| 1 | //! Markdown rendering for TUI transcript lines. |
| 2 | //! |
| 3 | //! ## Width-independent parse vs width-dependent render (CX#6) |
| 4 | //! |
| 5 | //! The previous renderer was a single function `render_markdown(content, width)` |
| 6 | //! that scanned the source, classified each line (heading / list / code-fence / |
| 7 | //! paragraph / link), and word-wrapped to `Line<'static>` in one pass. That meant |
| 8 | //! every terminal resize forced a full re-parse of the source for every visible |
| 9 | //! cell — wasted work on the streaming cell whose content is changing anyway. |
| 10 | //! |
| 11 | //! The codex tui solves this by splitting parse from render. We mirror that: |
| 12 | //! |
| 13 | //! * [`parse`] turns the markdown source into a [`ParsedMarkdown`] AST: a vector |
| 14 | //! of width-independent [`Block`]s. The block kind already records all the |
| 15 | //! classification decisions (heading level, list bullet, code block membership) |
| 16 | //! that don't depend on width. |
| 17 | //! * [`render_parsed`] takes a `ParsedMarkdown` plus a width and a base style and |
| 18 | //! produces `Vec<Line<'static>>`. It only does word-wrap and span styling. |
| 19 | //! |
| 20 | //! [`render_markdown`] is kept as a thin convenience that does both — useful for |
| 21 | //! callers (Thinking body, message body) that don't want to manage the cache. |
| 22 | //! |
| 23 | //! The transcript cache layer (see `tui/transcript.rs`) caches the parsed AST per |
| 24 | //! cell and re-runs only the render step on width changes. That makes resize a |
| 25 | //! re-flow operation rather than a re-parse + re-flow operation. |
| 26 | |
| 27 | #[cfg(test)] |
| 28 | use std::cell::Cell; |
| 29 | |
| 30 | use ratatui::style::{Modifier, Style}; |
| 31 | use ratatui::text::{Line, Span}; |
| 32 | use unicode_width::{UnicodeWidthChar, UnicodeWidthStr}; |
| 33 | |
| 34 | use crate::palette; |
| 35 | use crate::tui::osc8; |
| 36 | |
| 37 | // Thread-local counter incremented every time `parse` runs. Used by tests to |
| 38 | // prove that width-only changes hit the cached-AST path and skip parsing. |
| 39 | // Thread-local (not global atomic) so concurrent tests calling `parse()` can't |
| 40 | // pollute each other's counters. |
| 41 | #[cfg(test)] |
| 42 | thread_local! { |
| 43 | static PARSE_INVOCATIONS: Cell<u64> = const { Cell::new(0) }; |
| 44 | } |
| 45 | |
| 46 | #[cfg(test)] |
| 47 | #[must_use] |
| 48 | pub fn parse_invocation_count() -> u64 { |
| 49 | PARSE_INVOCATIONS.with(|c| c.get()) |
| 50 | } |
| 51 | |
| 52 | #[cfg(test)] |
| 53 | pub fn reset_parse_invocation_count() { |
| 54 | PARSE_INVOCATIONS.with(|c| c.set(0)); |
| 55 | } |
| 56 | |
| 57 | /// One classified line of markdown source, width-independent. |
| 58 | /// |
| 59 | /// All decisions that depend only on the source text (heading level, bullet |
| 60 | /// kind, whether we're inside a fenced code block, paragraph text) are made at |
| 61 | /// parse time. Width-dependent layout (word-wrap, prefix indent) is deferred to |
| 62 | /// the render step. |
| 63 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 64 | pub enum Block { |
| 65 | /// `# heading text`. Includes the heading level (1..6). |
| 66 | Heading { level: usize, text: String }, |
| 67 | /// A horizontal rule emitted under a level-1 heading. |
| 68 | HeadingRule, |
| 69 | /// A standalone `---` / `***` / `___` horizontal rule. |
| 70 | HorizontalRule, |
| 71 | /// A bullet (`-`/`*`) or ordered (`1.`) list item with its prefix and body. |
| 72 | ListItem { bullet: String, text: String }, |
| 73 | /// A line inside a fenced code block. Fences themselves are dropped. |
| 74 | Code { line: String }, |
| 75 | /// A table row: cells split on `|`. |
| 76 | TableRow(Vec<String>), |
| 77 | /// A table separator row (`|---|---|`). Kept so the renderer can draw |
| 78 | /// horizontal rules at the correct positions. |
| 79 | TableSeparator, |
| 80 | /// A non-empty paragraph line that may contain inline links. |
| 81 | Paragraph { text: String }, |
| 82 | /// An empty source line, preserved so paragraph spacing survives. |
| 83 | Blank, |
| 84 | } |
| 85 | |
| 86 | /// Width-independent parsed-markdown AST for one cell's source. |
| 87 | /// |
| 88 | /// Wrapped in `Arc` at the cache layer so the cache can hand the same AST to |
| 89 | /// many render calls without copying. |
| 90 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 91 | pub struct ParsedMarkdown { |
