git.giacomolli.uk - onechat/blob - markdown.go

   1 package main
   2
   3 import (
   4         "slices"
   5         "strings"
   6         "unicode"
   7         "unicode/utf8"
   8 )
   9
  10 // markdownPunctuation holds the ASCII characters that a backslash makes plain.
  11 const markdownPunctuation = "!\"#$%&'()*+,-./:;<=>?@[\\]^_`{|}~"
  12
  13 // markdownText is the plain text of a Markdown reply and its Telegram entities,
  14 // while the conversion builds it. Units counts the UTF-16 code units of text,
  15 // the unit of the entity offsets and lengths.
  16 type markdownText struct {
  17         text     strings.Builder
  18         units    int
  19         entities []botEntity
  20 }
  21
  22 // markdownSpan is an inline mark of a line: the entity type, the content inside
  23 // the marks, the index after the closing mark, and the URL of a link.
  24 type markdownSpan struct {
  25         kind    string
  26         content string
  27         end     int
  28         url     string
  29 }
  30
  31 // markdownEntities converts the Markdown of a reply to plain text and Telegram
  32 // message entities. It knows bold (** and __), italic (* and _),
  33 // strikethrough (~~), inline code, fenced code blocks with an optional
  34 // language, links [text](url), and ATX headings, which become bold. Lists,
  35 // quotes, and tables stay as their text. A mark that does not close stays as
  36 // text with no entity, and code holds no entity.
  37 //
  38 // Inline marks do not cross a line break, so a mark that does not close cannot
  39 // change the rest of the reply.
  40 func markdownEntities(markdown string) (string, []botEntity) {
  41         var out markdownText
  42         lines := strings.Split(markdown, "\n")
  43         for index := 0; index < len(lines); index++ {
  44                 if index > 0 {
  45                         out.add("\n")
  46                 }
  47
  48                 line := lines[index]
  49                 language, fence := codeFence(line)
  50                 end := closingFence(lines, index, fence)
  51                 heading := headingText(line)
  52                 switch {
  53                 case end > index:
  54                         start := out.units
  55                         out.add(strings.Join(lines[index+1:end], "\n"))
  56                         out.entity(botEntity{
  57                                 Type:     "pre",
  58                                 Offset:   start,
  59                                 Length:   out.units - start,
  60                                 Language: language,
  61                         })
  62
  63                         index = end
  64                 case heading != "":
  65                         start := out.units
  66                         out.inline(heading)
  67                         out.entity(botEntity{
  68                                 Type:   "bold",
  69                                 Offset: start,
  70                                 Length: out.units - start,
  71                         })
  72                 default:
  73                         out.inline(line)
  74                 }
  75         }
  76
  77         slices.SortStableFunc(out.entities, func(a, b botEntity) int {
  78                 if a.Offset != b.Offset {
  79                         return a.Offset - b.Offset
  80                 }
  81
  82                 return b.Length - a.Length
  83         })
  84
  85         return out.text.String(), out.entities
  86 }
  87
  88 // add appends text.
  89 func (m *markdownText) add(text string) {
  90         m.text.WriteString(text)
  91         m.units += utf16Units(text)
  92 }
  93
  94 // entity keeps entity, unless it is empty, because Telegram refuses an entity
  95 // of length 0.
  96 func (m *markdownText) entity(entity botEntity) {
  97         if entity.Length > 0 {
  98                 m.entities = append(m.entities, entity)
  99         }
 100 }
 101
 102 // inline appends one line of Markdown with its inline marks as entities. A
 103 // backslash before an ASCII punctuation character makes the character plain
 104 // text.
 105 func (m *markdownText) inline(line string) {
 106         plain := 0
 107         index := 0
 108         for index < len(line) {
 109                 if line[index] == '\\' && index+1 < len(line) && strings.IndexByte(markdownPunctuation, line[index+1]) >= 0 {
 110                         m.add(line[plain:index])
 111                         plain = index + 1
 112                         index += 2
 113
 114                         continue
 115                 }
 116
 117                 span, found := markSpan(line, index)
 118                 if !found {
 119                         index++
 120                         continue
 121                 }
 122
 123                 m.add(line[plain:index])
 124                 start := m.units
 125                 if span.kind == "code" {
 126                         m.add(span.content)
 127                 } else {
 128                         m.inline(span.content)
 129                 }
 130
 131                 m.entity(botEntity{
 132                         Type:   span.kind,
 133                         Offset: start,
 134                         Length: m.units - start,
 135                         URL:    span.url,
 136                 })
 137
 138                 index = span.end
 139                 plain = span.end
 140         }
 141
 142         m.add(line[plain:])
 143 }
 144
 145 // codeFence returns the language of a line that opens a fenced code block, and
 146 // the fence. The fence is "" when line does not open a block.
