all repos — grayfriday @ 4a7ff562a7e3c58311b4678169a0ae9b0ecc2e59

blackfriday fork with a few changes

markdown.go (view raw)

   1//
   2// Blackfriday Markdown Processor
   3// Available at http://github.com/russross/blackfriday
   4//
   5// Copyright © 2011 Russ Ross <russ@russross.com>.
   6// Distributed under the Simplified BSD License.
   7// See README.md for details.
   8//
   9
  10//
  11//
  12// Markdown parsing and processing
  13//
  14//
  15
  16// Blackfriday markdown processor.
  17//
  18// Translates plain text with simple formatting rules into HTML or LaTeX.
  19package blackfriday
  20
  21import (
  22	"bytes"
  23	"fmt"
  24	"strings"
  25	"unicode/utf8"
  26)
  27
  28const VERSION = "1.4"
  29
  30type Extensions int
  31
  32// These are the supported markdown parsing extensions.
  33// OR these values together to select multiple extensions.
  34const (
  35	NoExtensions           Extensions = 0
  36	NoIntraEmphasis        Extensions = 1 << iota // Ignore emphasis markers inside words
  37	Tables                                        // Render tables
  38	FencedCode                                    // Render fenced code blocks
  39	Autolink                                      // Detect embedded URLs that are not explicitly marked
  40	Strikethrough                                 // Strikethrough text using ~~test~~
  41	LaxHTMLBlocks                                 // Loosen up HTML block parsing rules
  42	SpaceHeaders                                  // Be strict about prefix header rules
  43	HardLineBreak                                 // Translate newlines into line breaks
  44	TabSizeEight                                  // Expand tabs to eight spaces instead of four
  45	Footnotes                                     // Pandoc-style footnotes
  46	NoEmptyLineBeforeBlock                        // No need to insert an empty line to start a (code, quote, ordered list, unordered list) block
  47	HeaderIDs                                     // specify header IDs  with {#id}
  48	Titleblock                                    // Titleblock ala pandoc
  49	AutoHeaderIDs                                 // Create the header ID from the text
  50	BackslashLineBreak                            // Translate trailing backslashes into line breaks
  51	DefinitionLists                               // Render definition lists
  52
  53	CommonHtmlFlags HtmlFlags = UseXHTML | UseSmartypants |
  54		SmartypantsFractions | SmartypantsDashes | SmartypantsLatexDashes
  55
  56	CommonExtensions Extensions = NoIntraEmphasis | Tables | FencedCode |
  57		Autolink | Strikethrough | SpaceHeaders | HeaderIDs |
  58		BackslashLineBreak | DefinitionLists
  59)
  60
  61var DefaultOptions = Options{
  62	Extensions: CommonExtensions,
  63}
  64
  65type LinkType int
  66
  67// These are the possible flag values for the link renderer.
  68// Only a single one of these values will be used; they are not ORed together.
  69// These are mostly of interest if you are writing a new output format.
  70const (
  71	LinkTypeNotAutolink LinkType = iota
  72	LinkTypeNormal
  73	LinkTypeEmail
  74)
  75
  76type ListType int
  77
  78// These are the possible flag values for the ListItem renderer.
  79// Multiple flag values may be ORed together.
  80// These are mostly of interest if you are writing a new output format.
  81const (
  82	ListTypeOrdered ListType = 1 << iota
  83	ListTypeDefinition
  84	ListTypeTerm
  85
  86	ListItemContainsBlock
  87	ListItemBeginningOfList
  88	ListItemEndOfList
  89)
  90
  91type TableFlags int
  92
  93// These are the possible flag values for the table cell renderer.
  94// Only a single one of these values will be used; they are not ORed together.
  95// These are mostly of interest if you are writing a new output format.
  96const (
  97	TableAlignmentLeft = 1 << iota
  98	TableAlignmentRight
  99	TableAlignmentCenter = (TableAlignmentLeft | TableAlignmentRight)
 100)
 101
 102// The size of a tab stop.
 103const (
 104	TabSizeDefault = 4
 105	TabSizeDouble  = 8
 106)
 107
 108// blockTags is a set of tags that are recognized as HTML block tags.
 109// Any of these can be included in markdown text without special escaping.