| 92 | blocks: Vec<Block>, |
| 93 | } |
| 94 | |
| 95 | /// Parse markdown source into a width-independent block AST. |
| 96 | /// |
| 97 | /// This is a small line-oriented parser tuned for the patterns we render: |
| 98 | /// fenced code blocks, ATX headings, dash/star/numbered list items, and plain |
| 99 | /// paragraphs with optional links. It does not attempt to handle every CommonMark |
| 100 | /// edge case — that's intentional. The renderer will treat anything we don't |
| 101 | /// classify as `Block::Paragraph`. |
| 102 | #[must_use] |
| 103 | pub fn parse(content: &str) -> ParsedMarkdown { |
| 104 | #[cfg(test)] |
| 105 | PARSE_INVOCATIONS.with(|c| c.set(c.get() + 1)); |
| 106 | |
| 107 | let mut blocks = Vec::new(); |
| 108 | let mut in_code_block = false; |
| 109 | |
| 110 | for raw_line in content.lines() { |
| 111 | let trimmed = raw_line.trim_start(); |
| 112 | if trimmed.starts_with("```") { |
| 113 | in_code_block = !in_code_block; |
| 114 | continue; |
| 115 | } |
| 116 | |
| 117 | if in_code_block { |
| 118 | blocks.push(Block::Code { |
| 119 | line: raw_line.to_string(), |
| 120 | }); |
| 121 | continue; |
| 122 | } |
| 123 | |
| 124 | if let Some((level, text)) = parse_heading(trimmed) { |
| 125 | blocks.push(Block::Heading { |
| 126 | level, |
| 127 | text: text.to_string(), |
| 128 | }); |
| 129 | if level == 1 { |
| 130 | blocks.push(Block::HeadingRule); |
| 131 | } |
| 132 | continue; |
| 133 | } |
| 134 | |
| 135 | if let Some((bullet, text)) = parse_list_item(trimmed) { |
| 136 | blocks.push(Block::ListItem { |
| 137 | bullet, |
| 138 | text: text.to_string(), |
| 139 | }); |
| 140 | continue; |
| 141 | } |
| 142 | |
| 143 | if is_horizontal_rule(trimmed) { |
| 144 | blocks.push(Block::HorizontalRule); |
| 145 | continue; |
| 146 | } |
| 147 | |
| 148 | match parse_table_row(trimmed) { |
| 149 | Some(cells) => { |
| 150 | blocks.push(Block::TableRow(cells)); |
| 151 | continue; |
| 152 | } |
| 153 | None if trimmed.starts_with('|') => { |
| 154 | blocks.push(Block::TableSeparator); |
| 155 | continue; |
| 156 | } |
| 157 | None => {} |
| 158 | } |
| 159 | |
| 160 | if raw_line.is_empty() { |
| 161 | blocks.push(Block::Blank); |
| 162 | continue; |
| 163 | } |
| 164 | |
| 165 | blocks.push(Block::Paragraph { |
| 166 | text: trimmed.to_string(), |
| 167 | }); |
| 168 | } |
| 169 | |
| 170 | ParsedMarkdown { blocks } |
| 171 | } |
| 172 | |
| 173 | /// Render a parsed-markdown AST at the given terminal width. |
| 174 | /// |
| 175 | /// This is the width-dependent half: word-wrapping, link styling, code-block |
| 176 | /// formatting. The AST is owned by the caller (typically the transcript cache), |
| 177 | /// so width-only changes can call `render_parsed` again with the same AST and |
| 178 | /// skip the parse step entirely. |
| 179 | #[must_use] |
| 180 | pub fn render_parsed(parsed: &ParsedMarkdown, width: u16, base_style: Style) -> Vec<Line<'static>> { |
| 181 | let width = width.max(1) as usize; |
| 182 | let mut out: Vec<Line<'static>> = Vec::with_capacity(parsed.blocks.len()); |
| 183 | |
| 184 | let mut i = 0; |
| 185 | while i < parsed.blocks.len() { |
| 186 | if matches!( |
| 187 | &parsed.blocks[i], |
| 188 | Block::TableRow(_) | Block::TableSeparator |
| 189 | ) { |
| 190 | let start = i; |
| 191 | while i < parsed.blocks.len() |
| 192 | && matches!( |
| 193 | &parsed.blocks[i], |
| 194 | Block::TableRow(_) | Block::TableSeparator |
| 195 | ) |
| 196 | { |
| 197 | i += 1; |
| 198 | } |
| 199 | out.extend(render_table_group( |
| 200 | &parsed.blocks[start..i], |
| 201 | width, |
| 202 | base_style, |
| 203 | )); |
| 204 | continue; |
| 205 | } |
| 206 | |
| 207 | match &parsed.blocks[i] { |
| 208 | Block::Heading { text, .. } => { |
| 209 | let style = Style::default() |
| 210 | .fg(palette::DEEPSEEK_SKY) |
| 211 | .add_modifier(Modifier::BOLD); |
| 212 | out.extend(render_wrapped_line(text, width, style, false)); |
| 213 | } |
| 214 | Block::HeadingRule => { |
| 215 | out.push(Line::from(Span::styled( |
| 216 | "─".repeat(width.min(40)), |
| 217 | Style::default().fg(palette::TEXT_DIM), |
| 218 | ))); |
| 219 | } |
| 220 | Block::HorizontalRule => { |
| 221 | out.push(Line::from(Span::styled( |
| 222 | "─".repeat(width.min(60)), |
| 223 | Style::default().fg(palette::TEXT_DIM), |
| 224 | ))); |
| 225 | } |
| 226 | Block::ListItem { bullet, text } => { |
| 227 | let bullet_style = Style::default().fg(palette::DEEPSEEK_SKY); |
