Go Streams API

Type safe Stream processing library inspired in the Java Streams API.

Table of contents

• Requirements
• Usage examples
• Limitations
• Performance
• Completion status
• Extra credits

Requirements

• Go 1.18 or higher

This library makes intensive usage of Type Parameters (generics) so it is not compatible with any Go version lower than 1.18.

Usage examples

For more details about the API, and until pkg.go.dev, is able to parse documentation for functions and types using generics, you can have a quick look to the generated go doc text descriptions, or just let that the embedded document browser of your IDE does the job.

Example 1: basic creation, transformation and iteration

1. Creates a literal stream containing all the integers from 1 to 11.
2. From the Stream, selects all the integers that are prime
3. For each filtered int, prints a message.

import (
  "fmt"
  "github.com/mariomac/gostream/stream"
)

func main() {
  stream.Of(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11).
    Filter(isPrime).
    ForEach(func(n int) {
      fmt.Printf("%d is a prime number\n", n)
    })
}

func isPrime(n int) bool {
  for i := 2; i < n/2; i++ {
    if n%i == 0 {
      return false
    }
  }
  return true
}

Output:

1 is a prime number
2 is a prime number
3 is a prime number
5 is a prime number
7 is a prime number
11 is a prime number

Example 2: generation, map, limit and slice conversion

1. Creates an infinite stream of random integers (no problem, streams are evaluated lazily!)
2. Divides the random integer to get a number between 1 and 6
3. Limits the infinite stream to 5 elements.
4. Collects the stream items as a slice.

rand.Seed(time.Now().UnixMilli())
fmt.Println("let me throw 5 times a dice for you")

results := stream.Generate(rand.Int).
    Map(func(n int) int {
        return n%6 + 1
    }).
    Limit(5).
    ToSlice()

fmt.Printf("results: %v\n", results)

Output:

let me throw 5 times a dice for you
results: [3 5 2 1 3]

Example 3: Generation from an iterator, Map to a different type

1. Generates an infinite stream composed by 1, double(1), double(double(1)), etc... and cut it to 6 elements.
2. Maps the numbers' stream to a strings' stream. Because, at the moment, go does not allow type parameters in methods, we need to invoke the stream.Map function instead of the numbers.Map method because the contained type of the output stream (string) is different than the type of the input stream (int).
3. Converts the words stream to a slice and prints it.

func main() {
    numbers := stream.Iterate(1, double).Limit(6)
    words := stream.Map(numbers, asWord).ToSlice()
    fmt.Println(words)
}

func double(n int) int {
    return 2 * n
}

func asWord(n int) string {
    if n < 10 {
        return []string{"zero", "one", "two", "three", "four", "five",
            "six", "seven", "eight", "nine"}[n]
    } else {
        return "many"
    }
}

Output:

[one two four eight many many]

Example 4: deduplication of elements

Following example requires to compare the elements of the Stream, so the Stream needs to be composed by comparable elements (this is, accepted by the the == and != operators):

1. Instantiate a Stream of comparable items.
2. Pass it to the Distinct method, that will return a copy of the original Stream without duplicates
3. Operating as any other stream.

words := stream.Distinct(
  stream.Of("hello", "hello", "!", "ho", "ho", "ho", "!"),
).ToSlice()

fmt.Printf("Deduplicated words: %v\n", words)

Output:

Deduplicated words: [hello ! ho]

Example 5: sorting from higher to lower

1. Generate a stream of uint32 numbers.
2. Picking up 5 elements.
3. Sorting them by the inverse natural order (from higher to lower)
   • It's important to limit the number of elements, avoiding invoking Sorted over an infinite stream (otherwise it would panic).

fmt.Println("picking up 5 random numbers from higher to lower")
stream.Generate(rand.Uint32).
    Limit(5).
    Sorted(order.Inverse(order.Natural[uint32])).
    ForEach(func(n uint32) {
        fmt.Println(n)
    })

Output:

picking up 5 random numbers from higher to lower
4039455774
2854263694
2596996162
1879968118
1823804162

Example 6: Reduce and helper functions

1. Generate an infinite incremental Stream (1, 2, 3, 4...) using the stream.Iterate instantiator and the item.Increment helper function.
2. Limit the generated to 8 elements
3. Reduce all the elements multiplying them using the item.Multiply helper function

fac8, _ := stream.Iterate(1, item.Increment[int]).
    Limit(8).
    Reduce(item.Multiply[int])
fmt.Println("The factorial of 8 is", fac8)

Output:

The factorial of 8 is 40320

Limitations

Due to the initial limitations of Go generics, the API has the following limitations. We will work on overcome them as long as new features are added to the Go type parameters specification.

• You can use Map and FlatMap as method as long as the output element has the same type of the input. If you need to map to a different type, you need to use stream.Map or stream.FlatMap as functions.
• There is no Distinct method. There is only a stream.Distinct function.
• There is no ToMap method. There is only a stream.ToMap function.

Performance

You might want to check: Performance comparison of Go functional stream libraries.

Streams aren't the fastest option. They are aimed for complex workflows where you can sacrifice few microseconds for the sake of code organization and readability. Also disclaimer: functional streams don't have to always be the most readable option.

The following results show the difference in performance for an arbitrary set of operations in an imperative form versus the functional form using streams (see stream/benchs_test.go file):

$ gotip test -bench=. -benchmem  ./...
goos: darwin
goarch: amd64
pkg: github.com/mariomac/gostream/stream
cpu: Intel(R) Core(TM) i5-5257U CPU @ 2.70GHz
BenchmarkImperative-4            2098518               550.6 ns/op          1016 B/op          7 allocs/op
BenchmarkFunctional-4             293095              3653 ns/op            2440 B/op         23 allocs/op

If you want a more performant, parallelizable alternative to create data processing pipelines (following a programming model focused on Extract-Transform-Load, ETL), you could give a try to my alternative project: PIPES: Processing In Pipeline-Embedded Stages.

Completion status

• Stream instantiation functions
  • Comparable
  • Concat
  • Empty
  • Generate
  • Iterate
  • Of
  • OfMap
  • OfSlice
  • OfChannel
• Stream transformers
  • Distinct
  • Filter
  • FlatMap
  • Limit
  • Map
  • Peek
  • Skip
  • Sorted
• Collectors/Terminals
  • ToMap
  • ToSlice
  • AllMatch
  • AnyMatch
  • Count
  • FindFirst
  • ForEach
  • Max
  • Min
  • NoneMatch
  • Reduce
• Auxiliary Functions
  • Add (for numbers)
  • Increment (for numbers)
  • IsZero
  • Multiply (for numbers)
  • Neg (for numbers)
  • Not (for bools)
• Future
  • Collectors for future standard generic data structures
    • E.g. [ ] Join (for strings)
  • Allow users implement their own Comparable or Ordered types
  • More operations inspired in the Kafka Streams API
  • Parallel streams
    • FindAny

Extra credits

The Stream processing and aggregation functions are heavily inspired in the Java Stream Specification.

Stream code documentation also used Stream Javadoc as an essential reference and might contain citations from it.