syntax: ParseDID performance (bluesky-social#853)

we are apparently burning a bunch of CPU on this regex in prod, so let's
bypass for the most common case (valid did:plc).

Would be nice to have a benchmark demonstrating perf gains (or not).

I did some manually testing with the CLI helper and inserted `panic()`
statements to confirm that valid DID PLC stays on fast path, and other
DIDs don't.

Feedback welcome if there is a more idiomatic/performant golang way to
do this... this is still doing unicode stuff, I assume, and maybe
casting to bytes would be faster? though probably doesn't impact the
actual hot path much. `strings.ContainsAny` or `strings.ContainsFunc`
could be used here I guess but 🤷

atproto/syntax/cmd/atp-syntax/main.go

@@ -22,6 +22,12 @@ func main() {
 			ArgsUsage: "<tid>",
 			Action:    runParseTID,
 		},
+		&cli.Command{
+			Name:      "parse-did",
+			Usage:     "parse a DID",
+			ArgsUsage: "<did>",
+			Action:    runParseDID,
+		},
 	}
 	h := slog.NewTextHandler(os.Stderr, &slog.HandlerOptions{Level: slog.LevelDebug})
 	slog.SetDefault(slog.New(h))
@@ -43,3 +49,18 @@ func runParseTID(cctx *cli.Context) error {
 
 	return nil
 }
+
+func runParseDID(cctx *cli.Context) error {
+	s := cctx.Args().First()
+	if s == "" {
+		return fmt.Errorf("need to provide identifier as an argument")
+	}
+
+	did, err := syntax.ParseDID(s)
+	if err != nil {
+		return err
+	}
+	fmt.Printf("%s\n", did)
+
+	return nil
+}

atproto/syntax/did.go

@@ -14,8 +14,30 @@ import (
 type DID string
 
 var didRegex = regexp.MustCompile(`^did:[a-z]+:[a-zA-Z0-9._:%-]*[a-zA-Z0-9._-]$`)
+var plcChars = ""
+
+func isASCIIAlphaNum(c rune) bool {
+	if (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') || (c >= '0' && c <= '9') {
+		return true
+	}
+	return false
+}
 
 func ParseDID(raw string) (DID, error) {
+	// fast-path for did:plc, avoiding regex
+	if len(raw) == 32 && strings.HasPrefix(raw, "did:plc:") {
+		// NOTE: this doesn't really check base32, just broader alphanumberic. might pass invalid PLC DIDs, but they still have overall valid DID syntax
+		isPlc := true
+		for _, c := range raw[8:32] {
+			if !isASCIIAlphaNum(c) {
+				isPlc = false
+				break
+			}
+		}
+		if isPlc {
+			return DID(raw), nil
+		}
+	}
 	if raw == "" {
 		return "", errors.New("expected DID, got empty string")
 	}