Yay! First version for which parsing long.toml drops below 100ms! Got an outcome of 93ms. Almost down to BurntSushi's speed level, but still with a generic parser backing. Looking good!!

examples/example_basiccalculator2_test.go

@@ -133,7 +133,7 @@ func (calc *calculator) factor(p *parse.API) {
 	var A, T = tokenize.A, tokenize.T
 	p.Accept(A.Blanks)
 	switch {
-	case p.Accept(T.Float64(nil, A.Signed(A.Float))):
+	case p.Accept(T.Float64(nil, A.Signed(A.Decimal))):
 		value := p.Result.Tokens[0].Value.(float64)
 		calc.interpreter.pushValue(value)
 	case p.Accept(A.LeftParen):

tokenize/api.go

@@ -170,6 +170,14 @@ func (i *API) Accept() {
 	i.acceptRunes(i.lastRuneWidth, i.lastRune)
 }
 
+func (i *API) skipBytes(bytes ...byte) {
+	for _, b := range bytes {
+		i.stackFrame.moveCursorByByte(b)
+	}
+	i.stackFrame.offset += len(bytes)
+	i.runeRead = false
+}
+
 func (i *API) acceptBytes(bytes ...byte) {
 	curRuneEnd := i.stackFrame.runeEnd
 	newRuneEnd := curRuneEnd + len(bytes)
@@ -190,6 +198,14 @@ func (i *API) acceptBytes(bytes ...byte) {
 	i.runeRead = false
 }
 
+func (i *API) skipRunes(width int, runes ...rune) {
+	for _, r := range runes {
+		i.stackFrame.moveCursorByRune(r)
+	}
+	i.stackFrame.offset += width
+	i.runeRead = false
+}
+
 func (i *API) acceptRunes(width int, runes ...rune) {
 	curRuneEnd := i.stackFrame.runeEnd
 	newRuneEnd := curRuneEnd + len(runes)

