Immutable

This repository contains generic immutable collection types for Go. It includes List, Map, SortedMap, Set and SortedSet implementations. Immutable collections can provide efficient, lock free sharing of data by requiring that edits to the collections return new collections.

The collection types in this library are meant to mimic Go built-in collections such asslice and map. The primary usage difference between Go collections and immutable collections is that immutable collections always return a new collection on mutation so you will need to save the new reference.

This project is a fork of github.com/benbjohnson/immutable with additional performance enhancements and builder APIs.

Performance: This library includes batch builders that provide high accelaration for bulk operations (vs. discreet insert), with optimized memory usage and automatic batching. Regular operations maintain ~2x overhead compared to Go's built-in collections while providing thread-safe immutability.

Immutable collections are not for every situation, however, as they can incur additional CPU and memory overhead. Please evaluate the cost/benefit for your particular project.

Special thanks to the Immutable.js team as the List & Map implementations are loose ports from that project.

Forked from https://github.com/benbjohnson/immutable with the following enhancements:

Performance Optimizations

Memory Architecture Improvements:

• Hybrid Data Structures:
  • List: Uses a simple slice for small lists (< 32 elements) for up to 2x faster operations and ~85% less memory usage in common cases, transparently converting to a HAMT for larger sizes.
  • Map: Small-structure fast paths via builder initial flush and tiny array-node updates (≤ 8 keys); core Map remains HAMT.
• Pointer-Based Array Sharing (planned): Reduce allocations in mapHashArrayNode.clone() via pointer-backed children with copy-on-write
• Lazy Copy-on-Write: Arrays shared via pointers until actual modification, reducing memory overhead by 6-8%
• Cache-Friendly Design: Improved memory layout for better CPU cache utilization

Concurrency with immutable collections

Immutable structures are snapshot-based: every mutation returns a new instance; the original remains unchanged. This makes concurrent reads safe without locks.

• Recommended pattern: a single writer goroutine owns the evolving collection and applies updates received via a channel. It then sends immutable snapshots to readers.
• Readers can safely use received snapshots without copying. Structural sharing ensures those snapshots are cheap to create and pass around.
• If you need a single, shared, evolving reference updated by multiple goroutines, synchronize the reference update (mutex or atomic CAS on a pointer). Without that, simultaneous Enqueue/Dequeue on the same snapshot may race logically (e.g., multiple consumers reading the same head, or lost enqueues).
• Builders are mutable conveniences and are not safe for concurrent use; keep them confined to one goroutine.

Batch Builders

Complete High-Performance Builder Suite:

• BatchListBuilder: up 19x faster in tests (vs. discreet ops) bulk list construction with configurable batch sizes
• BatchMapBuilder: Measured gains on bulk construction; biggest wins for initial tiny batches and small structures
• BatchSetBuilder & BatchSortedSetBuilder: Efficient bulk set construction
• StreamingListBuilder & StreamingMapBuilder: Auto-flush with functional operations (filter, transform)
• SortedBatchBuilder: Optimized sorted map construction with optional sort maintenance

Functional Programming Features:

• Stream processing with automatic memory management
• Filter and transform operations for bulk data processing
• Configurable auto-flush thresholds for memory efficiency

Enhanced Testing & Validation

Comprehensive Test Coverage:

• Extensive benchmark suite measuring performance improvements
• Memory profiling and allocation analysis
• Race condition testing (go test -race) for thread safety validation
• Edge case and error condition testing for all new builders
• Large-scale performance validation (100-100K elements)

Architectural Enhancements

Thread Safety & Immutability:

• Lock-free operations with atomic copying
• Structural sharing maintains thread safety
• Zero-overhead abstractions (Sets inherit Map optimizations)

List

The List type represents a sorted, indexed collection of values and operates similarly to a Go slice. It supports efficient append, prepend, update, and slice operations.

Adding list elements

Elements can be added to the end of the list with the Append() method or added to the beginning of the list with the Prepend() method. Unlike Go slices, prepending is as efficient as appending.

