rumdl_lib/rules/md013_line_length.rs
1/// Rule MD013: Line length
2///
3/// See [docs/md013.md](../../docs/md013.md) for full documentation, configuration, and examples.
4use crate::rule::{LintError, LintResult, LintWarning, Rule, RuleCategory, Severity};
5use crate::rule_config_serde::RuleConfig;
6use crate::utils::mkdocs_admonitions;
7use crate::utils::mkdocs_attr_list::is_standalone_attr_list;
8use crate::utils::mkdocs_snippets::is_snippet_block_delimiter;
9use crate::utils::mkdocs_tabs;
10use crate::utils::range_utils::LineIndex;
11use crate::utils::range_utils::calculate_excess_range;
12use crate::utils::regex_cache::{IMAGE_REF_PATTERN, LINK_REF_PATTERN, URL_PATTERN};
13use crate::utils::table_utils::TableUtils;
14use crate::utils::text_reflow::{
15 BlockquoteLineData, ReflowLengthMode, blockquote_continuation_style, dominant_blockquote_prefix,
16 reflow_blockquote_content, split_into_sentences,
17};
18use pulldown_cmark::LinkType;
19use toml;
20
21mod block_builder;
22mod helpers;
23pub mod md013_config;
24use crate::utils::is_template_directive_only;
25use block_builder::{Block, BlockBuilder};
26use helpers::{
27 extract_list_marker_and_content, has_hard_break, is_github_alert_marker, is_horizontal_rule, is_html_only_line,
28 is_list_item, is_standalone_link_or_image_line, is_unwrappable_line, split_into_segments,
29 trim_preserving_hard_break,
30};
31pub use md013_config::MD013Config;
32use md013_config::{LengthMode, ReflowMode};
33
34#[cfg(test)]
35mod tests;
36use unicode_width::UnicodeWidthStr;
37
38#[derive(Clone, Default)]
39pub struct MD013LineLength {
40 pub(crate) config: MD013Config,
41}
42
43/// Blockquote paragraph line collected for reflow, with original line index for range computation.
44struct CollectedBlockquoteLine {
45 line_idx: usize,
46 data: BlockquoteLineData,
47}
48
49impl MD013LineLength {
50 pub fn new(line_length: usize, code_blocks: bool, tables: bool, headings: bool, strict: bool) -> Self {
51 Self {
52 config: MD013Config {
53 line_length: crate::types::LineLength::new(line_length),
54 code_blocks,
55 code_spans: true,
56 tables,
57 headings,
58 paragraphs: true, // Default to true for backwards compatibility
59 blockquotes: true, // Default to true for backwards compatibility
60 strict,
61 stern: false,
62 heading_line_length: None,
63 code_block_line_length: None,
64 reflow: false,
65 reflow_mode: ReflowMode::default(),
66 length_mode: LengthMode::default(),
67 abbreviations: Vec::new(),
68 require_sentence_capital: true,
69 ignore_link_urls: true,
70 },
71 }
72 }
73
74 pub fn from_config_struct(config: MD013Config) -> Self {
75 Self { config }
76 }
77
78 /// Return a clone with code block checking disabled.
79 /// Used for doc comment linting where code blocks are Rust code managed by rustfmt.
80 pub fn with_code_blocks_disabled(&self) -> Self {
81 let mut clone = self.clone();
82 clone.config.code_blocks = false;
83 clone
84 }
85
86 /// Convert MD013 LengthMode to text_reflow ReflowLengthMode
87 fn reflow_length_mode(&self) -> ReflowLengthMode {
88 match self.config.length_mode {
89 LengthMode::Chars => ReflowLengthMode::Chars,
90 LengthMode::Visual => ReflowLengthMode::Visual,
91 LengthMode::Bytes => ReflowLengthMode::Bytes,
92 }
93 }
94
95 fn should_ignore_line(
96 &self,
97 line: &str,
98 _lines: &[&str],
99 current_line: usize,
100 ctx: &crate::lint_context::LintContext,
101 ) -> bool {
102 if self.config.strict {
103 return false;
104 }
105
106 // Quick check for common patterns before expensive regex
107 let trimmed = line.trim();
108
109 // Only skip if the entire line is a URL (quick check first)
110 if (trimmed.starts_with("http://") || trimmed.starts_with("https://")) && URL_PATTERN.is_match(trimmed) {
111 return true;
112 }
113
114 // Only skip if the entire line is an image reference (quick check first)
115 if trimmed.starts_with("![") && trimmed.ends_with(']') && IMAGE_REF_PATTERN.is_match(trimmed) {
116 return true;
117 }
118
119 // Note: link reference definitions are handled as always-exempt (even in strict mode)
120 // in the main check loop, so they don't need to be checked here.
121
122 // Code blocks with long strings (only check if in code block)
123 if ctx.line_info(current_line + 1).is_some_and(|info| info.in_code_block)
124 && !trimmed.is_empty()
125 && !line.contains(' ')
126 && !line.contains('\t')
127 {
128 return true;
129 }
130
131 false
132 }
133
134 /// Check if rule should skip based on provided config (used for inline config support)
135 fn should_skip_with_config(&self, ctx: &crate::lint_context::LintContext, config: &MD013Config) -> bool {
136 // Skip if content is empty
137 if ctx.content.is_empty() {
138 return true;
139 }
140
141 // For sentence-per-line, semantic-line-breaks, or normalize mode, never skip based on line length
142 if config.reflow
143 && (config.reflow_mode == ReflowMode::SentencePerLine
144 || config.reflow_mode == ReflowMode::SemanticLineBreaks
145 || config.reflow_mode == ReflowMode::Normalize)
146 {
147 return false;
148 }
149
150 // Use the smallest applicable budget across line/heading/code-block
151 // contexts so a stricter context-specific limit doesn't get masked by
152 // the document-wide budget.
153 let min_limit = config.min_effective_line_length();
154 if min_limit.is_unlimited() {
155 return true;
156 }
157 let min_limit_bytes = min_limit.get();
158
159 // Quick check: if total content is shorter than the smallest line limit,
160 // definitely skip.
161 if ctx.content.len() <= min_limit_bytes {
162 return true;
163 }
164
165 // Skip if no line exceeds the smallest applicable limit.
