new API: p := NewParser(&opts), p.Markdown(r, f) replaces d := Parse(r, opts), d.WriteHtml(w)

The new api allows to create a parser instance that can be
reused between calls to Markdown. Also, an interface `Formatter'
replaces the tight coupling between former Doc type and HTML
generation.

cmd/markdown/main.go

@@ -31,11 +31,12 @@ func main() {
 		r = f
 	}
 
+	p := markdown.NewParser(&opt)
+
 	startPProf()
 	defer stopPProf()
 
-	doc := markdown.Parse(r, opt)
 	w := bufio.NewWriter(os.Stdout)
-	doc.WriteHtml(w)
+	p.Markdown(r, markdown.ToHTML(w))
 	w.Flush()
 }

doc.go

@@ -6,16 +6,16 @@ Usage example:
 	package main
 
 	import (
-		md "github.com/knieriem/markdown"
+		"github.com/knieriem/markdown"
 		"os"
 		"bufio"
 	)
 
 	func main() {
-		doc := md.Parse(os.Stdin, md.Options{Smart: true})
+		p := markdown.NewParser(&markdown.Options{Smart: true})
 
 		w := bufio.NewWriter(os.Stdout)
-		doc.WriteHtml(w)	
+		p.Markdown(os.Stdin, markdown.ToHTML(w))
 		w.Flush()
 	}
 

markdown.go

@@ -17,8 +17,6 @@
 
 package markdown
 
-// implements Parse()
-
 import (
 	"bytes"
 	"io"
@@ -35,46 +33,73 @@ type Options struct {
 	Dlists       bool
 }
 
-// Parse converts a Markdown document into a tree for later output processing.
-func Parse(r io.Reader, opt Options) *Doc {
-	d := new(Doc)
-	d.extension = opt
-
-	d.parser = new(yyParser)
-	d.parser.Doc = d
-	d.parser.Init()
-
-	s := preformat(r)
-
-	d.parseRule(ruleReferences, s)
-	if opt.Notes {
-		d.parseRule(ruleNotes, s)
-	}
-	raw := d.parseMarkdown(s)
-	d.tree = d.processRawBlocks(raw)
-	return d
+type Parser struct {
+	yy           yyParser
+	preformatBuf *bytes.Buffer
 }
 
-func (d *Doc) parseRule(rule int, s string) {
-	m := d.parser
-	if m.ResetBuffer(s) != "" {
+// NewParser creates an instance of a parser. It can be reused
+// so that stacks and buffers need not be allocated anew for
+// each Markdown call.
+func NewParser(opt *Options) (p *Parser) {
+	p = new(Parser)
+	if opt != nil {
+		p.yy.state.extension = *opt
+	}
+	p.yy.Init()
+	p.preformatBuf = bytes.NewBuffer(make([]byte, 0, 32768))
+	return
+}
+
+type Formatter interface {
+	FormatBlock(*element)
+	Finish()
+}
+
+// Markdown parses input from an io.Reader into a tree, and sends
+// parsed blocks to a Formatter
+func (p *Parser) Markdown(src io.Reader, f Formatter) {
+	s := p.preformat(src)
+
+	p.parseRule(ruleReferences, s)
+	if p.yy.extension.Notes {
+		p.parseRule(ruleNotes, s)
+	}
+
+L:
+	for {
+		tree := p.parseRule(ruleDocblock, s)
+		s = p.yy.ResetBuffer("")
+		tree = p.processRawBlocks(tree)
+		f.FormatBlock(tree)
+		switch s {
+		case "", "\n", "\n\n":
+			break L
+		}
+	}
+	f.Finish()
+}
+
+func (p *Parser) parseRule(rule int, s string) (tree *element) {
+	if p.yy.ResetBuffer(s) != "" {
 		log.Fatalf("Buffer not empty")
 	}
-	if err := m.Parse(rule); err != nil {
+	if err := p.yy.Parse(rule); err != nil {
 		log.Fatalln("markdown:", err)
 	}
-}
-
-func (d *Doc) parseMarkdown(text string) *element {
-	d.parseRule(ruleDoc, text)
-	return d.tree
+	switch rule {
+	case ruleDoc, ruleDocblock:
+		tree = p.yy.state.tree
+		p.yy.state.tree = nil
+	}
+	return
 }
 