tokenize/handlers_builtin.go

@@ -70,9 +70,11 @@ var A = struct {
 	Byte           func(byte) Handler
 	Bytes          func(...byte) Handler
 	ByteRange      func(byte, byte) Handler
+	ByteByCallback func(func(byte) bool) Handler
 	Rune           func(rune) Handler
 	Runes          func(...rune) Handler
 	RuneRange      func(rune, rune) Handler
+	RuneByCallback func(func(rune) bool) Handler
 	Str            func(string) Handler
 	StrNoCase      func(string) Handler
 	EndOfLine      Handler
@@ -136,11 +138,11 @@ var A = struct {
 	DigitNotZero   Handler
 	Digits         Handler
 	Zero           Handler
-	Float          Handler
 	Boolean        Handler
-	Integer        Handler
 	Signed         func(Handler) Handler
+	Integer        Handler
 	IntegerBetween func(min int64, max int64) Handler
+	Decimal        Handler
 	ASCII          Handler
 	ASCIILower     Handler
 	ASCIIUpper     Handler
@@ -160,9 +162,11 @@ var A = struct {
 	Byte:           MatchByte,
 	Bytes:          MatchBytes,
 	ByteRange:      MatchByteRange,
+	ByteByCallback: MatchByteByCallback,
 	Rune:           MatchRune,
 	Runes:          MatchRunes,
 	RuneRange:      MatchRuneRange,
+	RuneByCallback: MatchRuneByCallback,
 	Str:            MatchStr,
 	StrNoCase:      MatchStrNoCase,
 	EndOfFile:      MatchEndOfFile(),
@@ -172,51 +176,51 @@ var A = struct {
 	AnyRune:        MatchAnyRune(),
 	ValidRune:      MatchValidRune(),
 	InvalidRune:    MatchInvalidRune(),
-	Space:          MatchRune(' '),
+	Space:          MatchByte(' '),
-	Tab:            MatchRune('\t'),
+	Tab:            MatchByte('\t'),
-	CR:             MatchRune('\r'),
+	CR:             MatchByte('\r'),
-	LF:             MatchRune('\n'),
+	LF:             MatchByte('\n'),
 	CRLF:           MatchStr("\r\n"),
-	Excl:           MatchRune('!'),
+	Excl:           MatchByte('!'),
-	DoubleQuote:    MatchRune('"'),
+	DoubleQuote:    MatchByte('"'),
-	Hash:           MatchRune('#'),
+	Hash:           MatchByte('#'),
-	Dollar:         MatchRune('$'),
+	Dollar:         MatchByte('$'),
-	Percent:        MatchRune('%'),
+	Percent:        MatchByte('%'),
-	Amp:            MatchRune('&'),
+	Amp:            MatchByte('&'),
-	SingleQuote:    MatchRune('\''),
+	SingleQuote:    MatchByte('\''),
-	RoundOpen:      MatchRune('('),
+	RoundOpen:      MatchByte('('),
-	LeftParen:      MatchRune('('),
+	LeftParen:      MatchByte('('),
-	RoundClose:     MatchRune(')'),
+	RoundClose:     MatchByte(')'),
-	RightParen:     MatchRune(')'),
+	RightParen:     MatchByte(')'),
-	Asterisk:       MatchRune('*'),
+	Asterisk:       MatchByte('*'),
-	Multiply:       MatchRune('*'),
+	Multiply:       MatchByte('*'),
-	Plus:           MatchRune('+'),
+	Plus:           MatchByte('+'),
-	Add:            MatchRune('+'),
+	Add:            MatchByte('+'),
-	Comma:          MatchRune(','),
+	Comma:          MatchByte(','),
-	Minus:          MatchRune('-'),
+	Minus:          MatchByte('-'),
-	Subtract:       MatchRune('-'),
+	Subtract:       MatchByte('-'),
-	Dot:            MatchRune('.'),
+	Dot:            MatchByte('.'),
-	Slash:          MatchRune('/'),
+	Slash:          MatchByte('/'),
-	Divide:         MatchRune('/'),
+	Divide:         MatchByte('/'),
-	Colon:          MatchRune(':'),
+	Colon:          MatchByte(':'),
-	Semicolon:      MatchRune(';'),
+	Semicolon:      MatchByte(';'),
-	AngleOpen:      MatchRune('<'),
+	AngleOpen:      MatchByte('<'),
-	LessThan:       MatchRune('<'),
+	LessThan:       MatchByte('<'),
-	Equal:          MatchRune('='),
+	Equal:          MatchByte('='),
-	AngleClose:     MatchRune('>'),
+	AngleClose:     MatchByte('>'),
-	GreaterThan:    MatchRune('>'),
+	GreaterThan:    MatchByte('>'),
-	Question:       MatchRune('?'),
+	Question:       MatchByte('?'),
-	At:             MatchRune('@'),
+	At:             MatchByte('@'),
-	SquareOpen:     MatchRune('['),
+	SquareOpen:     MatchByte('['),
-	Backslash:      MatchRune('\\'),
+	Backslash:      MatchByte('\\'),
-	SquareClose:    MatchRune(']'),
+	SquareClose:    MatchByte(']'),
-	Caret:          MatchRune('^'),
+	Caret:          MatchByte('^'),
-	Underscore:     MatchRune('_'),
+	Underscore:     MatchByte('_'),
-	Backquote:      MatchRune('`'),
+	Backquote:      MatchByte('`'),
-	CurlyOpen:      MatchRune('{'),
+	CurlyOpen:      MatchByte('{'),
-	Pipe:           MatchRune('|'),
+	Pipe:           MatchByte('|'),
-	CurlyClose:     MatchRune('}'),
+	CurlyClose:     MatchByte('}'),
-	Tilde:          MatchRune('~'),
+	Tilde:          MatchByte('~'),
 	Newline:        MatchNewline(),
 	Blank:          MatchBlank(),
 	Blanks:         MatchBlanks(),
@@ -225,11 +229,11 @@ var A = struct {
 	Digit:          MatchDigit(),
 	DigitNotZero:   MatchDigitNotZero(),
 	Digits:         MatchDigits(),
-	Zero:           MatchRune('0'),
+	Zero:           MatchByte('0'),
-	Integer:        MatchInteger(),
 	Signed:         MatchSigned,
+	Integer:        MatchInteger(true),
 	IntegerBetween: MatchIntegerBetween,
-	Float:          MatchFloat(),
+	Decimal:        MatchDecimal(true),
 	Boolean:        MatchBoolean(),
 	ASCII:          MatchASCII(),
 	ASCIILower:     MatchASCIILower(),
@@ -355,7 +359,7 @@ func MatchByte(expected byte) Handler {
 