166 !ctx.lines.iter().any(|line| line.byte_len > min_limit_bytes)
167 }
168
169 fn normalize_mode_needs_reflow<'a, I>(&self, lines: I, config: &MD013Config) -> bool
170 where
171 I: IntoIterator<Item = &'a str>,
172 {
173 let mut line_count = 0;
174 let check_length = !config.line_length.is_unlimited();
175
176 for line in lines {
177 line_count += 1;
178 if check_length && self.calculate_effective_length(line) > config.line_length.get() {
179 return true;
180 }
181 }
182
183 line_count > 1
184 }
185}
186
187impl Rule for MD013LineLength {
188 fn name(&self) -> &'static str {
189 "MD013"
190 }
191
192 fn description(&self) -> &'static str {
193 "Line length should not be excessive"
194 }
195
196 fn check(&self, ctx: &crate::lint_context::LintContext) -> LintResult {
197 // Use pre-parsed inline config from LintContext
198 let config_override = ctx.inline_config().get_rule_config("MD013");
199
200 // Apply configuration override if present
201 let effective_config = if let Some(json_config) = config_override {
202 if let Some(obj) = json_config.as_object() {
203 let mut config = self.config.clone();
204 if let Some(line_length) = obj.get("line_length").and_then(serde_json::Value::as_u64) {
205 config.line_length = crate::types::LineLength::new(line_length as usize);
206 }
207 if let Some(code_blocks) = obj.get("code_blocks").and_then(serde_json::Value::as_bool) {
208 config.code_blocks = code_blocks;
209 }
210 if let Some(code_spans) = obj.get("code_spans").and_then(serde_json::Value::as_bool) {
211 config.code_spans = code_spans;
212 }
213 if let Some(tables) = obj.get("tables").and_then(serde_json::Value::as_bool) {
214 config.tables = tables;
215 }
216 if let Some(headings) = obj.get("headings").and_then(serde_json::Value::as_bool) {
217 config.headings = headings;
218 }
219 if let Some(blockquotes) = obj.get("blockquotes").and_then(serde_json::Value::as_bool) {
220 config.blockquotes = blockquotes;
221 }
222 if let Some(strict) = obj.get("strict").and_then(serde_json::Value::as_bool) {
223 config.strict = strict;
224 }
225 if let Some(stern) = obj.get("stern").and_then(serde_json::Value::as_bool) {
226 config.stern = stern;
227 }
228 if let Some(v) = obj
229 .get("ignore_link_urls")
230 .or_else(|| obj.get("ignore-link-urls"))
231 .or_else(|| obj.get("semantic_link_understanding"))
232 .or_else(|| obj.get("semantic-link-understanding"))
233 .and_then(serde_json::Value::as_bool)
234 {
235 config.ignore_link_urls = v;
236 }
237 if let Some(reflow) = obj.get("reflow").and_then(serde_json::Value::as_bool) {
238 config.reflow = reflow;
239 }
240 if let Some(reflow_mode) = obj.get("reflow_mode").and_then(|v| v.as_str()) {
241 config.reflow_mode = match reflow_mode {
242 "default" => ReflowMode::Default,
243 "normalize" => ReflowMode::Normalize,
244 "sentence-per-line" => ReflowMode::SentencePerLine,
245 "semantic-line-breaks" => ReflowMode::SemanticLineBreaks,
246 _ => ReflowMode::default(),
247 };
248 }
249 config
250 } else {
251 self.config.clone()
252 }
253 } else {
254 self.config.clone()
255 };
256
257 // Fast early return using should_skip with EFFECTIVE config (after inline overrides)
258 // But don't skip if we're in reflow mode with Normalize or SentencePerLine
259 if self.should_skip_with_config(ctx, &effective_config)
260 && !(effective_config.reflow
261 && (effective_config.reflow_mode == ReflowMode::Normalize
262 || effective_config.reflow_mode == ReflowMode::SentencePerLine
263 || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
264 {
265 return Ok(Vec::new());
266 }
267
268 // Direct implementation without DocumentStructure
269 let mut warnings = Vec::new();
270
271 // Special handling: line_length = 0 means "no line length limit"
272 // Skip all line length checks, but still allow reflow if enabled
273 let skip_length_checks = effective_config.line_length.is_unlimited();
274
275 // Pre-filter lines that could be problematic to avoid processing all lines.
276 // Use the smallest applicable budget across line/heading/code-block contexts
277 // so candidates aren't dropped when a stricter context-specific budget applies.
278 let prefilter_limit = effective_config.min_effective_line_length();
279 let prefilter_skip = prefilter_limit.is_unlimited();
280 let mut candidate_lines = Vec::new();
281 if !skip_length_checks && !prefilter_skip {
282 for (line_idx, line_info) in ctx.lines.iter().enumerate() {
283 // Skip front matter - it should never be linted
284 if line_info.in_front_matter {
285 continue;
286 }
287
288 // Quick length check first
289 if line_info.byte_len > prefilter_limit.get() {
290 candidate_lines.push(line_idx);
291 }
292 }
293 }
294
295 // If no candidate lines and not in normalize or sentence-per-line mode, early return
296 if candidate_lines.is_empty()
297 && !(effective_config.reflow
298 && (effective_config.reflow_mode == ReflowMode::Normalize
299 || effective_config.reflow_mode == ReflowMode::SentencePerLine
300 || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
301 {
302 return Ok(warnings);
303 }
304
305 let lines = ctx.raw_lines();
306
307 // Create a quick lookup set for heading lines
308 // We need this for both the heading skip check AND the paragraphs check
309 let heading_lines_set: std::collections::HashSet<usize> = ctx
310 .lines
311 .iter()
312 .enumerate()
313 .filter(|(_, line)| line.heading.is_some())
314 .map(|(idx, _)| idx + 1)
315 .collect();
316
317 // Use pre-computed table blocks from context
318 // We need this for both the table skip check AND the paragraphs check
319 let table_blocks = &ctx.table_blocks;
320 let mut table_lines_set = std::collections::HashSet::new();
321 for table in table_blocks {
322 table_lines_set.insert(table.header_line + 1);
323 table_lines_set.insert(table.delimiter_line + 1);
324 for &line in &table.content_lines {
325 table_lines_set.insert(line + 1);
326 }
327 }
328
329 // Process candidate lines for line length checks
330 'line_loop: for &line_idx in &candidate_lines {
331 let line_number = line_idx + 1;
332 let line = lines[line_idx];
333
334 // Calculate actual line length (used in warning messages)
335 let effective_length = self.calculate_effective_length(line);
336
337 // Pick the context-specific limit: heading > code-block > paragraph.
338 // Headings dominate over code-block context if a setext underline ever
339 // overlaps a fenced range (defensive — these are mutually exclusive in
340 // practice, but the explicit ordering documents intent).
341 let is_heading_line = heading_lines_set.contains(&line_number);
342 let in_code_block = ctx.line_info(line_number).is_some_and(|info| info.in_code_block);
343 let line_limit = if is_heading_line {
344 effective_config.effective_heading_line_length().get()
345 } else if in_code_block {
346 effective_config.effective_code_block_line_length().get()
347 } else {
348 effective_config.line_length.get()
349 };
350
351 // A context-specific limit of 0 means "unlimited for this context".
352 if line_limit == 0 {
353 continue;
354 }
355
356 // Stern mode: like default, but the trailing-token forgiveness is
357 // disabled — a line with whitespace that exceeds the limit is a
358 // violation even if the excess is the final token. The "unwrappable"
359 // line exemption (single token, optionally prefixed by # or >) is
360 // still honored. Strict overrides stern entirely.
361 if effective_config.stern && !effective_config.strict && is_unwrappable_line(line) {
362 continue;
363 }
364
365 // Trailing-token forgiveness: only in default mode (not strict, not stern).
366 // If the line only exceeds the limit because of a long token at the end
367 // (URL, link chain, identifier), it passes. This matches markdownlint's
368 // behavior: line.replace(/\S*$/u, "#")
369 let check_length = if effective_config.strict || effective_config.stern {
370 effective_length
371 } else {
372 match line.rfind(char::is_whitespace) {
373 Some(pos) => {
374 let ws_char = line[pos..].chars().next().unwrap();
375 let prefix_end = pos + ws_char.len_utf8();
376 self.calculate_string_length(&line[..prefix_end]) + 1
377 }
378 None => 1, // No whitespace — entire line is a single token
379 }
380 };
381
382 // Skip lines where the check length is within the limit
383 if check_length <= line_limit {
384 continue;
385 }
386
387 // Ignore inline link/image URLs: suppress when excess comes entirely from inline URLs.
388 // Disabled by `strict` (all forgiveness off) and by `ignore_link_urls = false`
389 // (count link/image URLs toward the line length so such lines are flagged).
390 if !effective_config.strict && effective_config.ignore_link_urls {
391 let text_only_length = self.calculate_text_only_length(effective_length, line_number, ctx);
392 if text_only_length <= line_limit {
393 continue;
394 }
395 }
396
397 // Inline code spans cannot be wrapped, so reflow cannot shorten a line
398 // whose excess length is one. When code-span checking is disabled,
399 // suppress a violation that would fit once inline code spans are excluded.
400 if !effective_config.code_spans {
401 let code_span_width: usize = ctx
402 .code_spans()
403 .iter()
404 .filter(|span| span.line == line_number && span.end_line == line_number)
405 .map(|span| self.calculate_string_length(&ctx.content[span.byte_offset..span.byte_end]))
406 .sum();
407 if effective_length.saturating_sub(code_span_width) <= line_limit {
408 continue;
409 }
410 }
411
412 // Skip mkdocstrings and pymdown blocks (already handled by LintContext)
413 if ctx.lines[line_idx].in_mkdocstrings || ctx.lines[line_idx].in_pymdown_block {
414 continue;
415 }
416
417 // Skip MyST comments (% comment) — structural lines, not prose
418 if ctx.lines[line_idx].is_myst_comment {
419 continue;
420 }
421
422 // Link reference definitions are always exempt, even in strict mode.