 110var blockTags = map[string]struct{}{
 111	"blockquote": struct{}{},
 112	"del":        struct{}{},
 113	"div":        struct{}{},
 114	"dl":         struct{}{},
 115	"fieldset":   struct{}{},
 116	"form":       struct{}{},
 117	"h1":         struct{}{},
 118	"h2":         struct{}{},
 119	"h3":         struct{}{},
 120	"h4":         struct{}{},
 121	"h5":         struct{}{},
 122	"h6":         struct{}{},
 123	"iframe":     struct{}{},
 124	"ins":        struct{}{},
 125	"math":       struct{}{},
 126	"noscript":   struct{}{},
 127	"ol":         struct{}{},
 128	"pre":        struct{}{},
 129	"p":          struct{}{},
 130	"script":     struct{}{},
 131	"style":      struct{}{},
 132	"table":      struct{}{},
 133	"ul":         struct{}{},
 134
 135	// HTML5
 136	"address":    struct{}{},
 137	"article":    struct{}{},
 138	"aside":      struct{}{},
 139	"canvas":     struct{}{},
 140	"figcaption": struct{}{},
 141	"figure":     struct{}{},
 142	"footer":     struct{}{},
 143	"header":     struct{}{},
 144	"hgroup":     struct{}{},
 145	"main":       struct{}{},
 146	"nav":        struct{}{},
 147	"output":     struct{}{},
 148	"progress":   struct{}{},
 149	"section":    struct{}{},
 150	"video":      struct{}{},
 151}
 152
 153// Renderer is the rendering interface.
 154// This is mostly of interest if you are implementing a new rendering format.
 155//
 156// When a byte slice is provided, it contains the (rendered) contents of the
 157// element.
 158//
 159// When a callback is provided instead, it will write the contents of the
 160// respective element directly to the output buffer and return true on success.
 161// If the callback returns false, the rendering function should reset the
 162// output buffer as though it had never been called.
 163//
 164// Currently Html and Latex implementations are provided
 165type Renderer interface {
 166	// block-level callbacks
 167	BlockCode(text []byte, lang string)
 168	BlockQuote(text []byte)
 169	BlockHtml(text []byte)
 170	BeginHeader(level int, id string)
 171	EndHeader(level int, id string, header []byte)
 172	HRule()
 173	BeginList(flags ListType)
 174	EndList(flags ListType)
 175	ListItem(text []byte, flags ListType)
 176	BeginParagraph()
 177	EndParagraph()
 178	Table(header []byte, body []byte, columnData []int)
 179	TableRow(text []byte)
 180	TableHeaderCell(out *bytes.Buffer, text []byte, flags int)
 181	TableCell(out *bytes.Buffer, text []byte, flags int)
 182	BeginFootnotes()
 183	EndFootnotes()
 184	FootnoteItem(name, text []byte, flags ListType)
 185	TitleBlock(text []byte)
 186
 187	// Span-level callbacks
 188	AutoLink(link []byte, kind LinkType)
 189	CodeSpan(text []byte)
 190	DoubleEmphasis(text []byte)
 191	Emphasis(text []byte)
 192	Image(link []byte, title []byte, alt []byte)
 193	LineBreak()
 194	Link(link []byte, title []byte, content []byte)
 195	RawHtmlTag(tag []byte)
 196	TripleEmphasis(text []byte)
 197	StrikeThrough(text []byte)
 198	FootnoteRef(ref []byte, id int)
 199
 200	// Low-level callbacks
 201	Entity(entity []byte)
 202	NormalText(text []byte)
 203
 204	// Header and footer
 205	DocumentHeader()
 206	DocumentFooter()
 207
 208	GetFlags() HtmlFlags
 209	Write(b []byte) (int, error)
 210
 211	Render(ast *Node) []byte
 212}
 213
 214// Callback functions for inline parsing. One such function is defined
 215// for each character that triggers a response when parsing inline data.
 216type inlineParser func(p *parser, data []byte, offset int) int
 217
 218// Parser holds runtime state used by the parser.
 219// This is constructed by the Markdown function.
 220type parser struct {
 221	refOverride    ReferenceOverrideFunc
 222	refs           map[string]*reference
 223	inlineCallback [256]inlineParser
 224	flags          Extensions
 225	nesting        int
 226	maxNesting     int
 227	insideLink     bool
 228
 229	// Footnotes need to be ordered as well as available to quickly check for
 230	// presence. If a ref is also a footnote, it's stored both in refs and here
 231	// in notes. Slice is nil if footnotes not enabled.