| 228 | out.extend(render_list_line( |
| 229 | bullet, |
| 230 | text, |
| 231 | width, |
| 232 | bullet_style, |
| 233 | base_style, |
| 234 | )); |
| 235 | } |
| 236 | Block::Code { line } => { |
| 237 | let code_style = Style::default() |
| 238 | .fg(palette::DEEPSEEK_SKY) |
| 239 | .add_modifier(Modifier::ITALIC); |
| 240 | out.extend(render_wrapped_line(line, width, code_style, true)); |
| 241 | } |
| 242 | Block::Paragraph { text } => { |
| 243 | let link_style = Style::default() |
| 244 | .fg(palette::DEEPSEEK_BLUE) |
| 245 | .add_modifier(Modifier::UNDERLINED); |
| 246 | out.extend(render_line_with_links(text, width, base_style, link_style)); |
| 247 | } |
| 248 | Block::Blank => { |
| 249 | out.push(Line::from("")); |
| 250 | } |
| 251 | Block::TableRow(_) | Block::TableSeparator => unreachable!(), |
| 252 | } |
| 253 | i += 1; |
| 254 | } |
| 255 | |
| 256 | if out.is_empty() { |
| 257 | out.push(Line::from("")); |
| 258 | } |
| 259 | |
| 260 | out |
| 261 | } |
| 262 | |
| 263 | /// Convenience wrapper: parse + render in one call. |
| 264 | /// |
| 265 | /// Equivalent to `render_parsed(&parse(content), width, base_style)`. Callers |
| 266 | /// that don't manage their own cache (the Thinking body, the immediate message |
| 267 | /// body) use this. |
| 268 | #[must_use] |
| 269 | pub fn render_markdown(content: &str, width: u16, base_style: Style) -> Vec<Line<'static>> { |
| 270 | let parsed = parse(content); |
| 271 | render_parsed(&parsed, width, base_style) |
| 272 | } |
| 273 | |
| 274 | fn parse_heading(line: &str) -> Option<(usize, &str)> { |
| 275 | let trimmed = line.trim_start(); |
| 276 | let hashes = trimmed.chars().take_while(|c| *c == '#').count(); |
| 277 | if hashes == 0 { |
| 278 | return None; |
| 279 | } |
| 280 | let text = trimmed[hashes..].trim(); |
| 281 | if text.is_empty() { |
| 282 | None |
| 283 | } else { |
| 284 | Some((hashes, text)) |
| 285 | } |
| 286 | } |
| 287 | |
| 288 | fn parse_list_item(line: &str) -> Option<(String, &str)> { |
| 289 | let trimmed = line.trim_start(); |
| 290 | if trimmed.starts_with("- ") || trimmed.starts_with("* ") { |
| 291 | return Some(("-".to_string(), trimmed[2..].trim())); |
| 292 | } |
| 293 | let bytes = trimmed.as_bytes(); |
| 294 | let mut idx = 0; |
| 295 | while idx < bytes.len() && bytes[idx].is_ascii_digit() { |
| 296 | idx += 1; |
| 297 | } |
| 298 | if idx == 0 || idx >= bytes.len() || bytes[idx] != b'.' { |
| 299 | return None; |
| 300 | } |
| 301 | let rest = &trimmed[idx + 1..]; |
| 302 | if !rest.starts_with(' ') { |
| 303 | return None; |
| 304 | } |
| 305 | Some((format!("{}.", &trimmed[..idx]), rest.trim_start())) |
| 306 | } |
| 307 | |
| 308 | fn render_wrapped_line( |
| 309 | line: &str, |
| 310 | width: usize, |
| 311 | style: Style, |
| 312 | indent_code: bool, |
| 313 | ) -> Vec<Line<'static>> { |
| 314 | let prefix = if indent_code { " " } else { "" }; |
| 315 | let prefix_width = prefix.width(); |
| 316 | let available = width.saturating_sub(prefix_width).max(1); |
| 317 | let wrapped = wrap_text(line, available); |
| 318 | let mut out = Vec::new(); |
| 319 | |
| 320 | for (idx, chunk) in wrapped.into_iter().enumerate() { |
| 321 | if idx == 0 { |
| 322 | out.push(Line::from(vec![ |
| 323 | Span::raw(prefix), |
| 324 | Span::styled(chunk, style), |
| 325 | ])); |
| 326 | } else { |
| 327 | out.push(Line::from(vec![ |
| 328 | Span::raw(" ".repeat(prefix_width)), |
| 329 | Span::styled(chunk, style), |
| 330 | ])); |
| 331 | } |
| 332 | } |
| 333 | |
| 334 | out |
| 335 | } |
| 336 | |
| 337 | fn render_list_line( |
| 338 | bullet: &str, |
| 339 | text: &str, |
| 340 | width: usize, |
| 341 | bullet_style: Style, |
| 342 | text_style: Style, |
| 343 | ) -> Vec<Line<'static>> { |
| 344 | let bullet_prefix = format!("{bullet} "); |
| 345 | let bullet_width = bullet_prefix.width(); |
| 346 | let available = width.saturating_sub(bullet_width).max(1); |
| 347 | let wrapped = render_line_with_links(text, available, text_style, link_style()); |
| 348 | |
| 349 | let mut out = Vec::new(); |
| 350 | for (idx, line) in wrapped.into_iter().enumerate() { |
| 351 | if idx == 0 { |
| 352 | let mut spans = vec![Span::styled(bullet_prefix.clone(), bullet_style)]; |
| 353 | spans.extend(line.spans); |
| 354 | out.push(Line::from(spans)); |
| 355 | } else { |
| 356 | let mut spans = vec![Span::raw(" ".repeat(bullet_width))]; |
| 357 | spans.extend(line.spans); |
| 358 | out.push(Line::from(spans)); |
| 359 | } |
| 360 | } |