423 // There's no way to shorten them without breaking the URL.
424 // Also check after stripping list markers, since list items may
425 // contain link ref defs as their content.
426 {
427 let trimmed = line.trim();
428 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
429 continue;
430 }
431 if is_list_item(trimmed) {
432 let (_, content) = extract_list_marker_and_content(trimmed);
433 let content_trimmed = content.trim();
434 if content_trimmed.starts_with('[')
435 && content_trimmed.contains("]:")
436 && LINK_REF_PATTERN.is_match(content_trimmed)
437 {
438 continue;
439 }
440 }
441 }
442
443 // Skip various block types efficiently
444 if !effective_config.strict {
445 // Lines whose only content is a link/image are exempt.
446 // After stripping list markers, blockquote markers, and emphasis,
447 // if only a link or image remains, there is no way to shorten it.
448 if is_standalone_link_or_image_line(line) {
449 continue;
450 }
451
452 // Lines consisting entirely of HTML tags are exempt.
453 // Badge lines, images with attributes, and similar inline HTML
454 // are long due to URLs in attributes and can't be meaningfully shortened.
455 if is_html_only_line(line) {
456 continue;
457 }
458
459 // Skip setext heading underlines
460 if !line.trim().is_empty() && line.trim().chars().all(|c| c == '=' || c == '-') {
461 continue;
462 }
463
464 // Skip block elements according to config flags
465 // The flags mean: true = check these elements, false = skip these elements
466 // So we skip when the flag is FALSE and the line is in that element type
467 if (!effective_config.headings && heading_lines_set.contains(&line_number))
468 || (!effective_config.code_blocks
469 && ctx.line_info(line_number).is_some_and(|info| info.in_code_block))
470 || (!effective_config.tables && table_lines_set.contains(&line_number))
471 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
472 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
473 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
474 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
475 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
476 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
477 || ctx.line_info(line_number).is_some_and(|info| info.in_pymdown_block)
478 {
479 continue;
480 }
481
482 // Check if this is a paragraph/regular text line
483 // If paragraphs = false, skip lines that are NOT in special blocks
484 // Blockquote content is treated as paragraph text, so it's not
485 // included in the special blocks list here.
486 if !effective_config.paragraphs {
487 let is_special_block = heading_lines_set.contains(&line_number)
488 || ctx.line_info(line_number).is_some_and(|info| info.in_code_block)
489 || table_lines_set.contains(&line_number)
490 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
491 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
492 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
493 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
494 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
495 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
496 || ctx
497 .line_info(line_number)
498 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container);
499
500 // Skip regular paragraph text when paragraphs = false
501 if !is_special_block {
502 continue;
503 }
504 }
505
506 // Skip blockquote lines when blockquotes = false.
507 // Also skip lazy continuation lines that belong to a blockquote
508 // (lines without `>` prefix that follow a blockquote line).
509 if !effective_config.blockquotes {
510 if ctx.lines[line_number - 1].blockquote.is_some() {
511 continue;
512 }
513 // Check for lazy continuation: scan backwards through
514 // non-blank lines to find if this paragraph started with
515 // a blockquote marker
516 if !line.trim().is_empty() {
517 let mut scan = line_number.saturating_sub(2);
518 loop {
519 if ctx.lines[scan].blockquote.is_some() {
520 // Found a blockquote ancestor — this is a lazy continuation
521 continue 'line_loop;
522 }
523 if lines[scan].trim().is_empty() || scan == 0 {
524 break;
525 }
526 scan -= 1;
527 }
528 }
529 }
530
531 // Skip lines that are only a URL, image ref, or link ref
532 if self.should_ignore_line(line, lines, line_idx, ctx) {
533 continue;
534 }
535 }
536
537 // In sentence-per-line mode, check if this is a single long sentence
538 // If so, emit a warning without a fix (user must manually rephrase)
539 if effective_config.reflow_mode == ReflowMode::SentencePerLine {
540 let sentences = split_into_sentences(line.trim());
541 if sentences.len() == 1 {
542 // Single sentence that's too long - warn but don't auto-fix
543 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
544
545 let (start_line, start_col, end_line, end_col) =
546 calculate_excess_range(line_number, line, line_limit);
547
548 warnings.push(LintWarning {
549 rule_name: Some(self.name().to_string()),
550 message,
551 line: start_line,
552 column: start_col,
553 end_line,
554 end_column: end_col,
555 severity: Severity::Warning,
556 fix: None, // No auto-fix for long single sentences
557 });
558 continue;
559 }
560 // Multiple sentences will be handled by paragraph-based reflow
561 continue;
562 }
563
564 // In semantic-line-breaks mode, skip per-line checks —
565 // all reflow is handled at the paragraph level with cascading splits
566 if effective_config.reflow_mode == ReflowMode::SemanticLineBreaks {
567 continue;
568 }
569
570 // Don't provide fix for individual lines when reflow is enabled
571 // Paragraph-based fixes will be handled separately
572 let fix = None;
573
574 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
575
576 // Calculate precise character range for the excess portion
577 let (start_line, start_col, end_line, end_col) = calculate_excess_range(line_number, line, line_limit);
578
579 warnings.push(LintWarning {
580 rule_name: Some(self.name().to_string()),
581 message,
582 line: start_line,
583 column: start_col,
584 end_line,
585 end_column: end_col,
586 severity: Severity::Warning,
587 fix,
588 });
589 }
590
591 // If reflow is enabled, generate paragraph-based fixes
592 if effective_config.reflow {
593 let paragraph_warnings = self.generate_paragraph_fixes(ctx, &effective_config, lines);
594 // Merge paragraph warnings with line warnings, removing duplicates
595 for pw in paragraph_warnings {
596 // Remove any line warnings that overlap with this paragraph
597 warnings.retain(|w| w.line < pw.line || w.line > pw.end_line);
598 warnings.push(pw);
599 }
600 }
601
602 Ok(warnings)
603 }
604
605 fn fix(&self, ctx: &crate::lint_context::LintContext) -> Result<String, LintError> {
606 // For CLI usage, apply fixes from warnings
607 // LSP will use the warning-based fixes directly
608 let warnings = self.check(ctx)?;
609 let warnings =
610 crate::utils::fix_utils::filter_warnings_by_inline_config(warnings, ctx.inline_config(), self.name());
611
612 // If there are no fixes, return content unchanged
613 if !warnings.iter().any(|w| w.fix.is_some()) {
614 return Ok(ctx.content.to_string());
615 }
616
617 // Apply warning-based fixes
618 crate::utils::fix_utils::apply_warning_fixes(ctx.content, &warnings)
619 .map_err(|e| LintError::FixFailed(format!("Failed to apply fixes: {e}")))
620 }
621
622 fn as_any(&self) -> &dyn std::any::Any {
623 self
624 }
625
626 fn category(&self) -> RuleCategory {
627 RuleCategory::Whitespace
628 }
629
630 fn should_skip(&self, ctx: &crate::lint_context::LintContext) -> bool {
631 self.should_skip_with_config(ctx, &self.config)
632 }
633
634 fn default_config_section(&self) -> Option<(String, toml::Value)> {
635 let table = crate::rule_config_serde::config_schema_table(&MD013Config::default())?;
636 if table.is_empty() {
637 None
638 } else {
639 Some((MD013Config::RULE_NAME.to_string(), toml::Value::Table(table)))
640 }
641 }
642
643 fn config_aliases(&self) -> Option<std::collections::HashMap<String, String>> {
644 let mut aliases = std::collections::HashMap::new();
645 aliases.insert("enable_reflow".to_string(), "reflow".to_string());
646 aliases.insert("strict_sentences".to_string(), "require-sentence-capital".to_string());
647 aliases.insert("strict-sentences".to_string(), "require-sentence-capital".to_string());
648 Some(aliases)
649 }
650
651 fn from_config(config: &crate::config::Config) -> Box<dyn Rule>
652 where
653 Self: Sized,
654 {
655 let mut rule_config = crate::rule_config_serde::load_rule_config::<MD013Config>(config);
656 // Use global line_length if rule-specific config still has default value
657 if rule_config.line_length.get() == 80 {
658 rule_config.line_length = config.global.line_length;
659 }
660 Box::new(Self::from_config_struct(rule_config))
661 }
662}
663
664impl MD013LineLength {
665 /// True when `line_num` (1-based) sits inside a structure whose lines must be
666 /// preserved verbatim (code block, front matter, HTML/JSX/MDX block, MkDocs
667 /// container, div marker, ...). Used to keep blockquote reflow from touching
668 /// quoted-looking text embedded in such structures.