 // MatchRune creates a Handler function that matches against the provided rune.
 func MatchRune(expected rune) Handler {
-	if expected <= 127 {
+	if expected <= '\x7F' {
 		return MatchByte(byte(expected))
 	}
 	return func(t *API) bool {
@@ -392,7 +396,7 @@ func MatchRunes(expected ...rune) Handler {
 	onlyBytes := true
 	expectedBytes := make([]byte, len(expected))
 	for i, r := range expected {
-		if r > 255 {
+		if r > '\x7F' {
 			onlyBytes = false
 			break
 		}
@@ -448,7 +452,7 @@ func MatchRuneRange(start rune, end rune) Handler {
 	if end < start {
 		callerPanic("MatchRuneRange", "Handler: {name} definition error at {caller}: start %q must not be < end %q", start, end)
 	}
-	if end <= 127 {
+	if end <= '\x7F' {
 		return MatchByteRange(byte(start), byte(end))
 	}
 	return func(t *API) bool {
@@ -574,6 +578,23 @@ func MatchUnicodeSpace() Handler {
 	return MatchOneOrMore(MatchRuneByCallback(unicode.IsSpace))
 }
 
+// MatchByteByCallback creates a Handler that matches a single byte from the
+// input against the provided callback function. When the callback returns true,
+// it is considered a match.
+//
+// Note that the callback function matches the signature of the unicode.Is* functions,
+// so those can be used. E.g. MatchRuneByCallback(unicode.IsLower).
+func MatchByteByCallback(callback func(byte) bool) Handler {
+	return func(t *API) bool {
+		b, err := t.PeekByte(0)
+		if err == nil && callback(b) {
+			t.acceptBytes(b)
+			return true
+		}
+		return false
+	}
+}
+
 // MatchRuneByCallback creates a Handler that matches a single rune from the
 // input against the provided callback function. When the callback returns true,
 // it is considered a match.
@@ -621,7 +642,7 @@ func MatchStr(expected string) Handler {
 	return func(t *API) bool {
 		offset := 0
 		for _, e := range expectedRunes {
-			if e <= 127 {
+			if e <= '\x7F' {
 				b, err := t.PeekByte(offset)
 				if err != nil || b != byte(e) {
 					return false
@@ -650,7 +671,7 @@ func MatchStrNoCase(expected string) Handler {
 		width := 0
 		i := 0
 		for _, e := range expected {
-			if e <= 127 {
+			if e <= '\x7F' {
 				b, err := t.PeekByte(width)
 				if err != nil || (b != byte(e) && unicode.ToUpper(rune(b)) != unicode.ToUpper(e)) {
 					return false
@@ -732,9 +753,9 @@ func MatchNot(handler Handler) Handler {
 			return false
 		}
 		t.Dispose(child)
-		_, err := t.NextRune()
+		r, w, err := t.PeekRune(0)
 		if err == nil {
-			t.Accept()
+			t.acceptRunes(w, r)
 			return true
 		}
 		return false
@@ -924,8 +945,24 @@ func MakeInputFlusher(handler Handler) Handler {
 //
 //     C.Signed(A.Integer)
 func MatchSigned(handler Handler) Handler {
-	sign := MatchOptional(MatchAny(MatchRune('+'), MatchRune('-')))
+	return func(t *API) bool {
-	return MatchSeq(sign, handler)
+		child := t.Fork()
+		b, err := t.PeekByte(0)
+		if err != nil {
+			t.Dispose(child)
+			return false
+		}
+		if b == '-' || b == '+' {
+			t.acceptBytes(b)
+		}
+		if handler(t) {
+			t.Merge(child)
+			t.Dispose(child)
+			return true
+		}
+		t.Dispose(child)
+		return false
+	}
 }
 