| 361 | out |
| 362 | } |
| 363 | |
| 364 | fn render_line_with_links( |
| 365 | line: &str, |
| 366 | width: usize, |
| 367 | base_style: Style, |
| 368 | link_style: Style, |
| 369 | ) -> Vec<Line<'static>> { |
| 370 | if line.trim().is_empty() { |
| 371 | return vec![Line::from("")]; |
| 372 | } |
| 373 | |
| 374 | // Flatten inline tokens into (word, style) pairs preserving inter-token spaces. |
| 375 | let tokens = parse_inline_spans(line, base_style, link_style); |
| 376 | let mut words: Vec<(String, Style)> = Vec::new(); |
| 377 | for (text, style) in tokens { |
| 378 | let mut first = true; |
| 379 | for part in text.split(' ') { |
| 380 | if !first { |
| 381 | // The space consumed by split — attach as a plain space word |
| 382 | // so the wrap loop can decide whether to keep or break it. |
| 383 | words.push((" ".to_string(), style)); |
| 384 | } |
| 385 | if !part.is_empty() { |
| 386 | words.push((part.to_string(), style)); |
| 387 | } |
| 388 | first = false; |
| 389 | } |
| 390 | } |
| 391 | |
| 392 | let mut lines = Vec::new(); |
| 393 | let mut current_spans: Vec<Span> = Vec::new(); |
| 394 | let mut current_width = 0usize; |
| 395 | |
| 396 | for (word, style) in words { |
| 397 | let ww = word.width(); |
| 398 | if word == " " { |
| 399 | // Space: emit only if we're mid-line and it fits; otherwise drop |
| 400 | // (it's a potential wrap point, not content). |
| 401 | if !current_spans.is_empty() && current_width < width { |
| 402 | current_spans.push(Span::raw(" ")); |
| 403 | current_width += 1; |
| 404 | } |
| 405 | continue; |
| 406 | } |
| 407 | // Wrap before this word if it doesn't fit. |
| 408 | if current_width > 0 && current_width + ww > width { |
| 409 | // Trim trailing space span before breaking. |
| 410 | if let Some(last) = current_spans.last() |
| 411 | && last.content.as_ref() == " " |
| 412 | { |
| 413 | current_spans.pop(); |
| 414 | } |
| 415 | lines.push(Line::from(current_spans)); |
| 416 | current_spans = Vec::new(); |
| 417 | current_width = 0; |
| 418 | } |
| 419 | current_spans.push(Span::styled(word, style)); |
| 420 | current_width += ww; |
| 421 | } |
| 422 | |
| 423 | if !current_spans.is_empty() { |
| 424 | lines.push(Line::from(current_spans)); |
| 425 | } |
| 426 | if lines.is_empty() { |
| 427 | lines.push(Line::from("")); |
| 428 | } |
| 429 | lines |
| 430 | } |
| 431 | |
| 432 | /// Parse an entire line into (text, style) segments, handling **bold**, |
| 433 | /// *italic*, `code`, ~~strikethrough~~, [text](url) links, and bare URLs. |
| 434 | fn parse_inline_spans(line: &str, base_style: Style, link_style: Style) -> Vec<(String, Style)> { |
| 435 | let bold_style = base_style.add_modifier(Modifier::BOLD); |
| 436 | let italic_style = base_style.add_modifier(Modifier::ITALIC); |
| 437 | let code_style = base_style |
| 438 | .add_modifier(Modifier::ITALIC) |
| 439 | .bg(palette::SURFACE_ELEVATED); |
| 440 | let strike_style = base_style.add_modifier(Modifier::CROSSED_OUT); |
| 441 | let mut out = Vec::new(); |
| 442 | let mut rest = line; |
| 443 | |
| 444 | while !rest.is_empty() { |
| 445 | // **bold** |
| 446 | if let Some(end) = rest.strip_prefix("**").and_then(|s| s.find("**")) { |
| 447 | let inner = &rest[2..2 + end]; |
| 448 | out.push((inner.to_string(), bold_style)); |
| 449 | rest = &rest[2 + end + 2..]; |
| 450 | continue; |
| 451 | } |
| 452 | // __bold__ |
| 453 | if let Some(end) = rest.strip_prefix("__").and_then(|s| s.find("__")) { |
| 454 | let inner = &rest[2..2 + end]; |
| 455 | out.push((inner.to_string(), bold_style)); |
| 456 | rest = &rest[2 + end + 2..]; |
| 457 | continue; |
| 458 | } |
| 459 | // *italic* |
| 460 | if rest.starts_with('*') |
| 461 | && !rest.starts_with("**") |
| 462 | && let Some(end) = rest[1..].find('*') |
| 463 | { |
| 464 | let inner = &rest[1..1 + end]; |
| 465 | out.push((inner.to_string(), italic_style)); |
| 466 | rest = &rest[1 + end + 1..]; |
| 467 | continue; |
| 468 | } |
| 469 | // _italic_ |
| 470 | if rest.starts_with('_') |
| 471 | && !rest.starts_with("__") |
| 472 | && let Some(end) = rest[1..].find('_') |
| 473 | { |
| 474 | let inner = &rest[1..1 + end]; |
| 475 | out.push((inner.to_string(), italic_style)); |
| 476 | rest = &rest[1 + end + 1..]; |
| 477 | continue; |
| 478 | } |
| 479 | // `inline code` |
| 480 | if let Some(end) = rest.strip_prefix('`').and_then(|s| s.find('`')) { |
| 481 | let inner = &rest[1..1 + end]; |
| 482 | out.push((inner.to_string(), code_style)); |
