Skip to content

Commit

Permalink
insert_unique_unchecked operation
Browse files Browse the repository at this point in the history
Sometimes a map is constructed when it is known that all keys are
unique (e. e. if keys are coming from another map or from a
sorted/deduplicated iterator). In this case we can make insertion
faster by skipping a check that a key already exists in the map.

`insert_unique_unchecked` is guaranteed to be memory-safe, but does
not guarantee anything beyond that: if inserted key is not unique,
`HashMap` can panic, loop forever, return incorrect entry etc.

Added simple benchmark. `insert_unique_unchecked` is about 30%
faster than `insert`.  Your mileage may vary of course.

Similar PR was
[added to `indexmap` crate](indexmap-rs/indexmap#200)
and they asked to discuss the name of the operation with `hashbrown`
crate owners to come to the same naming convention (if `hashbrown`
is willing to have the same operation).
  • Loading branch information
stepancheg committed Sep 12, 2021
1 parent 728f9e8 commit 8d6c278
Show file tree
Hide file tree
Showing 3 changed files with 89 additions and 0 deletions.
32 changes: 32 additions & 0 deletions benches/insert_unique_unchecked.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1,32 @@
//! Compare `insert` and `insert_unique_unchecked` operations performance.
#![feature(test)]

extern crate test;

use hashbrown::HashMap;
use test::Bencher;

#[bench]
fn insert(b: &mut Bencher) {
let keys: Vec<String> = (0..1000).map(|i| format!("xxxx{}yyyy", i)).collect();
b.iter(|| {
let mut m = HashMap::with_capacity(1000);
for k in &keys {
m.insert(k, k);
}
m
});
}

#[bench]
fn insert_unique_unchecked(b: &mut Bencher) {
let keys: Vec<String> = (0..1000).map(|i| format!("xxxx{}yyyy", i)).collect();
b.iter(|| {
let mut m = HashMap::with_capacity(1000);
for k in &keys {
m.insert_unique_unchecked(k, k);
}
m
});
}
35 changes: 35 additions & 0 deletions src/map.rs
Original file line number Diff line number Diff line change
Expand Up @@ -1278,6 +1278,31 @@ where
}
}

/// Insert a key-value pair into the map without checking
/// if the key already exists in the map.
///
/// This operation is safe if a key does not exist in the map.
///
/// However, if a key exists in the map already, the behavior is unspecified:
/// this operation may panic, loop forever, or any following operation with the map
/// may panic, loop forever or return arbitrary result.
///
/// That said, this operation (and following operations) are guaranteed to
/// not violate memory safety.
///
/// This operation is faster than regular insert, because it does not perform
/// lookup before insertion.
///
/// This operation is useful during initial population of the map.
/// For example, when constructing a map from another map, we know
/// that keys are unique.
#[cfg_attr(feature = "inline-more", inline)]
pub fn insert_unique_unchecked(&mut self, k: K, v: V) {
let hash = make_insert_hash::<K, S>(&self.hash_builder, &k);
self.table
.insert(hash, (k, v), make_hasher::<K, _, V, S>(&self.hash_builder));
}

/// Tries to insert a key-value pair into the map, and returns
/// a mutable reference to the value in the entry.
///
Expand Down Expand Up @@ -3898,6 +3923,16 @@ mod test_map {
assert_eq!(*m.get(&5).unwrap(), 3);
}

#[test]
fn test_insert_unique_unchecked() {
let mut map = HashMap::new();
map.insert_unique_unchecked(10, 11);
map.insert_unique_unchecked(20, 21);
assert_eq!(Some(&11), map.get(&10));
assert_eq!(Some(&21), map.get(&20));
assert_eq!(None, map.get(&30));
}

#[test]
fn test_is_empty() {
let mut m = HashMap::with_capacity(4);
Expand Down
22 changes: 22 additions & 0 deletions src/set.rs
Original file line number Diff line number Diff line change
Expand Up @@ -991,6 +991,28 @@ where
self.map.insert(value, ()).is_none()
}

/// Insert a value the set without checking if the value already exists in the set.
///
/// This operation is safe if a value does not exist in the set.
///
/// However, if a value exists in the set already, the behavior is unspecified:
/// this operation may panic, loop forever, or any following operation with the set
/// may panic, loop forever or return arbitrary result.
///
/// That said, this operation (and following operations) are guaranteed to
/// not violate memory safety.
///
/// This operation is faster than regular insert, because it does not perform
/// lookup before insertion.
///
/// This operation is useful during initial population of the set.
/// For example, when constructing a set from another set, we know
/// that values are unique.
#[cfg_attr(feature = "inline-more", inline)]
pub fn insert_unique_unchecked(&mut self, value: T) {
self.map.insert_unique_unchecked(value, ());
}

/// Adds a value to the set, replacing the existing value, if any, that is equal to the given
/// one. Returns the replaced value.
///
Expand Down

0 comments on commit 8d6c278

Please sign in to comment.