669 fn line_in_verbatim_context(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
670 ctx.line_info(line_num).is_some_and(|info| {
671 info.in_code_block
672 || info.in_front_matter
673 || info.in_html_block
674 || info.in_html_comment
675 || info.in_esm_block
676 || info.in_jsx_expression
677 || info.in_jsx_block
678 || info.in_mdx_comment
679 || info.in_mkdocstrings
680 || info.in_pymdown_block
681 || info.in_mkdocs_container()
682 || info.is_div_marker
683 })
684 }
685
686 fn is_blockquote_content_boundary(
687 &self,
688 content: &str,
689 line_num: usize,
690 ctx: &crate::lint_context::LintContext,
691 strict: bool,
692 ) -> bool {
693 let trimmed = content.trim();
694
695 trimmed.is_empty()
696 || self.line_in_verbatim_context(line_num, ctx)
697 || trimmed.starts_with('#')
698 || trimmed.starts_with("```")
699 || trimmed.starts_with("~~~")
700 || trimmed.starts_with('>')
701 || TableUtils::is_potential_table_row(content)
702 || is_list_item(trimmed)
703 || is_horizontal_rule(content)
704 || (trimmed.starts_with('[') && content.contains("]:"))
705 || is_template_directive_only(content)
706 || is_standalone_attr_list(content)
707 || is_snippet_block_delimiter(content)
708 || is_github_alert_marker(trimmed)
709 || is_html_only_line(content)
710 // A standalone link/image line is exempt from MD013 (non-strict mode),
711 // so it must end the blockquote paragraph rather than be absorbed into
712 // it, mirroring the top-level paragraph reflow boundary.
713 || (!strict && is_standalone_link_or_image_line(content))
714 }
715
716 fn generate_blockquote_paragraph_fix(
717 &self,
718 ctx: &crate::lint_context::LintContext,
719 config: &MD013Config,
720 lines: &[&str],
721 line_index: &LineIndex,
722 start_idx: usize,
723 line_ending: &str,
724 ) -> (Option<LintWarning>, usize) {
725 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
726 return (None, start_idx + 1);
727 };
728 let target_level = start_bq.nesting_level;
729
730 let mut collected: Vec<CollectedBlockquoteLine> = Vec::new();
731 let mut i = start_idx;
732
733 while i < lines.len() {
734 if !collected.is_empty() && has_hard_break(&collected[collected.len() - 1].data.content) {
735 break;
736 }
737
738 let line_num = i + 1;
739 if line_num > ctx.lines.len() {
740 break;
741 }
742
743 if lines[i].trim().is_empty() {
744 break;
745 }
746
747 let line_bq = ctx.lines[i].blockquote.as_deref();
748 if let Some(bq) = line_bq {
749 if bq.nesting_level != target_level {
750 break;
751 }
752
753 if self.is_blockquote_content_boundary(&bq.content, line_num, ctx, config.strict) {
754 break;
755 }
756
757 collected.push(CollectedBlockquoteLine {
758 line_idx: i,
759 data: BlockquoteLineData::explicit(trim_preserving_hard_break(&bq.content), bq.prefix.clone()),
760 });
761 i += 1;
762 continue;
763 }
764
765 let lazy_content = lines[i].trim_start();
766 if self.is_blockquote_content_boundary(lazy_content, line_num, ctx, config.strict) {
767 break;
768 }
769
770 collected.push(CollectedBlockquoteLine {
771 line_idx: i,
772 data: BlockquoteLineData::lazy(trim_preserving_hard_break(lazy_content)),
773 });
774 i += 1;
775 }
776
777 if collected.is_empty() {
778 return (None, start_idx + 1);
779 }
780
781 let next_idx = i;
782 let paragraph_start = collected[0].line_idx;
783 let end_line = collected[collected.len() - 1].line_idx;
784 let line_data: Vec<BlockquoteLineData> = collected.iter().map(|l| l.data.clone()).collect();
785 let paragraph_text = line_data
786 .iter()
787 .map(|d| d.content.as_str())
788 .collect::<Vec<_>>()
789 .join(" ");
790
791 let contains_definition_list = line_data
792 .iter()
793 .any(|d| crate::utils::is_definition_list_item(&d.content));
794 if contains_definition_list {
795 return (None, next_idx);
796 }
797
798 let contains_snippets = line_data.iter().any(|d| is_snippet_block_delimiter(&d.content));
799 if contains_snippets {
800 return (None, next_idx);
801 }
802
803 let needs_reflow = match config.reflow_mode {
804 ReflowMode::Normalize => {
805 self.normalize_mode_needs_reflow(line_data.iter().map(|d| d.content.as_str()), config)
806 }
807 ReflowMode::SentencePerLine => {
808 let sentences = split_into_sentences(¶graph_text);
809 sentences.len() > 1 || line_data.len() > 1
810 }
811 ReflowMode::SemanticLineBreaks => {
812 let sentences = split_into_sentences(¶graph_text);
813 sentences.len() > 1
814 || line_data.len() > 1
815 || collected
816 .iter()
817 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get())
818 }
819 ReflowMode::Default => collected
820 .iter()
821 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get()),
822 };
823
824 if !needs_reflow {
825 return (None, next_idx);
826 }
827
828 let fallback_prefix = start_bq.prefix.clone();
829 let explicit_prefix = dominant_blockquote_prefix(&line_data, &fallback_prefix);
830 let continuation_style = blockquote_continuation_style(&line_data);
831
832 let reflow_line_length = if config.line_length.is_unlimited() {
833 usize::MAX
834 } else {
835 config
836 .line_length
837 .get()
838 .saturating_sub(self.calculate_string_length(&explicit_prefix))
839 .max(1)
840 };
841
842 let reflow_options = crate::utils::text_reflow::ReflowOptions {
843 line_length: reflow_line_length,
844 break_on_sentences: true,
845 preserve_breaks: false,
846 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
847 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
848 abbreviations: config.abbreviations_for_reflow(),
849 length_mode: self.reflow_length_mode(),
850 attr_lists: ctx.flavor.supports_attr_lists(),
851 myst_roles: ctx.flavor.supports_myst_roles(),
852 require_sentence_capital: config.require_sentence_capital,
853 max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
854 Some(4)
855 } else {
856 None
857 },
858 };
859
860 let reflowed_with_style =
861 reflow_blockquote_content(&line_data, &explicit_prefix, continuation_style, &reflow_options);
862
863 if reflowed_with_style.is_empty() {
864 return (None, next_idx);
865 }
866
867 let reflowed_text = reflowed_with_style.join(line_ending);
868
869 let start_range = line_index.whole_line_range(paragraph_start + 1);
870 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
871 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
872 } else {
873 line_index.whole_line_range(end_line + 1)
874 };
875 let byte_range = start_range.start..end_range.end;
876
877 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
878 format!("{reflowed_text}{line_ending}")
879 } else {
880 reflowed_text
881 };
882
883 let original_text = &ctx.content[byte_range.clone()];
884 if original_text == replacement {
885 return (None, next_idx);
886 }
887
888 let (warning_line, warning_end_line) = match config.reflow_mode {
889 ReflowMode::Normalize => (paragraph_start + 1, end_line + 1),
890 ReflowMode::SentencePerLine | ReflowMode::SemanticLineBreaks => (paragraph_start + 1, end_line + 1),
891 ReflowMode::Default => {
892 let violating_line = collected
893 .iter()
894 .find(|line| self.calculate_effective_length(lines[line.line_idx]) > config.line_length.get())
895 .map_or(paragraph_start + 1, |line| line.line_idx + 1);
896 (violating_line, violating_line)
897 }
898 };
899
900 let warning = LintWarning {
901 rule_name: Some(self.name().to_string()),
902 message: match config.reflow_mode {
903 ReflowMode::Normalize => format!(
904 "Paragraph could be normalized to use line length of {} characters",
905 config.line_length.get()
906 ),
907 ReflowMode::SentencePerLine => {
908 let num_sentences = split_into_sentences(¶graph_text).len();
909 if line_data.len() == 1 {
910 format!("Line contains {num_sentences} sentences (one sentence per line required)")
911 } else {
912 let num_lines = line_data.len();
913 format!(
914 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
915 )
916 }
917 }
918 ReflowMode::SemanticLineBreaks => {
919 let num_sentences = split_into_sentences(¶graph_text).len();
920 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
921 }
922 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
923 },
924 line: warning_line,
925 column: 1,
926 end_line: warning_end_line,
927 end_column: lines[warning_end_line.saturating_sub(1)].chars().count() + 1,
928 severity: Severity::Warning,
929 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
930 };
931
932 (Some(warning), next_idx)
933 }
934
935 /// Reflow a single list item that lives inside a blockquote.
