all repos — grayfriday @ 4bed88b4fd00fbb66b49b0f38ed3dd0b902ab515

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	"strings"
 24	"unicode/utf8"
 25)
 26
 27const VERSION = "1.1"
 28
 29// These are the supported markdown parsing extensions.
 30// OR these values together to select multiple extensions.
 31const (
 32	EXTENSION_NO_INTRA_EMPHASIS          = 1 << iota // ignore emphasis markers inside words
 33	EXTENSION_TABLES                                 // render tables
 34	EXTENSION_FENCED_CODE                            // render fenced code blocks
 35	EXTENSION_AUTOLINK                               // detect embedded URLs that are not explicitly marked
 36	EXTENSION_STRIKETHROUGH                          // strikethrough text using ~~test~~
 37	EXTENSION_LAX_HTML_BLOCKS                        // loosen up HTML block parsing rules
 38	EXTENSION_SPACE_HEADERS                          // be strict about prefix header rules
 39	EXTENSION_HARD_LINE_BREAK                        // translate newlines into line breaks
 40	EXTENSION_TAB_SIZE_EIGHT                         // expand tabs to eight spaces instead of four
 41	EXTENSION_FOOTNOTES                              // Pandoc-style footnotes
 42	EXTENSION_NO_EMPTY_LINE_BEFORE_BLOCK             // No need to insert an empty line to start a (code, quote, order list, unorder list)block
 43	EXTENSION_HEADER_IDS                             // specify header IDs  with {#id}
 44	EXTENSION_TITLEBLOCK                             // Titleblock ala pandoc
 45	EXTENSION_AUTO_HEADER_IDS                        // Create the header ID from the text
 46	EXTENSION_BACKSLASH_LINE_BREAK                   // translate trailing backslashes into line breaks
 47
 48	commonHtmlFlags = 0 |
 49		HTML_USE_XHTML |
 50		HTML_USE_SMARTYPANTS |
 51		HTML_SMARTYPANTS_FRACTIONS |
 52		HTML_SMARTYPANTS_LATEX_DASHES
 53
 54	commonExtensions = 0 |
 55		EXTENSION_NO_INTRA_EMPHASIS |
 56		EXTENSION_TABLES |
 57		EXTENSION_FENCED_CODE |
 58		EXTENSION_AUTOLINK |
 59		EXTENSION_STRIKETHROUGH |
 60		EXTENSION_SPACE_HEADERS |
 61		EXTENSION_HEADER_IDS |
 62		EXTENSION_BACKSLASH_LINE_BREAK
 63)
 64
 65// These are the possible flag values for the link renderer.
 66// Only a single one of these values will be used; they are not ORed together.
 67// These are mostly of interest if you are writing a new output format.
 68const (
 69	LINK_TYPE_NOT_AUTOLINK = iota
 70	LINK_TYPE_NORMAL
 71	LINK_TYPE_EMAIL
 72)
 73
 74// These are the possible flag values for the ListItem renderer.
 75// Multiple flag values may be ORed together.
 76// These are mostly of interest if you are writing a new output format.
 77const (
 78	LIST_TYPE_ORDERED = 1 << iota
 79	LIST_ITEM_CONTAINS_BLOCK
 80	LIST_ITEM_BEGINNING_OF_LIST
 81	LIST_ITEM_END_OF_LIST
 82)
 83
 84// These are the possible flag values for the table cell renderer.
 85// Only a single one of these values will be used; they are not ORed together.
 86// These are mostly of interest if you are writing a new output format.
 87const (
 88	TABLE_ALIGNMENT_LEFT = 1 << iota
 89	TABLE_ALIGNMENT_RIGHT
 90	TABLE_ALIGNMENT_CENTER = (TABLE_ALIGNMENT_LEFT | TABLE_ALIGNMENT_RIGHT)
 91)
 92
 93// The size of a tab stop.
 94const (
 95	TAB_SIZE_DEFAULT = 4
 96	TAB_SIZE_EIGHT   = 8
 97)
 98
 99// These are the tags that are recognized as HTML block tags.
100// Any of these can be included in markdown text without special escaping.