| 483 | rest = &rest[1 + end + 1..]; |
| 484 | continue; |
| 485 | } |
| 486 | // ~~strikethrough~~ |
| 487 | if let Some(end) = rest.strip_prefix("~~").and_then(|s| s.find("~~")) { |
| 488 | let inner = &rest[2..2 + end]; |
| 489 | out.push((inner.to_string(), strike_style)); |
| 490 | rest = &rest[2 + end + 2..]; |
| 491 | continue; |
| 492 | } |
| 493 | // [text](url) |
| 494 | if rest.starts_with('[') |
| 495 | && let Some(bracket_end) = rest.find(']') |
| 496 | { |
| 497 | let text = &rest[1..bracket_end]; |
| 498 | let after_bracket = &rest[bracket_end + 1..]; |
| 499 | if after_bracket.starts_with('(') |
| 500 | && let Some(paren_end) = after_bracket.find(')') |
| 501 | { |
| 502 | let url = &after_bracket[1..paren_end]; |
| 503 | let content = if osc8::enabled() { |
| 504 | osc8::wrap_link(url, text) |
| 505 | } else { |
| 506 | format!("{text} ({url})") |
| 507 | }; |
| 508 | out.push((content, link_style)); |
| 509 | rest = &after_bracket[paren_end + 1..]; |
| 510 | continue; |
| 511 | } |
| 512 | } |
| 513 | // URL: consume until whitespace |
| 514 | if rest.starts_with("http://") || rest.starts_with("https://") { |
| 515 | let end = rest.find(char::is_whitespace).unwrap_or(rest.len()); |
| 516 | let url = &rest[..end]; |
| 517 | let content = if osc8::enabled() { |
| 518 | osc8::wrap_link(url, url) |
| 519 | } else { |
| 520 | url.to_string() |
| 521 | }; |
| 522 | out.push((content, link_style)); |
| 523 | rest = &rest[end..]; |
| 524 | continue; |
| 525 | } |
| 526 | // Plain text: consume until next marker or URL; always advance at least 1 char. |
| 527 | let next = find_next_marker(rest).max(rest.chars().next().map_or(1, |c| c.len_utf8())); |
| 528 | out.push((rest[..next].to_string(), base_style)); |
| 529 | rest = &rest[next..]; |
| 530 | } |
| 531 | out |
| 532 | } |
| 533 | |
| 534 | /// Find the index of the next inline marker (`**`, `__`, `*`, `_`, `http`) |
| 535 | /// in `s`, or `s.len()` if none found. |
| 536 | fn find_next_marker(s: &str) -> usize { |
| 537 | let mut i = 0; |
| 538 | let bytes = s.as_bytes(); |
| 539 | while i < bytes.len() { |
| 540 | let ch_len = s[i..].chars().next().map_or(1, |c| c.len_utf8()); |
| 541 | let slice = &s[i..]; |
| 542 | if slice.starts_with("**") |
| 543 | || slice.starts_with("__") |
| 544 | || slice.starts_with("~~") |
| 545 | || slice.starts_with('`') |
| 546 | || slice.starts_with('[') |
| 547 | || (slice.starts_with('*') && !slice.starts_with("**")) |
| 548 | || (slice.starts_with('_') && !slice.starts_with("__")) |
| 549 | || slice.starts_with("http://") |
| 550 | || slice.starts_with("https://") |
| 551 | { |
| 552 | return i; |
| 553 | } |
| 554 | i += ch_len; |
| 555 | } |
| 556 | s.len() |
| 557 | } |
| 558 | |
| 559 | fn is_horizontal_rule(line: &str) -> bool { |
| 560 | let stripped: String = line.chars().filter(|c| !c.is_whitespace()).collect(); |
| 561 | (stripped.chars().all(|c| c == '-') |
| 562 | || stripped.chars().all(|c| c == '*') |
| 563 | || stripped.chars().all(|c| c == '_')) |
| 564 | && stripped.len() >= 3 |
| 565 | } |
| 566 | |
| 567 | /// Parse a markdown table row like `| foo | bar |` into trimmed cell strings. |
| 568 | /// Returns `None` for separator rows (`|---|---|`). |
| 569 | fn parse_table_row(line: &str) -> Option<Vec<String>> { |
| 570 | if !line.starts_with('|') { |
| 571 | return None; |
| 572 | } |
| 573 | let inner = line.trim_matches('|'); |
| 574 | let cells: Vec<String> = inner.split('|').map(|c| c.trim().to_string()).collect(); |
| 575 | // Separator row: every non-empty cell is only dashes/colons/spaces |
| 576 | if cells |
| 577 | .iter() |
| 578 | .all(|c| c.is_empty() || c.chars().all(|ch| ch == '-' || ch == ':' || ch == ' ')) |
| 579 | { |
| 580 | return None; |
| 581 | } |
| 582 | Some(cells) |
| 583 | } |
| 584 | |
| 585 | fn render_table_row(cells: &[String], width: usize, base_style: Style) -> Vec<Line<'static>> { |
| 586 | if cells.is_empty() { |
| 587 | return vec![Line::from("")]; |
| 588 | } |
| 589 | let col_width = (width.saturating_sub(3 * cells.len() + 1)) / cells.len(); |
| 590 | let col_width = col_width.max(4); |
| 591 | let sep_style = Style::default().fg(palette::TEXT_DIM); |
| 592 | let mut spans: Vec<Span> = vec![Span::styled("│ ".to_string(), sep_style)]; |
| 593 | for (i, cell) in cells.iter().enumerate() { |
| 594 | let truncated = if cell.width() > col_width { |
| 595 | let mut s = String::new(); |
| 596 | let mut w = 0; |
| 597 | for ch in cell.chars() { |
