tune slice append performance

[hjweddie]

I find some codes abount binlog, which is doing slices assignment or append could be simply changed, so that we can get some perfomance tuning.
Hope that it could help

some corresponding benchmark code may be as follow:

package main_test

import (
	"testing"
)

func BenchmarkAppend(b *testing.B) {
	strs := []string{"a", "b", "c"}
	tmp := make([]string, 0, len(strs))

	b.ResetTimer()
	for i := 0; i < b.N; i++ {
		for _, str := range strs {
			tmp = append(tmp, str)
		}
	}
}

func BenchmarkAssign(b *testing.B) {
	strs := []string{"a", "b", "c"}
	tmp := make([]string, len(strs))

	b.ResetTimer()
	for i := 0; i < b.N; i++ {
		for i, str := range strs {
			tmp[i] = str
		}
	}
}

[hjweddie]

umm, I will check ci failure result

[lance6716]

func BenchmarkAppend(b *testing.B) {
	strs := []string{"a", "b", "c"}
	tmp := make([]string, 0, len(strs))

	b.ResetTimer()
	for i := 0; i < b.N; i++ {
		for _, str := range strs {
			tmp = append(tmp, str)
		}
	}
}

for b.N larger than 1, your code will let tmp grow so it's slower.

[hjweddie]

func BenchmarkAppend(b *testing.B) {
	strs := []string{"a", "b", "c"}
	tmp := make([]string, 0, len(strs))

	b.ResetTimer()
	for i := 0; i < b.N; i++ {
		for _, str := range strs {
			tmp = append(tmp, str)
		}
	}
}

for b.N larger than 1, your code will let tmp grow so it's slower.

Agree it! I intend to reduce the overhead of slice growth

[hjweddie]

ci jobs pass

[lance6716]

I mean your benchmark is not correct

func BenchmarkAppend(b *testing.B) {
	strs := []string{"a", "b", "c"}
	tmp := make([]string, 0, len(strs))

	b.ResetTimer()
	for i := 0; i < b.N; i++ {
+		tmp = tmp[:0]
		for _, str := range strs {
			tmp = append(tmp, str)
		}
	}
}

goos: darwin
goarch: arm64
pkg: awesomeProject
BenchmarkAppend
BenchmarkAppend-8   	373262626	         3.161 ns/op
BenchmarkAssign
BenchmarkAssign-8   	541436124	         2.219 ns/op
PASS

[hjweddie]

package main_test

import (
"testing"
)

func BenchmarkAppend(b *testing.B) {
strs := []string{"a", "b", "c"}
tmp := make([]string, 0, len(strs))

b.ResetTimer()
for i := 0; i < b.N; i++ {
for _, str := range strs {
tmp = append(tmp, str)
}
}
}

func BenchmarkAssign(b *testing.B) {
strs := []string{"a", "b", "c"}
tmp := make([]string, len(strs))

b.ResetTimer()
for i := 0; i < b.N; i++ {
for i, str := range strs {
tmp[i] = str
}
}
}

You are right. I put slice intialization in the loop and find slice assignment gets just a little more performance than append.

Because function parseSingleEvent could be called with high frequency in binlog parsing, I think it may be acceptable

func BenchmarkAppend(b *testing.B) {
	strs := []string{"a", "b", "c"}

	b.ResetTimer()
	for i := 0; i < b.N; i++ {
		tmp := make([]string, 0, len(strs))
		for _, str := range strs {
			tmp = append(tmp, str)
		}
	}
}

func BenchmarkAssign(b *testing.B) {
	strs := []string{"a", "b", "c"}

	b.ResetTimer()
	for i := 0; i < b.N; i++ {
		tmp := make([]string, len(strs))
		for i, str := range strs {
			tmp[i] = str
		}
	}
}

goos: darwin
goarch: arm64
pkg: test
BenchmarkAppend-8   	29462312	        40.81 ns/op
BenchmarkAssign-8   	30202609	        39.69 ns/op
PASS
ok  	test	2.600s

[hjweddie]

If I try to make strs be a slice with a larger size, performance difference between assign and append will be larger too

func BenchmarkAppend(b *testing.B) {
	strs := make([]string, 0)
	for i := 0; i < 10000; i++ {
		strs = append(strs, "d")
	}

	b.ResetTimer()
	for i := 0; i < b.N; i++ {
		tmp := make([]string, 0, len(strs))
		for _, str := range strs {
			tmp = append(tmp, str)
		}
	}
}

func BenchmarkAssign(b *testing.B) {
	strs := make([]string, 0)
	for i := 0; i < 10000; i++ {
		strs = append(strs, "d")
	}

	b.ResetTimer()
	for i := 0; i < b.N; i++ {
		tmp := make([]string, len(strs))
		for i, str := range strs {
			tmp[i] = str
		}
	}
}

goos: darwin
goarch: arm64
pkg: test
BenchmarkAppend-8   	   42304	     31042 ns/op
BenchmarkAssign-8   	   37588	     29779 ns/op
PASS
ok  	test	3.158s

[lance6716]

rest lgtm

[hjweddie]

I will check it later

[hjweddie]

I have changed the code as suggestion