 232	notes []*reference
 233
 234	doc                  *Node
 235	tip                  *Node // = doc
 236	oldTip               *Node
 237	lastMatchedContainer *Node // = doc
 238	allClosed            bool
 239	currBlock            *Node // a block node currently being parsed by inline parser
 240}
 241
 242func (p *parser) getRef(refid string) (ref *reference, found bool) {
 243	if p.refOverride != nil {
 244		r, overridden := p.refOverride(refid)
 245		if overridden {
 246			if r == nil {
 247				return nil, false
 248			}
 249			return &reference{
 250				link:     []byte(r.Link),
 251				title:    []byte(r.Title),
 252				noteId:   0,
 253				hasBlock: false,
 254				text:     []byte(r.Text)}, true
 255		}
 256	}
 257	// refs are case insensitive
 258	ref, found = p.refs[strings.ToLower(refid)]
 259	return ref, found
 260}
 261
 262func (p *parser) finalize(block *Node) {
 263	above := block.Parent
 264	block.open = false
 265	p.tip = above
 266}
 267
 268func (p *parser) addChild(node NodeType, offset uint32) *Node {
 269	for !p.tip.canContain(node) {
 270		p.finalize(p.tip)
 271	}
 272	newNode := NewNode(node)
 273	newNode.content = []byte{}
 274	p.tip.appendChild(newNode)
 275	p.tip = newNode
 276	return newNode
 277}
 278
 279func (p *parser) closeUnmatchedBlocks() {
 280	if !p.allClosed {
 281		for p.oldTip != p.lastMatchedContainer {
 282			parent := p.oldTip.Parent
 283			p.finalize(p.oldTip)
 284			p.oldTip = parent
 285		}
 286		p.allClosed = true
 287	}
 288}
 289
 290//
 291//
 292// Public interface
 293//
 294//
 295
 296// Reference represents the details of a link.
 297// See the documentation in Options for more details on use-case.
 298type Reference struct {
 299	// Link is usually the URL the reference points to.
 300	Link string
 301	// Title is the alternate text describing the link in more detail.
 302	Title string
 303	// Text is the optional text to override the ref with if the syntax used was
 304	// [refid][]
 305	Text string
 306}
 307
 308// ReferenceOverrideFunc is expected to be called with a reference string and
 309// return either a valid Reference type that the reference string maps to or
 310// nil. If overridden is false, the default reference logic will be executed.
 311// See the documentation in Options for more details on use-case.
 312type ReferenceOverrideFunc func(reference string) (ref *Reference, overridden bool)
 313
 314// Options represents configurable overrides and callbacks (in addition to the
 315// extension flag set) for configuring a Markdown parse.
 316type Options struct {
 317	// Extensions is a flag set of bit-wise ORed extension bits. See the
 318	// Extensions flags defined in this package.
 319	Extensions Extensions
 320
 321	// ReferenceOverride is an optional function callback that is called every
 322	// time a reference is resolved.
 323	//
 324	// In Markdown, the link reference syntax can be made to resolve a link to
 325	// a reference instead of an inline URL, in one of the following ways:
 326	//
 327	//  * [link text][refid]
 328	//  * [refid][]
 329	//
 330	// Usually, the refid is defined at the bottom of the Markdown document. If
 331	// this override function is provided, the refid is passed to the override
 332	// function first, before consulting the defined refids at the bottom. If
 333	// the override function indicates an override did not occur, the refids at
 334	// the bottom will be used to fill in the link details.
 335	ReferenceOverride ReferenceOverrideFunc
 336}
 337
 338// MarkdownBasic is a convenience function for simple rendering.
 339// It processes markdown input with no extensions enabled.
 340func MarkdownBasic(input []byte) []byte {
 341	// set up the HTML renderer
 342	htmlFlags := UseXHTML
 343	renderer := HtmlRenderer(htmlFlags, "", "")
 344
 345	// set up the parser
 346	return MarkdownOptions(input, renderer, Options{Extensions: 0})
 347}
 348
 349// Call Markdown with most useful extensions enabled
 350// MarkdownCommon is a convenience function for simple rendering.