| 598 | let cw = ch.width().unwrap_or(1); |
| 599 | if w + cw + 1 > col_width { |
| 600 | s.push('…'); |
| 601 | break; |
| 602 | } |
| 603 | s.push(ch); |
| 604 | w += cw; |
| 605 | } |
| 606 | s |
| 607 | } else { |
| 608 | cell.clone() |
| 609 | }; |
| 610 | let cell_spans: Vec<(String, Style)> = |
| 611 | parse_inline_spans(&truncated, base_style, link_style()); |
| 612 | let cell_width: usize = cell_spans.iter().map(|(t, _)| t.width()).sum(); |
| 613 | let pad = col_width.saturating_sub(cell_width); |
| 614 | for (text, style) in cell_spans { |
| 615 | spans.push(Span::styled(text, style)); |
| 616 | } |
| 617 | spans.push(Span::raw(" ".repeat(pad))); |
| 618 | if i + 1 < cells.len() { |
| 619 | spans.push(Span::styled(" │ ".to_string(), sep_style)); |
| 620 | } else { |
| 621 | spans.push(Span::styled(" │".to_string(), sep_style)); |
| 622 | } |
| 623 | } |
| 624 | vec![Line::from(spans)] |
| 625 | } |
| 626 | |
| 627 | fn table_col_width(num_cols: usize, term_width: usize) -> usize { |
| 628 | let col_width = (term_width.saturating_sub(3 * num_cols + 1)) / num_cols; |
| 629 | col_width.max(4) |
| 630 | } |
| 631 | |
| 632 | fn render_table_border( |
| 633 | num_cols: usize, |
| 634 | col_width: usize, |
| 635 | sep_style: Style, |
| 636 | left: &str, |
| 637 | mid: &str, |
| 638 | right: &str, |
| 639 | ) -> Line<'static> { |
| 640 | let fill = "\u{2500}".repeat(col_width); |
| 641 | let mut s = String::new(); |
| 642 | s.push_str(left); |
| 643 | for i in 0..num_cols { |
| 644 | s.push_str(&fill); |
| 645 | if i + 1 < num_cols { |
| 646 | s.push_str(mid); |
| 647 | } else { |
| 648 | s.push_str(right); |
| 649 | } |
| 650 | } |
| 651 | Line::from(Span::styled(s, sep_style)) |
| 652 | } |
| 653 | |
| 654 | fn render_table_group(blocks: &[Block], width: usize, base_style: Style) -> Vec<Line<'static>> { |
| 655 | let sep_style = Style::default().fg(palette::TEXT_DIM); |
| 656 | |
| 657 | let num_cols = blocks |
| 658 | .iter() |
| 659 | .filter_map(|b| match b { |
| 660 | Block::TableRow(cells) => Some(cells.len()), |
| 661 | _ => None, |
| 662 | }) |
| 663 | .max() |
| 664 | .unwrap_or(1); |
| 665 | |
| 666 | let col_width = table_col_width(num_cols, width); |
| 667 | |
| 668 | let mut lines = Vec::new(); |
| 669 | |
| 670 | // Top border |
| 671 | lines.push(render_table_border( |
| 672 | num_cols, |
| 673 | col_width, |
| 674 | sep_style, |
| 675 | "\u{250C}\u{2500}", |
| 676 | "\u{2500}\u{252C}\u{2500}", |
| 677 | "\u{2500}\u{2510}", |
| 678 | )); |
| 679 | |
| 680 | let mid_border = || { |
| 681 | render_table_border( |
| 682 | num_cols, |
| 683 | col_width, |
| 684 | sep_style, |
| 685 | "\u{251C}\u{2500}", |
| 686 | "\u{2500}\u{253C}\u{2500}", |
| 687 | "\u{2500}\u{2524}", |
| 688 | ) |
| 689 | }; |
| 690 | |
| 691 | for i in 0..blocks.len() { |
| 692 | match &blocks[i] { |
| 693 | Block::TableRow(cells) => { |
| 694 | lines.extend(render_table_row(cells, width, base_style)); |
| 695 | if i + 1 < blocks.len() && matches!(&blocks[i + 1], Block::TableRow(_)) { |
| 696 | lines.push(mid_border()); |
| 697 | } |
| 698 | } |
| 699 | Block::TableSeparator => { |
| 700 | lines.push(mid_border()); |
| 701 | } |
| 702 | _ => {} |
| 703 | } |
| 704 | } |
| 705 | |
| 706 | // Bottom border |
| 707 | lines.push(render_table_border( |
| 708 | num_cols, |
| 709 | col_width, |
| 710 | sep_style, |
| 711 | "\u{2514}\u{2500}", |
| 712 | "\u{2500}\u{2534}\u{2500}", |
| 713 | "\u{2500}\u{2518}", |
| 714 | )); |
| 715 | |
| 716 | lines |
| 717 | } |
| 718 | |
| 719 | fn link_style() -> Style { |
| 720 | Style::default() |
| 721 | .fg(palette::DEEPSEEK_BLUE) |
| 722 | .add_modifier(Modifier::UNDERLINED) |
| 723 | } |
| 724 | |
| 725 | fn wrap_text(text: &str, width: usize) -> Vec<String> { |
| 726 | if width == 0 { |
| 727 | return vec![text.to_string()]; |
| 728 | } |
| 729 | let mut lines = Vec::new(); |
| 730 | let mut current = String::new(); |
| 731 | let mut current_width = 0; |
| 732 | |
| 733 | for word in text.split_whitespace() { |
| 734 | let word_width = word.width(); |
| 735 | let additional = if current.is_empty() { |
| 736 | word_width |
| 737 | } else { |
| 738 | word_width + 1 |
| 739 | }; |
| 740 | if current_width + additional > width && !current.is_empty() { |
| 741 | lines.push(current); |
| 742 | current = word.to_string(); |
| 743 | current_width = word_width; |
| 744 | } else { |
| 745 | if !current.is_empty() { |
| 746 | current.push(' '); |
