Patricia Tree API - BASE_MAP
Base map signature: a generic 'b map storing bindings of 'a key to ('a,'b) values. All maps and set are a variation of this type, sometimes with a simplified interface.
HETEROGENEOUS_MAPis just aBASE_MAPwith a functorHETEROGENEOUS_MAP.WithForeignfor building operations that operate on two maps of different base types;MAPspecializes the interface for non-generic keys (keyinstead of'a key);HETEROGENEOUS_SETspecializesBASE_MAPfor sets (('a,'b) value = unit) and removes the value argument from most operations;SETspecializesHETEROGENEOUS_SETfurther by making elements (keys) non-generic (eltinstead of'a elt).
include NODE_WITH_FIND
include NODE
Types
The type of value, which depends on the type of the key and the type of the map.
Constructors: build values
val empty : 'map tThe empty map
A singleton leaf, similar to BASE_MAP.singleton
A branch node. This shouldn't be called externally unless you know what you're doing! Doing so could easily break the data structure's invariants.
When called, it assumes that:
- Neither
tree0nortree1should be empty. branching_bitshould have a single bit setprefixshould be normalized (bits belowbranching_bitset to zero)- All elements of
tree0should have theirto_intstart byprefixfollowed by 0 at positionbranching_bit). - All elements of
tree1should have theirto_intstart byprefixfollowed by 0 at positionbranching_bit).
Destructors: access the value
type 'map view = private | Empty : 'map view(*Can happen only at the toplevel: there is no empty interior node.
*)| Branch : {} -> 'map view(*Same constraints as
branch:branching_bitcontains only one bit set; the corresponding mask is (branching_bit - 1).prefixis normalized: the bits below thebranching_bitare set to zero (i.e.prefix & (branching_bit - 1) = 0).- All elements of
tree0should have theirto_intstart byprefixfollowed by 0 at positionbranching_bit). - All elements of
tree1should have theirto_intstart byprefixfollowed by 0 at positionbranching_bit).
| Leaf : {} -> 'map view(*A key -> value mapping.
*)
This makes the map nodes accessible to the pattern matching algorithm; this corresponds 1:1 to the SimpleNode implementation. This just needs to be copy-and-pasted for every node type.
val is_empty : 'map t -> boolCheck if the map is empty. Should be constant time.
find key map returns the value associated with key in map if present.
Existential wrapper for the 'a parameter in a 'a key, ('a,'map) value pair
Basic functions
val unsigned_min_binding : 'a t -> 'a key_value_pairunsigned_min_binding m is minimal binding KeyValue(k,v) of the map, using the unsigned order on KEY.to_int.
val unsigned_max_binding : 'a t -> 'a key_value_pairunsigned_max_binding m is maximal binding KeyValue(k,v) of the map, using the unsigned order on KEY.to_int.
val cardinal : 'a t -> intThe size of the map, O(n) complexity
val is_singleton : 'a t -> 'a key_value_pair optionis_singleton m returns Some(KeyValue(k,v)) if and only if m contains a unique binding k->v.
mem key map returns true iff key is bound in map, O(log(n)) complexity.
Returns a map with the element removed, O(log(n)) complexity. Returns a physically equal map if the element is absent.
val pop_unsigned_minimum : 'map t -> ('map key_value_pair * 'map t) optionpop_unsigned_minimum m returns None if is_empty m, or Some(key,value,m') where (key,value) = unsigned_min_binding m and m' = remove m key. Uses the unsigned order on KEY.to_int. O(log(n)) complexity.
val pop_unsigned_maximum : 'map t -> ('map key_value_pair * 'map t) optionpop_unsigned_maximum m returns None if is_empty m, or Some(key,value,m') where (key,value) = unsigned_max_binding m and m' = remove m key. Uses the unsigned order on KEY.to_int. O(log(n)) complexity.
insert key f map modifies or insert an element of the map; f takes None if the value was not previously bound, and Some old where old is the previously bound value otherwise. The function preserves physical equality when possible. O(log(n)) complexity. Preserves physical equality if the new value is physically equal to the old.
update key f map modifies, insert, or remove an element from the map; f takes None if the value was not previously bound, and Some old where old is the previously bound value otherwise. The function preserves physical equality when possible. It returns None if the element should be removed O(log(n)) complexity. Preserves physical equality if the new value is physically equal to the old.
