Struct immutable_chunkmap::map::Map [−][src]
Expand description
This Map uses a similar strategy to BTreeMap to ensure cache efficient performance on modern hardware while still providing log(N) get, insert, and remove operations.
For good performance, it is very important to understand that clone is a fundamental operation, it needs to be fast for your key and data types, because it’s going to be called a lot whenever you change the map.
Why
-
Multiple threads can read this structure even while one thread is updating it. Using a library like arc_swap you can avoid ever blocking readers.
-
Snapshotting this structure is free.
Examples
use std::string::String;
use self::immutable_chunkmap::map::MapM;
let m =
MapM::new()
.insert(String::from("1"), 1).0
.insert(String::from("2"), 2).0
.insert(String::from("3"), 3).0;
assert_eq!(m.get("1"), Option::Some(&1));
assert_eq!(m.get("2"), Option::Some(&2));
assert_eq!(m.get("3"), Option::Some(&3));
assert_eq!(m.get("4"), Option::None);
for (k, v) in &m {
println!("key {}, val: {}", k, v)
}Implementations
Create a weak reference to this map
Return the number of strong references to this map (see Arc)
Return the number of weak references to this map (see Arc)
This will insert many elements at once, and is potentially a lot faster than inserting one by one, especially if the data is sorted. It is just a wrapper around the more general update_many method.
#Examples
use self::immutable_chunkmap::map::MapM;
let mut v = vec![(1, 3), (10, 1), (-12, 2), (44, 0), (50, -1)];
v.sort_unstable_by_key(|&(k, _)| k);
let m = MapM::new().insert_many(v.iter().map(|(k, v)| (*k, *v)));
for (k, v) in &v {
assert_eq!(m.get(k), Option::Some(v))
}This method updates multiple bindings in one call. Given an iterator of an arbitrary type (Q, D), where Q is any borrowed form of K, an update function taking Q, D, the current binding in the map, if any, and producing the new binding, if any, this method will produce a new map with updated bindings of many elements at once. It will skip intermediate node allocations where possible. If the data in elts is sorted, it will be able to skip many more intermediate allocations, and can produce a speedup of about 10x compared to inserting/updating one by one. In any case it should always be faster than inserting elements one by one, even with random unsorted keys.
#Examples
use std::iter::FromIterator;
use self::immutable_chunkmap::map::MapM;
let m = MapM::from_iter((0..4).map(|k| (k, k)));
let m = m.update_many(
(0..4).map(|x| (x, ())),
|k, (), cur| cur.map(|(_, c)| (k, c + 1))
);
assert_eq!(
m.into_iter().map(|(k, v)| (*k, *v)).collect::<Vec<_>>(),
vec![(0, 1), (1, 2), (2, 3), (3, 4)]
);return a new map with (k, v) inserted into it. If k already exists in the old map, the old binding will be returned, and the new map will contain the new binding. In fact this method is just a wrapper around update.
insert in place using copy on write semantics if self is not a
unique reference to the map. see update_cow.
return a new map with the binding for q, which can be any
borrowed form of k, updated to the result of f. If f returns
None, the binding will be removed from the new map, otherwise
it will be inserted. This function is more efficient than
calling get and then insert, since it makes only one tree
traversal instead of two. This method runs in log(N) time and
space where N is the size of the map.
Examples
use self::immutable_chunkmap::map::MapM;
let (m, _) = MapM::new().update(0, 0, |k, d, _| Some((k, d)));
let (m, _) = m.update(1, 1, |k, d, _| Some((k, d)));
let (m, _) = m.update(2, 2, |k, d, _| Some((k, d)));
assert_eq!(m.get(&0), Some(&0));
assert_eq!(m.get(&1), Some(&1));
assert_eq!(m.get(&2), Some(&2));
let (m, _) = m.update(0, (), |k, (), v| v.map(move |(_, v)| (k, v + 1)));
assert_eq!(m.get(&0), Some(&1));
assert_eq!(m.get(&1), Some(&1));
assert_eq!(m.get(&2), Some(&2));
let (m, _) = m.update(1, (), |_, (), _| None);
assert_eq!(m.get(&0), Some(&1));
assert_eq!(m.get(&1), None);
assert_eq!(m.get(&2), Some(&2));Perform a copy on write update to the map. In the case that self is a unique reference to the map, then the update will be performed completly in place. self will be mutated, and no previous version will be available. In the case that self has a clone, or clones, then only the parts of the map that need to be mutated will be copied before the update is performed. self will reference the mutated copy, and previous versions of the map will not be modified. self will still share all the parts of the map that did not need to be mutated with any pre existing clones.
COW semantics are a flexible middle way between full peristance and full mutability. Needless to say in the case where you have a unique reference to the map, using update_cow is a lot faster than using update, and a lot more flexible than update_many.
Other than copy on write the semanics of this method are identical to those of update.