936 ///
937 /// The blockquote paragraph reflow treats a list marker as a content boundary,
938 /// so `> - long item ...` is never wrapped even though the identical top-level
939 /// item is. This handles that case: it collects one tight, prose-only list item
940 /// at the starting blockquote level, reflows the item body to the configured
941 /// width, and re-emits it with the blockquote prefix preserved and continuation
942 /// lines aligned under the list content.
943 ///
944 /// Sibling and nested list items end collection and are reflowed independently
945 /// by the caller's outer loop (a nested item carries its indent folded into the
946 /// blockquote prefix, so it reflows correctly on its own). Items that are not
947 /// simple tight prose - those embedding a code block, table, fence, or hard
948 /// break - are left untouched (`None`) and the cursor still advances past the
949 /// whole item so its inner lines are never reprocessed as loose prose.
950 fn generate_blockquote_list_item_fix(
951 &self,
952 ctx: &crate::lint_context::LintContext,
953 config: &MD013Config,
954 lines: &[&str],
955 line_index: &LineIndex,
956 start_idx: usize,
957 line_ending: &str,
958 ) -> (Option<LintWarning>, usize) {
959 use crate::utils::blockquote::effective_indent_in_blockquote;
960
961 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
962 return (None, start_idx + 1);
963 };
964
965 // A `>`-prefixed line can be marked as a blockquote even inside a fenced code
966 // block (or other verbatim structure); such content must never be reflowed.
967 if self.line_in_verbatim_context(start_idx + 1, ctx) {
968 return (None, start_idx + 1);
969 }
970
971 let target_level = start_bq.nesting_level;
972
973 // The marker line carries the canonical blockquote prefix: its content begins
974 // with the list marker, so no list indent has been folded into the prefix.
975 let bq_prefix = start_bq.prefix.clone();
976
977 let (marker, first_body) = extract_list_marker_and_content(&start_bq.content);
978 if marker.is_empty() {
979 return (None, start_idx + 1);
980 }
981 let marker_width = marker.chars().count();
982
983 // Continuation lines of a checkbox item align under the bullet+checkbox, but
984 // are recognized from the bullet width, matching the top-level list reflow.
985 let base_marker_width = ["[ ] ", "[x] ", "[X] "]
986 .iter()
987 .find_map(|cb| marker.find(*cb))
988 .unwrap_or(marker_width);
989
990 // Collect the item: the marker line plus its tight prose continuation lines.
991 // `end_idx` always tracks the last consumed line so the cursor advances past
992 // the entire item, even when it turns out to be too complex to reflow safely.
993 let first_piece = trim_preserving_hard_break(&first_body);
994 let mut simple = !has_hard_break(&first_piece);
995 let mut body_pieces: Vec<String> = vec![first_piece];
996 let mut end_idx = start_idx;
997 let mut i = start_idx + 1;
998
999 while i < lines.len() {
1000 let Some(bq) = ctx.lines[i].blockquote.as_deref() else {
1001 // A blank line ends the item.
1002 if lines[i].trim().is_empty() {
1003 break;
1004 }
1005 // A lazy continuation (no `>` marker) is too ambiguous to reflow
1006 // safely. Consume the whole lazy run into this item's span and leave
1007 // the item untouched, so the caller does not reflow the continuation
1008 // in isolation and leave the marker line partially fixed.
1009 simple = false;
1010 while i < lines.len() && ctx.lines[i].blockquote.is_none() && !lines[i].trim().is_empty() {
1011 end_idx = i;
1012 i += 1;
1013 }
1014 break;
1015 };
1016 if bq.nesting_level != target_level {
1017 break;
1018 }
1019
1020 let content = bq.content.as_str();
1021 if content.trim().is_empty() {
1022 // Blank quoted line ends the tight paragraph. A following indented
1023 // paragraph (loose item) is reflowed on its own by the prose path.
1024 break;
1025 }
1026
1027 let eff_indent = effective_indent_in_blockquote(lines[i], target_level, 0);
1028 if eff_indent < base_marker_width {
1029 // Dedented: a sibling list item or text outside this item. Stop here
1030 // and let the outer loop classify it.
1031 break;
1032 }
1033 if is_list_item(content) {
1034 // A nested list item: its own item, handled independently.
1035 break;
1036 }
1037
1038 // An embedded structure (code block, table, fence, nested quote, ...)
1039 // means the item is not simple prose: keep consuming so the cursor clears
1040 // the whole structure, but do not produce a fix.
1041 if self.is_blockquote_content_boundary(content, i + 1, ctx, config.strict) {
1042 simple = false;
1043 }
1044
1045 let piece = trim_preserving_hard_break(content);
1046 if has_hard_break(&piece) {
1047 simple = false;
1048 }
1049 body_pieces.push(piece);
1050 end_idx = i;
1051 i += 1;
1052 }
1053
1054 let next_idx = end_idx + 1;
1055
1056 if !simple {
1057 return (None, next_idx);
1058 }
1059
1060 let exceeds_limit =
1061 || (start_idx..=end_idx).any(|idx| self.calculate_effective_length(lines[idx]) > config.line_length.get());
1062 let body_text = body_pieces.join(" ");
1063 let body_text = body_text.trim();
1064
1065 // Some bodies cannot be shortened and must stay verbatim, matching the
1066 // exemptions the top-level list reflow applies: link reference definitions
1067 // always, and (in non-strict mode) standalone links/images and HTML-only
1068 // lines. Reflowing a link reference definition would split it after the
1069 // colon/URL and break the definition.