| 747 | current_width += 1; |
| 748 | } |
| 749 | current.push_str(word); |
| 750 | current_width += word_width; |
| 751 | } |
| 752 | } |
| 753 | |
| 754 | if current.is_empty() { |
| 755 | lines.push(String::new()); |
| 756 | } else { |
| 757 | lines.push(current); |
| 758 | } |
| 759 | |
| 760 | lines |
| 761 | } |
| 762 | |
| 763 | #[cfg(test)] |
| 764 | mod tests { |
| 765 | use super::*; |
| 766 | use ratatui::style::Style; |
| 767 | |
| 768 | #[test] |
| 769 | fn render_markdown_matches_parse_then_render() { |
| 770 | // Both calls run in the same thread under the same OSC8 lock so the |
| 771 | // flag is identical for both paths. |
| 772 | let source = "# Title\n\nA paragraph with a https://example.com link.\n\n- one\n- two\n```\ncode\n```"; |
| 773 | let direct = render_with_osc8(false, source); |
| 774 | let two_step = with_osc8(false, || { |
| 775 | let parsed = parse(source); |
| 776 | render_parsed(&parsed, 80, Style::default()) |
| 777 | .iter() |
| 778 | .flat_map(|l| l.spans.iter().map(|s| s.content.as_ref())) |
| 779 | .collect::<String>() |
| 780 | }); |
| 781 | assert_eq!(direct, two_step); |
| 782 | } |
| 783 | |
| 784 | #[test] |
| 785 | fn parse_is_width_independent() { |
| 786 | // Same source, two parses, must produce identical AST. (Sanity: |
| 787 | // parse must not depend on hidden global state like terminal width.) |
| 788 | let source = "Hello\n\n## Heading\n- list\n"; |
| 789 | let a = parse(source); |
| 790 | let b = parse(source); |
| 791 | assert_eq!(a, b); |
| 792 | } |
| 793 | |
| 794 | #[test] |
| 795 | fn render_parsed_word_wrap_changes_with_width() { |
| 796 | // The same AST must produce different layouts at different widths; |
| 797 | // otherwise the split is decorative, not functional. |
| 798 | let parsed = parse("alpha beta gamma delta epsilon zeta"); |
| 799 | let wide = render_parsed(&parsed, 80, Style::default()); |
| 800 | let narrow = render_parsed(&parsed, 10, Style::default()); |
| 801 | assert!( |
| 802 | narrow.len() > wide.len(), |
| 803 | "narrow should produce more lines" |
| 804 | ); |
| 805 | } |
| 806 | |
| 807 | #[test] |
| 808 | fn parse_invocations_increment() { |
| 809 | // Counter is thread-local, so concurrent tests calling `parse()` |
| 810 | // can't pollute each other. |
| 811 | reset_parse_invocation_count(); |
| 812 | let _ = parse("hello\n"); |
| 813 | let _ = parse("world\n"); |
| 814 | assert_eq!(parse_invocation_count(), 2); |
| 815 | } |
| 816 | |
| 817 | #[test] |
| 818 | fn render_parsed_does_not_call_parse() { |
| 819 | // Width-only changes must hit only the render path. This is the |
| 820 | // perf invariant CX#6 was filed for. |
| 821 | let parsed = parse("multiline\nsource\nwith several\nlines\n"); |
| 822 | reset_parse_invocation_count(); |
| 823 | let _ = render_parsed(&parsed, 80, Style::default()); |
| 824 | let _ = render_parsed(&parsed, 40, Style::default()); |
| 825 | let _ = render_parsed(&parsed, 20, Style::default()); |
| 826 | assert_eq!( |
| 827 | parse_invocation_count(), |
| 828 | 0, |
| 829 | "render_parsed must not call parse" |
| 830 | ); |
| 831 | } |
| 832 | |
| 833 | #[test] |
| 834 | fn fenced_code_block_collected_in_parse() { |
| 835 | let parsed = parse("text\n```\ncode line one\ncode line two\n```\nmore\n"); |
| 836 | let blocks = &parsed.blocks; |
| 837 | // text paragraph, two code lines, more paragraph (fences are dropped) |
| 838 | let code_lines: Vec<_> = blocks |
| 839 | .iter() |
| 840 | .filter_map(|b| match b { |
| 841 | Block::Code { line } => Some(line.as_str()), |
| 842 | _ => None, |
| 843 | }) |
| 844 | .collect(); |
| 845 | assert_eq!(code_lines, vec!["code line one", "code line two"]); |
| 846 | } |
| 847 | |
| 848 | #[test] |
| 849 | fn ordered_and_unordered_list_items_parse() { |
| 850 | let parsed = parse("- alpha\n* beta\n1. gamma\n"); |
| 851 | let items: Vec<_> = parsed |
| 852 | .blocks |
| 853 | .iter() |
| 854 | .filter_map(|b| match b { |
| 855 | Block::ListItem { bullet, text } => Some((bullet.as_str(), text.as_str())), |
| 856 | _ => None, |
| 857 | }) |
| 858 | .collect(); |
| 859 | assert_eq!(items, vec![("-", "alpha"), ("-", "beta"), ("1.", "gamma")]); |
| 860 | } |
| 861 | |
| 862 | /// Render with the OSC 8 flag pinned to `enabled`, then restore the prior |
| 863 | /// value. We serialize through a static mutex because `osc8::ENABLED` is |
| 864 | /// process-wide state and other tests touching it would race otherwise. |
| 865 | fn render_with_osc8(enabled: bool, source: &str) -> String { |