101var blockTags = map[string]bool{
102	"p":          true,
103	"dl":         true,
104	"h1":         true,
105	"h2":         true,
106	"h3":         true,
107	"h4":         true,
108	"h5":         true,
109	"h6":         true,
110	"ol":         true,
111	"ul":         true,
112	"del":        true,
113	"div":        true,
114	"ins":        true,
115	"pre":        true,
116	"form":       true,
117	"math":       true,
118	"table":      true,
119	"iframe":     true,
120	"script":     true,
121	"fieldset":   true,
122	"noscript":   true,
123	"blockquote": true,
124
125	// HTML5
126	"video":      true,
127	"aside":      true,
128	"canvas":     true,
129	"figure":     true,
130	"footer":     true,
131	"header":     true,
132	"hgroup":     true,
133	"output":     true,
134	"article":    true,
135	"section":    true,
136	"progress":   true,
137	"figcaption": true,
138}
139
140// Renderer is the rendering interface.
141// This is mostly of interest if you are implementing a new rendering format.
142//
143// When a byte slice is provided, it contains the (rendered) contents of the
144// element.
145//
146// When a callback is provided instead, it will write the contents of the
147// respective element directly to the output buffer and return true on success.
148// If the callback returns false, the rendering function should reset the
149// output buffer as though it had never been called.
150//
151// Currently Html and Latex implementations are provided
152type Renderer interface {
153	// block-level callbacks
154	BlockCode(out *bytes.Buffer, text []byte, lang string)
155	BlockQuote(out *bytes.Buffer, text []byte)
156	BlockHtml(out *bytes.Buffer, text []byte)
157	Header(out *bytes.Buffer, text func() bool, level int, id string)
158	HRule(out *bytes.Buffer)
159	List(out *bytes.Buffer, text func() bool, flags int)
160	ListItem(out *bytes.Buffer, text []byte, flags int)
161	Paragraph(out *bytes.Buffer, text func() bool)
162	Table(out *bytes.Buffer, header []byte, body []byte, columnData []int)
163	TableRow(out *bytes.Buffer, text []byte)
164	TableHeaderCell(out *bytes.Buffer, text []byte, flags int)
165	TableCell(out *bytes.Buffer, text []byte, flags int)
166	Footnotes(out *bytes.Buffer, text func() bool)
167	FootnoteItem(out *bytes.Buffer, name, text []byte, flags int)
168	TitleBlock(out *bytes.Buffer, text []byte)
169
170	// Span-level callbacks
171	AutoLink(out *bytes.Buffer, link []byte, kind int)
172	CodeSpan(out *bytes.Buffer, text []byte)
173	DoubleEmphasis(out *bytes.Buffer, text []byte)
174	Emphasis(out *bytes.Buffer, text []byte)
175	Image(out *bytes.Buffer, link []byte, title []byte, alt []byte)
176	LineBreak(out *bytes.Buffer)
177	Link(out *bytes.Buffer, link []byte, title []byte, content []byte)
178	RawHtmlTag(out *bytes.Buffer, tag []byte)
179	TripleEmphasis(out *bytes.Buffer, text []byte)
180	StrikeThrough(out *bytes.Buffer, text []byte)
181	FootnoteRef(out *bytes.Buffer, ref []byte, id int)
182
183	// Low-level callbacks
184	Entity(out *bytes.Buffer, entity []byte)
185	NormalText(out *bytes.Buffer, text []byte)
186
187	// Header and footer
188	DocumentHeader(out *bytes.Buffer)
189	DocumentFooter(out *bytes.Buffer)
190
191	GetFlags() int
192}
193
194// Callback functions for inline parsing. One such function is defined
195// for each character that triggers a response when parsing inline data.
196type inlineParser func(p *parser, out *bytes.Buffer, data []byte, offset int) int
197
198// Parser holds runtime state used by the parser.
199// This is constructed by the Markdown function.
200type parser struct {
201	r              Renderer
202	refOverride    ReferenceOverrideFunc
203	refs           map[string]*reference
204	inlineCallback [256]inlineParser
205	flags          int
206	nesting        int
207	maxNesting     int
208	insideLink     bool
209
210	// Footnotes need to be ordered as well as available to quickly check for
211	// presence. If a ref is also a footnote, it's stored both in refs and here
212	// in notes. Slice is nil if footnotes not enabled.