 // MatchIntegerBetween creates a Handler that checks for an integer
@@ -956,7 +993,7 @@ func MatchIntegerBetween(min int64, max int64) Handler {
 func MatchEndOfFile() Handler {
 	return func(t *API) bool {
 		child := t.Fork()
-		_, err := t.NextRune()
+		_, err := t.PeekByte(0)
 		t.Dispose(child)
 		return err == io.EOF
 	}
@@ -1024,37 +1061,157 @@ func MatchInvalidRune() Handler {
 // MatchDigit creates a Handler that checks if a single digit can be read
 // from the input.
 func MatchDigit() Handler {
-	return MatchRuneRange('0', '9')
+	return MatchByteRange('0', '9')
 }
 
 // MatchDigits creates a Handler that checks if one or more digits can be read
 // from the input.
 func MatchDigits() Handler {
-	return MatchOneOrMore(MatchDigit())
+	return func(t *API) bool {
+		// Check if the first character is a digit.
+		b, err := t.PeekByte(0)
+		if err != nil || b < '0' || b > '9' {
+			return false
+		}
+		t.acceptBytes(b)
+
+		// Continue accepting bytes as long as they are digits.
+		for {
+			b, err := t.PeekByte(0)
+			if err != nil || b < '0' || b > '9' {
+				return true
+			}
+			t.acceptBytes(b)
+		}
+	}
 }
 
 // MatchDigitNotZero creates a Handler that checks if a single digit not equal
 // to zero '0' can be read from the input.
 func MatchDigitNotZero() Handler {
-	return MatchRuneRange('1', '9')
+	return MatchByteRange('1', '9')
 }
 
 // MatchInteger creates a Handler function that checks if a valid integer
-// can be read from the input. In line with Go, an integer cannot start with
+// can be read from the input.
-// a zero. Starting with a zero is used to indicate other bases, like octal or
+//
-// hexadecimal.
+// Leading zeroes are allowed. When the normalize parameter is true, these
-func MatchInteger() Handler {
+// will be stripped from the input.
-	justZero := MatchRune('0')
+func MatchInteger(normalize bool) Handler {
-	integer := MatchSeq(MatchDigitNotZero(), MatchZeroOrMore(MatchDigit()))
+	return func(t *API) bool {
-	return MatchAny(integer, justZero)
+		// Check if the first character is a digit.
+		b, err := t.PeekByte(0)
+		if err != nil || b < '0' || b > '9' {
+			return false
+		}
+
+		// When normalization is requested, drop leading zeroes.
+		if normalize && b == '0' {
+			for {
+				b2, err := t.PeekByte(1)
+
+				// The next character is a zero, skip the leading zero and check again.
+				if err == nil && b2 == b {
+					t.skipBytes('0')
+					continue
+				}
+				// The next character is not a zero, nor a digit at all.
+				// We're looking at a zero on its own here.
+				if err != nil || b2 < '1' || b2 > '9' {
+					t.acceptBytes('0')
+					return true
+				}
+				// The next character is a digit. SKip the leading zero and go with the digit.
+				t.skipBytes('0')
+				t.acceptBytes(b2)
+				break
+			}
+		}
+
+		// Continue accepting bytes as long as they are digits.
+		for {
+			b, err := t.PeekByte(0)
+			if err != nil || b < '0' || b > '9' {
+				return true
+			}
+			t.acceptBytes(b)
+		}
+	}
 }
 