// Create a list with 3 elements.
l := immutable.NewList[string]()
l = l.Append("foo")
l = l.Append("bar")
l = l.Prepend("baz")

fmt.Println(l.Len())  // 3
fmt.Println(l.Get(0)) // "baz"
fmt.Println(l.Get(1)) // "foo"
fmt.Println(l.Get(2)) // "bar"

Note that each change to the list results in a new list being created. These lists are all snapshots at that point in time and cannot be changed so they are safe to share between multiple goroutines.

Updating list elements

You can also overwrite existing elements by using the Set() method. In the following example, we'll update the third element in our list and return the new list to a new variable. You can see that our old l variable retains a snapshot of the original value.

l := immutable.NewList[string]()
l = l.Append("foo")
l = l.Append("bar")
newList := l.Set(2, "baz")

fmt.Println(l.Get(1))       // "bar"
fmt.Println(newList.Get(1)) // "baz"

Deriving sublists

You can create a sublist by using the Slice() method. This method works with the same rules as subslicing a Go slice:

l = l.Slice(0, 2)

fmt.Println(l.Len())  // 2
fmt.Println(l.Get(0)) // "baz"
fmt.Println(l.Get(1)) // "foo"

Please note that since List follows the same rules as slices, it will panic if you try to Get(), Set(), or Slice() with indexes that are outside of the range of the List.

Searching list elements

You can check if a value exists in the list using Contains(). For comparable types, equality uses ==. For non-comparable types (e.g., []byte), it falls back to reflect.DeepEqual.

// Basic usage
l := immutable.NewList[int]()
for i := 0; i < 5; i++ { l = l.Append(i) }

fmt.Println(l.Contains(3))  // true
fmt.Println(l.Contains(10)) // false

// Non-comparable example uses DeepEqual under the hood
b := immutable.NewList[[]byte]([]byte("foo"), []byte("bar"))
fmt.Println(b.Contains([]byte("foo"))) // true

For full control and best performance on custom types, use ContainsFunc() to provide an equality function:

type node struct{ v int }
l := immutable.NewList[*node](&node{v: 1}, &node{v: 2})

eq := func(a, b *node) bool { return a != nil && b != nil && a.v == b.v }
fmt.Println(l.ContainsFunc(&node{v: 2}, eq)) // true
fmt.Println(l.ContainsFunc(&node{v: 3}, eq)) // false

Iterating lists

Iterators provide a clean, simple way to iterate over the elements of the list in order. This is more efficient than simply calling Get() for each index.

Below is an example of iterating over all elements of our list from above:

itr := l.Iterator()
for !itr.Done() {
	index, value, _ := itr.Next()
	fmt.Printf("Index %d equals %v\n", index, value)
}

// Index 0 equals baz
// Index 1 equals foo

By default iterators start from index zero, however, the Seek() method can be used to jump to a given index.

Efficiently building lists

If you are building large lists, it is significantly more efficient to use the ListBuilder. It uses nearly the same API as List except that it updates a list in-place until you are ready to use it. This can improve bulk list building by 10x or more.

For even better performance with bulk operations (100+ elements), see the Advanced Batch Builders section which provides up to 19x performance improvements.

b := immutable.NewListBuilder[string]()
b.Append("foo")
b.Append("bar")
b.Set(2, "baz")

l := b.List()
fmt.Println(l.Get(0)) // "foo"
fmt.Println(l.Get(1)) // "baz"

Builders are invalid after the call to List().

Queue

An immutable FIFO queue with amortized O(1) Enqueue, Dequeue, and Peek, implemented using the classic Okasaki two-list representation backed by cons-lists. Each operation creates at most one cons cell, and the periodic O(n) reversal is amortized across n operations.

q := immutable.NewQueue[int]()
q = q.Enqueue(1).Enqueue(2).Enqueue(3)

v, ok := q.Peek() // v=1, ok=true

q, v, ok = q.Dequeue() // v=1, ok=true
q, v, ok = q.Dequeue() // v=2, ok=true

itr := q.Iterator()
for !itr.Done() {
    idx, x, _ := itr.Next()
    _ = idx
    _ = x
}

// Builder
b := immutable.NewQueueBuilder[int]()
b.Enqueue(10)
b.EnqueueSlice([]int{20, 30})
finalQ := b.Queue()

Thread safety: operations return new queues; existing instances are never mutated, so concurrent reads are safe.