1070 let is_link_ref_def =
1071 body_text.starts_with('[') && body_text.contains("]:") && LINK_REF_PATTERN.is_match(body_text);
1072 let raw_marker_line = lines[start_idx];
1073 let body_is_unwrappable = is_link_ref_def
1074 || (!config.strict && is_standalone_link_or_image_line(raw_marker_line))
1075 || (!config.strict && is_html_only_line(raw_marker_line));
1076 if body_is_unwrappable {
1077 return (None, next_idx);
1078 }
1079
1080 let needs_reflow = match config.reflow_mode {
1081 ReflowMode::Normalize => body_pieces.len() > 1 || exceeds_limit(),
1082 ReflowMode::Default => exceeds_limit(),
1083 ReflowMode::SentencePerLine => split_into_sentences(body_text).len() > 1 || body_pieces.len() > 1,
1084 ReflowMode::SemanticLineBreaks => split_into_sentences(body_text).len() > 1 || exceeds_limit(),
1085 };
1086 if !needs_reflow {
1087 return (None, next_idx);
1088 }
1089
1090 let prefix_width = self.calculate_string_length(&bq_prefix) + self.calculate_string_length(&marker);
1091 let reflow_line_length = if config.line_length.is_unlimited() {
1092 usize::MAX
1093 } else {
1094 config.line_length.get().saturating_sub(prefix_width).max(1)
1095 };
1096
1097 let reflow_options = crate::utils::text_reflow::ReflowOptions {
1098 line_length: reflow_line_length,
1099 break_on_sentences: true,
1100 preserve_breaks: false,
1101 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
1102 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
1103 abbreviations: config.abbreviations_for_reflow(),
1104 length_mode: self.reflow_length_mode(),
1105 attr_lists: ctx.flavor.supports_attr_lists(),
1106 myst_roles: ctx.flavor.supports_myst_roles(),
1107 require_sentence_capital: config.require_sentence_capital,
1108 max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
1109 Some(4)
1110 } else {
1111 None
1112 },
1113 };
1114
1115 let reflowed = crate::utils::text_reflow::reflow_line(body_text, &reflow_options);
1116 if reflowed.is_empty() {
1117 return (None, next_idx);
1118 }
1119
1120 let continuation_indent = " ".repeat(marker_width);
1121 let reflowed_text = reflowed
1122 .iter()
1123 .enumerate()
1124 .map(|(idx, line)| {
1125 if idx == 0 {
1126 format!("{bq_prefix}{marker}{line}")
1127 } else {
1128 format!("{bq_prefix}{continuation_indent}{line}")
1129 }
1130 })
1131 .collect::<Vec<_>>()
1132 .join(line_ending);
1133
1134 let start_range = line_index.whole_line_range(start_idx + 1);
1135 let end_range = if end_idx == lines.len() - 1 && !ctx.content.ends_with('\n') {
1136 line_index.line_text_range(end_idx + 1, 1, lines[end_idx].len() + 1)
1137 } else {
1138 line_index.whole_line_range(end_idx + 1)
1139 };
1140 let byte_range = start_range.start..end_range.end;
1141
1142 let replacement = if end_idx < lines.len() - 1 || ctx.content.ends_with('\n') {
1143 format!("{reflowed_text}{line_ending}")
1144 } else {
1145 reflowed_text
1146 };
1147
1148 let original_text = &ctx.content[byte_range.clone()];
1149 if original_text == replacement {
1150 return (None, next_idx);
1151 }
1152
1153 let message = match config.reflow_mode {
1154 ReflowMode::Normalize => format!(
1155 "Paragraph could be normalized to use line length of {} characters",
1156 config.line_length.get()
1157 ),
1158 ReflowMode::SentencePerLine => {
1159 let num_sentences = split_into_sentences(body_text).len();
1160 format!("List item should have one sentence per line (found {num_sentences} sentences)")
1161 }
1162 ReflowMode::SemanticLineBreaks => {
1163 let num_sentences = split_into_sentences(body_text).len();
1164 format!("List item should use semantic line breaks ({num_sentences} sentences)")
1165 }
1166 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
1167 };
1168
1169 let warning = LintWarning {
1170 rule_name: Some(self.name().to_string()),
1171 message,
1172 line: start_idx + 1,
1173 column: 1,
1174 end_line: end_idx + 1,
1175 end_column: lines[end_idx].chars().count() + 1,
1176 severity: Severity::Warning,
1177 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1178 };
1179
1180 (Some(warning), next_idx)
1181 }
1182
1183 /// Generate paragraph-based fixes
1184 fn generate_paragraph_fixes(
1185 &self,
1186 ctx: &crate::lint_context::LintContext,
1187 config: &MD013Config,
1188 lines: &[&str],
1189 ) -> Vec<LintWarning> {
1190 let mut warnings = Vec::new();
1191 let line_index = LineIndex::new(ctx.content);
1192
1193 // Detect the content's line ending style to preserve it in replacements.
1194 // The LSP receives content from editors which may use CRLF (Windows).
1195 // Replacements must match the original line endings to avoid false positives.
1196 let line_ending = crate::utils::line_ending::detect_line_ending(ctx.content);
1197
1198 let mut i = 0;
1199 while i < lines.len() {
1200 let line_num = i + 1;
1201
1202 // Handle blockquote paragraphs with style-preserving reflow.
1203 // Skip blockquotes when blockquotes=false or paragraphs=false
1204 if line_num > 0 && line_num <= ctx.lines.len() && ctx.lines[line_num - 1].blockquote.is_some() {
1205 if !config.blockquotes || !config.paragraphs {
1206 // Skip past all blockquote lines (explicit and lazy continuations).
1207 // A lazy continuation is a non-blank line without `>` that follows
1208 // a blockquote line and isn't a structural element.
1209 let mut saw_explicit_bq = false;
1210 while i < lines.len() && i < ctx.lines.len() {
1211 if ctx.lines[i].blockquote.is_some() {
1212 saw_explicit_bq = true;
1213 i += 1;
1214 } else if saw_explicit_bq
1215 && !lines[i].trim().is_empty()
1216 && !lines[i].trim_start().starts_with('#')
1217 && !lines[i].trim_start().starts_with('>')
1218 {
1219 // Lazy continuation of preceding blockquote
1220 i += 1;
1221 } else {
1222 break;
1223 }
1224 }
1225 continue;
1226 }
1227 // A list item inside the blockquote needs list-aware reflow (marker +
1228 // continuation indent); plain prose goes through the paragraph path.
1229 let is_bq_list_item = ctx.lines[i]
1230 .blockquote
1231 .as_deref()
1232 .is_some_and(|bq| is_list_item(&bq.content));
1233 let (warning, next_idx) = if is_bq_list_item {
1234 self.generate_blockquote_list_item_fix(ctx, config, lines, &line_index, i, line_ending)
1235 } else {
1236 self.generate_blockquote_paragraph_fix(ctx, config, lines, &line_index, i, line_ending)
1237 };
1238 if let Some(warning) = warning {
1239 warnings.push(warning);
1240 }
1241 i = next_idx;
1242 continue;
1243 }
1244
1245 // Skip special structures (but NOT MkDocs containers - those get special handling)
1246 let should_skip_due_to_line_info = ctx.line_info(line_num).is_some_and(|info| {
1247 info.in_code_block
1248 || info.in_front_matter
1249 || info.in_html_block
1250 || info.in_html_comment
1251 || info.in_esm_block
1252 || info.in_jsx_expression
1253 || info.in_jsx_block
1254 || info.in_mdx_comment
1255 || info.in_mkdocstrings
1256 || info.in_pymdown_block
1257 });
1258
1259 // Skip link reference definitions but NOT footnote definitions.
1260 // Footnote definitions (`[^id]: prose`) contain reflowable text,
1261 // while link reference definitions (`[ref]: URL`) contain URLs
1262 // that cannot be shortened.
1263 let is_link_ref_def =
1264 lines[i].trim().starts_with('[') && !lines[i].trim().starts_with("[^") && lines[i].contains("]:");
1265
1266 if should_skip_due_to_line_info
1267 || lines[i].trim().starts_with('#')
1268 || TableUtils::is_potential_table_row(lines[i])
1269 || lines[i].trim().is_empty()
1270 || is_horizontal_rule(lines[i])
1271 || is_template_directive_only(lines[i])
1272 || is_link_ref_def
1273 || ctx.line_info(line_num).is_some_and(|info| info.is_div_marker)
1274 || is_html_only_line(lines[i])
1275 || (!config.strict && is_standalone_link_or_image_line(lines[i]))
1276 {
1277 i += 1;
1278 continue;
1279 }
1280
1281 // Handle footnote definitions: `[^id]: prose text that can be reflowed`
1282 // Supports multi-paragraph footnotes with code blocks, blockquotes,
1283 // tables, and lists preserved verbatim.
