in half.by_ref() {}
1846 DROP_VECTOR.with(|v| {
1847 let nk = (0..100).filter(|&i| {
1851 let nv = (0..100).filter(|&i| {
1852 v.borrow()[i+100] == 1
1860 DROP_VECTOR.with(|v| {
1862 assert_eq!(v.borrow()[i], 0);
1868 fn test_empty_pop() {
1869 let mut m: HashMap<isize, bool> = HashMap::new();
1870 assert_eq!(m.remove(&0), None);
1874 fn test_lots_of_insertions() {
1875 let mut m = HashMap::new();
1877 // Try this a few times to make sure we never screw up the hashmap's
1880 assert!(m.is_empty());
1883 assert!(m.insert(i, i).is_none());
1887 assert_eq!(r, Some(&j));
1890 for j in i+1..1001 {
1892 assert_eq!(r, None);
1896 for i in 1001..2001 {
1897 assert!(!m.contains_key(&i));
1902 assert!(m.remove(&i).is_some());
1905 assert!(!m.contains_key(&j));
1908 for j in i+1..1001 {
1909 assert!(m.contains_key(&j));
1914 assert!(!m.contains_key(&i));
1918 assert!(m.insert(i, i).is_none());
1922 for i in (1..1001).rev() {
1923 assert!(m.remove(&i).is_some());
1926 assert!(!m.contains_key(&j));
1930 assert!(m.contains_key(&j));
1937 fn test_find_mut() {
1938 let mut m = HashMap::new();
1939 assert!(m.insert(1, 12).is_none());
1940 assert!(m.insert(2, 8).is_none());
1941 assert!(m.insert(5, 14).is_none());
1943 match m.get_mut(&5) {
1944 None => panic!(), Some(x) => *x = new
1946 assert_eq!(m.get(&5), Some(&new));
1950 fn test_insert_overwrite() {
1951 let mut m = HashMap::new();
1952 assert!(m.insert(1, 2).is_none());
1953 assert_eq!(*m.get(&1).unwrap(), 2);
1954 assert!(!m.insert(1, 3).is_none());
1955 assert_eq!(*m.get(&1).unwrap(), 3);
1959 fn test_insert_conflicts() {
1960 let mut m = HashMap::with_capacity(4);
1961 assert!(m.insert(1, 2).is_none());
1962 assert!(m.insert(5, 3).is_none());
1963 assert!(m.insert(9, 4).is_none());
1964 assert_eq!(*m.get(&9).unwrap(), 4);
1965 assert_eq!(*m.get(&5).unwrap(), 3);
1966 assert_eq!(*m.get(&1).unwrap(), 2);
1970 fn test_conflict_remove() {
1971 let mut m = HashMap::with_capacity(4);
1972 assert!(m.insert(1, 2).is_none());
1973 assert_eq!(*m.get(&1).unwrap(), 2);
1974 assert!(m.insert(5, 3).is_none());
1975 assert_eq!(*m.get(&1).unwrap(), 2);
1976 assert_eq!(*m.get(&5).unwrap(), 3);
1977 assert!(m.insert(9, 4).is_none());
1978 assert_eq!(*m.get(&1).unwrap(), 2);
1979 assert_eq!(*m.get(&5).unwrap(), 3);
1980 assert_eq!(*m.get(&9).unwrap(), 4);
1981 assert!(m.remove(&1).is_some());
1982 assert_eq!(*m.get(&9).unwrap(), 4);
1983 assert_eq!(*m.get(&5).unwrap(), 3);
1987 fn test_is_empty() {