Unconditionally adds a value in the map (independently from whether the old value existed). O(log(n)) complexity. Preserves physical equality if the new value is physically equal to the old.
Iterators
split key map splits the map into:
- submap of
mapwhose keys are smaller thankey - value associated to
key(if present) - submap of
mapwhose keys are bigger thankey
Where the order is given by the unsigned order on KEY.to_int.
iter f m calls f.f on all bindings of m, in the unsigned order on KEY.to_int
fold f m acc returns f.f key_n value_n (... (f.f key_1 value_1 acc)) where (key_1, value_1) ... (key_n, value_n) are the bindings of m, in the unsigned order on KEY.to_int.
fold_on_nonequal_inter f m1 m2 acc returns f.f key_n value1_n value2n (... (f.f key_1 value1_1 value2_1 acc)) where (key_1, value1_1, value2_1) ... (key_n, value1_n, value2_n) are the bindings that exist in both maps (m1 ∩ m2) whose values are physically different. Calls to f.f are performed in the unsigned order of KEY.to_int.
val fold_on_nonequal_union :
('acc, 'map) polyfold2_union ->
'map t ->
'map t ->
'acc ->
'accfold_on_nonequal_union f m1 m2 acc returns f.f key_n value1_n value2n (... (f.f key_1 value1_1 value2_1 acc)) where (key_1, value1_1, value2_1) ... (key_n, value1_n, value2_n) are the bindings that exists in either map (m1 ∪ m2) whose values are physically different. Calls to f.f are performed in the unsigned order of KEY.to_int.
val filter : 'map polypredicate -> 'map t -> 'map tfilter f m returns the submap of m containing the bindings k->v such that f.f k v = true. f.f is called in the unsigned order of KEY.to_int
val for_all : 'map polypredicate -> 'map t -> boolfor_all f m checks that f holds on all bindings of m. Short-circuiting.
In the following, the *no_share function allows taking arguments of different types (but cannot share subtrees of the map), while the default functions attempt to preserve and benefit from sharing the subtrees (using physical equality to detect sharing).
map f m and map_no_share f m replace all bindings (k,v) by (k, f.f v). Bindings are examined in the unsigned order of KEY.to_int.
mapi f m and mapi_no_share f m replace all bindings (k,v) by (k, f.f k v). Bindings are examined in the unsigned order of KEY.to_int.
val filter_map : ('map, 'map) polyfilter_map -> 'map t -> 'map tfilter_map m f and filter_map_no_share m f remove the bindings (k,v) for which f.f k v is None, and replaces the bindings (k,v) for which f.f k v is Some v' by (k,v'). Bindings are examined in the unsigned order of KEY.to_int.
val pretty :
?pp_sep:(Stdlib.Format.formatter -> unit -> unit) ->
'map polypretty ->
Stdlib.Format.formatter ->
'map t ->
unitPretty-prints a map using the given formatter. pp_sep is called once between each binding, it defaults to Format.pp_print_cut. Bindings are printed in the unsigned order of KEY.to_int
Functions on pairs of maps
This section regroups functions that act on pairs of maps.
These functions are where Patricia trees offer substantial speedup compared to Stdlib's Maps:
- We can often avoid exploring physically equal subtrees (for equality tests, inclusion tests, union, intersection, difference). This yields important performance gains when combining maps that derive from a common ancestor or when using Hash-consed maps and sets maps which have a lot of elements in common.
- We can also avoid visiting a subtree when merging with
Empty(for union, intersection and difference).
To do so safely, we have specialized versions of these functions that assume properties of the function parameter (reflexive relation, idempotent operation, etc.)
When we cannot enjoy these properties, our functions explicitly say so (with a nonreflexive or nonidempotent prefix). The names are a bit long, but having these names avoids using an ineffective code by default, by forcing to know and choose between the fast and slow version.