213	notes []*reference
214}
215
216func (p *parser) getRef(refid string) (ref *reference, found bool) {
217	if p.refOverride != nil {
218		r, overridden := p.refOverride(refid)
219		if overridden {
220			if r == nil {
221				return nil, false
222			}
223			return &reference{
224				link:     []byte(r.Link),
225				title:    []byte(r.Title),
226				noteId:   0,
227				hasBlock: false,
228				text:     []byte(r.Text)}, true
229		}
230	}
231	// refs are case insensitive
232	ref, found = p.refs[strings.ToLower(refid)]
233	return ref, found
234}
235
236//
237//
238// Public interface
239//
240//
241
242// Reference represents the details of a link.
243// See the documentation in Options for more details on use-case.
244type Reference struct {
245	// Link is usually the URL the reference points to.
246	Link string
247	// Title is the alternate text describing the link in more detail.
248	Title string
249	// Text is the optional text to override the ref with if the syntax used was
250	// [refid][]
251	Text string
252}
253
254// ReferenceOverrideFunc is expected to be called with a reference string and
255// return either a valid Reference type that the reference string maps to or
256// nil. If overridden is false, the default reference logic will be executed.
257// See the documentation in Options for more details on use-case.
258type ReferenceOverrideFunc func(reference string) (ref *Reference, overridden bool)
259
260// Options represents configurable overrides and callbacks (in addition to the
261// extension flag set) for configuring a Markdown parse.
262type Options struct {
263	// Extensions is a flag set of bit-wise ORed extension bits. See the
264	// EXTENSION_* flags defined in this package.
265	Extensions int
266
267	// ReferenceOverride is an optional function callback that is called every
268	// time a reference is resolved.
269	//
270	// In Markdown, the link reference syntax can be made to resolve a link to
271	// a reference instead of an inline URL, in one of the following ways:
272	//
273	//  * [link text][refid]
274	//  * [refid][]
275	//
276	// Usually, the refid is defined at the bottom of the Markdown document. If
277	// this override function is provided, the refid is passed to the override
278	// function first, before consulting the defined refids at the bottom. If
279	// the override function indicates an override did not occur, the refids at
280	// the bottom will be used to fill in the link details.
281	ReferenceOverride ReferenceOverrideFunc
282}
283
284// MarkdownBasic is a convenience function for simple rendering.
285// It processes markdown input with no extensions enabled.
286func MarkdownBasic(input []byte) []byte {
287	// set up the HTML renderer
288	htmlFlags := HTML_USE_XHTML
289	renderer := HtmlRenderer(htmlFlags, "", "")
290
291	// set up the parser
292	return MarkdownOptions(input, renderer, Options{Extensions: 0})
293}
294
295// Call Markdown with most useful extensions enabled
296// MarkdownCommon is a convenience function for simple rendering.
297// It processes markdown input with common extensions enabled, including:
298//
299// * Smartypants processing with smart fractions and LaTeX dashes
300//
301// * Intra-word emphasis suppression
302//
303// * Tables
304//
305// * Fenced code blocks
306//
307// * Autolinking
308//
309// * Strikethrough support
310//
311// * Strict header parsing
312//
313// * Custom Header IDs
314func MarkdownCommon(input []byte) []byte {
315	// set up the HTML renderer
316	renderer := HtmlRenderer(commonHtmlFlags, "", "")
317	return MarkdownOptions(input, renderer, Options{
318		Extensions: commonExtensions})
319}
320
321// Markdown is the main rendering function.
322// It parses and renders a block of markdown-encoded text.
323// The supplied Renderer is used to format the output, and extensions dictates
324// which non-standard extensions are enabled.
325//
326// To use the supplied Html or LaTeX renderers, see HtmlRenderer and
327// LatexRenderer, respectively.
328func Markdown(input []byte, renderer Renderer, extensions int) []byte {
329	return MarkdownOptions(input, renderer, Options{
330		Extensions: extensions})
331}
332
333// MarkdownOptions is just like Markdown but takes additional options through
334// the Options struct.