#Examples
use self::immutable_chunkmap::map::MapM;
let mut m = MapM::new().update(0, 0, |k, d, _| Some((k, d))).0;
let orig = m.clone();
m.update_cow(1, 1, |k, d, _| Some((k, d))); // copies the original chunk
m.update_cow(2, 2, |k, d, _| Some((k, d))); // doesn't copy anything
assert_eq!(m.len(), 3);
assert_eq!(orig.len(), 1);
assert_eq!(m.get(&0), Some(&0));
assert_eq!(m.get(&1), Some(&1));
assert_eq!(m.get(&2), Some(&2));
assert_eq!(orig.get(&0), Some(&0));
assert_eq!(orig.get(&1), None);
assert_eq!(orig.get(&2), None);Merge two maps together. Bindings that exist in both maps will be passed to f, which may elect to remove the binding by returning None. This function runs in O(log(n) + m) time and space, where n is the size of the largest map, and m is the number of intersecting chunks. It will never be slower than calling update_many on the first map with an iterator over the second, and will be significantly faster if the intersection is minimal or empty.
Examples
use std::iter::FromIterator;
use self::immutable_chunkmap::map::MapM;
let m0 = MapM::from_iter((0..10).map(|k| (k, 1)));
let m1 = MapM::from_iter((10..20).map(|k| (k, 1)));
let m2 = m0.union(&m1, |_k, _v0, _v1| panic!("no intersection expected"));
for i in 0..20 {
assert!(m2.get(&i).is_some())
}
let m3 = MapM::from_iter((5..9).map(|k| (k, 1)));
let m4 = m3.union(&m2, |_k, v0, v1| Some(v0 + v1));
for i in 0..20 {
assert!(
*m4.get(&i).unwrap() ==
*m3.get(&i).unwrap_or(&0) + *m2.get(&i).unwrap_or(&0)
)
}Produce a map containing the mapping over F of the intersection (by key) of two maps. The function f runs on each intersecting element, and has the option to omit elements from the intersection by returning None, or change the value any way it likes. Runs in O(log(N) + M) time and space where N is the size of the smallest map, and M is the number of intersecting chunks.
Examples
use std::iter::FromIterator;
use self::immutable_chunkmap::map::MapM;
let m0 = MapM::from_iter((0..100000).map(|k| (k, 1)));
let m1 = MapM::from_iter((50..30000).map(|k| (k, 1)));
let m2 = m0.intersect(&m1, |_k, v0, v1| Some(v0 + v1));
for i in 0..100000 {
if i >= 30000 || i < 50 {
assert!(m2.get(&i).is_none());
} else {
assert!(*m2.get(&i).unwrap() == 2);
}
}Produce a map containing the second map subtracted from the first. The function F is called for each intersecting element, and ultimately decides whether it appears in the result, for example, to compute a classical set diff, the function should always return None.
Examples
use std::iter::FromIterator;
use self::immutable_chunkmap::map::MapM;
let m0 = MapM::from_iter((0..10000).map(|k| (k, 1)));
let m1 = MapM::from_iter((50..3000).map(|k| (k, 1)));
let m2 = m0.diff(&m1, |_k, _v0, _v1| None);
m2.invariant();
dbg!(m2.len());
assert!(m2.len() == 10000 - 2950);
for i in 0..10000 {
if i >= 3000 || i < 50 {
assert!(*m2.get(&i).unwrap() == 1);
} else {
assert!(m2.get(&i).is_none());
}
}lookup the mapping for k. If it doesn’t exist return None. Runs in log(N) time and constant space. where N is the size of the map.
lookup the mapping for k. Return the key. If it doesn’t exist return None. Runs in log(N) time and constant space. where N is the size of the map.
lookup the mapping for k. Return both the key and the value. If it doesn’t exist return None. Runs in log(N) time and constant space. where N is the size of the map.
return a new map with the mapping under k removed. If the binding existed in the old map return it. Runs in log(N) time and log(N) space, where N is the size of the map.
remove in place using copy on write semantics if self is not a
unique reference to the map. see update_cow.
return an iterator over the subset of elements in the map that are within the specified range.
The returned iterator runs in O(log(N) + M) time, and constant space. N is the number of elements in the tree, and M is the number of elements you examine.
if lbound >= ubound the returned iterator will be empty
Trait Implementations
impl<K, V, const SIZE: usize> PartialOrd<Map<K, V, SIZE>> for Map<K, V, SIZE> where
K: Ord + Clone,
V: PartialOrd + Clone,
impl<K, V, const SIZE: usize> PartialOrd<Map<K, V, SIZE>> for Map<K, V, SIZE> where
K: Ord + Clone,
V: PartialOrd + Clone,
This method returns an ordering between self and other values if one exists. Read more
This method tests less than (for self and other) and is used by the < operator. Read more
This method tests less than or equal to (for self and other) and is used by the <=
operator. Read more
This method tests greater than (for self and other) and is used by the > operator. Read more
Auto Trait Implementations
impl<K, V, const SIZE: usize> RefUnwindSafe for Map<K, V, SIZE> where
K: RefUnwindSafe,
V: RefUnwindSafe,
impl<K, V, const SIZE: usize> UnwindSafe for Map<K, V, SIZE> where
K: RefUnwindSafe,
V: RefUnwindSafe,
Blanket Implementations
Mutably borrows from an owned value. Read more