Map

The Map represents an associative array that maps unique keys to values. It is implemented to act similarly to the built-in Go map type. It is implemented as a Hash-Array Mapped Trie.

Maps require a Hasher to hash keys and check for equality. There are built-in hasher implementations for most primitive types such as int, uint, and string keys. You may pass in a nil hasher to NewMap() if you are using one of these key types.

Setting map key/value pairs

You can add a key/value pair to the map by using the Set() method. It will add the key if it does not exist or it will overwrite the value for the key if it does exist.

Values may be fetched for a key using the Get() method. This method returns the value as well as a flag indicating if the key existed. The flag is useful to check if a nil value was set for a key versus a key did not exist.

m := immutable.NewMap[string,int](nil)
m = m.Set("jane", 100)
m = m.Set("susy", 200)
m = m.Set("jane", 300) // overwrite

fmt.Println(m.Len())   // 2

v, ok := m.Get("jane")
fmt.Println(v, ok)     // 300 true

v, ok = m.Get("susy")
fmt.Println(v, ok)     // 200, true

v, ok = m.Get("john")
fmt.Println(v, ok)     // nil, false

Removing map keys

Keys may be removed from the map by using the Delete() method. If the key does not exist then the original map is returned instead of a new one.

m := immutable.NewMap[string,int](nil)
m = m.Set("jane", 100)
m = m.Delete("jane")

fmt.Println(m.Len())   // 0

v, ok := m.Get("jane")
fmt.Println(v, ok)     // nil false

Iterating maps

Maps are unsorted, however, iterators can be used to loop over all key/value pairs in the collection. Unlike Go maps, iterators are deterministic when iterating over key/value pairs.

m := immutable.NewMap[string,int](nil)
m = m.Set("jane", 100)
m = m.Set("susy", 200)

itr := m.Iterator()
for !itr.Done() {
	k, v := itr.Next()
	fmt.Println(k, v)
}

// susy 200
// jane 100

Note that you should not rely on two maps with the same key/value pairs to iterate in the same order. Ordering can be insertion order dependent when two keys generate the same hash.

Efficiently building maps

If you are executing multiple mutations on a map, it can be much more efficient to use the MapBuilder. It uses nearly the same API as Map except that it updates a map in-place until you are ready to use it.

For enhanced performance with bulk operations, see the Advanced Batch Builders section which provides additional optimizations and functional programming capabilities.

b := immutable.NewMapBuilder[string,int](nil)
b.Set("foo", 100)
b.Set("bar", 200)
b.Set("foo", 300)

m := b.Map()
fmt.Println(m.Get("foo")) // "300"
fmt.Println(m.Get("bar")) // "200"

Builders are invalid after the call to Map().

Implementing a custom Hasher

If you need to use a key type besides int, uint, or string then you'll need to create a custom Hasher implementation and pass it to NewMap() on creation.

Hashers are fairly simple. They only need to generate hashes for a given key and check equality given two keys.

Security Note: A poorly implemented Hasher can result in frequent hash collisions, which will degrade the Map's performance from O(log n) to O(n), making it vulnerable to algorithmic complexity attacks (a form of Denial of Service). Ensure your Hash function provides a good distribution.

type Hasher[K any] interface {
	Hash(key K) uint32
	Equal(a, b K) bool
}

Please see the internal intHasher, uintHasher, stringHasher, and byteSliceHasher for examples.

Sorted Map

The SortedMap represents an associative array that maps unique keys to values. Unlike the Map, however, keys can be iterated over in-order. It is implemented as a B+tree.

Sorted maps require a Comparer to sort keys and check for equality. There are built-in comparer implementations for int, uint, and string keys. You may pass a nil comparer to NewSortedMap() if you are using one of these key types.

The API is identical to the Map implementation. The sorted map also has a companion SortedMapBuilder for more efficiently building maps.

Implementing a custom Comparer

If you need to use a key type besides int, uint, or string or derived types, then you'll need to create a custom Comparer implementation and pass it to NewSortedMap() on creation.

Security Note: A slow Comparer implementation can severely degrade the performance of all SortedMap operations, making it vulnerable to Denial of Service attacks. Ensure your Compare function is efficient.