-// MatchFloat creates a Handler function that checks if a valid float value
+// MatchDecimal creates a Handler function that checks if a valid decimal value
-// can be read from the input. In case the fractional part is missing, this
+// can be read from the input. In case the fractional part is missing (which is
-// Handler will report a match, so both "123" and "123.123" will match.
+// a valid decimal number), this Handler will report a match, so both "123" and
-func MatchFloat() Handler {
+// "123.123" will match.
-	digits := MatchDigits()
+//
-	return MatchSeq(digits, MatchOptional(MatchSeq(MatchRune('.'), digits)))
+// Leading zeroes are allowed. When the normalize parameter is true, these
+// will be stripped from the input.
+func MatchDecimal(normalize bool) Handler {
+	return func(t *API) bool {
+		// Check if the first character is a digit.
+		b, err := t.PeekByte(0)
+		if err != nil || b < '0' || b > '9' {
+			return false
+		}
+
+		// When normalization is requested, drop leading zeroes.
+		if normalize && b == '0' {
+			for {
+				b2, err := t.PeekByte(1)
+
+				// The next character is a zero, skip the leading zero and check again.
+				if err == nil && b2 == b {
+					t.skipBytes('0')
+					continue
+				}
+				// The next character is a dot, go with the zero before the dot and
+				// let the upcoming code handle the dot.
+				if err == nil && b2 == '.' {
+					t.acceptBytes('0')
+					break
+				}
+				// The next character is not a zero, nor a digit at all.
+				// We're looking at a zero on its own here.
+				if err != nil || b2 < '1' || b2 > '9' {
+					t.acceptBytes('0')
+					return true
+				}
+				// The next character is a digit. SKip the leading zero and go with the digit.
+				t.skipBytes('0')
+				t.acceptBytes(b2)
+				break
+			}
+		}
+
+		// Continue accepting bytes as long as they are digits.
+		for {
+			b, err = t.PeekByte(0)
+			if err != nil || b < '0' || b > '9' {
+				break
+			}
+			t.acceptBytes(b)
+		}
+
+		// No dot or no digit after a dot? Then we're done.
+		if b != '.' {
+			return true
+		}
+		b, err = t.PeekByte(1)
+		if err != nil || b < '0' || b > '9' {
+			return true
+		}
+
+		// Continue accepting bytes as long as they are digits.
+		t.acceptBytes('.', b)
+		for {
+			b, err = t.PeekByte(0)
+			if err != nil || b < '0' || b > '9' {
+				break
+			}
+			t.acceptBytes(b)
+		}
+		return true
+	}
 }
 
 // MatchBoolean creates a Handler function that checks if a boolean
@@ -1075,7 +1232,11 @@ func MatchBoolean() Handler {
 			return true
 		}
 		if b1 == 't' || b1 == 'T' {
-			b2, _ := t.PeekByte(1)
+			b2, err := t.PeekByte(1)
+			if err != nil || (b2 != 'R' && b2 != 'r') {
+				t.acceptBytes(b1)
+				return true
+			}
 			b3, _ := t.PeekByte(2)
 			b4, err := t.PeekByte(3)
 			if err == nil && b2 == 'r' && b3 == 'u' && b4 == 'e' {
@@ -1091,11 +1252,14 @@ func MatchBoolean() Handler {
 		}
 
 		if b1 == 'f' || b1 == 'F' {
-			b2, _ := t.PeekByte(1)
+			b2, err := t.PeekByte(1)
+			if err != nil || (b2 != 'A' && b2 != 'a') {
+				t.acceptBytes(b1)
+				return true
+			}
 			b3, _ := t.PeekByte(2)
 			b4, _ := t.PeekByte(3)
 			b5, err := t.PeekByte(4)
-
 			if err == nil && b2 == 'a' && b3 == 'l' && b4 == 's' && b5 == 'e' {
 				t.acceptBytes(b1, b2, b3, b4, b5)
 				return true
@@ -1114,19 +1278,19 @@ func MatchBoolean() Handler {
 // MatchASCII creates a Handler function that matches against any
 // ASCII value on the input.
 func MatchASCII() Handler {
-	return MatchRuneRange('\x00', '\x7F')
+	return MatchByteRange('\x00', '\x7F')
 }
 
 // MatchASCIILower creates a Handler function that matches against any
 // lower case ASCII letter on the input (a - z).
 func MatchASCIILower() Handler {
-	return MatchRuneRange('a', 'z')
+	return MatchByteRange('a', 'z')
 }
 
 // MatchASCIIUpper creates a Handler function that matches against any
 // upper case ASCII letter on the input (a - z).
 func MatchASCIIUpper() Handler {
-	return MatchRuneRange('A', 'Z')
+	return MatchByteRange('A', 'Z')
 }
 
