parallelize dir traversal

walk/BUILD.bazel

@@ -14,6 +14,7 @@ go_library(
         "//flag",
         "//rule",
         "@com_github_bmatcuk_doublestar_v4//:doublestar",
+        "@org_golang_x_sync//errgroup",
     ],
 )
 

walk/walk.go

@@ -25,9 +25,11 @@ import (
 	"path/filepath"
 	"sort"
 	"strings"
+	"sync"
 
 	"github.com/bazelbuild/bazel-gazelle/config"
 	"github.com/bazelbuild/bazel-gazelle/rule"
+	"golang.org/x/sync/errgroup"
 )
 
 // Mode determines which directories Walk visits and which directories
@@ -366,25 +368,33 @@ func resolveFileInfo(wc *walkConfig, dir, rel string, ent fs.DirEntry) fs.DirEnt
 
 func buildTrie(c *config.Config, isBazelIgnored isIgnoredFunc) (*pathTrie, error) {
 	trie := &pathTrie{}
+	mu := sync.Mutex{}
+
+	eg := errgroup.Group{}
+	eg.SetLimit(1000) // TODO: make this configurable? calculate based on number of CPUs?
+	eg.Go(func() error {
+		return walkDir(c.RepoRoot, "", &eg, func(rel string, d fs.DirEntry) error {
+			if isBazelIgnored(rel) {
+				return walkSkipDir
+			}
 
-	err := walkDir(c.RepoRoot, "", func(rel string, d fs.DirEntry) error {
-		if isBazelIgnored(rel) {
-			return walkSkipDir
-		}
+			mu.Lock()
+			defer mu.Unlock()
 
-		trie.Put(rel, &d)
-		return nil
+			trie.Put(rel, &d)
+			return nil
+		})
 	})
 
-	return trie, err
+	return trie, eg.Wait()
 }
 
 var walkSkipDir error = fs.SkipDir
 
 type walkDirFunc func(rel string, d fs.DirEntry) error
 
 // walkDir recursively descends path, calling walkDirFn.
-func walkDir(root, rel string, walkDirFn walkDirFunc) error {
+func walkDir(root, rel string, eg *errgroup.Group, walkDirFn walkDirFunc) error {
 	dirs, err := os.ReadDir(filepath.Join(root, rel))
 	if err != nil {
 		return err
@@ -400,10 +410,9 @@ func walkDir(root, rel string, walkDirFn walkDirFunc) error {
 		}
 
 		if d1.IsDir() {
-			// TODO: parallelize
-			if err := walkDir(root, path1, walkDirFn); err != nil {
-				return err
-			}
+			eg.Go(func() error {
+				return walkDir(root, path1, eg, walkDirFn)
+			})
 		}
 	}
 	return nil