B+ tree

B+ trees are ordered collections of key-value pairs, sorted by key.

This is a fast B+ tree implementation, largely compatible with the standard Map, but with a much more diverse and powerful API. To use it, import BTree from 'sorted-btree'.

BTree is faster and/or uses less memory than other popular JavaScript sorted trees (see Benchmarks). However, data structures in JavaScript tend to be slower than the built-in Array and Map data structures in typical cases, because the built-in data structures are mostly implemented in a faster language such as C++. Even so, if you have a large amount of data that you want to keep sorted, the built-in data structures will not serve you well, and BTree offers features like fast cloning that the built-in types don't.

Use npm install sorted-btree in a terminal to install it in your npm-based project.

Features

Requires ES5 only (Symbol.iterator is not required but is used if defined.)
Includes typings (BTree was written in TypeScript)
API similar to ES6 Map with methods such as size(), clear(), forEach((v,k,tree)=>{}), get(K), set(K,V), has(K), delete(K), plus iterator functions keys(), values() and entries().
Supports keys that are numbers, strings, arrays of numbers/strings, Date, and objects that have a valueOf() method that returns a number or string.
Other data types can also be supported with a custom comparator (second
constructor argument).
Supports O(1) fast cloning with subtree sharing. This works by marking the root node as "shared between instances". This makes the tree read-only with copy-on-edit behavior; both copies of the tree remain mutable. I call this category of data structure "dynamically persistent" because AFAIK no one else has given it a name; it walks the line between mutating and persistent.
Includes persistent methods such as with and without, which return a modified tree without changing the original (in O(log(size)) time).
When a node fills up, items are shifted to siblings when possible to keep nodes near their capacity, to improve memory utilization.
Efficiently supports sets (keys without values). The collection does not allocate memory for values if the value undefined is associated with all keys in a given node.
Includes neat stuff such as Range methods for batch operations
Throws an exception if you try to use NaN as a key, but infinity is allowed.
No dependencies. 19.5K minified.
Includes a lattice of interfaces for TypeScript users (see below)
Supports diffing computation between two trees that is highly optimized for the case in which a majority of nodes are shared (such as when persistent methods are used).

Additional operations supported on this B+ tree

Set a value only if the key does not already exist: t.setIfNotPresent(k,v)
Set a value only if the key already exists: t.changeIfPresent(k,v)
Iterate in backward order: for (pair of t.entriesReversed()) {}
Iterate from a particular first element: for (let p of t.entries(first)) {}
Convert to an array: t.toArray(), t.keysArray(), t.valuesArray()
Get pairs for a range of keys ([K,V][]): t.getRange(loK, hiK, includeHi)
Delete a range of keys and their values: t.deleteRange(loK, hiK, includeHi)
Scan all items: t.forEachPair((key, value, index) => {...})
Scan a range of items: t.forRange(lowKey, highKey, includeHiFlag, (k,v) => {...})
Count the number of keys in a range: c = t.forRange(loK, hiK, includeHi, undefined)
Get smallest or largest key: t.minKey(), t.maxKey()
Get next larger key/pair than k: t.nextHigherKey(k), t.nextHigherPair(k)
Get largest key/pair that is lower than k: t.nextLowerKey(k), t.nextLowerPair(k)
Freeze to prevent modifications: t.freeze() (you can also t.unfreeze())
Fast clone: t.clone()
Compute a diff between two trees (quickly skipping shared subtrees): t.diffAgainst(otherTree, ...)
For more information, see full documentation in the source code.

Note: Confusingly, the ES6 Map.forEach(c) method calls c(value,key) instead of c(key,value), in contrast to other methods such as set() and entries() which put the key first. I can only assume that they reversed the order on the hypothesis that users would usually want to examine values and ignore keys. BTree's forEach() therefore works the same way, but there is a second method .forEachPair((key,value)=>{...}) which sends you the key first and the value second; this method is slightly faster because it is the "native" for-each method for this class.