In general, the fast versions of these function will be on O(log n + d) where n is the size of the maps being joined and d the size of their difference (number of keys bound in both maps to non-physically equal values). The slow version is O(n).
Many of these are high-order functions, taking as argument a function f that operates on elements. Due to restrictions with higher-order polymorphism, we need to wrap the function f in a record, which has a single field f. These is what the polyXXX types are for.
Comparing two maps
Functions for equality, inclusion, and test for disjointness.
val reflexive_same_domain_for_all2 :
('map, 'map) polysame_domain_for_all2 ->
'map t ->
'map t ->
boolreflexive_same_domain_for_all2 f m1 m2 is true if and only if
m1andm2have the same domain (set of keys)- for all bindings
(k, v1)inm1and(k, v2)inm2,f.f k v1 v2holds
Assumes f.f is reflexive, i.e. f.f k v v = true to skip calls to equal subtrees. Calls f.f in ascending unsigned order of KEY.to_int. Exits early if the domains mismatch or if f.f returns false.
It is useful to implement equality on maps:
# let equal m1 m2 = MyMap.reflexive_same_domain_for_all2
{ f = fun _ v1 v2 -> MyValue.equal v1 v2}
m1 m2;;
val equal : 'a MyMap.t -> 'a MyMap.t -> bool = <fun>val nonreflexive_same_domain_for_all2 :
('map1, 'map2) polysame_domain_for_all2 ->
'map1 t ->
'map2 t ->
boolnonreflexive_same_domain_for_all2 f m1 m2 is the same as reflexive_same_domain_for_all2, but doesn't assume f.f is reflexive. It thus calls f.f on every binding, in ascending unsigned order of KEY.to_int. Exits early if the domains mismatch or if f.f returns false.
val reflexive_subset_domain_for_all2 :
('map, 'map) polysame_domain_for_all2 ->
'map t ->
'map t ->
boolreflexive_subset_domain_for_all2 f m1 m2 is true if and only if
m1's domain is a subset ofm2's. (all keys defined inm1are also defined inm2)- for all bindings
(k, v1)inm1and(k, v2)inm2,f.f k v1 v2holds
Assumes f.f is reflexive, i.e. f.f k v v = true to skip calls to equal subtrees. Calls f.f in ascending unsigned order of KEY.to_int. Exits early if the domains mismatch.
It is useful to implement inclusion test on maps:
# let is_submap m1 m2 = MyMap.reflexive_subset_domain_for_all2
{ f = fun _ v1 v2 -> MyValue.equal v1 v2}
m1 m2;;
val is_submap : 'a MyMap.t -> 'a MyMap.t -> bool = <fun>val reflexive_compare : 'a polycompare -> 'a t -> 'a t -> intreflexive_compare f m1 m2 is an order relation on maps. m1 and m2 are equal (return 0) if they have the same domain and for all bindings (k,v) in m1, (k,v') in m2, we have f v v' = 0.
m1 is considered striclty smaller than m2 (return a negative integer) when the first difference (lowest key in the unsigned order of KEY.to_int) is either a shared binding (k,v) in m1, (k,v') in m2 with f v v' < 0, or a binding that only occurs in m2.
Assumes that f v v = 0.
Combining two maps
We provide many functions that operate on pairs of maps for computing intersection, union, difference... Here is a short summary of what each of one returns when applied to two maps m1 and m2. Here y is physically the same value in m1 and m2.
Keys |
|
|
|
|
|---|---|---|---|---|
|
|
|
| |
|
|
|
| |
|
|
|
| |
|
|
|
|
|
|
| |||
|
| |||
|
| |||
|
| |||
|
|
|
[1]: Here f returns an optional value, returning None removes the binding.
[2]: Here the function f actually takes option as arguments, omitted for brevity. _ is None.
idempotent_union f map1 map2 returns a map whose keys is the union of the keys of map1 and map2. f.f is used to combine the values of keys mapped in both maps.
Assumes f.f idempotent (i.e. f key value value == value) f.f is called in the unsigned order of KEY.to_int. f.f is never called on physically equal values. Preserves physical equality as much as possible. Complexity is O(log(n)*Delta) where Delta is the number of different keys between map1 and map2.
idempotent_inter f map1 map2 returns a map whose keys is the intersection of the keys of map1 and map2. f.f is used to combine the values a key is mapped in both maps.
