refactor: use strings.Builder to improve performance

[zjumathcode]

Assistance Disclosure

strings.Builder has fewer memory allocations and better performance.
More info: golang/go#75190

[CLAassistant]

All committers have signed the CLA.

[vnxme]

Interested to see any tests showing how much the performance gain is, if any, in ns per request or requests per second.

[zjumathcode]

@vnxme I wrote a benchmark that shows the performance advantage of using strings.Builder, as follows:

package caddy_test

import (
	"fmt"
	"strings"
	"testing"
)

type Route struct {
	Group          string
	MatcherSetsRaw string
	HandlersRaw    []string
	Terminal       bool
}

func (r Route) StringOld() string {
	handlersRaw := "["
	for _, hr := range r.HandlersRaw {
		handlersRaw += " " + string(hr)
	}
	handlersRaw += "]"

	return fmt.Sprintf(`{Group:"%s" MatcherSetsRaw:%s HandlersRaw:%s Terminal:%t}`,
		r.Group, r.MatcherSetsRaw, handlersRaw, r.Terminal)
}

func (r Route) StringNew() string {
	var handlersRaw strings.Builder
	handlersRaw.WriteString("[")
	for _, hr := range r.HandlersRaw {
		handlersRaw.WriteString(" " + string(hr))
	}
	handlersRaw.WriteString("]")

	return fmt.Sprintf(`{Group:"%s" MatcherSetsRaw:%s HandlersRaw:%s Terminal:%t}`,
		r.Group, r.MatcherSetsRaw, handlersRaw.String(), r.Terminal)
}

func BenchmarkRouteString_Old(b *testing.B) {
	r := Route{
		Group:          "api",
		MatcherSetsRaw: "[matcher]",
		HandlersRaw:    []string{"foo", "bar", "baz", "qux", "quux"},
		Terminal:       true,
	}
	b.ReportAllocs()
	for i := 0; i < b.N; i++ {
		_ = r.StringOld()
	}
}

func BenchmarkRouteString_New(b *testing.B) {
	r := Route{
		Group:          "api",
		MatcherSetsRaw: "[matcher]",
		HandlersRaw:    []string{"foo", "bar", "baz", "qux", "quux"},
		Terminal:       true,
	}
	b.ReportAllocs()
	for i := 0; i < b.N; i++ {
		_ = r.StringNew()
	}
}

➜  benchmark go test -bench=. -benchmem
goos: darwin
goarch: arm64
pkg: demo/stringsbuilder/benchmark
cpu: Apple M4
BenchmarkRouteString_Old-10      5239779               211.3 ns/op           248 B/op         10 allocs/op
BenchmarkRouteString_New-10      6258066               193.5 ns/op           200 B/op          7 allocs/op
PASS
ok      demo/stringsbuilder/benchmark 3.482s

[vnxme]

@zjumathcode Thanks, so you have an 8% gain in ns/op, a 19% decrease in B/op and a 30% drop in allocs/op in your benchmark which is quite good. Does your PR cover all the places in the code where strings are concatenated?

[boqishan]

@zjumathcode Thanks, so you have an 8% gain in ns/op, a 19% decrease in B/op and a 30% drop in allocs/op in your benchmark which is quite good. Does your PR cover all the places in the code where strings are concatenated?

This is a generic performance improvement plan, based on suggestions and support from the Go team.

More info: golang/go#75190.

[francislavoie]

The PR had fallen inactive so I went ahead and fixed up the remainder as per review comments.