335func MarkdownOptions(input []byte, renderer Renderer, opts Options) []byte {
336	// no point in parsing if we can't render
337	if renderer == nil {
338		return nil
339	}
340
341	extensions := opts.Extensions
342
343	// fill in the render structure
344	p := new(parser)
345	p.r = renderer
346	p.flags = extensions
347	p.refOverride = opts.ReferenceOverride
348	p.refs = make(map[string]*reference)
349	p.maxNesting = 16
350	p.insideLink = false
351
352	// register inline parsers
353	p.inlineCallback['*'] = emphasis
354	p.inlineCallback['_'] = emphasis
355	if extensions&EXTENSION_STRIKETHROUGH != 0 {
356		p.inlineCallback['~'] = emphasis
357	}
358	p.inlineCallback['`'] = codeSpan
359	p.inlineCallback['\n'] = lineBreak
360	p.inlineCallback['['] = link
361	p.inlineCallback['<'] = leftAngle
362	p.inlineCallback['\\'] = escape
363	p.inlineCallback['&'] = entity
364
365	if extensions&EXTENSION_AUTOLINK != 0 {
366		p.inlineCallback[':'] = autoLink
367	}
368
369	if extensions&EXTENSION_FOOTNOTES != 0 {
370		p.notes = make([]*reference, 0)
371	}
372
373	first := firstPass(p, input)
374	second := secondPass(p, first)
375	return second
376}
377
378// first pass:
379// - extract references
380// - expand tabs
381// - normalize newlines
382// - copy everything else
383// - add missing newlines before fenced code blocks
384func firstPass(p *parser, input []byte) []byte {
385	var out bytes.Buffer
386	tabSize := TAB_SIZE_DEFAULT
387	if p.flags&EXTENSION_TAB_SIZE_EIGHT != 0 {
388		tabSize = TAB_SIZE_EIGHT
389	}
390	beg, end := 0, 0
391	lastLineWasBlank := false
392	lastFencedCodeBlockEnd := 0
393	for beg < len(input) { // iterate over lines
394		if end = isReference(p, input[beg:], tabSize); end > 0 {
395			beg += end
396		} else { // skip to the next line
397			end = beg
398			for end < len(input) && input[end] != '\n' && input[end] != '\r' {
399				end++
400			}
401
402			if p.flags&EXTENSION_FENCED_CODE != 0 {
403				// when last line was none blank and a fenced code block comes after
404				if beg >= lastFencedCodeBlockEnd {
405					if i := p.fencedCode(&out, input[beg:], false); i > 0 {
406						if !lastLineWasBlank {
407							out.WriteByte('\n') // need to inject additional linebreak
408						}
409						lastFencedCodeBlockEnd = beg + i
410					}
411				}
412				lastLineWasBlank = end == beg
413			}
414
415			// add the line body if present
416			if end > beg {
417				if end < lastFencedCodeBlockEnd { // Do not expand tabs while inside fenced code blocks.
418					out.Write(input[beg:end])
419				} else {
420					expandTabs(&out, input[beg:end], tabSize)
421				}
422			}
423			out.WriteByte('\n')
424
425			if end < len(input) && input[end] == '\r' {
426				end++
427			}
428			if end < len(input) && input[end] == '\n' {
429				end++
430			}
431
432			beg = end
433		}
434	}
435
436	// empty input?
437	if out.Len() == 0 {
438		out.WriteByte('\n')
439	}
440
441	return out.Bytes()
442}
443
444// second pass: actual rendering
445func secondPass(p *parser, input []byte) []byte {
446	var output bytes.Buffer
447
448	p.r.DocumentHeader(&output)
449	p.block(&output, input)
450
451	if p.flags&EXTENSION_FOOTNOTES != 0 && len(p.notes) > 0 {
452		p.r.Footnotes(&output, func() bool {
453			flags := LIST_ITEM_BEGINNING_OF_LIST
454			for _, ref := range p.notes {
455				var buf bytes.Buffer
456				if ref.hasBlock {
457					flags |= LIST_ITEM_CONTAINS_BLOCK
458					p.block(&buf, ref.title)
459				} else {
460					p.inline(&buf, ref.title)
461				}
462				p.r.FootnoteItem(&output, ref.link, buf.Bytes(), flags)
463				flags &^= LIST_ITEM_BEGINNING_OF_LIST | LIST_ITEM_CONTAINS_BLOCK
464			}
465
466			return true
467		})
468	}
469
470	p.r.DocumentFooter(&output)
471
472	if p.nesting != 0 {
473		panic("Nesting level did not end at zero")
474	}
475
476	return output.Bytes()
477}
478
479//
480// Link references
481//
482// This section implements support for references that (usually) appear
483// as footnotes in a document, and can be referenced anywhere in the document.