Comparers on have one method—Compare(). It works the same as the strings.Compare() function. It returns -1 if a is less than b, returns 1 if a is greater than b, and returns 0 if a is equal to b.

type Comparer[K any] interface {
	Compare(a, b K) int
}

Please see the internal defaultComparer for an example, bearing in mind that it works for several types.

Set

The Set represents a collection of unique values, and it is implemented as a wrapper around a Map[T, struct{}].

Like Maps, Sets require a Hasher to hash keys and check for equality. There are built-in hasher implementations for most primitive types such as int, uint, and string keys. You may pass in a nil hasher to NewSet() if you are using one of these key types.

Sorted Set

The SortedSet represents a sorted collection of unique values. Unlike the Set, however, keys can be iterated over in-order. It is implemented as a B+tree.

Sorted sets require a Comparer to sort values and check for equality. There are built-in comparer implementations for int, uint, and string keys. You may pass a nil comparer to NewSortedSet() if you are using one of these key types.

The API is identical to the Set implementation.

Advanced Batch Builders

For high-performance bulk operations, this library provides advanced batch builders that can dramatically improve performance for large-scale data construction. These builders use internal batching and mutable operations to minimize allocations and provide up to 19x performance improvements for bulk operations.

Batch List Builder

The BatchListBuilder provides batched list construction with configurable batch sizes for optimal performance:

// Create a batch builder with batch size of 64
builder := immutable.NewBatchListBuilder[int](64)

// Add many elements efficiently
for i := 0; i < 10000; i++ {
    builder.Append(i)
}

// Or add slices efficiently
values := []int{1, 2, 3, 4, 5}
builder.AppendSlice(values)

list := builder.List() // 19x faster than individual Append() calls

Performance: Up to 19x faster than direct construction for large lists.

Batch Map Builder

The BatchMapBuilder provides batched map construction with automatic flushing:

// Create a batch map builder with batch size of 32
builder := immutable.NewBatchMapBuilder[string, int](nil, 32)

// Add many entries efficiently
for i := 0; i < 10000; i++ {
    builder.Set(fmt.Sprintf("key-%d", i), i)
}

// Or add from existing maps
entries := map[string]int{"a": 1, "b": 2, "c": 3}
builder.SetMap(entries)

m := builder.Map() // 8% faster + 5.8% less memory than regular builder

Performance: 8% faster with 5.8% memory reduction compared to regular builders.

Streaming Builders

Streaming builders provide auto-flush capabilities and functional operations:

// Streaming list builder with auto-flush at 1000 elements
builder := immutable.NewStreamingListBuilder[int](32, 1000)

// Functional operations
data := []int{1, 2, 3, 4, 5, 6, 7, 8, 9, 10}

// Filter even numbers
builder.Filter(data, func(x int) bool { return x%2 == 0 })

// Transform by doubling
builder.Transform(data, func(x int) int { return x * 2 })

list := builder.List() // Contains processed elements

// Streaming map builder with auto-flush and bulk operations
builder := immutable.NewStreamingMapBuilder[int, string](nil, 32, 500)

// Add individual entries (auto-flushes at 500 elements)
for i := 0; i < 1000; i++ {
    builder.Set(i, fmt.Sprintf("value-%d", i))
}

// Bulk operations with auto-flush
builder.SetMany(map[int]string{10: "ten", 20: "twenty", 30: "thirty"})

m := builder.Map()

Batch Set Builders

Set builders provide efficient bulk set construction:

// Batch set builder
builder := immutable.NewBatchSetBuilder[string](nil, 64)

values := []string{"apple", "banana", "cherry", "apple"} // "apple" duplicate
builder.AddSlice(values)

set := builder.Set() // Contains 3 unique values

// Sorted set builder with sort maintenance
sortedBuilder := immutable.NewBatchSortedSetBuilder[int](nil, 32, true)
numbers := []int{5, 2, 8, 1, 9, 3}
sortedBuilder.AddSlice(numbers)

sortedSet := sortedBuilder.SortedSet() // Automatically sorted: [1, 2, 3, 5, 8, 9]

Sorted Batch Builder

For sorted maps, use SortedBatchBuilder with optional sort maintenance:

// Maintain sort order in buffer for optimal insertion
builder := immutable.NewSortedBatchBuilder[int, string](nil, 32, true)