 351// It processes markdown input with common extensions enabled, including:
 352//
 353// * Smartypants processing with smart fractions and LaTeX dashes
 354//
 355// * Intra-word emphasis suppression
 356//
 357// * Tables
 358//
 359// * Fenced code blocks
 360//
 361// * Autolinking
 362//
 363// * Strikethrough support
 364//
 365// * Strict header parsing
 366//
 367// * Custom Header IDs
 368func MarkdownCommon(input []byte) []byte {
 369	// set up the HTML renderer
 370	renderer := HtmlRenderer(CommonHtmlFlags, "", "")
 371	return MarkdownOptions(input, renderer, DefaultOptions)
 372}
 373
 374// Markdown is the main rendering function.
 375// It parses and renders a block of markdown-encoded text.
 376// The supplied Renderer is used to format the output, and extensions dictates
 377// which non-standard extensions are enabled.
 378//
 379// To use the supplied Html or LaTeX renderers, see HtmlRenderer and
 380// LatexRenderer, respectively.
 381func Markdown(input []byte, renderer Renderer, extensions Extensions) []byte {
 382	return MarkdownOptions(input, renderer, Options{
 383		Extensions: extensions})
 384}
 385
 386// MarkdownOptions is just like Markdown but takes additional options through
 387// the Options struct.
 388func MarkdownOptions(input []byte, renderer Renderer, opts Options) []byte {
 389	// no point in parsing if we can't render
 390	if renderer == nil {
 391		return nil
 392	}
 393	return renderer.Render(Parse(input, opts))
 394}
 395
 396func Parse(input []byte, opts Options) *Node {
 397	extensions := opts.Extensions
 398
 399	// fill in the render structure
 400	p := new(parser)
 401	p.flags = extensions
 402	p.refOverride = opts.ReferenceOverride
 403	p.refs = make(map[string]*reference)
 404	p.maxNesting = 16
 405	p.insideLink = false
 406
 407	docNode := NewNode(Document)
 408	p.doc = docNode
 409	p.tip = docNode
 410	p.oldTip = docNode
 411	p.lastMatchedContainer = docNode
 412	p.allClosed = true
 413
 414	// register inline parsers
 415	p.inlineCallback['*'] = emphasis
 416	p.inlineCallback['_'] = emphasis
 417	if extensions&Strikethrough != 0 {
 418		p.inlineCallback['~'] = emphasis
 419	}
 420	p.inlineCallback['`'] = codeSpan
 421	p.inlineCallback['\n'] = lineBreak
 422	p.inlineCallback['['] = link
 423	p.inlineCallback['<'] = leftAngle
 424	p.inlineCallback['\\'] = escape
 425	p.inlineCallback['&'] = entity
 426	p.inlineCallback['!'] = maybeImage
 427	p.inlineCallback['^'] = maybeInlineFootnote
 428
 429	if extensions&Autolink != 0 {
 430		p.inlineCallback['h'] = maybeAutoLink
 431		p.inlineCallback['m'] = maybeAutoLink
 432		p.inlineCallback['f'] = maybeAutoLink
 433		p.inlineCallback['H'] = maybeAutoLink
 434		p.inlineCallback['M'] = maybeAutoLink
 435		p.inlineCallback['F'] = maybeAutoLink
 436	}
 437
 438	if extensions&Footnotes != 0 {
 439		p.notes = make([]*reference, 0)
 440	}
 441
 442	first := firstPass(p, input)
 443	secondPass(p, first)
 444	// walk the tree and finish up some of unfinished blocks:
 445	for p.tip != nil {
 446		p.finalize(p.tip)
 447	}
 448	ForEachNode(p.doc, func(node *Node, entering bool) {
 449		if node.Type == Paragraph || node.Type == Header || node.Type == TableCell {
 450			p.currBlock = node
 451			p.inline(node.content)
 452			node.content = nil
 453		}
 454	})
 455	p.parseRefsToAST()
 456	return p.doc
 457}
 458
 459func (p *parser) parseRefsToAST() {
 460	if p.flags&Footnotes == 0 || len(p.notes) == 0 {
 461		return
 462	}
 463	p.tip = p.doc
 464	finalizeHtmlBlock(p.addBlock(HtmlBlock, []byte(`<div class="footnotes">`)))
 465	p.addBlock(HorizontalRule, nil)
 466	block := p.addBlock(List, nil)
 467	block.ListData = &ListData{ // TODO: fill in the real ListData
 468		Flags:      ListTypeOrdered,
 469		Tight:      false,
 470		BulletChar: '*',
 471		Delimiter:  0,
 472	}
 473	flags := ListItemBeginningOfList
 474	// Note: this loop is intentionally explicit, not range-form. This is
 475	// because the body of the loop will append nested footnotes to p.notes and
 476	// we need to process those late additions. Range form would only walk over
 477	// the fixed initial set.