484// The basic format is:
485//
486//    [1]: http://www.google.com/ "Google"
487//    [2]: http://www.github.com/ "Github"
488//
489// Anywhere in the document, the reference can be linked by referring to its
490// label, i.e., 1 and 2 in this example, as in:
491//
492//    This library is hosted on [Github][2], a git hosting site.
493//
494// Actual footnotes as specified in Pandoc and supported by some other Markdown
495// libraries such as php-markdown are also taken care of. They look like this:
496//
497//    This sentence needs a bit of further explanation.[^note]
498//
499//    [^note]: This is the explanation.
500//
501// Footnotes should be placed at the end of the document in an ordered list.
502// Inline footnotes such as:
503//
504//    Inline footnotes^[Not supported.] also exist.
505//
506// are not yet supported.
507
508// References are parsed and stored in this struct.
509type reference struct {
510	link     []byte
511	title    []byte
512	noteId   int // 0 if not a footnote ref
513	hasBlock bool
514	text     []byte
515}
516
517// Check whether or not data starts with a reference link.
518// If so, it is parsed and stored in the list of references
519// (in the render struct).
520// Returns the number of bytes to skip to move past it,
521// or zero if the first line is not a reference.
522func isReference(p *parser, data []byte, tabSize int) int {
523	// up to 3 optional leading spaces
524	if len(data) < 4 {
525		return 0
526	}
527	i := 0
528	for i < 3 && data[i] == ' ' {
529		i++
530	}
531
532	noteId := 0
533
534	// id part: anything but a newline between brackets
535	if data[i] != '[' {
536		return 0
537	}
538	i++
539	if p.flags&EXTENSION_FOOTNOTES != 0 {
540		if data[i] == '^' {
541			// we can set it to anything here because the proper noteIds will
542			// be assigned later during the second pass. It just has to be != 0
543			noteId = 1
544			i++
545		}
546	}
547	idOffset := i
548	for i < len(data) && data[i] != '\n' && data[i] != '\r' && data[i] != ']' {
549		i++
550	}
551	if i >= len(data) || data[i] != ']' {
552		return 0
553	}
554	idEnd := i
555
556	// spacer: colon (space | tab)* newline? (space | tab)*
557	i++
558	if i >= len(data) || data[i] != ':' {
559		return 0
560	}
561	i++
562	for i < len(data) && (data[i] == ' ' || data[i] == '\t') {
563		i++
564	}
565	if i < len(data) && (data[i] == '\n' || data[i] == '\r') {
566		i++
567		if i < len(data) && data[i] == '\n' && data[i-1] == '\r' {
568			i++
569		}
570	}
571	for i < len(data) && (data[i] == ' ' || data[i] == '\t') {
572		i++
573	}
574	if i >= len(data) {
575		return 0
576	}
577
578	var (
579		linkOffset, linkEnd   int
580		titleOffset, titleEnd int
581		lineEnd               int
582		raw                   []byte
583		hasBlock              bool
584	)
585
586	if p.flags&EXTENSION_FOOTNOTES != 0 && noteId != 0 {
587		linkOffset, linkEnd, raw, hasBlock = scanFootnote(p, data, i, tabSize)
588		lineEnd = linkEnd
589	} else {
590		linkOffset, linkEnd, titleOffset, titleEnd, lineEnd = scanLinkRef(p, data, i)
591	}
592	if lineEnd == 0 {
593		return 0
594	}
595
596	// a valid ref has been found
597
598	ref := &reference{
599		noteId:   noteId,
600		hasBlock: hasBlock,
601	}
602
603	if noteId > 0 {
604		// reusing the link field for the id since footnotes don't have links
605		ref.link = data[idOffset:idEnd]
606		// if footnote, it's not really a title, it's the contained text
607		ref.title = raw
608	} else {
609		ref.link = data[linkOffset:linkEnd]
610		ref.title = data[titleOffset:titleEnd]
611	}
612
613	// id matches are case-insensitive