Assumes f.f idempotent (i.e. f.f key value value == value) f.f is called in the unsigned order of KEY.to_int. f.f is never called on physically equal values. Preserves physical equality as much as possible. Complexity is O(log(n)*Delta) where Delta is the number of different keys between map1 and map2.
nonidempotent_inter_no_share f map1 map2 is the same as idempotent_inter but doesn't preverse physical equality, doesn't assume f.f is idempotent, and can change the type of values. f.f is called on every shared binding. f.f is called in increasing unsigned order of KEY.to_int. O(n) complexity
val idempotent_inter_filter :
('a, 'a, 'a) polyinterfilter ->
'a t ->
'a t ->
'a tidempotent_inter_filter f map1 map2 is the same as idempotent_inter but f.f can return None to remove a binding from the resutling map.
This is the same as Stdlib.Map.S.merge
type ('a, 'b) polydifference = ('a, 'b, 'a) polyinterfilterval symmetric_difference : ('a, 'a) polydifference -> 'a t -> 'a t -> 'a tsymmetric_difference f map1 map2 returns a map comprising of the bindings of map1 that aren't in map2, and the bindings of map2 that aren't in map1.
Bindings that are both in map1 and map2, but with non-physically equal values are passed to f.f. If f.f returns Some v then v is used as the new value, otherwise the binding is dropped.
Assumes f.f is none on equal values (i.e. f.f key value value == None) f.f is called in increasing unsigned order of KEY.to_int. f.f is never called on physically equal values.
Complexity is O(log n + d) where n is the size of the maps, and d the size of the difference.
val difference : ('a, 'a) polydifference -> 'a t -> 'a t -> 'a tdifference f map1 map2 returns the map containing the bindings of map1 that aren't in map2. For keys present in both maps but with different values, f.f is called. If it returns Some v, then binding k,v is kept, else k is dropped.
Assumes f.f is None on the diagonal: f.f k v v = None. f.f is called in the unsigned order of KEY.to_int. f.f is never called on physically equal values.
Min/max of intersection
type ('a, 'b) key_value_value = | KeyValueValue : 'k key * ('k, 'a) value * ('k, 'b) value -> ('a, 'b) key_value_value
Existential wrapper for a key with two values
val min_binding_inter : 'a t -> 'b t -> ('a, 'b) key_value_value optionmin_binding_inter m1 m2 is the minimal binding of the intersection. I.E. the KeyValueValue(k,v1,v2) such that (k,v1) is in m1, (k,v2) is in m2, and k is minimal using the unsigned order on keys.
Returns None if and only if the intersection is empty.
It is rougthly equivalent to calling unsigned_min_binding on nonindempotent_inter_no_share f m1 m2, but can be faster.
val max_binding_inter : 'a t -> 'b t -> ('a, 'b) key_value_value optionmax_binding_inter m1 m2 is the same as min_binding_inter, but returns the maximum key instead of the minimum.
Conversion functions
val to_seq : 'a t -> 'a key_value_pair Stdlib.Seq.tto_seq m iterates the whole map, in increasing unsigned order of KEY.to_int
val to_rev_seq : 'a t -> 'a key_value_pair Stdlib.Seq.tto_rev_seq m iterates the whole map, in decreasing unsigned order of KEY.to_int
val add_seq : 'a key_value_pair Stdlib.Seq.t -> 'a t -> 'a tadd_seq s m adds all bindings of the sequence s to m in order.
val of_seq : 'a key_value_pair Stdlib.Seq.t -> 'a tof_seq s creates a new map from the bindings of s. If a key is bound multiple times in s, the latest binding is kept
val of_list : 'a key_value_pair list -> 'a tof_list l creates a new map from the bindings of l. If a key is bound multiple times in l, the latest binding is kept
val to_list : 'a t -> 'a key_value_pair listto_list m returns the bindings of m as a list, in increasing unsigned order of KEY.to_int