forked from gitea/gitea
1
0
Fork 0
gitea/vendor/github.com/niklasfasching/go-org/org/inline.go

358 lines
11 KiB
Go
Raw Normal View History

package org
import (
"fmt"
"path"
"regexp"
"strings"
"time"
"unicode"
)
type Text struct {
Content string
IsRaw bool
}
type LineBreak struct{ Count int }
type ExplicitLineBreak struct{}
type StatisticToken struct{ Content string }
type Timestamp struct {
Time time.Time
IsDate bool
Interval string
}
type Emphasis struct {
Kind string
Content []Node
}
type LatexFragment struct {
OpeningPair string
ClosingPair string
Content []Node
}
type FootnoteLink struct {
Name string
Definition *FootnoteDefinition
}
type RegularLink struct {
Protocol string
Description []Node
URL string
AutoLink bool
}
var validURLCharacters = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-._~:/?#[]@!$&'()*+,;="
var autolinkProtocols = regexp.MustCompile(`^(https?|ftp|file)$`)
var imageExtensionRegexp = regexp.MustCompile(`^[.](png|gif|jpe?g|svg|tiff?)$`)
var videoExtensionRegexp = regexp.MustCompile(`^[.](webm|mp4)$`)
var subScriptSuperScriptRegexp = regexp.MustCompile(`^([_^]){([^{}]+?)}`)
var timestampRegexp = regexp.MustCompile(`^<(\d{4}-\d{2}-\d{2})( [A-Za-z]+)?( \d{2}:\d{2})?( \+\d+[dwmy])?>`)
var footnoteRegexp = regexp.MustCompile(`^\[fn:([\w-]*?)(:(.*?))?\]`)
var statisticsTokenRegexp = regexp.MustCompile(`^\[(\d+/\d+|\d+%)\]`)
var latexFragmentRegexp = regexp.MustCompile(`(?s)^\\begin{(\w+)}(.*)\\end{(\w+)}`)
var timestampFormat = "2006-01-02 Mon 15:04"
var datestampFormat = "2006-01-02 Mon"
var latexFragmentPairs = map[string]string{
`\(`: `\)`,
`\[`: `\]`,
`$$`: `$$`,
}
func (d *Document) parseInline(input string) (nodes []Node) {
previous, current := 0, 0
for current < len(input) {
rewind, consumed, node := 0, 0, (Node)(nil)
switch input[current] {
case '^':
consumed, node = d.parseSubOrSuperScript(input, current)
case '_':
consumed, node = d.parseSubScriptOrEmphasis(input, current)
case '*', '/', '+':
consumed, node = d.parseEmphasis(input, current, false)
case '=', '~':
consumed, node = d.parseEmphasis(input, current, true)
case '[':
consumed, node = d.parseOpeningBracket(input, current)
case '<':
consumed, node = d.parseTimestamp(input, current)
case '\\':
consumed, node = d.parseExplicitLineBreakOrLatexFragment(input, current)
case '$':
consumed, node = d.parseLatexFragment(input, current)
case '\n':
consumed, node = d.parseLineBreak(input, current)
case ':':
rewind, consumed, node = d.parseAutoLink(input, current)
current -= rewind
}
if consumed != 0 {
if current > previous {
nodes = append(nodes, Text{input[previous:current], false})
}
if node != nil {
nodes = append(nodes, node)
}
current += consumed
previous = current
} else {
current++
}
}
if previous < len(input) {
nodes = append(nodes, Text{input[previous:], false})
}
return nodes
}
func (d *Document) parseRawInline(input string) (nodes []Node) {
previous, current := 0, 0
for current < len(input) {
if input[current] == '\n' {
consumed, node := d.parseLineBreak(input, current)
if current > previous {
nodes = append(nodes, Text{input[previous:current], true})
}
nodes = append(nodes, node)
current += consumed
previous = current
} else {
current++
}
}
if previous < len(input) {
nodes = append(nodes, Text{input[previous:], true})
}
return nodes
}
func (d *Document) parseLineBreak(input string, start int) (int, Node) {
i := start
for ; i < len(input) && input[i] == '\n'; i++ {
}
return i - start, LineBreak{i - start}
}
func (d *Document) parseExplicitLineBreakOrLatexFragment(input string, start int) (int, Node) {