614	id := string(bytes.ToLower(data[idOffset:idEnd]))
615
616	p.refs[id] = ref
617
618	return lineEnd
619}
620
621func scanLinkRef(p *parser, data []byte, i int) (linkOffset, linkEnd, titleOffset, titleEnd, lineEnd int) {
622	// link: whitespace-free sequence, optionally between angle brackets
623	if data[i] == '<' {
624		i++
625	}
626	linkOffset = i
627	for i < len(data) && data[i] != ' ' && data[i] != '\t' && data[i] != '\n' && data[i] != '\r' {
628		i++
629	}
630	linkEnd = i
631	if data[linkOffset] == '<' && data[linkEnd-1] == '>' {
632		linkOffset++
633		linkEnd--
634	}
635
636	// optional spacer: (space | tab)* (newline | '\'' | '"' | '(' )
637	for i < len(data) && (data[i] == ' ' || data[i] == '\t') {
638		i++
639	}
640	if i < len(data) && data[i] != '\n' && data[i] != '\r' && data[i] != '\'' && data[i] != '"' && data[i] != '(' {
641		return
642	}
643
644	// compute end-of-line
645	if i >= len(data) || data[i] == '\r' || data[i] == '\n' {
646		lineEnd = i
647	}
648	if i+1 < len(data) && data[i] == '\r' && data[i+1] == '\n' {
649		lineEnd++
650	}
651
652	// optional (space|tab)* spacer after a newline
653	if lineEnd > 0 {
654		i = lineEnd + 1
655		for i < len(data) && (data[i] == ' ' || data[i] == '\t') {
656			i++
657		}
658	}
659
660	// optional title: any non-newline sequence enclosed in '"() alone on its line
661	if i+1 < len(data) && (data[i] == '\'' || data[i] == '"' || data[i] == '(') {
662		i++
663		titleOffset = i
664
665		// look for EOL
666		for i < len(data) && data[i] != '\n' && data[i] != '\r' {
667			i++
668		}
669		if i+1 < len(data) && data[i] == '\n' && data[i+1] == '\r' {
670			titleEnd = i + 1
671		} else {
672			titleEnd = i
673		}
674
675		// step back
676		i--
677		for i > titleOffset && (data[i] == ' ' || data[i] == '\t') {
678			i--
679		}
680		if i > titleOffset && (data[i] == '\'' || data[i] == '"' || data[i] == ')') {
681			lineEnd = titleEnd
682			titleEnd = i
683		}
684	}
685
686	return
687}
688
689// The first bit of this logic is the same as (*parser).listItem, but the rest
690// is much simpler. This function simply finds the entire block and shifts it
691// over by one tab if it is indeed a block (just returns the line if it's not).
692// blockEnd is the end of the section in the input buffer, and contents is the
693// extracted text that was shifted over one tab. It will need to be rendered at
694// the end of the document.
695func scanFootnote(p *parser, data []byte, i, indentSize int) (blockStart, blockEnd int, contents []byte, hasBlock bool) {
696	if i == 0 || len(data) == 0 {
697		return
698	}
699
700	// skip leading whitespace on first line
701	for i < len(data) && data[i] == ' ' {
702		i++
703	}
704
705	blockStart = i
706
707	// find the end of the line
708	blockEnd = i
709	for i < len(data) && data[i-1] != '\n' {
710		i++
711	}
712
713	// get working buffer
714	var raw bytes.Buffer
715
716	// put the first line into the working buffer
717	raw.Write(data[blockEnd:i])
718	blockEnd = i
719
720	// process the following lines
721	containsBlankLine := false
722
723gatherLines:
724	for blockEnd < len(data) {
725		i++
726
727		// find the end of this line
728		for i < len(data) && data[i-1] != '\n' {
729			i++
730		}
731
732		// if it is an empty line, guess that it is part of this item
733		// and move on to the next line
734		if p.isEmpty(data[blockEnd:i]) > 0 {
735			containsBlankLine = true
736			blockEnd = i
737			continue
738		}
739
740		n := 0
741		if n = isIndented(data[blockEnd:i], indentSize); n == 0 {
742			// this is the end of the block.