Note: Duplicate keys are not allowed (supporting duplicates properly is complex).

The "scanning" methods (forEach, forRange, editRange, deleteRange) will normally return the number of elements that were scanned. However, the callback can return {break:R} to stop iterating early and return a value R from the scanning method.

Functional methods

Get a copy of the tree including only items fitting a criteria: t.filter((k,v) => k.fitsCriteria())
Get a copy of the tree with all values modified: t.mapValues((v,k) => v.toString())
Reduce a tree (see below): t.reduce((acc, pair) => acc+pair[1], 0)

Persistent methods

Get a new tree with one pair changed: t.with(key, value)
Get a new tree with multiple pairs changed: t.withPairs([[k1,v1], [k2,v2]])
Ensure that specified keys exist in a new tree: t.withKeys([k1,k2])
Get a new tree with one pair removed: t.without(key)
Get a new tree with specific pairs removed: t.withoutKeys(keys)
Get a new tree with a range of keys removed: t.withoutRange(low, high, includeHi)

Things to keep in mind: I ran a test which suggested t.with is three times slower than t.set. These methods do not return a frozen tree even if the original tree was frozen (for performance reasons, e.g. frozen trees use slightly more memory.)

Examples

Custom comparator

Given a set of {name: string, age: number} objects, you can create a tree sorted by name and then by age like this:

  // First constructor argument is an optional list of pairs ([K,V][])
  var tree = new BTree(undefined, (a, b) => {
    if (a.name > b.name)
      return 1; // Return a number >0 when a > b
    else if (a.name < b.name)
      return -1; // Return a number <0 when a < b
    else // names are equal (or incomparable)
      return a.age - b.age; // Return >0 when a.age > b.age
  });

  tree.set({name:"Bill", age:17}, "happy");
  tree.set({name:"Fran", age:40}, "busy & stressed");
  tree.set({name:"Bill", age:55}, "recently laid off");
  tree.forEachPair((k, v) => {
    console.log(`Name: ${k.name} Age: ${k.age} Status: ${v}`);
  });

reduce

The reduce method performs a reduction operation, like the reduce method of Array. It is used to combine all keys, values or pairs into a single value, or to perform type conversions conversions. reduce is best understood by example. So here's how you can multiply all the keys in a tree together:

var product = tree.reduce((p, pair) => p * pair[0], 1)

It means "start with p=1, and for each pair change p to p * pair[0]" (pair[0] is the key). You may be thinking "hey, wouldn't it make more sense if the 1 argument came first?" Yes it would, but in Array the parameter is second, so it must also be second in BTree for consistency.

Here's a similar example that adds all values together:

var total = tree.reduce((sum, pair) => sum + pair[1], 0)

This final example converts the tree to a Map:

var map = tree.reduce((m, pair) => m.set(pair[0], pair[1]), new Map())`

Remember that m.set returns m, which is different from BTree where tree.set returns a boolean indicating whether a new key was added.

editRange

You can scan a range of items and selectively delete or change some of them using t.editRange. For example, the following code adds an exclamation mark to each non-boring value and deletes key number 4:

var t = new BTree().setRange([[1,"fun"],[2,"yay"],[4,"whee"],[8,"zany"],[10,"boring"]);
t.editRange(t.minKey(), t.maxKey(), true, (k, v) => {
  if (k === 4) 
    return {delete: true};
  if (v !== "boring")
    return {value: v + '!'};
})

Interface lattice

BTree includes a lattice of interface types representing subsets of BTree's interface. I would encourage other authors of map/dictionary/tree/hashtable types to utilize these interfaces. These interfaces can be divided along three dimensions:

1. Read/write access

I have defined several kinds of interfaces along the read/write access dimension:

Source: A "source" is a read-only interface (ISetSource<K> and IMapSource<K,V>). At minimum, sources include a size property and methods get, has, forEach, and keys.
Sink: A "sink" is a write-only interface (ISetSink<K> and IMapSink<K,V>). At minimum, sinks have set, delete and clear methods.
Mutable: An interface that combines the source and sink interfaces (ISet<K> and IMap<K,V>).
Functional: An interface for persistent data structures. It combines a read-only interface with methods that return a modified copy of the collection. The functional interfaces end with F (ISetF<K> and IMapF<K,V>).