switch {
case start+2 >= len(input):
case input[start+1] == '\\' && start != 0 && input[start-1] != '\n':
2019-11-05 16:39:03 +08:00
for i := start + 2; i <= len(input)-1 && unicode.IsSpace(rune(input[i])); i++ {
if input[i] == '\n' {
return i + 1 - start, ExplicitLineBreak{}
}
}
case input[start+1] == '(' || input[start+1] == '[':
return d.parseLatexFragment(input, start)
case strings.Index(input[start:], `\begin{`) == 0:
if m := latexFragmentRegexp.FindStringSubmatch(input[start:]); m != nil {
if open, content, close := m[1], m[2], m[3]; open == close {
openingPair, closingPair := `\begin{`+open+`}`, `\end{`+close+`}`
i := strings.Index(input[start:], closingPair)
return i + len(closingPair), LatexFragment{openingPair, closingPair, d.parseRawInline(content)}
}
}
}
return 0, nil
}
func (d *Document) parseLatexFragment(input string, start int) (int, Node) {
if start+2 >= len(input) {
return 0, nil
}
openingPair := input[start : start+2]
closingPair := latexFragmentPairs[openingPair]
if i := strings.Index(input[start+2:], closingPair); i != -1 {
content := d.parseRawInline(input[start+2 : start+2+i])
return i + 2 + 2, LatexFragment{openingPair, closingPair, content}
}
return 0, nil
}
func (d *Document) parseSubOrSuperScript(input string, start int) (int, Node) {
if m := subScriptSuperScriptRegexp.FindStringSubmatch(input[start:]); m != nil {
return len(m[2]) + 3, Emphasis{m[1] + "{}", []Node{Text{m[2], false}}}
}
return 0, nil
}
func (d *Document) parseSubScriptOrEmphasis(input string, start int) (int, Node) {
if consumed, node := d.parseSubOrSuperScript(input, start); consumed != 0 {
return consumed, node
}
return d.parseEmphasis(input, start, false)
}
func (d *Document) parseOpeningBracket(input string, start int) (int, Node) {
if len(input[start:]) >= 2 && input[start] == '[' && input[start+1] == '[' {
return d.parseRegularLink(input, start)
} else if footnoteRegexp.MatchString(input[start:]) {
return d.parseFootnoteReference(input, start)
} else if statisticsTokenRegexp.MatchString(input[start:]) {
return d.parseStatisticToken(input, start)
}
return 0, nil
}
func (d *Document) parseFootnoteReference(input string, start int) (int, Node) {
if m := footnoteRegexp.FindStringSubmatch(input[start:]); m != nil {
name, definition := m[1], m[3]
if name == "" && definition == "" {
return 0, nil
}
link := FootnoteLink{name, nil}
if definition != "" {
link.Definition = &FootnoteDefinition{name, []Node{Paragraph{d.parseInline(definition)}}, true}
}
return len(m[0]), link
}
return 0, nil
}
func (d *Document) parseStatisticToken(input string, start int) (int, Node) {
if m := statisticsTokenRegexp.FindStringSubmatch(input[start:]); m != nil {
return len(m[1]) + 2, StatisticToken{m[1]}
}
return 0, nil
}
func (d *Document) parseAutoLink(input string, start int) (int, int, Node) {
if !d.AutoLink || start == 0 || len(input[start:]) < 3 || input[start:start+3] != "://" {
return 0, 0, nil
}
protocolStart, protocol := start-1, ""
for ; protocolStart > 0; protocolStart-- {
if !unicode.IsLetter(rune(input[protocolStart])) {
protocolStart++
break
}
}
if m := autolinkProtocols.FindStringSubmatch(input[protocolStart:start]); m != nil {
protocol = m[1]
} else {
return 0, 0, nil
}
end := start
for ; end < len(input) && strings.ContainsRune(validURLCharacters, rune(input[end])); end++ {
}
path := input[start:end]
if path == "://" {
return 0, 0, nil
}
return len(protocol), len(path + protocol), RegularLink{protocol, nil, protocol + path, true}
}
