all repos — grayfriday @ 4b26653fe00067c22322249d58654737e44b0dfb

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