 478	for i := 0; i < len(p.notes); i++ {
 479		ref := p.notes[i]
 480		block := p.addBlock(Item, nil)
 481		block.ListData = &ListData{ // TODO: fill in the real ListData
 482			Flags:      ListTypeOrdered,
 483			Tight:      false,
 484			BulletChar: '*',
 485			Delimiter:  0,
 486			RefLink:    ref.link,
 487		}
 488		if ref.hasBlock {
 489			flags |= ListItemContainsBlock
 490			p.block(ref.title)
 491		} else {
 492			p.currBlock = block
 493			p.inline(ref.title)
 494		}
 495		flags &^= ListItemBeginningOfList | ListItemContainsBlock
 496	}
 497	above := block.Parent
 498	finalizeList(block)
 499	p.tip = above
 500	finalizeHtmlBlock(p.addBlock(HtmlBlock, []byte("</div>")))
 501	ForEachNode(block, func(node *Node, entering bool) {
 502		if node.Type == Paragraph || node.Type == Header {
 503			p.currBlock = node
 504			p.inline(node.content)
 505			node.content = nil
 506		}
 507	})
 508}
 509
 510// first pass:
 511// - extract references
 512// - expand tabs
 513// - normalize newlines
 514// - copy everything else
 515func firstPass(p *parser, input []byte) []byte {
 516	var out bytes.Buffer
 517	tabSize := TabSizeDefault
 518	if p.flags&TabSizeEight != 0 {
 519		tabSize = TabSizeDouble
 520	}
 521	beg, end := 0, 0
 522	lastFencedCodeBlockEnd := 0
 523	for beg < len(input) { // iterate over lines
 524		if end = isReference(p, input[beg:], tabSize); end > 0 {
 525			beg += end
 526		} else { // skip to the next line
 527			end = beg
 528			for end < len(input) && input[end] != '\n' && input[end] != '\r' {
 529				end++
 530			}
 531
 532			if p.flags&FencedCode != 0 {
 533				// track fenced code block boundaries to suppress tab expansion
 534				// inside them:
 535				if beg >= lastFencedCodeBlockEnd {
 536					if i := p.fencedCode(input[beg:], false); i > 0 {
 537						lastFencedCodeBlockEnd = beg + i
 538					}
 539				}
 540			}
 541
 542			// add the line body if present
 543			if end > beg {
 544				if end < lastFencedCodeBlockEnd { // Do not expand tabs while inside fenced code blocks.
 545					out.Write(input[beg:end])
 546				} else {
 547					expandTabs(&out, input[beg:end], tabSize)
 548				}
 549			}
 550			out.WriteByte('\n')
 551
 552			if end < len(input) && input[end] == '\r' {
 553				end++
 554			}
 555			if end < len(input) && input[end] == '\n' {
 556				end++
 557			}
 558
 559			beg = end
 560		}
 561	}
 562
 563	// empty input?
 564	if out.Len() == 0 {
 565		out.WriteByte('\n')
 566	}
 567
 568	return out.Bytes()
 569}
 570
 571// second pass: actual rendering
 572func secondPass(p *parser, input []byte) {
 573	p.block(input)
 574
 575	if p.flags&Footnotes != 0 && len(p.notes) > 0 {
 576		flags := ListItemBeginningOfList
 577		for i := 0; i < len(p.notes); i += 1 {
 578			ref := p.notes[i]
 579			if ref.hasBlock {
 580				flags |= ListItemContainsBlock
 581				p.block(ref.title)
 582			} else {
 583				p.inline(ref.title)
 584			}
 585			flags &^= ListItemBeginningOfList | ListItemContainsBlock
 586		}
 587	}
 588
 589	if p.nesting != 0 {
 590		panic("Nesting level did not end at zero")
 591	}
 592}
 593
 594//
 595// Link references
 596//
 597// This section implements support for references that (usually) appear
 598// as footnotes in a document, and can be referenced anywhere in the document.
 599// The basic format is:
 600//
 601//    [1]: http://www.google.com/ "Google"
 602//    [2]: http://www.github.com/ "Github"
 603//
 604// Anywhere in the document, the reference can be linked by referring to its
 605// label, i.e., 1 and 2 in this example, as in:
 606//
 607//    This library is hosted on [Github][2], a git hosting site.
