Algorithm library for golang1.18+ using type parameters.
# Or any other way to install go1.18 beta.
# https://go.dev/blog/go1.18beta2
go install golang.org/dl/go1.18beta2@latest
go1.18beta2 download
# Install library
go1.18beta2 get github.com/datosh/algopackage main
import (
"fmt"
"github.com/datosh/algo"
)
func main() {
sum := algo.Fold(func(a, b int) int { return a + b })
numbers := []int{1, 2, 3, 4, 5, 6, 7, 8, 9}
fmt.Println(sum(numbers)) // 45
}package main
import (
"fmt"
"strings"
"github.com/datosh/algo"
)
func main() {
onlyShortWords := algo.Filter(func(s string) bool { return len(s) <= 5 })
shout := algo.Map(func(s string) string { return strings.ToUpper(s) })
concat := algo.Fold(func(t1, t2 string) string { return t1 + " " + t2 })
words := []string{"Hello", "gophers", "World"}
sentence := concat(shout(onlyShortWords(words)))
fmt.Println(sentence) // HELLO WORLD
}Explore how a go library for algorithms COULD look like.
To get an idea what is usually implemented in such a library I looked at
- Currently only support slices
- "Type cast" a slice of build-in types
- "Type cast" a slice of complex types
- Iterator based approach to support any data structure
⚠️ No performance optimizations were implemented, yet!
These benchmarks were created to get a feeling for the rough performance characteristics of golangs new type parameters.
Baseline is a custom implementation for a specific type.
Filter/Map is the generic implementation returning a filter/map function.
Filter2/Map2 is the generic implementation doing the operation in one call.
Interfaced is a interface{} based implementations.
cpu: 11th Gen Intel(R) Core(TM) i7-11800H @ 2.30GHz
| Test | Iterations | Speed |
|---|---|---|
| Filter Baseline | 506 | 2989602 ns/op |
| Filter | 172 | 6455209 ns/op |
| Filter2 | 451 | 8465454 ns/op |
| Filter Interfaced | 48 | 25998184 ns/op |
| Map Baseline | 97946026 | 12.43 ns/op |
| Map | 6051519 | 182.3 ns/op |
| Map2 | 4241458 | 292.0 ns/op |
My take-away is that a performance hit is to be expected when using type
paremters compared to a custom implementation. On the other hand, if you
relied on interface{} & reflection based implementations in the past,
type parameters will provide you both with a performance as well as
safety (compiler checks) improvement.