 // MatchUnicodeLetter creates a Handler function that matches against any
@@ -1365,19 +1529,15 @@ func MatchIPv6Net(normalize bool) Handler {
 // In both cases, it would match the first form.
 func ModifyDrop(handler Handler) Handler {
 	return func(t *API) bool {
-		child := t.Fork()
+		runeEnd := t.stackFrame.runeEnd
+		tokenEnd := t.stackFrame.tokenEnd
 		if handler(t) {
-			// Do a partial merge: only move the cursor and read offset forward.
+			// We keep offset and cursor updates, but rollback any runes / tokens
-			// Any produced runes and tokens are ignored and not merged to the parent
+			// that were added by the handler.
-			// (since we're dropping those here).
+			t.stackFrame.runeEnd = runeEnd
-			parent := &t.stackFrames[t.stackLevel-1]
+			t.stackFrame.tokenEnd = tokenEnd
-			parent.offset = t.stackFrame.offset
-			parent.line = t.stackFrame.line
-			parent.column = t.stackFrame.column
-			t.Dispose(child)
 			return true
 		}
-		t.Dispose(child)
 		return false
 	}
 }

tokenize/handlers_builtin_test.go

@@ -231,22 +231,27 @@ func TestAtoms(t *testing.T) {
 		{"F", a.HexDigit, true, "F"},
 		{"g", a.HexDigit, false, "g"},
 		{"G", a.HexDigit, false, "G"},
+		{"09", a.Integer, true, "9"},
+		{"0000129", a.Integer, true, "129"},
 		{"0", a.Integer, true, "0"},
-		{"09", a.Integer, true, "0"}, // following Go: 09 is invalid octal, so only 0 is valid for the integer
+		{"00000", a.Integer, true, "0"},
 		{"1", a.Integer, true, "1"},
 		{"-10X", a.Integer, false, ""},
 		{"+10X", a.Integer, false, ""},
 		{"-10X", a.Signed(a.Integer), true, "-10"},
 		{"+10X", a.Signed(a.Integer), true, "+10"},
 		{"+10.1X", a.Signed(a.Integer), true, "+10"},
-		{"0X", a.Float, true, "0"},
+		{"0X", a.Decimal, true, "0"},
-		{"0X", a.Float, true, "0"},
+		{"0000X", a.Decimal, true, "0"},
-		{"1X", a.Float, true, "1"},
+		{"1X", a.Decimal, true, "1"},
-		{"1.", a.Float, true, "1"}, // incomplete float, so only the 1 is picked up
+		{"01X", a.Decimal, true, "1"},
-		{"123.321X", a.Float, true, "123.321"},
+		{"000001X", a.Decimal, true, "1"},
-		{"-3.14X", a.Float, false, ""},
+		{"1.", a.Decimal, true, "1"}, // incomplete float, so only the 1 is picked up
-		{"-3.14X", a.Signed(a.Float), true, "-3.14"},
+		{"123.321X", a.Decimal, true, "123.321"},
-		{"-003.0014X", a.Signed(a.Float), true, "-003.0014"},
+		{"0.6X", a.Decimal, true, "0.6"},
+		{"-3.14X", a.Decimal, false, ""},
+		{"-3.14X", a.Signed(a.Decimal), true, "-3.14"},
+		{"-003.0014X", a.Signed(a.Decimal), true, "-3.0014"},
 		{"-11", a.IntegerBetween(-10, 10), false, "0"},
 		{"-10", a.IntegerBetween(-10, 10), true, "-10"},
 		{"0", a.IntegerBetween(-10, 10), true, "0"},
@@ -430,8 +435,8 @@ func TestTokenMakers(t *testing.T) {
 		{`4294967295XYZ`, tok.Uint32("L", a.Integer), []tokenize.Token{{Type: "L", Value: uint32(4294967295)}}},
 		{`18446744073709551615XYZ`, tok.Uint64("M", a.Integer), []tokenize.Token{{Type: "M", Value: uint64(18446744073709551615)}}},
 
-		{`3.1415=PI`, tok.Float32("N", a.Float), []tokenize.Token{{Type: "N", Value: float32(3.1415)}}},
+		{`3.1415=PI`, tok.Float32("N", a.Decimal), []tokenize.Token{{Type: "N", Value: float32(3.1415)}}},
-		{`24.19287=PI`, tok.Float64("O", a.Float), []tokenize.Token{{Type: "O", Value: float64(24.19287)}}},
+		{`24.19287=PI`, tok.Float64("O", a.Decimal), []tokenize.Token{{Type: "O", Value: float64(24.19287)}}},
 
 		{`1tTtrueTRUETrue`, c.OneOrMore(tok.Boolean("P", a.Boolean)), []tokenize.Token{
 			{Type: "P", Value: true},