Lex performance improvement (#176)

* Use []token instead of chan token

name             old time/op    new time/op    delta
ParseToml-8        1.18ms ± 0%    0.91ms ± 0%  -22.98%
UnmarshalToml-8    1.29ms ± 0%    0.95ms ± 0%  -25.96%

name             old alloc/op   new alloc/op   delta
ParseToml-8         429kB ± 0%     444kB ± 0%   +3.49%
UnmarshalToml-8     451kB ± 0%     466kB ± 0%   +3.32%

name             old allocs/op  new allocs/op  delta
ParseToml-8         14.1k ± 0%     13.7k ± 0%   -2.31%
UnmarshalToml-8     15.1k ± 0%     14.7k ± 0%   -2.16%

* Lex on []byte instead of io.Reader

name             old time/op    new time/op    delta
ParseToml-8        1.18ms ± 0%    0.29ms ± 0%  -75.18%
UnmarshalToml-8    1.27ms ± 0%    0.38ms ± 0%  -70.38%

name             old alloc/op   new alloc/op   delta
ParseToml-8         429kB ± 0%     135kB ± 0%  -68.53%
UnmarshalToml-8     451kB ± 0%     157kB ± 0%  -65.22%

name             old allocs/op  new allocs/op  delta
ParseToml-8         14.1k ± 0%      3.2k ± 0%  -77.20%
UnmarshalToml-8     15.1k ± 0%      4.2k ± 0%  -72.00%

lexer.go

@@ -9,12 +9,9 @@ import (
 	"bytes"
 	"errors"
 	"fmt"
-	"io"
 	"regexp"
 	"strconv"
 	"strings"
-
-	"github.com/pelletier/go-buffruneio"
 )
 
 var dateRegexp *regexp.Regexp
@@ -24,43 +21,43 @@ type tomlLexStateFn func() tomlLexStateFn
 
 // Define lexer
 type tomlLexer struct {
-	input         *buffruneio.Reader // Textual source
-	buffer        bytes.Buffer       // Runes composing the current token
-	tokens        chan token
-	depth         int
-	line          int
-	col           int
-	endbufferLine int
-	endbufferCol  int
+	inputIdx          int
+	input             []rune // Textual source
+	currentTokenStart int
+	currentTokenStop  int
+	tokens            []token
+	depth             int
+	line              int
+	col               int
+	endbufferLine     int
+	endbufferCol      int
 }
 
 // Basic read operations on input
 
 func (l *tomlLexer) read() rune {
-	r, _, err := l.input.ReadRune()
-	if err != nil {
-		panic(err)
-	}
+	r := l.peek()
 	if r == '\n' {
 		l.endbufferLine++
 		l.endbufferCol = 1
 	} else {
 		l.endbufferCol++
 	}
+	l.inputIdx++
 	return r
 }
 
 func (l *tomlLexer) next() rune {
 	r := l.read()
 
 	if r != eof {
-		l.buffer.WriteRune(r)
+		l.currentTokenStop++
 	}
 	return r
 }
 
 func (l *tomlLexer) ignore() {
-	l.buffer.Reset()
+	l.currentTokenStart = l.currentTokenStop
 	l.line = l.endbufferLine
 	l.col = l.endbufferCol
 }
@@ -77,49 +74,46 @@ func (l *tomlLexer) fastForward(n int) {
 }
 
 func (l *tomlLexer) emitWithValue(t tokenType, value string) {
-	l.tokens <- token{
+	l.tokens = append(l.tokens, token{
 		Position: Position{l.line, l.col},
 		typ:      t,
 		val:      value,
-	}
+	})
 	l.ignore()
 }
 
 func (l *tomlLexer) emit(t tokenType) {
-	l.emitWithValue(t, l.buffer.String())
+	l.emitWithValue(t, string(l.input[l.currentTokenStart:l.currentTokenStop]))
 }
 
 func (l *tomlLexer) peek() rune {
-	r, _, err := l.input.ReadRune()
-	if err != nil {
-		panic(err)
+	if l.inputIdx >= len(l.input) {
+		return eof
 	}
-	l.input.UnreadRune()
-	return r
+	return l.input[l.inputIdx]
 }
 