1988 let mut m = HashMap::with_capacity(4);
1989 assert!(m.insert(1, 2).is_none());
1990 assert!(!m.is_empty());
1991 assert!(m.remove(&1).is_some());
1992 assert!(m.is_empty());
1997 let mut m = HashMap::new();
1999 assert_eq!(m.remove(&1), Some(2));
2000 assert_eq!(m.remove(&1), None);
2005 let mut m = HashMap::with_capacity(4);
2007 assert!(m.insert(i, i*2).is_none());
2009 assert_eq!(m.len(), 32);
2011 let mut observed: u32 = 0;
2014 assert_eq!(*v, *k * 2);
2015 observed |= 1 << *k;
2017 assert_eq!(observed, 0xFFFF_FFFF);
2022 let vec = vec![(1, 'a'), (2, 'b'), (3, 'c')];
2023 let map: HashMap<_, _> = vec.into_iter().collect();
2024 let keys: Vec<_> = map.keys().cloned().collect();
2025 assert_eq!(keys.len(), 3);
2026 assert!(keys.contains(&1));
2027 assert!(keys.contains(&2));
2028 assert!(keys.contains(&3));
2033 let vec = vec![(1, 'a'), (2, 'b'), (3, 'c')];
2034 let map: HashMap<_, _> = vec.into_iter().collect();
2035 let values: Vec<_> = map.values().cloned().collect();
2036 assert_eq!(values.len(), 3);
2037 assert!(values.contains(&'a'));
2038 assert!(values.contains(&'b'));
2039 assert!(values.contains(&'c'));
2044 let mut m = HashMap::new();
2045 assert!(m.get(&1).is_none());
2049 Some(v) => assert_eq!(*v, 2)
2055 let mut m1 = HashMap::new();
2060 let mut m2 = HashMap::new();
2073 let mut map = HashMap::new();
2074 let empty: HashMap<i32, i32> = HashMap::new();
2079 let map_str = format!("{:?}", map);
2081 assert!(map_str == "{1: 2, 3: 4}" ||
2082 map_str == "{3: 4, 1: 2}");
2083 assert_eq!(format!("{:?}", empty), "{}");
2088 let mut m = HashMap::new();
2090 assert_eq!(m.len(), 0);
2091 assert!(m.is_empty());
2094 let old_cap = m.table.capacity();
2095 while old_cap == m.table.capacity() {
2100 assert_eq!(m.len(), i);
2101 assert!(!m.is_empty());
2105 fn test_behavior_resize_policy() {
2106 let mut m = HashMap::new();
2108 assert_eq!(m.len(), 0);
2109 assert_eq!(m.table.capacity(), 0);
2110 assert!(m.is_empty());
2114 assert!(m.is_empty());
2115 let initial_cap = m.table.capacity();
2116 m.reserve(initial_cap);
2117 let cap = m.table.capacity();
2119 assert_eq!(cap, initial_cap * 2);
2122 for _ in 0..cap * 3 / 4 {
2126 // three quarters full
2128 assert_eq!(m.len(), i);
2129 assert_eq!(m.table.capacity(), cap);
2131 for _ in 0..cap / 4 {
2137 let new_cap = m.table.capacity();
2138 assert_eq!(new_cap, cap * 2);
2140 for _ in 0..cap / 2 - 1 {
2143 assert_eq!(m.table.capacity(), new_cap);
2145 // A little more than one quarter full.