 608//
 609// Actual footnotes as specified in Pandoc and supported by some other Markdown
 610// libraries such as php-markdown are also taken care of. They look like this:
 611//
 612//    This sentence needs a bit of further explanation.[^note]
 613//
 614//    [^note]: This is the explanation.
 615//
 616// Footnotes should be placed at the end of the document in an ordered list.
 617// Inline footnotes such as:
 618//
 619//    Inline footnotes^[Not supported.] also exist.
 620//
 621// are not yet supported.
 622
 623// References are parsed and stored in this struct.
 624type reference struct {
 625	link     []byte
 626	title    []byte
 627	noteId   int // 0 if not a footnote ref
 628	hasBlock bool
 629	text     []byte
 630}
 631
 632func (r *reference) String() string {
 633	return fmt.Sprintf("{link: %q, title: %q, text: %q, noteId: %d, hasBlock: %v}",
 634		r.link, r.title, r.text, r.noteId, r.hasBlock)
 635}
 636
 637// Check whether or not data starts with a reference link.
 638// If so, it is parsed and stored in the list of references
 639// (in the render struct).
 640// Returns the number of bytes to skip to move past it,
 641// or zero if the first line is not a reference.
 642func isReference(p *parser, data []byte, tabSize int) int {
 643	// up to 3 optional leading spaces
 644	if len(data) < 4 {
 645		return 0
 646	}
 647	i := 0
 648	for i < 3 && data[i] == ' ' {
 649		i++
 650	}
 651
 652	noteId := 0
 653
 654	// id part: anything but a newline between brackets
 655	if data[i] != '[' {
 656		return 0
 657	}
 658	i++
 659	if p.flags&Footnotes != 0 {
 660		if i < len(data) && data[i] == '^' {
 661			// we can set it to anything here because the proper noteIds will
 662			// be assigned later during the second pass. It just has to be != 0
 663			noteId = 1
 664			i++
 665		}
 666	}
 667	idOffset := i
 668	for i < len(data) && data[i] != '\n' && data[i] != '\r' && data[i] != ']' {
 669		i++
 670	}
 671	if i >= len(data) || data[i] != ']' {
 672		return 0
 673	}
 674	idEnd := i
 675
 676	// spacer: colon (space | tab)* newline? (space | tab)*
 677	i++
 678	if i >= len(data) || data[i] != ':' {
 679		return 0
 680	}
 681	i++
 682	for i < len(data) && (data[i] == ' ' || data[i] == '\t') {
 683		i++
 684	}
 685	if i < len(data) && (data[i] == '\n' || data[i] == '\r') {
 686		i++
 687		if i < len(data) && data[i] == '\n' && data[i-1] == '\r' {
 688			i++
 689		}
 690	}
 691	for i < len(data) && (data[i] == ' ' || data[i] == '\t') {
 692		i++
 693	}
 694	if i >= len(data) {
 695		return 0
 696	}
 697
 698	var (
 699		linkOffset, linkEnd   int
 700		titleOffset, titleEnd int
 701		lineEnd               int
 702		raw                   []byte
 703		hasBlock              bool
 704	)
 705
 706	if p.flags&Footnotes != 0 && noteId != 0 {
 707		linkOffset, linkEnd, raw, hasBlock = scanFootnote(p, data, i, tabSize)
 708		lineEnd = linkEnd
 709	} else {
 710		linkOffset, linkEnd, titleOffset, titleEnd, lineEnd = scanLinkRef(p, data, i)
 711	}
 712	if lineEnd == 0 {
 713		return 0
 714	}
 715
 716	// a valid ref has been found
 717
 718	ref := &reference{
 719		noteId:   noteId,
 720		hasBlock: hasBlock,
 721	}
 722
 723	if noteId > 0 {
 724		// reusing the link field for the id since footnotes don't have links
 725		ref.link = data[idOffset:idEnd]
 726		// if footnote, it's not really a title, it's the contained text
 727		ref.title = raw
 728	} else {
 729		ref.link = data[linkOffset:linkEnd]
 730		ref.title = data[titleOffset:titleEnd]
 731	}
 732
 733	// id matches are case-insensitive
 734	id := string(bytes.ToLower(data[idOffset:idEnd]))
 735
 736	p.refs[id] = ref
 737
 738	return lineEnd
 739}
 740