2. Sorted versus unsorted

The Sorted interfaces extend the non-sorted interfaces with queries that only a sorted collection can perform efficiently, such as minKey() and nextHigherKey(k). At minimum, sorted interfaces add methods minKey, maxKey, nextHigherKey, nextLowerKey, and forRange, plus iterators that return keys/values/pairs in sorted order and accept a firstKey parameter to control the starting point of iteration.

3. Set versus map

A map is a collection of keys with values, while a set is a collection of keys without values.

For the most part, each Set interface is a subset of the corresponding Map interface with "values" removed. For example, MapF<K,V> extends SetF<K>. An exception to this is that IMapSink<K, V> could not be derived from ISetSink<K> (and thus IMap<K,V> is not derived from ISet<K>) because the type V does not necessarily include undefined. Therefore you can write set.set(key) to add a key to a set, but you cannot write map.set(key) without specifying a value (in TypeScript this is true even if V includes undefined.)

List of interfaces

All of these interfaces use any as the default type of K and V.

ISetSource<K>
ISetSink<K>
ISet<K> extends ISetSource<K>, ISetSink<K>
IMapSource<K, V> extends ISetSource<K>
IMapSink<K, V>
IMap<K, V> extends IMapSource<K,V>, IMapSink<K,V>
ISortedSetSource<K> extends ISetSource<K>
ISortedSet<K> extends ISortedSetSource<K>, ISetSink<K>
ISortedMapSource<K,V> extends IMapSource<K, V>, ISortedSetSource<K>
ISortedMap<K,V> extends IMap<K,V>, ISortedMapSource<K,V>
ISetF<K> extends ISetSource<K>
IMapF<K, V> extends IMapSource<K,V>, ISetF<K>
ISortedSetF<K> extends ISetF<K>, ISortedSetSource<K>
ISortedMapF<K,V> extends ISortedSetF<K>, IMapF<K,V>, ISortedMapSource<K,V>

If the lattice were complete there would be 16 interfaces (422). In fact there are only 14 interfaces because ISortedMapSink<K,V> and ISortedSetSink<K, V> don't exist, because sorted sinks are indistinguishable from unsorted sinks.

BTree<K,V> implements all of these interfaces except ISetSink<K>, ISet<K>, and ISortedSet<K>.

ES6 Map/Set compatibility

The IMap<K,V> interface is compatible with the ES6 Map<K,V> type as well as BTree<K,V>. In order to accomplish this, compromises had to be made:

The set(k,v) method returns any for compatibility with both BTree and Map, since BTree returns boolean (true if an item was added or false if it already existed), while Map returns this.
ES6's Map.forEach(c) method calls c(value,key) instead of c(key,value), unlike all other methods which put the key first. Therefore IMap works the same way. Unfortunately, this means that ISetSource<K>, the supertype of IMapSource<K,V>, cannot sanely have a forEach method because if it did, the first parameter to the callback would be unused.
The batch operations setPairs, deletePairs and reduce are left out because they are not defined by Map. Instead, these methods are defined in ISortedMap<K,V>.
Likewise, the functional operations reduce, filter and mapValues are not included in IMap, but they are defined in IMapF<K,V> and (except mapValues) ISetF<K>.

Similarly, ISet<K> is compatible with ES6 Set. Again there are compromises:

The set method is renamed add in Set and ISet<K>, so add exists on BTree.prototype as a synonym for set.
There is no forEach method for reasons alluded to above. Use keys() instead.
There is no filter or reduce because Set doesn't support them.

Although BTree<K,V> doesn't directly implement ISet<K>, it does implement ISetSource<K> and it is safe to cast BTree<K,V> to ISet<K> or ISortedSet<K> provided that V is allowed to be undefined.