| 866 | with_osc8(enabled, || { |
| 867 | render_markdown(source, 80, Style::default()) |
| 868 | .iter() |
| 869 | .flat_map(|l| l.spans.iter().map(|s| s.content.as_ref())) |
| 870 | .collect::<String>() |
| 871 | }) |
| 872 | } |
| 873 | |
| 874 | fn with_osc8<T>(enabled: bool, f: impl FnOnce() -> T) -> T { |
| 875 | use std::sync::Mutex; |
| 876 | static OSC8_GUARD: Mutex<()> = Mutex::new(()); |
| 877 | let _guard = OSC8_GUARD.lock().unwrap_or_else(|e| e.into_inner()); |
| 878 | let prior = osc8::enabled(); |
| 879 | osc8::set_enabled(enabled); |
| 880 | let result = f(); |
| 881 | osc8::set_enabled(prior); |
| 882 | result |
| 883 | } |
| 884 | |
| 885 | #[test] |
| 886 | fn http_links_get_osc_8_wrapped_when_enabled() { |
| 887 | let joined = render_with_osc8(true, "see https://example.com for details"); |
| 888 | assert!( |
| 889 | joined.contains("\x1b]8;;https://example.com\x1b\\https://example.com\x1b]8;;\x1b\\"), |
| 890 | "expected OSC 8 wrapper around URL; got {joined:?}" |
| 891 | ); |
| 892 | } |
| 893 | |
| 894 | #[test] |
| 895 | fn osc_8_disabled_emits_plain_url() { |
| 896 | let joined = render_with_osc8(false, "see https://example.com for details"); |
| 897 | assert!( |
| 898 | !joined.contains("\x1b]8;;"), |
| 899 | "expected no OSC 8 wrapper when disabled; got {joined:?}" |
| 900 | ); |
| 901 | assert!(joined.contains("https://example.com")); |
| 902 | } |
| 903 | |
| 904 | #[test] |
| 905 | fn table_separator_row_is_kept() { |
| 906 | // Separator rows are now kept as TableSeparator blocks so the |
| 907 | // renderer can draw horizontal rules at the correct positions. |
| 908 | let src = "| 项目属性 | 详情 |\n|----------|------|\n| **语言** | Rust 1.88+ |\n"; |
| 909 | let parsed = parse(src); |
| 910 | let blocks: Vec<_> = parsed.blocks.iter().collect(); |
| 911 | // Should have 2 TableRow blocks (header + data) + 1 TableSeparator |
| 912 | let table_rows: Vec<_> = blocks |
| 913 | .iter() |
| 914 | .filter(|b| matches!(b, Block::TableRow(_))) |
| 915 | .collect(); |
| 916 | assert_eq!(table_rows.len(), 2, "expected 2 table rows: {blocks:?}"); |
| 917 | let separators: Vec<_> = blocks |
| 918 | .iter() |
| 919 | .filter(|b| matches!(b, Block::TableSeparator)) |
| 920 | .collect(); |
| 921 | assert_eq!( |
| 922 | separators.len(), |
| 923 | 1, |
| 924 | "expected 1 table separator: {blocks:?}" |
| 925 | ); |
| 926 | } |
| 927 | |
| 928 | #[test] |
| 929 | fn bold_markers_stripped_in_render() { |
| 930 | let src = "这是一个 **Rust 工作区项目**，包含多个 crate。\n"; |
| 931 | let lines = render_markdown(src, 80, Style::default()); |
| 932 | let text: String = lines |
| 933 | .iter() |
| 934 | .flat_map(|l| l.spans.iter().map(|s| s.content.as_ref())) |
| 935 | .collect(); |
| 936 | assert!( |
| 937 | !text.contains("**"), |
| 938 | "bold markers leaked into output: {text:?}" |
| 939 | ); |
| 940 | assert!(text.contains("Rust"), "bold content missing: {text:?}"); |
| 941 | } |
| 942 | |
| 943 | #[test] |
| 944 | fn table_renders_with_box_drawing_borders() { |
| 945 | let src = "| 文件 | 改动 |\n|---|---|\n| foo.rs | 重写 |\n"; |
| 946 | let lines = render_markdown(src, 60, Style::default()); |
| 947 | let text: String = lines |
| 948 | .iter() |
| 949 | .flat_map(|l| l.spans.iter().map(|s| s.content.as_ref())) |
| 950 | .collect(); |
| 951 | // Column pipes still present |
| 952 | assert!(text.contains('│'), "table pipe separator missing: {text:?}"); |
| 953 | // Separator row rendered as middle border, not raw markdown |
| 954 | assert!( |
| 955 | !text.contains("|---|"), |
| 956 | "raw separator row leaked: {text:?}" |
| 957 | ); |
| 958 | // Top and bottom borders present |
| 959 | assert!( |
| 960 | text.contains('\u{250C}'), |
| 961 | "top-left corner missing: {text:?}" |
| 962 | ); |
| 963 | assert!( |
| 964 | text.contains('\u{2510}'), |
| 965 | "top-right corner missing: {text:?}" |
| 966 | ); |
| 967 | assert!( |
| 968 | text.contains('\u{2514}'), |
| 969 | "bottom-left corner missing: {text:?}" |
| 970 | ); |
| 971 | assert!( |
| 972 | text.contains('\u{2518}'), |
| 973 | "bottom-right corner missing: {text:?}" |
| 974 | ); |
| 975 | // Middle separator present (at the |---|---| position) |
| 976 | assert!( |
| 977 | text.contains('\u{251C}'), |
| 978 | "middle-left junction missing: {text:?}" |
| 979 | ); |
| 980 | assert!( |
| 981 | text.contains('\u{2524}'), |
| 982 | "middle-right junction missing: {text:?}" |
| 983 | ); |
| 984 | } |
| 985 | } |
| 986 |