all repos — grayfriday @ 7c95b7a1891c77d04a0f9571a455ba4d229f2d1a

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