Benchmarks (in milliseconds for integer keys/values)

These benchmark results were gathered on my PC in Node v10.4.1, July 2018
BTree is 3 to 5 times faster than SortedMap and SortedSet in the collections package
BTree has similar speed to RBTree at smaller sizes, but is faster at very large sizes and uses less memory because it packs many keys into one array instead of allocating an extra heap object for every key.
If you need functional persistence, functional-red-black-tree is remarkably fast for a persistent tree, but BTree should require less memory unless you frequently use clone/with/without and are saving snapshots of the old tree to prevent garbage collection.
B+ trees normally use less memory than hashtables (such as the standard Map), although in JavaScript this is not guaranteed because the B+ tree's memory efficiency depends on avoiding wasted space in the arrays for each node, and JavaScript provides no way to detect or control the capacity of an array's underlying memory area. Also, Map should be faster because it does not sort its keys.
"Sorted array" refers to SortedArray<K,V>, a wrapper class for an array of [K,V] pairs. Benchmark results were not gathered for sorted arrays with one million elements (it takes too long)

Insertions at random locations: sorted-btree vs the competition

0.8     Insert 1000 pairs in sorted-btree's BTree
0.4     Insert 1000 pairs in sorted-btree's BTree set (no values)
2.5     Insert 1000 pairs in collections' SortedMap
1.6     Insert 1000 pairs in collections' SortedSet (no values)
0.7     Insert 1000 pairs in functional-red-black-tree
0.5     Insert 1000 pairs in bintrees' RBTree (no values)

8.6     Insert 10000 pairs in sorted-btree's BTree
5.1     Insert 10000 pairs in sorted-btree's BTree set (no values)
37.8    Insert 10000 pairs in collections' SortedMap
25.8    Insert 10000 pairs in collections' SortedSet (no values)
8.7     Insert 10000 pairs in functional-red-black-tree
5.4     Insert 10000 pairs in bintrees' RBTree (no values)

95.9    Insert 100000 pairs in sorted-btree's BTree
69.1    Insert 100000 pairs in sorted-btree's BTree set (no values)
564     Insert 100000 pairs in collections' SortedMap
366.5   Insert 100000 pairs in collections' SortedSet (no values)
192.5   Insert 100000 pairs in functional-red-black-tree
107.3   Insert 100000 pairs in bintrees' RBTree (no values)

1363    Insert 1000000 pairs in sorted-btree's BTree
909     Insert 1000000 pairs in sorted-btree's BTree set (no values)
8783    Insert 1000000 pairs in collections' SortedMap
5443    Insert 1000000 pairs in collections' SortedSet (no values)
3356    Insert 1000000 pairs in functional-red-black-tree
1581    Insert 1000000 pairs in bintrees' RBTree (no values)

Insert in order, delete: sorted-btree vs the competition

0.6     Insert 1000 sorted pairs in B+ tree
0.4     Insert 1000 sorted keys in B+ tree set (no values)
0.6     Insert 1000 sorted pairs in collections' SortedMap
0.4     Insert 1000 sorted keys in collections' SortedSet (no values)
0.7     Insert 1000 sorted pairs in functional-red-black-tree
0.5     Insert 1000 sorted keys in bintrees' RBTree (no values)
1       Delete every second item in B+ tree
3       Delete every second item in B+ tree set
1       Bulk-delete every second item in B+ tree set
1       Delete every second item in collections' SortedMap
1       Delete every second item in collections' SortedSet
5       Delete every second item in functional-red-black-tree
10      Delete every second item in bintrees' RBTree

6.5     Insert 10000 sorted pairs in B+ tree
3.9     Insert 10000 sorted keys in B+ tree set (no values)
6.5     Insert 10000 sorted pairs in collections' SortedMap
3.9     Insert 10000 sorted keys in collections' SortedSet (no values)
12.4    Insert 10000 sorted pairs in functional-red-black-tree
5.8     Insert 10000 sorted keys in bintrees' RBTree (no values)
4       Delete every second item in B+ tree
4       Delete every second item in B+ tree set
3       Bulk-delete every second item in B+ tree set
13      Delete every second item in collections' SortedMap
7       Delete every second item in collections' SortedSet
8       Delete every second item in functional-red-black-tree
6       Delete every second item in bintrees' RBTree