 741func scanLinkRef(p *parser, data []byte, i int) (linkOffset, linkEnd, titleOffset, titleEnd, lineEnd int) {
 742	// link: whitespace-free sequence, optionally between angle brackets
 743	if data[i] == '<' {
 744		i++
 745	}
 746	linkOffset = i
 747	for i < len(data) && data[i] != ' ' && data[i] != '\t' && data[i] != '\n' && data[i] != '\r' {
 748		i++
 749	}
 750	if i == len(data) {
 751		return
 752	}
 753	linkEnd = i
 754	if data[linkOffset] == '<' && data[linkEnd-1] == '>' {
 755		linkOffset++
 756		linkEnd--
 757	}
 758
 759	// optional spacer: (space | tab)* (newline | '\'' | '"' | '(' )
 760	for i < len(data) && (data[i] == ' ' || data[i] == '\t') {
 761		i++
 762	}
 763	if i < len(data) && data[i] != '\n' && data[i] != '\r' && data[i] != '\'' && data[i] != '"' && data[i] != '(' {
 764		return
 765	}
 766
 767	// compute end-of-line
 768	if i >= len(data) || data[i] == '\r' || data[i] == '\n' {
 769		lineEnd = i
 770	}
 771	if i+1 < len(data) && data[i] == '\r' && data[i+1] == '\n' {
 772		lineEnd++
 773	}
 774
 775	// optional (space|tab)* spacer after a newline
 776	if lineEnd > 0 {
 777		i = lineEnd + 1
 778		for i < len(data) && (data[i] == ' ' || data[i] == '\t') {
 779			i++
 780		}
 781	}
 782
 783	// optional title: any non-newline sequence enclosed in '"() alone on its line
 784	if i+1 < len(data) && (data[i] == '\'' || data[i] == '"' || data[i] == '(') {
 785		i++
 786		titleOffset = i
 787
 788		// look for EOL
 789		for i < len(data) && data[i] != '\n' && data[i] != '\r' {
 790			i++
 791		}
 792		if i+1 < len(data) && data[i] == '\n' && data[i+1] == '\r' {
 793			titleEnd = i + 1
 794		} else {
 795			titleEnd = i
 796		}
 797
 798		// step back
 799		i--
 800		for i > titleOffset && (data[i] == ' ' || data[i] == '\t') {
 801			i--
 802		}
 803		if i > titleOffset && (data[i] == '\'' || data[i] == '"' || data[i] == ')') {
 804			lineEnd = titleEnd
 805			titleEnd = i
 806		}
 807	}
 808
 809	return
 810}
 811
 812// The first bit of this logic is the same as (*parser).listItem, but the rest
 813// is much simpler. This function simply finds the entire block and shifts it
 814// over by one tab if it is indeed a block (just returns the line if it's not).
 815// blockEnd is the end of the section in the input buffer, and contents is the
 816// extracted text that was shifted over one tab. It will need to be rendered at
 817// the end of the document.
 818func scanFootnote(p *parser, data []byte, i, indentSize int) (blockStart, blockEnd int, contents []byte, hasBlock bool) {
 819	if i == 0 || len(data) == 0 {
 820		return
 821	}
 822
 823	// skip leading whitespace on first line
 824	for i < len(data) && data[i] == ' ' {
 825		i++
 826	}
 827
 828	blockStart = i
 829
 830	// find the end of the line
 831	blockEnd = i
 832	for i < len(data) && data[i-1] != '\n' {
 833		i++
 834	}
 835
 836	// get working buffer
 837	var raw bytes.Buffer
 838
 839	// put the first line into the working buffer
 840	raw.Write(data[blockEnd:i])
 841	blockEnd = i
 842
 843	// process the following lines
 844	containsBlankLine := false
 845
 846gatherLines:
 847	for blockEnd < len(data) {
 848		i++
 849
 850		// find the end of this line
 851		for i < len(data) && data[i-1] != '\n' {
 852			i++
 853		}
 854
 855		// if it is an empty line, guess that it is part of this item
 856		// and move on to the next line
 857		if p.isEmpty(data[blockEnd:i]) > 0 {
 858			containsBlankLine = true
 859			blockEnd = i
 860			continue
 861		}
 862
 863		n := 0
 864		if n = isIndented(data[blockEnd:i], indentSize); n == 0 {
 865			// this is the end of the block.
 866			// we don't want to include this last line in the index.