 147 func codeFence(line string) (string, string) {
 148         trimmed := strings.TrimLeft(line, " ")
 149         if len(line)-len(trimmed) > 3 || !strings.HasPrefix(trimmed, "```") {
 150                 return "", ""
 151         }
 152
 153         run := len(trimmed) - len(strings.TrimLeft(trimmed, "`"))
 154         info := strings.TrimSpace(trimmed[run:])
 155         if strings.Contains(info, "`") {
 156                 return "", ""
 157         }
 158
 159         language, _, _ := strings.Cut(info, " ")
 160
 161         return language, trimmed[:run]
 162 }
 163
 164 // closingFence returns the index of the line after index that closes the code
 165 // block of fence, or -1 when fence is "" or no line closes it.
 166 func closingFence(lines []string, index int, fence string) int {
 167         if fence == "" {
 168                 return -1
 169         }
 170
 171         for end := index + 1; end < len(lines); end++ {
 172                 trimmed := strings.TrimSpace(lines[end])
 173                 if strings.HasPrefix(trimmed, fence) && strings.Trim(trimmed, "`") == "" {
 174                         return end
 175                 }
 176         }
 177
 178         return -1
 179 }
 180
 181 // headingText returns the text of an ATX heading line without its marks, or ""
 182 // when line is not a heading or has no text.
 183 func headingText(line string) string {
 184         trimmed := strings.TrimLeft(line, " ")
 185         if len(line)-len(trimmed) > 3 {
 186                 return ""
 187         }
 188
 189         level := len(trimmed) - len(strings.TrimLeft(trimmed, "#"))
 190         if level == 0 || level > 6 {
 191                 return ""
 192         }
 193
 194         rest := trimmed[level:]
 195         if rest != "" && rest[0] != ' ' && rest[0] != '\t' {
 196                 return ""
 197         }
 198
 199         rest = strings.TrimSpace(rest)
 200         closed := strings.TrimRight(rest, "#")
 201         if closed == "" || strings.HasSuffix(closed, " ") {
 202                 rest = strings.TrimSpace(closed)
 203         }
 204
 205         return rest
 206 }
 207
 208 // utf16Units returns the number of UTF-16 code units of text.
 209 func utf16Units(text string) int {
 210         units := 0
 211         for _, char := range text {
 212                 units++
 213                 if char > 0xFFFF {
 214                         units++
 215                 }
 216         }
 217
 218         return units
 219 }
 220
 221 // markSpan returns the inline mark that starts at index of line and closes in
 222 // line. It returns false when no mark starts there or the mark does not close.
 223 func markSpan(line string, index int) (markdownSpan, bool) {
 224         switch line[index] {
 225         case '`':
 226                 return codeSpan(line, index)
 227         case '[':
 228                 return linkSpan(line, index)
 229         case '~':
 230                 return delimited(line, index, "~~", "strikethrough")
 231         case '*', '_':
 232                 double := line[index : index+1]
 233                 double += double
 234                 span, found := delimited(line, index, double, "bold")
 235                 if found {
 236                         return span, true
 237                 }
 238
 239                 return delimited(line, index, line[index:index+1], "italic")
 240         default:
 241                 return markdownSpan{}, false
 242         }
 243 }
 244
 245 // codeSpan returns the inline code that starts at index: a run of backticks,
 246 // the code, and a run of the same length. One space on each side of the code
 247 // is left out, as in Markdown.
 248 func codeSpan(line string, index int) (markdownSpan, bool) {
 249         run := backticks(line[index:])
 250         for search := index + run; search < len(line); {
 251                 next := strings.IndexByte(line[search:], '`')
 252                 if next < 0 {
 253                         break
 254                 }
 255
 256                 next += search
 257                 length := backticks(line[next:])
 258                 if length != run {
 259                         search = next + length
 260                         continue
 261                 }
 262
 263                 content := line[index+run : next]
 264                 if strings.TrimSpace(content) == "" {
 265                         break
 266                 }
 267
 268                 if len(content) > 1 && content[0] == ' ' && content[len(content)-1] == ' ' {
 269                         content = content[1 : len(content)-1]
 270                 }
 271
 272                 return markdownSpan{
 273                         kind:    "code",
 274                         content: content,
 275                         end:     next + run,
 276                 }, true
 277         }
 278
 279         return markdownSpan{}, false
 280 }
 281
 282 // linkSpan returns the link [text](url) that starts at index. The text can
 283 // hold brackets in pairs, and the URL holds no space.