75.9    Insert 100000 sorted pairs in B+ tree
45      Insert 100000 sorted keys in B+ tree set (no values)
98.7    Insert 100000 sorted pairs in collections' SortedMap
61.4    Insert 100000 sorted keys in collections' SortedSet (no values)
145.8   Insert 100000 sorted pairs in functional-red-black-tree
82.6    Insert 100000 sorted keys in bintrees' RBTree (no values)
79      Delete every second item in B+ tree
52      Delete every second item in B+ tree set
18      Bulk-delete every second item in B+ tree set
166     Delete every second item in collections' SortedMap
58      Delete every second item in collections' SortedSet
64      Delete every second item in functional-red-black-tree
74      Delete every second item in bintrees' RBTree

700     Insert 1000000 sorted pairs in B+ tree
452.5   Insert 1000000 sorted keys in B+ tree set (no values)
1069    Insert 1000000 sorted pairs in collections' SortedMap
864     Insert 1000000 sorted keys in collections' SortedSet (no values)
1531    Insert 1000000 sorted pairs in functional-red-black-tree
966     Insert 1000000 sorted keys in bintrees' RBTree (no values)
435     Delete every second item in B+ tree
291     Delete every second item in B+ tree set
159     Bulk-delete every second item in B+ tree set
1447    Delete every second item in collections' SortedMap
796     Delete every second item in collections' SortedSet
573     Delete every second item in functional-red-black-tree
537     Delete every second item in bintrees' RBTree

Insertions at random locations: sorted-btree vs Array vs Map

0.5     Insert 1000 pairs in sorted array
0.6     Insert 1000 pairs in B+ tree
0.1     Insert 1000 pairs in ES6 Map (hashtable)

13.2    Insert 10000 pairs in sorted array
7.2     Insert 10000 pairs in B+ tree
1.3     Insert 10000 pairs in ES6 Map (hashtable)

56980   Insert 100000 pairs in sorted array
122     Insert 100000 pairs in B+ tree
17.7    Insert 100000 pairs in ES6 Map (hashtable)

SLOW!   Insert 1000000 pairs in sorted array
1354    Insert 1000000 pairs in B+ tree
304.5   Insert 1000000 pairs in ES6 Map (hashtable)

Insert in order, scan, delete: sorted-btree vs Array vs Map

0.4     Insert 1000 sorted pairs in array
0.6     Insert 1000 sorted pairs in B+ tree
0.1     Insert 1000 sorted pairs in Map hashtable
0       Sum of all values with forEach in sorted array: 27350180
0       Sum of all values with forEachPair in B+ tree: 27350180
0       Sum of all values with forEach in B+ tree: 27350180
0       Sum of all values with iterator in B+ tree: 27350180
0       Sum of all values with forEach in Map: 27350180
0.1     Delete every second item in sorted array
0.1     Delete every second item in B+ tree
0       Delete every second item in Map hashtable

3.9     Insert 10000 sorted pairs in array
6.7     Insert 10000 sorted pairs in B+ tree
1.3     Insert 10000 sorted pairs in Map hashtable
0.2     Sum of all values with forEach in sorted array: 2716659330
0.3     Sum of all values with forEachPair in B+ tree: 2716659330
0.4     Sum of all values with forEach in B+ tree: 2716659330
0.3     Sum of all values with iterator in B+ tree: 2716659330
0.2     Sum of all values with forEach in Map: 2716659330
1.2     Delete every second item in sorted array
1.1     Delete every second item in B+ tree
0.3     Delete every second item in Map hashtable