// Add in random order - automatically maintained in sorted buffer
builder.Set(3, "three")
builder.Set(1, "one")
builder.Set(2, "two")

sm := builder.SortedMap() // Efficiently constructed sorted map

Performance Guidelines

When to use batch builders:

• Building collections with 100+ elements
• Bulk data import/export operations
• Processing large datasets
• When memory efficiency is critical

Batch size recommendations:

• Small operations (< 1K elements): 16-32
• Medium operations (1K-10K elements): 32-64
• Large operations (> 10K elements): 64-128
• Memory-constrained environments: 16-32

Performance improvements:

• List construction: Up to 19x faster for bulk operations
• Map construction: 8% faster with 5.8% memory reduction
• Set construction: Inherits map performance benefits
• Streaming operations: Automatic memory management with functional programming

Contributing

The goal of immutable is to provide stable, reasonably performant, immutable collections library for Go that has a simple, idiomatic API. As such, additional features and minor performance improvements will generally not be accepted. If you have a suggestion for a clearer API or substantial performance improvement, please open an issue first to discuss. All pull requests without a related issue will be closed immediately.

Please submit issues relating to bugs & documentation improvements.

What's New (2025-09)

• zero dependencies
• Small-structure fast paths:
  • List: Batch flush extends slice-backed lists in a single allocation
  • Map: Empty-map batch flush builds an array node in one shot (last-write-wins); tiny array-node updates applied in-place when under threshold
• New builder APIs:
  • (*BatchListBuilder).Reset() and (*BatchMapBuilder).Reset() for builder reuse without reallocations
• Concurrency:
  • Added concurrent read benchmarks and mixed read/write benchmarks (immutable Map vs sync.Map)
  • Added concurrency correctness tests (copy-on-write isolation under concurrent readers)

Current Performance Snapshot

• Map Get (10K): immutable ~14.5 ns/op (0 allocs); builtin map ~6.8 ns/op; sync.Map Load ~20.3 ns/op
• Map RandomSet (10K): ~595–687 ns/op, 1421 B/op, 7 allocs/op (after tuning)
• Concurrent reads (ns/op, lower is better):
  • 1G: immutable 3.53 vs sync.Map 6.03
  • 4G: immutable 2.31 vs sync.Map 3.21
  • 16G: immutable 2.39 vs sync.Map 3.24
• Mixed read/write (ns/op):
  • 90/10 (9R/1W): immutable 26.0 vs sync.Map 38.4
  • 70/30 (7R/3W): immutable 24.6 vs sync.Map 65.0
  • 50/50 (5R/5W): immutable 27.3 vs sync.Map 47.4

New APIs (Builders)

// Reuse list builder across batches without reallocations
lb := immutable.NewBatchListBuilder[int](64)
// ... append/flush ...
lb.Reset() // clears state, keeps capacity

// Reuse map builder and retain hasher
mb := immutable.NewBatchMapBuilder[string,int](nil, 64)
// ... set/flush ...
mb.Reset() // clears state, preserves hasher & buffer capacity

Benchmarking & Profiling

• Run all benchmarks with allocations:

go test -bench=. -benchmem -count=3 ./...

• Profile a representative write-heavy benchmark:

# CPU and memory profiles (example: Map RandomSet, size=10K)
go test -bench=BenchmarkMap_RandomSet/size-10000 -benchmem -run="^$" \
  -cpuprofile=cpu.prof -memprofile=mem.prof -count=1

# Inspect hotspots
go tool pprof -top cpu.prof
go tool pprof -top -sample_index=alloc_space mem.prof

Optional: Enable PGO locally

# Generate a profile and write default.pgo
go test -bench=. -run="^$" -cpuprofile=cpu.prof -count=1
go tool pprof -proto cpu.prof > cpu.pb.gz
go tool pgo -compile=local -o default.pgo cpu.pb.gz

# Use the profile for builds/tests (Go 1.21+)
go test -bench=. -benchmem -count=3 -pgo=auto ./...

• Compare immutable vs sync.Map concurrent reads:

go test -bench=BenchmarkConcurrentReads -benchmem -run="^$" -count=1

• Mixed workload (reads/writes):

go test -bench=BenchmarkConcurrentMixed -benchmem -run="^$" -count=1