func (d *Document) parseRegularLink(input string, start int) (int, Node) {
input = input[start:]
if len(input) < 3 || input[:2] != "[[" || input[2] == '[' {
return 0, nil
}
end := strings.Index(input, "]]")
if end == -1 {
return 0, nil
}
rawLinkParts := strings.Split(input[2:end], "][")
description, link := ([]Node)(nil), rawLinkParts[0]
if len(rawLinkParts) == 2 {
link, description = rawLinkParts[0], d.parseInline(rawLinkParts[1])
}
if strings.ContainsRune(link, '\n') {
return 0, nil
}
consumed := end + 2
protocol, linkParts := "", strings.SplitN(link, ":", 2)
if len(linkParts) == 2 {
protocol = linkParts[0]
}
return consumed, RegularLink{protocol, description, link, false}
}
func (d *Document) parseTimestamp(input string, start int) (int, Node) {
if m := timestampRegexp.FindStringSubmatch(input[start:]); m != nil {
ddmmyy, hhmm, interval, isDate := m[1], m[3], strings.TrimSpace(m[4]), false
if hhmm == "" {
hhmm, isDate = "00:00", true
}
t, err := time.Parse(timestampFormat, fmt.Sprintf("%s Mon %s", ddmmyy, hhmm))
if err != nil {
return 0, nil
}
timestamp := Timestamp{t, isDate, interval}
return len(m[0]), timestamp
}
return 0, nil
}
func (d *Document) parseEmphasis(input string, start int, isRaw bool) (int, Node) {
marker, i := input[start], start
if !hasValidPreAndBorderChars(input, i) {
return 0, nil
}
for i, consumedNewLines := i+1, 0; i < len(input) && consumedNewLines <= d.MaxEmphasisNewLines; i++ {
if input[i] == '\n' {
consumedNewLines++
}
if input[i] == marker && i != start+1 && hasValidPostAndBorderChars(input, i) {
if isRaw {
return i + 1 - start, Emphasis{input[start : start+1], d.parseRawInline(input[start+1 : i])}
}
return i + 1 - start, Emphasis{input[start : start+1], d.parseInline(input[start+1 : i])}
}
}
return 0, nil
}
// see org-emphasis-regexp-components (emacs elisp variable)
func hasValidPreAndBorderChars(input string, i int) bool {
return (i+1 >= len(input) || isValidBorderChar(rune(input[i+1]))) && (i == 0 || isValidPreChar(rune(input[i-1])))
}
func hasValidPostAndBorderChars(input string, i int) bool {
return (i == 0 || isValidBorderChar(rune(input[i-1]))) && (i+1 >= len(input) || isValidPostChar(rune(input[i+1])))
}
func isValidPreChar(r rune) bool {
return unicode.IsSpace(r) || strings.ContainsRune(`-({'"`, r)
}
func isValidPostChar(r rune) bool {
return unicode.IsSpace(r) || strings.ContainsRune(`-.,:!?;'")}[`, r)
}
func isValidBorderChar(r rune) bool { return !unicode.IsSpace(r) }
func (l RegularLink) Kind() string {
if p := l.Protocol; l.Description != nil || (p != "" && p != "file" && p != "http" && p != "https") {
return "regular"
}
if imageExtensionRegexp.MatchString(path.Ext(l.URL)) {
return "image"
}
if videoExtensionRegexp.MatchString(path.Ext(l.URL)) {
return "video"
}
return "regular"
}
2019-11-05 16:39:03 +08:00
func (n Text) String() string { return orgWriter.WriteNodesAsString(n) }
func (n LineBreak) String() string { return orgWriter.WriteNodesAsString(n) }
func (n ExplicitLineBreak) String() string { return orgWriter.WriteNodesAsString(n) }
func (n StatisticToken) String() string { return orgWriter.WriteNodesAsString(n) }
func (n Emphasis) String() string { return orgWriter.WriteNodesAsString(n) }
func (n LatexFragment) String() string { return orgWriter.WriteNodesAsString(n) }
func (n FootnoteLink) String() string { return orgWriter.WriteNodesAsString(n) }
func (n RegularLink) String() string { return orgWriter.WriteNodesAsString(n) }
func (n Timestamp) String() string { return orgWriter.WriteNodesAsString(n) }