743			// we don't want to include this last line in the index.
744			break gatherLines
745		}
746
747		// if there were blank lines before this one, insert a new one now
748		if containsBlankLine {
749			raw.WriteByte('\n')
750			containsBlankLine = false
751		}
752
753		// get rid of that first tab, write to buffer
754		raw.Write(data[blockEnd+n : i])
755		hasBlock = true
756
757		blockEnd = i
758	}
759
760	if data[blockEnd-1] != '\n' {
761		raw.WriteByte('\n')
762	}
763
764	contents = raw.Bytes()
765
766	return
767}
768
769//
770//
771// Miscellaneous helper functions
772//
773//
774
775// Test if a character is a punctuation symbol.
776// Taken from a private function in regexp in the stdlib.
777func ispunct(c byte) bool {
778	for _, r := range []byte("!\"#$%&'()*+,-./:;<=>?@[\\]^_`{|}~") {
779		if c == r {
780			return true
781		}
782	}
783	return false
784}
785
786// Test if a character is a whitespace character.
787func isspace(c byte) bool {
788	return c == ' ' || c == '\t' || c == '\n' || c == '\r' || c == '\f' || c == '\v'
789}
790
791// Test if a character is letter.
792func isletter(c byte) bool {
793	return (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z')
794}
795
796// Test if a character is a letter or a digit.
797// TODO: check when this is looking for ASCII alnum and when it should use unicode
798func isalnum(c byte) bool {
799	return (c >= '0' && c <= '9') || isletter(c)
800}
801
802// Replace tab characters with spaces, aligning to the next TAB_SIZE column.
803// always ends output with a newline
804func expandTabs(out *bytes.Buffer, line []byte, tabSize int) {
805	// first, check for common cases: no tabs, or only tabs at beginning of line
806	i, prefix := 0, 0
807	slowcase := false
808	for i = 0; i < len(line); i++ {
809		if line[i] == '\t' {
810			if prefix == i {
811				prefix++
812			} else {
813				slowcase = true
814				break
815			}
816		}
817	}
818
819	// no need to decode runes if all tabs are at the beginning of the line
820	if !slowcase {
821		for i = 0; i < prefix*tabSize; i++ {
822			out.WriteByte(' ')
823		}
824		out.Write(line[prefix:])
825		return
826	}
827
828	// the slow case: we need to count runes to figure out how
829	// many spaces to insert for each tab
830	column := 0
831	i = 0
832	for i < len(line) {
833		start := i
834		for i < len(line) && line[i] != '\t' {
835			_, size := utf8.DecodeRune(line[i:])
836			i += size
837			column++
838		}
839
840		if i > start {
841			out.Write(line[start:i])
842		}
843
844		if i >= len(line) {
845			break
846		}
847
848		for {
849			out.WriteByte(' ')
850			column++
851			if column%tabSize == 0 {
852				break
853			}
854		}
855
856		i++
857	}
858}
859
860// Find if a line counts as indented or not.
861// Returns number of characters the indent is (0 = not indented).
862func isIndented(data []byte, indentSize int) int {
863	if len(data) == 0 {
864		return 0
865	}
866	if data[0] == '\t' {
867		return 1
868	}
869	if len(data) < indentSize {
870		return 0
871	}
872	for i := 0; i < indentSize; i++ {
873		if data[i] != ' ' {
874			return 0
875		}
876	}
877	return indentSize
878}
879
880// Create a url-safe slug for fragments
881func slugify(in []byte) []byte {
882	if len(in) == 0 {
883		return in
884	}
885	out := make([]byte, 0, len(in))
886	sym := false
887
888	for _, ch := range in {
889		if isalnum(ch) {
890			sym = false
891			out = append(out, ch)
892		} else if sym {
893			continue
894		} else {
895			out = append(out, '-')
896			sym = true
897		}
898	}
899	var a, b int
900	var ch byte
901	for a, ch = range out {
902		if ch != '-' {
903			break
904		}
905	}
906	for b = len(out) - 1; b > 0; b-- {
907		if out[b] != '-' {
908			break
909		}
910	}
911	return out[a : b+1]
912}