go/scanner: 17% faster scanning

- Changed the Scan API semantics slightly:
The token literal string is only returned
if the token is a literal, comment, semicolon,
or illegal character. In all other cases, the
token literal value is determined by the token
value.

Clients that care about the token literal value
when not present can always use the following
piece of code:

pos, tok, lit := scanner.Scan()
if lit == "" {
   lit = tok.String()
}

- Changed token.Lookup API to use a string instead
of a []byte argument.

- Both these changes were long-standing TODOs.

- Added BenchmarkScan.

This change permits a faster implementation of Scan
with much fewer string creations:

benchmark                old ns/op    new ns/op    delta
scanner.BenchmarkScan        74404        61457  -17.40%

R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/5532076

src/cmd/cgo/gcc.go

@@ -1374,7 +1374,7 @@ func (c *typeConv) Struct(dt *dwarf.StructType) (expr *ast.StructType, csyntax s
 
 	if !*godefs && !*cdefs {
 		for cid, goid := range ident {
-			if token.Lookup([]byte(goid)).IsKeyword() {
+			if token.Lookup(goid).IsKeyword() {
 				// Avoid keyword
 				goid = "_" + goid
 

src/pkg/go/scanner/scanner.go

@@ -157,7 +157,7 @@ func (S *Scanner) interpretLineComment(text []byte) {
 	}
 }
 
-func (S *Scanner) scanComment() {
+func (S *Scanner) scanComment() string {
 	// initial '/' already consumed; S.ch == '/' || S.ch == '*'
 	offs := S.offset - 1 // position of initial '/'
 
@@ -171,7 +171,7 @@ func (S *Scanner) scanComment() {
 			// comment starts at the beginning of the current line
 			S.interpretLineComment(S.src[offs:S.offset])
 		}
-		return
+		goto exit
 	}
 
 	/*-style comment */
@@ -181,11 +181,14 @@ func (S *Scanner) scanComment() {
 		S.next()
 		if ch == '*' && S.ch == '/' {
 			S.next()
-			return
+			goto exit
 		}
 	}
 
 	S.error(offs, "comment not terminated")
+
+exit:
+	return string(S.src[offs:S.offset])
 }
 
 func (S *Scanner) findLineEnd() bool {
@@ -240,12 +243,12 @@ func isDigit(ch rune) bool {
 	return '0' <= ch && ch <= '9' || ch >= 0x80 && unicode.IsDigit(ch)
 }
 
-func (S *Scanner) scanIdentifier() token.Token {
+func (S *Scanner) scanIdentifier() string {
 	offs := S.offset
 	for isLetter(S.ch) || isDigit(S.ch) {
 		S.next()
 	}
-	return token.Lookup(S.src[offs:S.offset])
+	return string(S.src[offs:S.offset])
 }
 
 func digitVal(ch rune) int {
@@ -266,11 +269,13 @@ func (S *Scanner) scanMantissa(base int) {
 	}
 }
 
-func (S *Scanner) scanNumber(seenDecimalPoint bool) token.Token {
+func (S *Scanner) scanNumber(seenDecimalPoint bool) (token.Token, string) {
 	// digitVal(S.ch) < 10
+	offs := S.offset
 	tok := token.INT
 
 	if seenDecimalPoint {
+		offs--
 		tok = token.FLOAT
 		S.scanMantissa(10)
 		goto exponent
@@ -334,7 +339,7 @@ exponent:
 	}
 
 exit:
-	return tok
+	return tok, string(S.src[offs:S.offset])
 }
 
 func (S *Scanner) scanEscape(quote rune) {
@@ -381,7 +386,7 @@ func (S *Scanner) scanEscape(quote rune) {
 	}
 }
 
-func (S *Scanner) scanChar() {
+func (S *Scanner) scanChar() string {
 	// '\'' opening already consumed
 	offs := S.offset - 1
 
@@ -405,9 +410,11 @@ func (S *Scanner) scanChar() {
 	if n != 1 {
 		S.error(offs, "illegal character literal")
 	}
+
+	return string(S.src[offs:S.offset])
 }
 
-func (S *Scanner) scanString() {
+func (S *Scanner) scanString() string {
 	// '"' opening already consumed
 	offs := S.offset - 1
 
@@ -424,12 +431,27 @@ func (S *Scanner) scanString() {
 	}
 
 	S.next()
+
+	return string(S.src[offs:S.offset])
 }
 
-func (S *Scanner) scanRawString() (hasCR bool) {
+func stripCR(b []byte) []byte {
+	c := make([]byte, len(b))
+	i := 0
+	for _, ch := range b {
+		if ch != '\r' {
+			c[i] = ch
+			i++
+		}
+	}
+	return c[:i]
+}
+
+func (S *Scanner) scanRawString() string {
 	// '`' opening already consumed
 	offs := S.offset - 1
 