 /* process_raw_blocks - traverses an element list, replacing any RAW elements with
  * the result of parsing them as markdown text, and recursing into the children
  * of parent elements.  The result should be a tree of elements without any RAWs.
  */
-func (d *Doc) processRawBlocks(input *element) *element {
+func (p *Parser) processRawBlocks(input *element) *element {
 
 	for current := input; current != nil; current = current.next {
 		if current.key == RAW {
@@ -86,7 +111,7 @@ func (d *Doc) processRawBlocks(input *element) *element {
 			current.children = nil
 			listEnd := &current.children
 			for _, contents := range strings.Split(current.contents.str, "\001") {
-				if list := d.parseMarkdown(contents); list != nil {
+				if list := p.parseRule(ruleDoc, contents); list != nil {
 					*listEnd = list
 					for list.next != nil {
 						list = list.next
@@ -97,7 +122,7 @@ func (d *Doc) processRawBlocks(input *element) *element {
 			current.contents.str = ""
 		}
 		if current.children != nil {
-			current.children = d.processRawBlocks(current.children)
+			current.children = p.processRawBlocks(current.children)
 		}
 	}
 	return input
@@ -110,11 +135,12 @@ const (
 /* preformat - allocate and copy text buffer while
  * performing tab expansion.
  */
-func preformat(r io.Reader) (s string) {
+func (p *Parser) preformat(r io.Reader) (s string) {
 	charstotab := TABSTOP
 	buf := make([]byte, 32768)
 
-	b := bytes.NewBuffer(make([]byte, 0, 32768))
+	b := p.preformatBuf
+	b.Reset()
 	for {
 		n, err := r.Read(buf)
 		if err != nil {

output.go

@@ -27,6 +27,7 @@ import (
 )
 
 type Writer interface {
+	Write([]byte) (int, error)
 	WriteString(string) (int, error)
 	WriteRune(rune) (int, error)
 	WriteByte(byte) error
@@ -34,26 +35,28 @@ type Writer interface {
 
 type htmlOut struct {
 	Writer
-	padded int
+	padded    int
 	obfuscate bool
 
 	notenum  int
 	endNotes []*element /* List of endnotes to print after main content. */
 }
 
-// WriteHtml prints a document tree in HTML format using the specified Writer.
-//
-func (d *Doc) WriteHtml(w Writer) int {
-	out := new(htmlOut)
-	out.Writer = w
-	out.padded = 2
-	out.elist(d.tree)
-	if len(out.endNotes) != 0 {
-		out.pad(2)
-		out.printEndnotes()
+func ToHTML(w Writer) Formatter {
+	f := new(htmlOut)
+	f.Writer = w
+	f.padded = 2
+	return f
+}
+func (f *htmlOut) FormatBlock(tree *element) {
+	f.elist(tree)
+}
+func (f *htmlOut) Finish() {
+	if len(f.endNotes) != 0 {
+		f.pad(2)
+		f.printEndnotes()
 	}
-	out.WriteByte('\n')
-	return 0
+	f.WriteByte('\n')
 }
 
 // pad - add newlines if needed

parser.leg

@@ -91,10 +91,8 @@ const (
 	numVAL
 )
 
-type Doc struct {
-	parser		*yyParser
-	extension	Options
-
+type state struct {
+	extension		Options
 	tree				*element	/* Results of parse. */
 	references			*element	/* List of link references found. */
 	notes				*element	/* List of footnotes found. */
@@ -102,7 +100,9 @@ type Doc struct {
 
 %}
 
-%userstate *Doc
+%userstate state
+
+%noexport
 
 %YYSTYPE *element
 
@@ -110,6 +110,8 @@ Doc =       a:StartList ( Block { a = cons($$, a) } )*
             { p.tree = reverse(a) }
             commit
 
+Docblock = Block { p.tree = $$ } commit
+
 Block =     BlankLine*
             ( BlockQuote
             | Verbatim
@@ -965,8 +967,8 @@ func match_inlines(l1, l2 *element) bool {
 /* find_reference - return true if link found in references matching label.
  * 'link' is modified with the matching url and title.
  */
-func (d *Doc) findReference(label *element) (*link, bool) {
-	for cur := d.references; cur != nil; cur = cur.next {
+func (p *yyParser) findReference(label *element) (*link, bool) {
+	for cur := p.references; cur != nil; cur = cur.next {
 		l := cur.contents.link
 		if match_inlines(label, l.label) {
 			return l, true
@@ -979,8 +981,8 @@ func (d *Doc) findReference(label *element) (*link, bool) {
 /* find_note - return true if note found in notes matching label.
  * if found, 'result' is set to point to matched note.
  */
-func (d *Doc) find_note(label string) (*element, bool) {
-	for el := d.notes; el != nil; el = el.next {
+func (p *yyParser) find_note(label string) (*element, bool) {
+	for el := p.notes; el != nil; el = el.next {
 		if label == el.contents.str {
 			return el, true
 		}