Module Lru

Scalable LRU caches

Lru provides weight-bounded finite maps that can remove the least-recently-used (LRU) bindings in order to maintain a weight constraint. Two implementations are provided: one is functional, the other imperative.

The functional map is backed by a priority search queue. Operations on individual elements are O(log n).

The mutable map is backed by the standard Hashtbl paired with a doubly-linked list. Operations on individual elements incur an O(1) overhead on top of hash table access.

Both versions support differentially weighted bindings, and have a capacity parameter that limits the combined weight of the bindings. To limit the maps by the number of bindings, use let weight _ = 1.

v0.3.0 — homepage

Semantics

A pretty accurate model of a functional k -> v map is an association list ((k * v) list) with unique keys.

Adding a bindings k -> v to kvs means List.remove_assoc k kvs @ [(k, v)], finding a k means List.assoc_opt k kvs, and removing it means List.remove_assoc k kvs.

The LRU binding is then the first element of the list.

Promoting a binding k -> v means removing, and then re-adding it.

Trimming kvs means retaining the longest suffix with the sum of weight v not larger than capacity.

The imperative LRU map is like the above, but kept in a reference cell.

Lru

module type Weighted = sig ... end

Signature of types with measurable weight.

module F : sig ... end

Functional LRU map.

module M : sig ... end

Mutable LRU map.

One-off memoization

val memo : ?⁠hashed:(('a -> int) * ('a -> 'a -> bool)) -> ?⁠weight:('b -> int) -> cap:int -> (('a -> 'b) -> 'a -> 'b) -> 'a -> 'b

memo ?hashed ?weight ~cap f is a new memoized instance of f, using LRU caching. f is an open recursive function of one parameter.

~hashed are hashing and equality over the arguments 'a. It defaults to (Hashtbl.hash, Pervasives.(=)).

~weight is the weighting function over the results 'b. It defaults to fun _ -> 1.

~cap is the total cache capacity.

raises Invalid_argument

when cap < 0.