+	hasCR := false
 	for S.ch != '`' {
 		ch := S.ch
 		S.next()
@@ -443,7 +465,13 @@ func (S *Scanner) scanRawString() (hasCR bool) {
 	}
 
 	S.next()
-	return
+
+	lit := S.src[offs:S.offset]
+	if hasCR {
+		lit = stripCR(lit)
+	}
+
+	return string(lit)
 }
 
 func (S *Scanner) skipWhitespace() {
@@ -494,27 +522,24 @@ func (S *Scanner) switch4(tok0, tok1 token.Token, ch2 rune, tok2, tok3 token.Tok
 	return tok0
 }
 
-func stripCR(b []byte) []byte {
-	c := make([]byte, len(b))
-	i := 0
-	for _, ch := range b {
-		if ch != '\r' {
-			c[i] = ch
-			i++
-		}
-	}
-	return c[:i]
-}
-
-// Scan scans the next token and returns the token position,
-// the token, and the literal string corresponding to the
-// token. The source end is indicated by token.EOF.
+// Scan scans the next token and returns the token position, the token,
+// and its literal string if applicable. The source end is indicated by
+// token.EOF.
+//
+// If the returned token is a literal (token.IDENT, token.INT, token.FLOAT,
+// token.IMAG, token.CHAR, token.STRING) or token.COMMENT, the literal string
+// has the corresponding value.
 //
 // If the returned token is token.SEMICOLON, the corresponding
 // literal string is ";" if the semicolon was present in the source,
 // and "\n" if the semicolon was inserted because of a newline or
 // at EOF.
 //
+// If the returned token is token.ILLEGAL, the literal string is the
+// offending character.
+//
+// In all other cases, Scan returns an empty literal string.
+//
 // For more tolerant parsing, Scan will return a valid token if
 // possible even if a syntax error was encountered. Thus, even
 // if the resulting token sequence contains no illegal tokens,
@@ -526,34 +551,33 @@ func stripCR(b []byte) []byte {
 // set with Init. Token positions are relative to that file
 // and thus relative to the file set.
 //
-func (S *Scanner) Scan() (token.Pos, token.Token, string) {
+func (S *Scanner) Scan() (pos token.Pos, tok token.Token, lit string) {
 scanAgain:
 	S.skipWhitespace()
 
 	// current token start
-	insertSemi := false
-	offs := S.offset
-	tok := token.ILLEGAL
-	hasCR := false
+	pos = S.file.Pos(S.offset)
 
 	// determine token value
+	insertSemi := false
 	switch ch := S.ch; {
 	case isLetter(ch):
-		tok = S.scanIdentifier()
+		lit = S.scanIdentifier()
+		tok = token.Lookup(lit)
 		switch tok {
 		case token.IDENT, token.BREAK, token.CONTINUE, token.FALLTHROUGH, token.RETURN:
 			insertSemi = true
 		}
 	case digitVal(ch) < 10:
 		insertSemi = true
-		tok = S.scanNumber(false)
+		tok, lit = S.scanNumber(false)
 	default:
 		S.next() // always make progress
 		switch ch {
 		case -1:
 			if S.insertSemi {
 				S.insertSemi = false // EOF consumed
-				return S.file.Pos(offs), token.SEMICOLON, "\n"
+				return pos, token.SEMICOLON, "\n"
 			}
 			tok = token.EOF
 		case '\n':
@@ -561,25 +585,25 @@ scanAgain:
 			// set in the first place and exited early
 			// from S.skipWhitespace()
 			S.insertSemi = false // newline consumed
-			return S.file.Pos(offs), token.SEMICOLON, "\n"
+			return pos, token.SEMICOLON, "\n"
 		case '"':
 			insertSemi = true
 			tok = token.STRING
-			S.scanString()
+			lit = S.scanString()
 		case '\'':
 			insertSemi = true
 			tok = token.CHAR
-			S.scanChar()
+			lit = S.scanChar()
 		case '`':
 			insertSemi = true
 			tok = token.STRING
-			hasCR = S.scanRawString()
+			lit = S.scanRawString()
 		case ':':
 			tok = S.switch2(token.COLON, token.DEFINE)
 		case '.':
 			if digitVal(S.ch) < 10 {
 				insertSemi = true
-				tok = S.scanNumber(true)
+				tok, lit = S.scanNumber(true)
 			} else if S.ch == '.' {
 				S.next()
 				if S.ch == '.' {
@@ -593,6 +617,7 @@ scanAgain:
 			tok = token.COMMA
 		case ';':
 			tok = token.SEMICOLON
+			lit = ";"
 		case '(':
 			tok = token.LPAREN
 		case ')':
@@ -626,12 +651,12 @@ scanAgain:
 				if S.insertSemi && S.findLineEnd() {
 					// reset position to the beginning of the comment
 					S.ch = '/'
-					S.offset = offs
-					S.rdOffset = offs + 1
+					S.offset = S.file.Offset(pos)
+					S.rdOffset = S.offset + 1
 					S.insertSemi = false // newline consumed
-					return S.file.Pos(offs), token.SEMICOLON, "\n"
+					return pos, token.SEMICOLON, "\n"
 				}
-				S.scanComment()
+				lit = S.scanComment()
 				if S.mode&ScanComments == 0 {
 					// skip comment
 					S.insertSemi = false // newline consumed
@@ -668,21 +693,15 @@ scanAgain:
 		case '|':
 			tok = S.switch3(token.OR, token.OR_ASSIGN, '|', token.LOR)
 		default:
-			S.error(offs, fmt.Sprintf("illegal character %#U", ch))
+			S.error(S.file.Offset(pos), fmt.Sprintf("illegal character %#U", ch))
 			insertSemi = S.insertSemi // preserve insertSemi info
+			tok = token.ILLEGAL
+			lit = string(ch)
 		}
 	}
-
 	if S.mode&dontInsertSemis == 0 {
 		S.insertSemi = insertSemi
 	}
 