2147 assert_eq!(m.table.capacity(), cap);
2148 // again, a little more than half full
2149 for _ in 0..cap / 2 - 1 {
2155 assert_eq!(m.len(), i);
2156 assert!(!m.is_empty());
2157 assert_eq!(m.table.capacity(), initial_cap);
2161 fn test_reserve_shrink_to_fit() {
2162 let mut m = HashMap::new();
2165 assert!(m.capacity() >= m.len());
2171 let usable_cap = m.capacity();
2172 for i in 128..(128 + 256) {
2174 assert_eq!(m.capacity(), usable_cap);
2177 for i in 100..(128 + 256) {
2178 assert_eq!(m.remove(&i), Some(i));
2182 assert_eq!(m.len(), 100);
2183 assert!(!m.is_empty());
2184 assert!(m.capacity() >= m.len());
2187 assert_eq!(m.remove(&i), Some(i));
2192 assert_eq!(m.len(), 1);
2193 assert!(m.capacity() >= m.len());
2194 assert_eq!(m.remove(&0), Some(0));
2198 fn test_from_iter() {
2199 let xs = [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)];
2201 let map: HashMap<_, _> = xs.iter().cloned().collect();
2203 for &(k, v) in &xs {
2204 assert_eq!(map.get(&k), Some(&v));
2209 fn test_size_hint() {
2210 let xs = [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)];
2212 let map: HashMap<_, _> = xs.iter().cloned().collect();
2214 let mut iter = map.iter();
2216 for _ in iter.by_ref().take(3) {}
2218 assert_eq!(iter.size_hint(), (3, Some(3)));
2222 fn test_iter_len() {
2223 let xs = [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)];
2225 let map: HashMap<_, _> = xs.iter().cloned().collect();
2227 let mut iter = map.iter();
2229 for _ in iter.by_ref().take(3) {}
2231 assert_eq!(iter.len(), 3);
2235 fn test_mut_size_hint() {
2236 let xs = [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)];
2238 let mut map: HashMap<_, _> = xs.iter().cloned().collect();
2240 let mut iter = map.iter_mut();
2242 for _ in iter.by_ref().take(3) {}
2244 assert_eq!(iter.size_hint(), (3, Some(3)));
2248 fn test_iter_mut_len() {
2249 let xs = [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)];
2251 let mut map: HashMap<_, _> = xs.iter().cloned().collect();
2253 let mut iter = map.iter_mut();
2255 for _ in iter.by_ref().take(3) {}
2257 assert_eq!(iter.len(), 3);
2262 let mut map = HashMap::new();
2268 assert_eq!(map[&2], 1);
2273 fn test_index_nonexistent() {
2274 let mut map = HashMap::new();
2285 let xs = [(1, 10), (2, 20), (3, 30), (4, 40), (5, 50), (6, 60)];
2287 let mut map: HashMap<_, _> = xs.iter().cloned().collect();
2289 // Existing key (insert)
2290 match map.entry(1) {
2291 Vacant(_) => unreachable!(),
2292 Occupied(mut view) => {
2293 assert_eq!(view.get(), &10);
2294 assert_eq!(view.insert(100), 10);
2297 assert_eq!(map.get(&1).unwrap(), &100);
2298 assert_eq!(map.len(), 6);
2301 // Existing key (update)
2302 match map.entry(2) {
2303 Vacant(_) => unreachable!(),
2304 Occupied(mut view) => {
2305 let v = view.get_mut();
2306 let new_v = (*v) * 10;
2310 assert_eq!(map.get(&2).unwrap(), &200);
2311 assert_eq!(map.len(), 6);
2313 // Existing key (take)
2314 match map.entry(3) {
2315 Vacant(_) => unreachable!(),
2317 assert_eq!(view.remove(), 30);
2320 assert_eq!(map.get(&3), None);
2321 assert_eq!(map.len(), 5);
2324 // Inexistent key (insert)
2325 match map.entry(10) {
2326 Occupied(_) => unreachable!(),
2328 assert_eq!(*view.insert(1000), 1000);
2331 assert_eq!(map.get(&10).unwrap(), &1000);
2332 assert_eq!(map.len(), 6);
2336 fn test_entry_take_doesnt_corrupt() {
2337 #![allow(deprecated)] //rand
2339 fn check(m: &HashMap<isize, ()>) {
2341 assert!(m.contains_key(k),
2342 "{} is in keys() but not in the map?", k);
2346 let mut m = HashMap::new();
2347 let mut rng = thread_rng();
2349 // Populate the map with some items.
2351 let x = rng.gen_range(-10, 10);
2356 let x = rng.gen_range(-10, 10);
2360 println!("{}: remove {}", i, x);
2370 fn test_extend_ref() {
2371 let mut a = HashMap::new();
2373 let mut b = HashMap::new();
2375 b.insert(3, "three");
2379 assert_eq!(a.len(), 3);
2380 assert_eq!(a[&1], "one");
2381 assert_eq!(a[&2], "two");
2382 assert_eq!(a[&3], "three");