68.4    Insert 100000 sorted pairs in array
72.7    Insert 100000 sorted pairs in B+ tree
18.4    Insert 100000 sorted pairs in Map hashtable
2.5     Sum of all values with forEach in sorted array: 275653049020
3.3     Sum of all values with forEachPair in B+ tree: 275653049020
4.5     Sum of all values with forEach in B+ tree: 275653049020
2.8     Sum of all values with iterator in B+ tree: 275653049020
2.2     Sum of all values with forEach in Map: 275653049020
2420    Delete every second item in sorted array
14.4    Delete every second item in B+ tree
3.7     Delete every second item in Map hashtable

969     Insert 1000000 sorted pairs in array
773     Insert 1000000 sorted pairs in B+ tree
305.5   Insert 1000000 sorted pairs in Map hashtable
25.3    Sum of all values with forEach in sorted array: 27510295368690
32.4    Sum of all values with forEachPair in B+ tree: 27510295368690
46.1    Sum of all values with forEach in B+ tree: 27510295368690
29.9    Sum of all values with iterator in B+ tree: 27510295368690
22      Sum of all values with forEach in Map: 27510295368690
SLOW!   Delete every second item in sorted array
305.5   Delete every second item in B+ tree
95.6    Delete every second item in Map hashtable

Version history

v1.6.0

Added BTree.getPairOrNextLower and BTree.getPairOrNextHigher methods (PR #23)
Added optional second parameter reusedArray to nextHigherPair and nextLowerPair (PR #23)
Optimizations added in diffAgainst (PR #24) and nextLowerPair (PR #23)

v1.5.0

Added BTree.diffAgainst method (PR #16)
Added simpleComparator function (PR #15)
Improved defaultComparator (PR #15) to support edge cases better. Most notably, heterogenous key types will no longer cause trouble such as failure to find keys that are, in fact, present in the tree. BTree is slightly slower using the new default comparator, but the benchmarks above have not been refreshed. For maximum performance, use simpleComparator or a custom comparator as the second constructor parameter. The simplest possible comparator is (a, b) => a - b, which works for finite numbers only.

v1.4.0

Now built as CommonJS module instead of UMD module, for better compatibility with webpack. No semantic change.

v1.3.0

Now built with TypeScript v3.8.3. No semantic change.

v1.2.4

Issue #9 fixed: nextLowerPair(0) was being treated like nextLowerPair(undefined), and nextLowerPair(undefined) was returning the second-highest pair when it should have returned the highest pair.

v1.2.3

Important bug fix in deletion code avoids occasional tree corruption that can occur after a series of delete operations
Add typings option in package.json so that tsc works for end-users

v1.2

Added a complete lattice of interfaces as described above.
Interfaces have been moved to a separate interfaces.d.ts file which is re-exported by the main module in b+tree.d.ts.

v1.1

Added isEmpty property getter
Added nextHigherPair, nextHigherKey, nextLowerPair, nextLowerKey methods
Added editAll, which is like editRange but touches all keys
Added deleteKeys for deleting a sequence of keys (iterable)
Added persistent methods with, withPairs, withKeys, without, withoutKeys, withoutRange
Added functional methods filter, reduce, mapValues
Added greedyClone for cloning nodes immediately, to avoid marking the original tree as shared which slows it down.
Relaxed type constraint on second parameter of entries/entriesReversed
Renamed setRange to setPairs for logical consistency with withoutPairs and withoutRange. The old name is deprecated but added to the prototype as a synonym. setPairs returns the number of pairs added instead of this.
Added export EmptyBTree, a frozen empty tree

v1.0: Initial version

With fast cloning and all that good stuff

Endnote

♥ This package was made to help people learn TypeScript & React.

Are you a C# developer? You might like the similar data structures I made for C# (BDictionary, BList, etc.), and other dynamically persistent collection types.

You might think that the package name "sorted btree" is overly redundant, but I did make a data structure similar to B+ Tree that is not sorted. I called it the A-List (C#). But yeah, the names btree and bplustree were already taken, so what was I supposed to do, right?

sorted-btree