 867			break gatherLines
 868		}
 869
 870		// if there were blank lines before this one, insert a new one now
 871		if containsBlankLine {
 872			raw.WriteByte('\n')
 873			containsBlankLine = false
 874		}
 875
 876		// get rid of that first tab, write to buffer
 877		raw.Write(data[blockEnd+n : i])
 878		hasBlock = true
 879
 880		blockEnd = i
 881	}
 882
 883	if data[blockEnd-1] != '\n' {
 884		raw.WriteByte('\n')
 885	}
 886
 887	contents = raw.Bytes()
 888
 889	return
 890}
 891
 892//
 893//
 894// Miscellaneous helper functions
 895//
 896//
 897
 898// Test if a character is a punctuation symbol.
 899// Taken from a private function in regexp in the stdlib.
 900func ispunct(c byte) bool {
 901	for _, r := range []byte("!\"#$%&'()*+,-./:;<=>?@[\\]^_`{|}~") {
 902		if c == r {
 903			return true
 904		}
 905	}
 906	return false
 907}
 908
 909// Test if a character is a whitespace character.
 910func isspace(c byte) bool {
 911	return c == ' ' || c == '\t' || c == '\n' || c == '\r' || c == '\f' || c == '\v'
 912}
 913
 914// Test if a character is letter.
 915func isletter(c byte) bool {
 916	return (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z')
 917}
 918
 919// Test if a character is a letter or a digit.
 920// TODO: check when this is looking for ASCII alnum and when it should use unicode
 921func isalnum(c byte) bool {
 922	return (c >= '0' && c <= '9') || isletter(c)
 923}
 924
 925// Replace tab characters with spaces, aligning to the next TAB_SIZE column.
 926// always ends output with a newline
 927func expandTabs(out *bytes.Buffer, line []byte, tabSize int) {
 928	// first, check for common cases: no tabs, or only tabs at beginning of line
 929	i, prefix := 0, 0
 930	slowcase := false
 931	for i = 0; i < len(line); i++ {
 932		if line[i] == '\t' {
 933			if prefix == i {
 934				prefix++
 935			} else {
 936				slowcase = true
 937				break
 938			}
 939		}
 940	}
 941
 942	// no need to decode runes if all tabs are at the beginning of the line
 943	if !slowcase {
 944		for i = 0; i < prefix*tabSize; i++ {
 945			out.WriteByte(' ')
 946		}
 947		out.Write(line[prefix:])
 948		return
 949	}
 950
 951	// the slow case: we need to count runes to figure out how
 952	// many spaces to insert for each tab
 953	column := 0
 954	i = 0
 955	for i < len(line) {
 956		start := i
 957		for i < len(line) && line[i] != '\t' {
 958			_, size := utf8.DecodeRune(line[i:])
 959			i += size
 960			column++
 961		}
 962
 963		if i > start {
 964			out.Write(line[start:i])
 965		}
 966
 967		if i >= len(line) {
 968			break
 969		}
 970
 971		for {
 972			out.WriteByte(' ')
 973			column++
 974			if column%tabSize == 0 {
 975				break
 976			}
 977		}
 978
 979		i++
 980	}
 981}
 982
 983// Find if a line counts as indented or not.
 984// Returns number of characters the indent is (0 = not indented).
 985func isIndented(data []byte, indentSize int) int {
 986	if len(data) == 0 {
 987		return 0
 988	}
 989	if data[0] == '\t' {
 990		return 1
 991	}
 992	if len(data) < indentSize {
 993		return 0
 994	}
 995	for i := 0; i < indentSize; i++ {
 996		if data[i] != ' ' {
 997			return 0
 998		}
 999	}
1000	return indentSize
1001}
1002
1003// Create a url-safe slug for fragments
1004func slugify(in []byte) []byte {
1005	if len(in) == 0 {
1006		return in
1007	}
1008	out := make([]byte, 0, len(in))
1009	sym := false
1010
1011	for _, ch := range in {
1012		if isalnum(ch) {
1013			sym = false
1014			out = append(out, ch)
1015		} else if sym {
1016			continue
1017		} else {
1018			out = append(out, '-')
1019			sym = true
1020		}
1021	}
1022	var a, b int
1023	var ch byte
1024	for a, ch = range out {
1025		if ch != '-' {
1026			break
1027		}
1028	}
1029	for b = len(out) - 1; b > 0; b-- {
1030		if out[b] != '-' {
1031			break
1032		}
1033	}
1034	return out[a : b+1]
1035}