 284 func linkSpan(line string, index int) (markdownSpan, bool) {
 285         depth := 0
 286         for closing := index; closing < len(line); closing++ {
 287                 switch line[closing] {
 288                 case '\\':
 289                         closing++
 290                 case '[':
 291                         depth++
 292                 case ']':
 293                         depth--
 294                 }
 295
 296                 if depth > 0 {
 297                         continue
 298                 }
 299
 300                 rest := line[closing+1:]
 301                 if closing == index+1 || !strings.HasPrefix(rest, "(") {
 302                         return markdownSpan{}, false
 303                 }
 304
 305                 url, _, found := strings.Cut(rest[1:], ")")
 306                 if !found || url == "" || strings.ContainsAny(url, " \t") {
 307                         return markdownSpan{}, false
 308                 }
 309
 310                 return markdownSpan{
 311                         kind:    "text_link",
 312                         content: line[index+1 : closing],
 313                         end:     closing + 2 + len(url) + 1,
 314                         url:     url,
 315                 }, true
 316         }
 317
 318         return markdownSpan{}, false
 319 }
 320
 321 // delimited returns the span of kind between the mark at index and the same
 322 // mark later in line. An opening mark is followed by a character that is not a
 323 // space, and a closing mark follows one. A single mark is not a part of a
 324 // double one. A mark of underscores does not open or close inside a word, so
 325 // snake_case stays as text. A code span inside the content is skipped whole.
 326 func delimited(line string, index int, mark, kind string) (markdownSpan, bool) {
 327         char := mark[0]
 328         after := index + len(mark)
 329         if !strings.HasPrefix(line[index:], mark) || after >= len(line) || isSpace(line[after]) {
 330                 return markdownSpan{}, false
 331         }
 332
 333         if len(mark) == 1 && (line[after] == char || index > 0 && line[index-1] == char) {
 334                 return markdownSpan{}, false
 335         }
 336
 337         if char == '_' && index > 0 && wordCharBefore(line[:index]) {
 338                 return markdownSpan{}, false
 339         }
 340
 341         for closing := after + 1; closing < len(line); closing++ {
 342                 switch {
 343                 case line[closing] == '\\':
 344                         closing++
 345                         continue
 346                 case line[closing] == '`':
 347                         span, found := codeSpan(line, closing)
 348                         if found {
 349                                 closing = span.end - 1
 350                         }
 351
 352                         continue
 353                 case !strings.HasPrefix(line[closing:], mark) || isSpace(line[closing-1]):
 354                         continue
 355                 }
 356
 357                 end := closing + len(mark)
 358                 if len(mark) == 1 && (line[closing-1] == char || end < len(line) && line[end] == char) {
 359                         continue
 360                 }
 361
 362                 if char == '_' && wordChar(line[end:]) {
 363                         continue
 364                 }
 365
 366                 return markdownSpan{
 367                         kind:    kind,
 368                         content: line[after:closing],
 369                         end:     end,
 370                 }, true
 371         }
 372
 373         return markdownSpan{}, false
 374 }
 375
 376 // backticks returns the length of the run of backticks at the start of text.
 377 func backticks(text string) int {
 378         return len(text) - len(strings.TrimLeft(text, "`"))
 379 }
 380
 381 // isSpace tells if char is a space, a tab, or a line break.
 382 func isSpace(char byte) bool {
 383         return char == ' ' || char == '\t' || char == '\n' || char == '\r'
 384 }
 385
 386 // wordCharBefore tells if the character that ends text is a letter or a digit.
 387 func wordCharBefore(text string) bool {
 388         char, _ := utf8.DecodeLastRuneInString(text)
 389         return text != "" && (unicode.IsLetter(char) || unicode.IsDigit(char))
 390 }
 391
 392 // wordChar tells if the character that starts text is a letter or a digit. An
 393 // underscore next to one is a part of a word, such as in snake_case.
 394 func wordChar(text string) bool {
 395         char, _ := utf8.DecodeRuneInString(text)
 396         return text != "" && (unicode.IsLetter(char) || unicode.IsDigit(char))
 397 }