timsort is a Go implementation of Tim Peters' mergesort sorting algorithm. It's stable and runs in O(n) time for presorted inputs and O(n log n) otherwise.
For many input types it is 2-3 times faster than Go's built-in sorting.
The main drawback of this sort method is that it is not in-place (as any mergesort), and may put extra strain on garbage collector.
This implementation was ported to Go by Mike Kroutikov and derived from Java's TimSort object by Josh Bloch, which, in turn, was based on the original code by Tim Peters.
$ go get -u github.com/psilva261/timsort/v2
Inside the source directory, type
go test
to run test harness.
Inside the source directory, type
go test -test.bench=.*
to run benchmarks. Each combination of input type/size is presented to timsort, and, for comparison, to the standard Go sort (sort.Sort for ints or sort.Stable otherwise). See BENCHMARKS.md for more info and some benchmarking results.
package main
import (
"github.com/psilva261/timsort/v2"
"fmt"
"sort"
)
func main() {
l := []string{"c", "a", "b"}
timsort.TimSort(sort.StringSlice(l)
fmt.Printf("sorted array: %+v\n", l)
}
package main
import (
"github.com/psilva261/timsort/v2"
"fmt"
)
type Record struct {
ssn int
name string
}
func BySsn(a, b interface{}) bool {
return a.(Record).ssn < b.(Record).ssn
}
func ByName(a, b interface{}) bool {
return a.(Record).name < b.(Record).name
}
func main() {
db := make([]interface{}, 3)
db[0] = Record{123456789, "joe"}
db[1] = Record{101765430, "sue"}
db[2] = Record{345623452, "mary"}
// sorts array by ssn (ascending)
timsort.Sort(db, BySsn)
fmt.Printf("sorted by ssn: %v\n", db)
// now re-sort same array by name (ascending)
timsort.Sort(db, ByName)
fmt.Printf("sorted by name: %v\n", db)
}