In Go, what's a concise/well-performing way to deep copy a slice? I need to copy the slice to a new backing array, because the other array is owned by something else and may be modified after the copy.
I'm currently doing it like this:
copy := append([]T{}, orig...)
where T
is the element type of orig
.
Not sure which solution is fastest without a benchmark, but an alternative is using the built in copy
:
cpy := make([]T, len(orig))
copy(cpy, orig)
From the documentation:
func copy(dst, src []Type) int
The copy built-in function copies elements from a source slice into a destination slice. (As a special case, it also will copy bytes from a string to a slice of bytes.) The source and destination may overlap. Copy returns the number of elements copied, which will be the minimum of len(src) and len(dst).
Note
The solution will copy all the values in the slice. If the slice contains pointers or structs with pointer fields, these pointer values will still point to the same values as the orig
slice.
Benchmark
Benchmarking the two options, you can see they have very similar performance.
BenchmarkCopy 100000 24724 ns/op
BenchmarkAppend 100000 24967 ns/op
ok benchmark 5.478s
This is the benchmark code:
package main
import "testing"
var result []T
const size = 10000
type T int
func BenchmarkCopy(b *testing.B) {
orig := make([]T, size)
for n := 0; n < b.N; n++ {
cpy := make([]T, len(orig))
copy(cpy, orig)
orig = cpy
}
result = orig
}
func BenchmarkAppend(b *testing.B) {
orig := make([]T, size)
for n := 0; n < b.N; n++ {
cpy := append([]T{}, orig...)
orig = cpy
}
result = orig
}
I am not sure when/if the zero-fill is performed. But if you look at the assembly, in the append version you will have:
CALL ,runtime.growslice(SB)
while the copy will call:
CALL ,runtime.makeslice(SB)
and I would guess that both of these calls performs the zero-fill.