1284 // Validate structure: must start with `[^`, contain `]:`, and the ID
1285 // must not contain `[` or `]` (prevents false matches on nested brackets)
1286 if lines[i].trim().starts_with("[^") && lines[i].contains("]:") && {
1287 let after_caret = &lines[i].trim()[2..];
1288 after_caret
1289 .find("]:")
1290 .is_some_and(|pos| pos > 0 && !after_caret[..pos].contains(['[', ']']))
1291 } {
1292 let footnote_start = i;
1293 let line = lines[i];
1294
1295 // Extract the prefix `[^id]:`
1296 let Some(colon_pos) = line.find("]:") else {
1297 i += 1;
1298 continue;
1299 };
1300 let prefix_end = colon_pos + 2;
1301 let prefix = &line[..prefix_end];
1302
1303 // Content starts after `]: ` (with optional space)
1304 let content_start = if line[prefix_end..].starts_with(' ') {
1305 prefix_end + 1
1306 } else {
1307 prefix_end
1308 };
1309 let first_content = &line[content_start..];
1310
1311 // CommonMark footnotes use 4-space continuation indent
1312 const FN_INDENT: usize = 4;
1313
1314 // --- Line classification for footnote content ---
1315 #[derive(Debug, Clone)]
1316 enum FnLineType {
1317 Content(String),
1318 Verbatim(String, usize), // preserved text, original indent
1319 Empty,
1320 }
1321
1322 // Helper: compute visual indent (tabs = 4 spaces)
1323 let visual_indent = |s: &str| -> usize {
1324 s.chars()
1325 .take_while(|c| c.is_whitespace())
1326 .map(|c| if c == '\t' { 4 } else { 1 })
1327 .sum::<usize>()
1328 };
1329
1330 // Helper: check if a trimmed line is a fence marker (homogeneous chars)
1331 let is_fence = |s: &str| -> bool {
1332 let t = s.trim();
1333 let fence_char = t.chars().next();
1334 matches!(fence_char, Some('`') | Some('~'))
1335 && t.chars().take_while(|&c| c == fence_char.unwrap()).count() >= 3
1336 };
1337
1338 // Helper: check if a trimmed line is a setext underline
1339 let is_setext_underline = |s: &str| -> bool {
1340 let t = s.trim();
1341 !t.is_empty()
1342 && (t.chars().all(|c| c == '=' || c == ' ') || t.chars().all(|c| c == '-' || c == ' '))
1343 && t.contains(['=', '-'])
1344 };
1345
1346 // Deferred body: `[^id]:\n content` — first line has no content,
1347 // actual content starts on the next indented line
1348 let deferred_body = first_content.trim().is_empty();
1349
1350 // Collect all lines belonging to this footnote definition
1351 let mut fn_lines: Vec<FnLineType> = Vec::new();
1352 if !deferred_body {
1353 fn_lines.push(FnLineType::Content(first_content.to_string()));
1354 }
1355 let mut last_consumed = i;
1356 i += 1;
1357
1358 // Strip only the footnote continuation indent, preserving
1359 // internal indentation (e.g., code block body indent)
1360 let strip_fn_indent = |s: &str| -> String {
1361 let mut chars = s.chars();
1362 let mut stripped = 0;
1363 while stripped < FN_INDENT {
1364 match chars.next() {
1365 Some('\t') => stripped += 4,
1366 Some(c) if c.is_whitespace() => stripped += 1,
1367 _ => break,
1368 }
1369 }
1370 chars.as_str().to_string()
1371 };
1372
1373 let mut in_fenced_code = false;
1374 let mut consecutive_blanks = 0u32;
1375
1376 while i < lines.len() {
1377 let next = lines[i];
1378 let next_trimmed = next.trim();
1379
1380 // Blank line handling
1381 if next_trimmed.is_empty() {
1382 consecutive_blanks += 1;
1383 // 2+ consecutive blanks terminate the footnote
1384 if consecutive_blanks >= 2 {
1385 break;
1386 }
1387
1388 // Inside a fenced code block, blank lines are part of the code
1389 if in_fenced_code {
1390 consecutive_blanks = 0; // Don't count blanks inside code blocks
1391 fn_lines.push(FnLineType::Verbatim(String::new(), 0));
1392 last_consumed = i;
1393 i += 1;
1394 continue;
1395 }
1396
1397 // Peek ahead: if next non-blank line is indented >= FN_INDENT,
1398 // this blank is an internal paragraph separator
1399 if i + 1 < lines.len() {
1400 let peek = lines[i + 1];
1401 let peek_indent = visual_indent(peek);
1402 if !peek.trim().is_empty() && peek_indent >= FN_INDENT {
1403 fn_lines.push(FnLineType::Empty);
1404 last_consumed = i;
1405 i += 1;
1406 continue;
1407 }
1408 }
1409 // No valid continuation after blank — end of footnote
1410 break;
1411 }
1412
1413 consecutive_blanks = 0;
1414 let indent = visual_indent(next);
1415
1416 // Not indented enough — end of footnote
1417 if indent < FN_INDENT {
1418 break;
1419 }
1420
1421 // Inside a fenced code block: everything is verbatim until closing fence
1422 if in_fenced_code {
1423 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1424 if is_fence(next_trimmed) {
1425 in_fenced_code = false;
1426 }
1427 last_consumed = i;
1428 i += 1;
1429 continue;
1430 }
1431
1432 // Fence opener — start verbatim code block
1433 if is_fence(next_trimmed) {
1434 in_fenced_code = true;
1435 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1436 last_consumed = i;
1437 i += 1;
1438 continue;
1439 }
1440
1441 // Indented code block: indent >= FN_INDENT + 4 (= 8 spaces)
1442 if indent >= FN_INDENT + 4 {
1443 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1444 last_consumed = i;
1445 i += 1;
1446 continue;
1447 }
1448
1449 // Structural content that must be preserved verbatim
1450 if next_trimmed.starts_with('#')
1451 || is_list_item(next_trimmed)
1452 || next_trimmed.starts_with('>')
1453 || TableUtils::is_potential_table_row(next_trimmed)
1454 || is_setext_underline(next_trimmed)
1455 || is_horizontal_rule(next_trimmed)
1456 || crate::utils::mkdocs_footnotes::is_footnote_definition(next_trimmed)
1457 {
1458 // Preserve verbatim: blockquotes, tables, lists, setext
1459 // underlines, and horizontal rules inside the footnote
1460 if next_trimmed.starts_with('>')
1461 || TableUtils::is_potential_table_row(next_trimmed)
1462 || is_list_item(next_trimmed)
1463 || is_setext_underline(next_trimmed)
1464 || is_horizontal_rule(next_trimmed)
1465 {
1466 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1467 last_consumed = i;
1468 i += 1;
1469 continue;
1470 }
1471 // Headings, new footnote defs, link refs — end the footnote
1472 break;
1473 }
1474
1475 // Link reference definitions inside footnotes are not reflowable
1476 if next_trimmed.starts_with('[')
1477 && !next_trimmed.starts_with("[^")
1478 && next_trimmed.contains("]:")
1479 && LINK_REF_PATTERN.is_match(next_trimmed)
1480 {
1481 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1482 last_consumed = i;
1483 i += 1;
1484 continue;
1485 }
1486
1487 // HTML-only lines inside footnotes are not reflowable
1488 if is_html_only_line(next_trimmed) {
1489 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1490 last_consumed = i;
1491 i += 1;
1492 continue;
1493 }
1494
1495 // Regular prose content
1496 fn_lines.push(FnLineType::Content(next_trimmed.to_string()));
1497 last_consumed = i;
1498 i += 1;
1499 }
1500
1501 // Nothing collected or only empty lines
1502 if fn_lines.iter().all(|l| matches!(l, FnLineType::Empty)) || fn_lines.is_empty() {
1503 continue;
1504 }
1505
1506 // --- Group into blocks ---
1507 #[derive(Debug)]
1508 enum FnBlock {
1509 Paragraph(Vec<String>),
1510 Verbatim(Vec<(String, usize)>), // (content, indent) preserved as-is
1511 }
1512
1513 let mut blocks: Vec<FnBlock> = Vec::new();
1514 let mut current_para: Vec<String> = Vec::new();
1515 let mut current_verbatim: Vec<(String, usize)> = Vec::new();
1516
1517 for fl in &fn_lines {
1518 match fl {
1519 FnLineType::Content(s) => {
1520 if !current_verbatim.is_empty() {
1521 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
1522 }
1523 current_para.push(s.clone());
1524 }
1525 FnLineType::Verbatim(s, indent) => {
1526 if !current_para.is_empty() {
1527 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
1528 }
