krasun / trie

Missing Trie implementation for Go

trie

A missing trie implementation for Go.

Installation

go get github.com/krasun/trie

Usage

Known limitations:

1. Empty string keys are not supported. And functions won't return an error.
2. Keys are not ordered, so do not expect any ordering.

Without any stress:

t := trie.New() // or trie.NewRuneTrie()  

t.Insert("apple")
t.Insert("alphabet")
t.Insert("banana")

t.Contains("apple") // true
t.Contains("app") // false
t.Contains("banana") // true
t.Contains("ban") // false

t.SearchByPrefix("a") // []string{"apple", "alphabet"}

t.StartsWith("a") // true

A goroutine-safe (thread-safe) trie

By default, rune-wise trie are not safe to use concurrently, but it is easy to make them safe. You can wrap any trie into the safe version by:

safeTrie := trie.Safe(trie.NewTrie())

// the same interface as for a regular trie

Tests

To run tests, just execute:

$ go test . -race

Benchmarking

Zero heap allocations for Contains function:

$ go test -bench=. -benchmem -benchtime=1000x
goos: darwin
goarch: amd64
op	       0 allocs/op
BenchmarkRuneTrieInsert-8          	    1000	      7196 ns/op	    6984 B/op	     129 allocs/op
BenchmarkRuneTrieContains-8        	    1000	       517 ns/op	       0 B/op	       0 allocs/op
BenchmarkWordHashSetInsert-8       	    1000	      1406 ns/op	    1100 B/op	       4 allocs/op
BenchmarkWordHashSetContains-8     	    1000	       178 ns/op	       0 B/op	       0 allocs/op
PASS

But the hash set (map[string]{}struct) is much much more efficient than the trie. Carefully weigh when to use the trie.

Applications

Use the trie in cases when it only outperforms hash tables or hash sets (map[string]{}struct):

1. Search key by the prefix.

License

trie is released under the MIT license.