Performance needs some work

[nigeltao]

According to this crude micro-benchmark on 12 made up domains, github.com/weppos/publicsuffix-go performs 460x slower than golang.org/x/net/publicsuffix:

Program:

package main

import (
	"fmt"
	"time"

	wlib "github.com/weppos/publicsuffix-go/publicsuffix"
	xlib "golang.org/x/net/publicsuffix"
)

var testCases = []string{
	"example.com",
	"example.id.au",
	"www.ck",
	"foo.bar.xn--55qx5d.cn",
	"a.b.c.minami.fukuoka.jp",
	"posts-and-telecommunications.museum",
	"www.example.pvt.k12.ma.us",
	"many.lol",
	"the.russian.for.moscow.is.xn--80adxhks",
	"blah.blah.s3-us-west-1.amazonaws.com",
	"thing.dyndns.org",
	"nosuchtld",
}

var wants = map[string]string{
	"example.com":                            "example.com",
	"example.id.au":                          "example.id.au",
	"www.ck":                                 "www.ck",
	"foo.bar.xn--55qx5d.cn":                  "bar.xn--55qx5d.cn",
	"a.b.c.minami.fukuoka.jp":                "c.minami.fukuoka.jp",
	"posts-and-telecommunications.museum":    "",
	"www.example.pvt.k12.ma.us":              "example.pvt.k12.ma.us",
	"many.lol":                               "many.lol",
	"the.russian.for.moscow.is.xn--80adxhks": "is.xn--80adxhks",
	"blah.blah.s3-us-west-1.amazonaws.com":   "blah.s3-us-west-1.amazonaws.com",
	"thing.dyndns.org":                       "thing.dyndns.org",
	"nosuchtld":                              "",
}

func main() {
	do(false)
	do(true)
}

func do(xNetLibrary bool) {
	libraryName := "github.com/weppos/publicsuffix-go"
	if xNetLibrary {
		libraryName = "golang.org/x/net/publicsuffix"
	}

	now := time.Now()
	for i := 0; i < 100; i++ {
		for _, tc := range testCases {
			got := ""
			if xNetLibrary {
				got, _ = xlib.EffectiveTLDPlusOne(tc)
			} else {
				got, _ = wlib.Domain(tc)
			}

			if i != 0 {
				continue
			}
			if want := wants[tc]; got != want {
				panic(fmt.Sprintf("xNetLibrary=%t, tc=%q: got %q, want %q",
					xNetLibrary, tc, got, want))
			}
		}
	}
	since := time.Since(now)

	fmt.Printf("%16d nanos  %s\n", since, libraryName)
}

Output:

       172856110 nanos  github.com/weppos/publicsuffix-go
          375676 nanos  golang.org/x/net/publicsuffix

[nigeltao]

This library's (github.com/weppos/publicsuffix-go/publicsuffix) performance was noticed by a colleague. There's a little more discussion at golang/go#15518 (comment)

[weppos]

Thanks for opening the ticket @nigeltao. I confirm this is sort of expected due to the sequential lookup algorithm currently in use.

I'll try to allocate some time in August to switch to a more efficient lookup as mentioned in golang/go#15518 (comment)

[weppos]

Thanks for the work of #143 (kudos to @guliyevemil1 for contributing) performances have been drastically increased.

Before:

➜  publicsuffix-go git:(master) ✗ GOCACHE=off  go test ./... -bench=.
?   	github.com/weppos/publicsuffix-go/cmd/load	[no test files]
PASS
ok  	github.com/weppos/publicsuffix-go/net/publicsuffix	0.019s
goos: darwin
goarch: amd64
pkg: github.com/weppos/publicsuffix-go/publicsuffix
BenchmarkDomain-4   	    1000	   1522570 ns/op
BenchmarkXNet-4     	  500000	      3267 ns/op
PASS
ok  	github.com/weppos/publicsuffix-go/publicsuffix	3.383s

After

➜  publicsuffix-go git:(master) GOCACHE=off  go test ./... -bench=.
?   	github.com/weppos/publicsuffix-go/cmd/load	[no test files]
PASS
ok  	github.com/weppos/publicsuffix-go/net/publicsuffix	0.017s
goos: darwin
goarch: amd64
pkg: github.com/weppos/publicsuffix-go/publicsuffix
BenchmarkDomain-4   	  500000	      3140 ns/op
BenchmarkXNet-4     	  500000	      3313 ns/op
PASS
ok  	github.com/weppos/publicsuffix-go/publicsuffix	3.317s

@nigeltao No rush, but if you have some spare minutes, I'd appreciate your feedback whether you think I can consider this issue closed. Feel free to let me know if there's anything worth a review in the changes.

[nigeltao]

I'd have to ask my colleague whether they're happy, but if you don't hear back from me in, say, a week from now, go ahead and close the issue.

[vanbroup]

I can confirm that this has improved performance significantly!
Thanks @guliyevemil1

[weppos]

Thanks for the confirmation!