Go Data Structures

by Tim Henderson (tim.tadh@gmail.com)

Copyright 2013, Licensed under the GPL version 2. Please reach out to me directly if you require another licensing option. I am willing to work with you.

Purpose

To collect many important data structures for usage in go programs. Golang's standard library lacks many useful and important structures. This library attempts to fill the gap. I have implemented data-structure's as I have needed them. If there is a missing structure or even just a missing (or incorrect) method open an issue, send a pull request, or send an email patch.

The library also provides generic types to allow the user to swap out various data structures transparently. The interfaces provide operation for adding, removing, retrieving objects from collections as well as iterating over the collection using functional iterators.

The tree sub-package provides a variety of generic tree traversals. The tree traversals and other iterators in the package use a functional iteration technique detailed on my blog.

I hope you find my library useful. If you are using it drop me a line I would love to hear about it.

Current Collection

Lists

Doubly Linked List linked.LinkedList

A simple an extensible doubly linked list. It is Equatable Sortable, and Hashable as are the Nodes.

Array List list.List

Similar to a Java ArrayList or a Python or Ruby "list". There is a version (called Sortable) which integrates with the "sort" package from the standard library.

Sorted Array List list.Sorted

Keeps the ArrayList in sorted order for you.

Sorted Set set.SortedSet

Built on top of *list.Sorted, it provides basic set operations. With set.SortedSet you don't have to write code re-implementing sets with the map[type] datatype. Supports: intersection, union, set difference and overlap tests.

Map Set set.MapSet

Construct a types.Map from any types.Set.

Set Map set.SetMap

Construct a set from any types.Map.

Unique Deque linked.UniqueDeque

A double ended queue that only allows unique items inside. Constructed from a doubly linked list and a linear hash table.

Fixed Size Lists

Both list.List and list.Sorted have alternative constructors which make them fixed size. This prevents them from growing beyond a certain size bound and is useful for implementing other data structures on top of them.

Serialization

list.List, list.Sorted, and set.SortedSet all can be serialized provided their contents can be serialized. They are therefore suitable for being sent over the wire. See this example for how to use the serialization.

Heaps and Priority Queues

Binary Heap heap/Heap

This is a binary heap for usage as a priority queue. The priorities are given to items in the queue on insertion and cannot be changed after insertion. It can be used as both a min heap and a max heap.

Unique Priority Queue heap/UniquePQ

A priority queue which only allows unique entries.

Trees

AVL Tree tree/avl.AvlTree

An AVL Tree is a height balanced binary search tree. Insertion and retrieval are both O(log(n)) where n is the number items in the tree.

Immutable AVL Tree tree/avl.ImmutableAvlTree

This version of the classic is immutable and should be thread safe due to immutability. However, there is a performance hit:

BenchmarkAvlTree           10000            166657 ns/op
BenchmarkImmutableAvlTree   5000            333709 ns/op

Ternary Search Trie trie.TST

A ternary search trie is a symbol table specialized to byte strings. Ternary Search Tries (TSTs) are a particularly fast version of the more common R-Way Trie. They utilize less memory allowing them to store more data while still retaining all of the flexibility of the R-Way Trie. TSTs can be used to build a suffix tree for full text string indexing by storing every suffix of each string in addition to the string. However, even without storing all of the suffixes it is still a great structure for flexible prefix searches. For instance, TSTs can be used to implement extremely fast auto-complete functionality.

B+Tree tree/bptree.BpTree

A B+Tree is a general symbol table usually used for database indices. This implementation is not currently thread safe. However, unlike many B+Trees it fully supports duplicate keys making it suitable for use as a Multi-Map. There is also a variant which has unique keys, bptree.BpMap. B+Trees are storted and efficient to iterate over making them ideal choices for storing a large amount of data in sorted order. For storing a very large amount of data please utilize the fs2 version, fs2/bptree. fs2 utilizes memory mapped files in order to allow you to store more data than your computer has RAM.

Hash Tables

Separate Chaining Hash Table hashtable.Hash

See hashtable/hashtable.go. An implementation of the classic hash table with separate chaining to handle collisions.

Linear Hash Table with AVL Tree Buckets hashtable.LinearHash

See hashtables/linhash.go. An implementation of Linear Hashing, a technique usually used for secondary storage hash tables. Often employed by databases and file systems for hash indices. This version is mostly instructional see the accompanying blog post. If you want a disk backed version check out my file-structures repository. See the linhash directory.

Exceptions, Errors, and Testing

Errors errors

By default, most errors in Go programs to not track where they were created. Many programmers quickly discover they need to have stack traces associated with their errors. This is a light weight package which adds stack traces to errors. It also provides a very very simple logging function that reports where in your code your printed out the log. This is not a full featured logging solution but rather a replacement for using fmt.Printf when debugging.

Test Support test

The test package provides two minimal assertions and a way to get random strings and data. It also seeds the math/rand number generator. I consider this to the bare minimum of what is often needed when testing go code particularly data-structures. Since this package seeks to be entirely self contained with no dependencies no external testing package is used. This package is slowly being improved to encompass more common functionality between the different tests.

Exceptions as a Library exc

The exc package provides support for exceptions. They work very similarly to the way unchecked exceptions work in Java. They are built on top of the built-in panic and recover functions. See the README in the package for more information or checkout the documentation. They should play nice with the usual way of handling errors in Go and provide an easy way to create public APIs which return errors rather than throwing these non-standard exceptions.

Benchmarks

Note: these benchmarsk are fairly old and probably not easy to understand. Look at the relative difference not the absolute numbers as they are misleading. Each benchmark does many operations per "test" which makes it difficult to compare these numbers to numbers found elsewhere.

Benchmarks Put + Remove

$ go test -v -bench '.*' \
>   github.com/timtadh/data-structures/hashtable
>   github.com/timtadh/data-structures/tree/...
>   github.com/timtadh/data-structures/trie

BenchmarkGoMap             50000             30051 ns/op
BenchmarkMLHash            20000             78840 ns/op
BenchmarkHash              20000             81012 ns/op
BenchmarkTST               10000            149985 ns/op
BenchmarkBpTree            10000            185134 ns/op
BenchmarkAvlTree           10000            193069 ns/op
BenchmarkImmutableAvlTree   5000            367602 ns/op
BenchmarkLHash              1000           2743693 ns/op

Benchmarks Put

BenchmarkGoMap            100000             22036 ns/op
BenchmarkMLHash            50000             52104 ns/op
BenchmarkHash              50000             53426 ns/op
BenchmarkTST               50000             69852 ns/op
BenchmarkBpTree            20000             76124 ns/op
BenchmarkAvlTree           10000            142104 ns/op
BenchmarkImmutableAvlTree  10000            302196 ns/op
BenchmarkLHash              1000           1739710 ns/op

The performance of the in memory linear hash (MLHash) is slightly improved since the blog post do to the usage of an AVL Tree tree/avltree.go instead of an unbalanced binary search tree.

Related Projects

• fs2 Memory mapped datastructures. A B+Tree, a list, and a platform for implementing more.

• file-structures The previous version of fs2 of disk based file-structures. Also includes a linear virtual hashing implementation.