-func (l *tomlLexer) follow(next string) bool {
-	for _, expectedRune := range next {
-		r, _, err := l.input.ReadRune()
-		defer l.input.UnreadRune()
-		if err != nil {
-			panic(err)
-		}
-		if expectedRune != r {
-			return false
-		}
+func (l *tomlLexer) peekString(size int) string {
+	maxIdx := len(l.input)
+	upperIdx := l.inputIdx + size // FIXME: potential overflow
+	if upperIdx > maxIdx {
+		upperIdx = maxIdx
 	}
-	return true
+	return string(l.input[l.inputIdx:upperIdx])
+}
+
+func (l *tomlLexer) follow(next string) bool {
+	return next == l.peekString(len(next))
 }
 
 // Error management
 
 func (l *tomlLexer) errorf(format string, args ...interface{}) tomlLexStateFn {
-	l.tokens <- token{
+	l.tokens = append(l.tokens, token{
 		Position: Position{l.line, l.col},
 		typ:      tokenError,
 		val:      fmt.Sprintf(format, args...),
-	}
+	})
 	return nil
 }
 
@@ -220,7 +214,7 @@ func (l *tomlLexer) lexRvalue() tomlLexStateFn {
 			break
 		}
 
-		possibleDate := string(l.input.PeekRunes(35))
+		possibleDate := l.peekString(35)
 		dateMatch := dateRegexp.FindString(possibleDate)
 		if dateMatch != "" {
 			l.fastForward(len(dateMatch))
@@ -537,7 +531,7 @@ func (l *tomlLexer) lexInsideTableArrayKey() tomlLexStateFn {
 	for r := l.peek(); r != eof; r = l.peek() {
 		switch r {
 		case ']':
-			if l.buffer.Len() > 0 {
+			if l.currentTokenStop > l.currentTokenStart {
 				l.emit(tokenKeyGroupArray)
 			}
 			l.next()
@@ -560,7 +554,7 @@ func (l *tomlLexer) lexInsideTableKey() tomlLexStateFn {
 	for r := l.peek(); r != eof; r = l.peek() {
 		switch r {
 		case ']':
-			if l.buffer.Len() > 0 {
+			if l.currentTokenStop > l.currentTokenStart {
 				l.emit(tokenKeyGroup)
 			}
 			l.next()
@@ -635,24 +629,23 @@ func (l *tomlLexer) run() {
 	for state := l.lexVoid; state != nil; {
 		state = state()
 	}
-	close(l.tokens)
 }
 
 func init() {
 	dateRegexp = regexp.MustCompile(`^\d{1,4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(\.\d{1,9})?(Z|[+-]\d{2}:\d{2})`)
 }
 
 // Entry point
-func lexToml(input io.Reader) chan token {
-	bufferedInput := buffruneio.NewReader(input)
+func lexToml(inputBytes []byte) []token {
+	runes := bytes.Runes(inputBytes)
 	l := &tomlLexer{
-		input:         bufferedInput,
-		tokens:        make(chan token),
+		input:         runes,
+		tokens:        make([]token, 0, 256),
 		line:          1,
 		col:           1,
 		endbufferLine: 1,
 		endbufferCol:  1,
 	}
-	go l.run()
+	l.run()
 	return l.tokens
 }

lexer_test.go

@@ -1,38 +1,14 @@
 package toml
 
 import (
-	"os"
-	"strings"
+	"reflect"
 	"testing"
 )
 
 func testFlow(t *testing.T, input string, expectedFlow []token) {
-	ch := lexToml(strings.NewReader(input))
-	for _, expected := range expectedFlow {
-		token := <-ch
-		if token != expected {
-			t.Log("While testing: ", input)
-			t.Log("compared (got)", token, "to (expected)", expected)
-			t.Log("\tvalue:", token.val, "<->", expected.val)
-			t.Log("\tvalue as bytes:", []byte(token.val), "<->", []byte(expected.val))
-			t.Log("\ttype:", token.typ.String(), "<->", expected.typ.String())
-			t.Log("\tline:", token.Line, "<->", expected.Line)
-			t.Log("\tcolumn:", token.Col, "<->", expected.Col)
-			t.Log("compared", token, "to", expected)
-			t.FailNow()
-		}
-	}
-
-	tok, ok := <-ch
-	if ok {
-		t.Log("channel is not closed!")
-		t.Log(len(ch)+1, "tokens remaining:")
-
-		t.Log("token ->", tok)
-		for token := range ch {
-			t.Log("token ->", token)
-		}
-		t.FailNow()
+	tokens := lexToml([]byte(input))
+	if !reflect.DeepEqual(tokens, expectedFlow) {
+		t.Fatal("Different flows. Expected\n", expectedFlow, "\nGot:\n", tokens)
 	}
 }
 