-	// TODO(gri): The scanner API should change such that the literal string
-	//            is only valid if an actual literal was scanned. This will
-	//            permit a more efficient implementation.
-	lit := S.src[offs:S.offset]
-	if hasCR {
-		lit = stripCR(lit)
-	}
-	return S.file.Pos(offs), tok, string(lit)
+	return
 }

src/pkg/go/scanner/scanner_test.go

@@ -177,6 +177,15 @@ var tokens = [...]elt{
 
 const whitespace = "  \t  \n\n\n" // to separate tokens
 
+var source = func() []byte {
+	var src []byte
+	for _, t := range tokens {
+		src = append(src, t.lit...)
+		src = append(src, whitespace...)
+	}
+	return src
+}()
+
 type testErrorHandler struct {
 	t *testing.T
 }
@@ -214,20 +223,20 @@ func checkPos(t *testing.T, lit string, p token.Pos, expected token.Position) {
 // Verify that calling Scan() provides the correct results.
 func TestScan(t *testing.T) {
 	// make source
-	var src string
-	for _, e := range tokens {
-		src += e.lit + whitespace
-	}
-	src_linecount := newlineCount(src)
+	src_linecount := newlineCount(string(source))
 	whitespace_linecount := newlineCount(whitespace)
 
 	// verify scan
 	var s Scanner
-	s.Init(fset.AddFile("", fset.Base(), len(src)), []byte(src), &testErrorHandler{t}, ScanComments|dontInsertSemis)
+	s.Init(fset.AddFile("", fset.Base(), len(source)), source, &testErrorHandler{t}, ScanComments|dontInsertSemis)
 	index := 0
 	epos := token.Position{"", 0, 1, 1} // expected position
 	for {
 		pos, tok, lit := s.Scan()
+		if lit == "" {
+			// no literal value for non-literal tokens
+			lit = tok.String()
+		}
 		e := elt{token.EOF, "", special}
 		if index < len(tokens) {
 			e = tokens[index]
@@ -659,3 +668,20 @@ func TestScanErrors(t *testing.T) {
 		checkError(t, e.src, e.tok, e.pos, e.err)
 	}
 }
+
+func BenchmarkScan(b *testing.B) {
+	b.StopTimer()
+	fset := token.NewFileSet()
+	file := fset.AddFile("", fset.Base(), len(source))
+	var s Scanner
+	b.StartTimer()
+	for i := b.N - 1; i >= 0; i-- {
+		s.Init(file, source, nil, ScanComments)
+		for {
+			_, tok, _ := s.Scan()
+			if tok == token.EOF {
+				break
+			}
+		}
+	}
+}

src/pkg/go/token/token.go

@@ -283,10 +283,8 @@ func init() {
 
 // Lookup maps an identifier to its keyword token or IDENT (if not a keyword).
 //
-func Lookup(ident []byte) Token {
-	// TODO Maps with []byte key are illegal because []byte does not
-	//      support == . Should find a more efficient solution eventually.
-	if tok, is_keyword := keywords[string(ident)]; is_keyword {
+func Lookup(ident string) Token {
+	if tok, is_keyword := keywords[ident]; is_keyword {
 		return tok
 	}
 	return IDENT
@@ -295,16 +293,16 @@ func Lookup(ident []byte) Token {
 // Predicates
 
 // IsLiteral returns true for tokens corresponding to identifiers
-// and basic type literals; returns false otherwise.
+// and basic type literals; it returns false otherwise.
 //
 func (tok Token) IsLiteral() bool { return literal_beg < tok && tok < literal_end }
 
 // IsOperator returns true for tokens corresponding to operators and
-// delimiters; returns false otherwise.
+// delimiters; it returns false otherwise.
 //
 func (tok Token) IsOperator() bool { return operator_beg < tok && tok < operator_end }
 
 // IsKeyword returns true for tokens corresponding to keywords;
-// returns false otherwise.
+// it returns false otherwise.
 //
 func (tok Token) IsKeyword() bool { return keyword_beg < tok && tok < keyword_end }