gopls/internal/test/integration/bench: improve completion benchmarks

Make the following improvements to the completion benchmarks:

- Run the "CompletionFollowingEdit" benchmarks with and without edits.
  Accordingly, rename it to just "BenchmarkCompletion", and shorten
  subtest names.
- BenchmarkCompletion is a cleaner way to specify completion tests.
  Deprecate other styles.
- Support running completion tests with a massive GOPATH, by passing
  -completion_gopath.
- Add a tools_unimportedselector test that seems to do a good job of
  exercising slow unimported completion. Other unimportedcompletion
  tests were short-circuited in one way or another (for example, because
  kubernetes uses vendoring, gopls doesn't scan GOPATH).

This will invalidate our benchmark dashboard, so we should merge this
change before subsequent CLs that optimize unimported completion.

For golang/go#63459

Change-Id: I9d7a06e3c1a7239b531ed8ff534f3174f4ac4ae8
Reviewed-on: https://go-review.googlesource.com/c/tools/+/560457
Reviewed-by: Alan Donovan <adonovan@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

Author: findleyr

gopls/internal/test/integration/bench/completion_test.go

@@ -15,7 +15,7 @@ import (
 	"golang.org/x/tools/gopls/internal/test/integration/fake"
 )
 
-// TODO(rfindley): update these completion tests to run on multiple repos.
+var completionGOPATH = flag.String("completion_gopath", "", "if set, use this GOPATH for BenchmarkCompletion")
 
 type completionBenchOptions struct {
 	file, locationRegexp string
@@ -25,6 +25,7 @@ type completionBenchOptions struct {
 	beforeCompletion func(*Env) // run before each completion
 }
 
+// Deprecated: new tests should be expressed in BenchmarkCompletion.
 func benchmarkCompletion(options completionBenchOptions, b *testing.B) {
 	repo := getRepo(b, "tools")
 	_ = repo.sharedEnv(b) // ensure cache is warm
@@ -146,15 +147,15 @@ func (c *completer) _() {
 	}, b)
 }
 
-type completionFollowingEditTest struct {
+type completionTest struct {
 	repo           string
 	name           string
 	file           string // repo-relative file to create
 	content        string // file content
 	locationRegexp string // regexp for completion
 }
 
-var completionFollowingEditTests = []completionFollowingEditTest{
+var completionTests = []completionTest{
 	{
 		"tools",
 		"selector",
@@ -168,6 +169,32 @@ func (c *completer) _() {
 `,
 		`func \(c \*completer\) _\(\) {\n\tc\.inference\.kindMatches\((c)`,
 	},
+	{
+		"tools",
+		"unimportedident",
+		"internal/lsp/source/completion/completion2.go",
+		`
+package completion
+
+func (c *completer) _() {
+	lo
+}
+`,
+		`lo()`,
+	},
+	{
+		"tools",
+		"unimportedselector",
+		"internal/lsp/source/completion/completion2.go",
+		`
+package completion
+
+func (c *completer) _() {
+	log.
+}
+`,
+		`log\.()`,
+	},
 	{
 		"kubernetes",
 		"selector",
@@ -213,14 +240,18 @@ func (p *Pivot) _() {
 //
 // Edits force type-checked packages to be invalidated, so we want to measure
 // how long it takes before completion results are available.
-func BenchmarkCompletionFollowingEdit(b *testing.B) {
-	for _, test := range completionFollowingEditTests {
+func BenchmarkCompletion(b *testing.B) {
+	for _, test := range completionTests {
 		b.Run(fmt.Sprintf("%s_%s", test.repo, test.name), func(b *testing.B) {
-			for _, completeUnimported := range []bool{true, false} {
-				b.Run(fmt.Sprintf("completeUnimported=%v", completeUnimported), func(b *testing.B) {
-					for _, budget := range []string{"0s", "100ms"} {
-						b.Run(fmt.Sprintf("budget=%s", budget), func(b *testing.B) {
-							runCompletionFollowingEdit(b, test, completeUnimported, budget)
+			for _, followingEdit := range []bool{true, false} {
+				b.Run(fmt.Sprintf("edit=%v", followingEdit), func(b *testing.B) {
+					for _, completeUnimported := range []bool{true, false} {
+						b.Run(fmt.Sprintf("unimported=%v", completeUnimported), func(b *testing.B) {
+							for _, budget := range []string{"0s", "100ms"} {
+								b.Run(fmt.Sprintf("budget=%s", budget), func(b *testing.B) {
+									runCompletion(b, test, followingEdit, completeUnimported, budget)
+								})
+							}
 						})
 					}
 				})
@@ -229,13 +260,20 @@ func BenchmarkCompletionFollowingEdit(b *testing.B) {
 	}
 }
 
+// For optimizing unimported completion, it can be useful to benchmark with a
+// huge GOMODCACHE.
 var gomodcache = flag.String("gomodcache", "", "optional GOMODCACHE for unimported completion benchmarks")
 
-func runCompletionFollowingEdit(b *testing.B, test completionFollowingEditTest, completeUnimported bool, budget string) {
+func runCompletion(b *testing.B, test completionTest, followingEdit, completeUnimported bool, budget string) {
 	repo := getRepo(b, test.repo)
-	sharedEnv := repo.sharedEnv(b) // ensure cache is warm
+	gopath := *completionGOPATH
+	if gopath == "" {
+		// use a warm GOPATH
+		sharedEnv := repo.sharedEnv(b)
+		gopath = sharedEnv.Sandbox.GOPATH()
+	}
 	envvars := map[string]string{
-		"GOPATH": sharedEnv.Sandbox.GOPATH(), // use the warm cache
+		"GOPATH": gopath,
 	}
 
 	if *gomodcache != "" {
@@ -248,7 +286,7 @@ func runCompletionFollowingEdit(b *testing.B, test completionFollowingEditTest,
 			"completeUnimported": completeUnimported,
 			"completionBudget":   budget,
 		},
-	}, "completionFollowingEdit", false)
+	}, "completion", false)
 	defer env.Close()
 
 	env.CreateBuffer(test.file, "// __TEST_PLACEHOLDER_0__\n"+test.content)
@@ -278,12 +316,14 @@ func runCompletionFollowingEdit(b *testing.B, test completionFollowingEditTest,
 
 	b.ResetTimer()
 
-	if stopAndRecord := startProfileIfSupported(b, env, qualifiedName(test.repo, "completionFollowingEdit")); stopAndRecord != nil {
+	if stopAndRecord := startProfileIfSupported(b, env, qualifiedName(test.repo, "completion")); stopAndRecord != nil {
 		defer stopAndRecord()
 	}
 
 	for i := 0; i < b.N; i++ {
-		editPlaceholder()
+		if followingEdit {
+			editPlaceholder()
+		}
 		loc := env.RegexpSearch(test.file, test.locationRegexp)
 		env.Completion(loc)
 	}