@@ -767,13 +743,8 @@ pluralizeListTitles = false
 	url = "https://github.com/spf13/hugo/releases"
 	weight = -200
 `
-	rd := strings.NewReader(sample)
-
 	b.ResetTimer()
 	for i := 0; i < b.N; i++ {
-		rd.Seek(0, os.SEEK_SET)
-		ch := lexToml(rd)
-		for _ = range ch {
-		}
+		lexToml([]byte(sample))
 	}
 }

parser.go

@@ -13,9 +13,9 @@ import (
 )
 
 type tomlParser struct {
-	flow          chan token
+	flowIdx       int
+	flow          []token
 	tree          *Tree
-	tokensBuffer  []token
 	currentTable  []string
 	seenTableKeys []string
 }
@@ -34,16 +34,10 @@ func (p *tomlParser) run() {
 }
 
 func (p *tomlParser) peek() *token {
-	if len(p.tokensBuffer) != 0 {
-		return &(p.tokensBuffer[0])
-	}
-
-	tok, ok := <-p.flow
-	if !ok {
+	if p.flowIdx >= len(p.flow) {
 		return nil
 	}
-	p.tokensBuffer = append(p.tokensBuffer, tok)
-	return &tok
+	return &p.flow[p.flowIdx]
 }
 
 func (p *tomlParser) assume(typ tokenType) {
@@ -57,16 +51,12 @@ func (p *tomlParser) assume(typ tokenType) {
 }
 
 func (p *tomlParser) getToken() *token {
-	if len(p.tokensBuffer) != 0 {
-		tok := p.tokensBuffer[0]
-		p.tokensBuffer = p.tokensBuffer[1:]
-		return &tok
-	}
-	tok, ok := <-p.flow
-	if !ok {
+	tok := p.peek()
+	if tok == nil {
 		return nil
 	}
-	return &tok
+	p.flowIdx++
+	return tok
 }
 
 func (p *tomlParser) parseStart() tomlParserStateFn {
@@ -374,13 +364,13 @@ func (p *tomlParser) parseArray() interface{} {
 	return array
 }
 
-func parseToml(flow chan token) *Tree {
+func parseToml(flow []token) *Tree {
 	result := newTree()
 	result.position = Position{1, 1}
 	parser := &tomlParser{
+		flowIdx:       0,
 		flow:          flow,
 		tree:          result,
-		tokensBuffer:  make([]token, 0),
 		currentTable:  make([]string, 0),
 		seenTableKeys: make([]string, 0),
 	}

toml.go

@@ -4,6 +4,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"io/ioutil"
 	"os"
 	"runtime"
 	"strings"
@@ -251,8 +252,8 @@ func (t *Tree) createSubTree(keys []string, pos Position) error {
 	return nil
 }
 
-// LoadReader creates a Tree from any io.Reader.
-func LoadReader(reader io.Reader) (tree *Tree, err error) {
+// LoadBytes creates a Tree from a []byte.
+func LoadBytes(b []byte) (tree *Tree, err error) {
 	defer func() {
 		if r := recover(); r != nil {
 			if _, ok := r.(runtime.Error); ok {
@@ -261,13 +262,23 @@ func LoadReader(reader io.Reader) (tree *Tree, err error) {
 			err = errors.New(r.(string))
 		}
 	}()
-	tree = parseToml(lexToml(reader))
+	tree = parseToml(lexToml(b))
+	return
+}
+
+// LoadReader creates a Tree from any io.Reader.
+func LoadReader(reader io.Reader) (tree *Tree, err error) {
+	inputBytes, err := ioutil.ReadAll(reader)
+	if err != nil {
+		return
+	}
+	tree, err = LoadBytes(inputBytes)
 	return
 }
 
 // Load creates a Tree from a string.
 func Load(content string) (tree *Tree, err error) {
-	return LoadReader(strings.NewReader(content))
+	return LoadBytes([]byte(content))
 }
 
 // LoadFile creates a Tree from a file.