1529 current_verbatim.push((s.clone(), *indent));
1530 }
1531 FnLineType::Empty => {
1532 if !current_para.is_empty() {
1533 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
1534 }
1535 if !current_verbatim.is_empty() {
1536 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
1537 }
1538 }
1539 }
1540 }
1541 if !current_para.is_empty() {
1542 blocks.push(FnBlock::Paragraph(current_para));
1543 }
1544 if !current_verbatim.is_empty() {
1545 blocks.push(FnBlock::Verbatim(current_verbatim));
1546 }
1547
1548 // --- Reflow paragraphs and reconstruct ---
1549 let prefix_display_width = prefix.chars().count() + 1; // +1 for space
1550 let reflow_line_length = if config.line_length.is_unlimited() {
1551 usize::MAX
1552 } else {
1553 config
1554 .line_length
1555 .get()
1556 .saturating_sub(FN_INDENT.max(prefix_display_width))
1557 .max(20)
1558 };
1559 let reflow_options = crate::utils::text_reflow::ReflowOptions {
1560 line_length: reflow_line_length,
1561 break_on_sentences: true,
1562 preserve_breaks: false,
1563 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
1564 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
1565 abbreviations: config.abbreviations_for_reflow(),
1566 length_mode: self.reflow_length_mode(),
1567 attr_lists: ctx.flavor.supports_attr_lists(),
1568 myst_roles: ctx.flavor.supports_myst_roles(),
1569 require_sentence_capital: config.require_sentence_capital,
1570 max_list_continuation_indent: None,
1571 };
1572
1573 let indent_str = " ".repeat(FN_INDENT);
1574 let mut result_lines: Vec<String> = Vec::new();
1575 let mut is_first_block = true;
1576
1577 for block in &blocks {
1578 match block {
1579 FnBlock::Paragraph(para_lines) => {
1580 let paragraph_text = para_lines.join(" ");
1581 let paragraph_text = paragraph_text.trim();
1582 if paragraph_text.is_empty() {
1583 continue;
1584 }
1585
1586 let reflowed = crate::utils::text_reflow::reflow_line(paragraph_text, &reflow_options);
1587 if reflowed.is_empty() {
1588 continue;
1589 }
1590
1591 // Blank line separator between blocks
1592 if !result_lines.is_empty() {
1593 result_lines.push(String::new());
1594 }
1595
1596 for (idx, rline) in reflowed.iter().enumerate() {
1597 if is_first_block && idx == 0 {
1598 result_lines.push(format!("{prefix} {rline}"));
1599 } else {
1600 result_lines.push(format!("{indent_str}{rline}"));
1601 }
1602 }
1603 is_first_block = false;
1604 }
1605 FnBlock::Verbatim(verb_lines) => {
1606 // Blank line separator between blocks
1607 if !result_lines.is_empty() {
1608 result_lines.push(String::new());
1609 }
1610
1611 if is_first_block {
1612 // Verbatim as first block in a deferred-body footnote
1613 if deferred_body {
1614 result_lines.push(prefix.to_string());
1615 }
1616 is_first_block = false;
1617 }
1618 for (content, _orig_indent) in verb_lines {
1619 result_lines.push(format!("{indent_str}{content}"));
1620 }
1621 }
1622 }
1623 }
1624
1625 // If nothing was produced, skip
1626 if result_lines.is_empty() {
1627 continue;
1628 }
1629
1630 let reflowed_text = result_lines.join(line_ending);
1631
1632 // Calculate byte range using last_consumed
1633 let start_range = line_index.whole_line_range(footnote_start + 1);
1634 let end_range = if last_consumed == lines.len() - 1 && !ctx.content.ends_with('\n') {
1635 line_index.line_text_range(last_consumed + 1, 1, lines[last_consumed].len() + 1)
1636 } else {
1637 line_index.whole_line_range(last_consumed + 1)
1638 };
1639 let byte_range = start_range.start..end_range.end;
1640
1641 let replacement = if last_consumed < lines.len() - 1 || ctx.content.ends_with('\n') {
1642 format!("{reflowed_text}{line_ending}")
1643 } else {
1644 reflowed_text
1645 };
1646
1647 let original_text = &ctx.content[byte_range.clone()];
1648 let max_length = (footnote_start..=last_consumed)
1649 .map(|idx| self.calculate_effective_length(lines[idx]))
1650 .max()
1651 .unwrap_or(0);
1652 let line_limit = if config.line_length.is_unlimited() {
1653 usize::MAX
1654 } else {
1655 config.line_length.get()
1656 };
1657 if original_text != replacement && max_length > line_limit {
1658 warnings.push(LintWarning {
1659 rule_name: Some(self.name().to_string()),
1660 message: format!(
1661 "Line length {} exceeds {} characters",
1662 max_length,
1663 config.line_length.get()
1664 ),
1665 line: footnote_start + 1,
1666 column: 1,
1667 end_line: last_consumed + 1,
1668 end_column: lines[last_consumed].chars().count() + 1,
1669 severity: Severity::Warning,
1670 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1671 });
1672 }
1673 continue;
1674 }
1675
1676 // Handle MkDocs container content (admonitions and tabs) with indent-preserving reflow
1677 if ctx
1678 .line_info(line_num)
1679 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
1680 {
1681 // Skip admonition/tab marker lines — only reflow their indented content
1682 let current_line = lines[i];
1683 if mkdocs_admonitions::is_admonition_start(current_line) || mkdocs_tabs::is_tab_marker(current_line) {
1684 i += 1;
1685 continue;
1686 }
1687
1688 let container_start = i;
1689
1690 // Detect the actual indent level from the first content line
1691 // (supports nested admonitions with 8+ spaces)
1692 let first_line = lines[i];
1693 let base_indent_len = first_line.len() - first_line.trim_start().len();
1694 let base_indent: String = " ".repeat(base_indent_len);
1695
1696 // Collect consecutive MkDocs container paragraph lines
1697 let mut container_lines: Vec<&str> = Vec::new();
1698 while i < lines.len() {
1699 let current_line_num = i + 1;
1700 let line_info = ctx.line_info(current_line_num);
1701
1702 // Stop if we leave the MkDocs container
1703 if !line_info.is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container) {
1704 break;
1705 }
1706
1707 let line = lines[i];
1708
1709 // Stop at paragraph boundaries within the container
1710 if line.trim().is_empty() {
1711 break;
1712 }
1713
1714 // Skip list items, code blocks, headings, HTML-only lines within containers
1715 if is_list_item(line.trim())
1716 || line.trim().starts_with("```")
1717 || line.trim().starts_with("~~~")
1718 || line.trim().starts_with('#')
1719 || is_html_only_line(line)
1720 {
1721 break;
1722 }
1723
1724 container_lines.push(line);
1725 i += 1;
1726 }
1727
1728 if container_lines.is_empty() {
1729 // Must advance i to avoid infinite loop when we encounter
1730 // non-paragraph content (code block, list, heading, empty line)
1731 // at the start of an MkDocs container
1732 i += 1;
1733 continue;
1734 }
1735
1736 // Strip the base indent from each line and join for reflow
1737 let stripped_lines: Vec<&str> = container_lines
1738 .iter()
1739 .map(|line| {
1740 if line.starts_with(&base_indent) {
1741 &line[base_indent_len..]
1742 } else {
1743 line.trim_start()
1744 }
1745 })
1746 .collect();
1747 let paragraph_text = stripped_lines.join(" ");
1748
1749 // Check if reflow is needed
1750 let needs_reflow = match config.reflow_mode {
1751 ReflowMode::Normalize => self.normalize_mode_needs_reflow(container_lines.iter().copied(), config),
1752 ReflowMode::SentencePerLine => {
1753 let sentences = split_into_sentences(¶graph_text);
1754 sentences.len() > 1 || container_lines.len() > 1
1755 }
1756 ReflowMode::SemanticLineBreaks => {
1757 let sentences = split_into_sentences(¶graph_text);
1758 sentences.len() > 1
1759 || container_lines.len() > 1
1760 || container_lines
1761 .iter()
1762 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
1763 }
1764 ReflowMode::Default => container_lines
1765 .iter()
1766 .any(|line| self.calculate_effective_length(line) > config.line_length.get()),
1767 };
1768
1769 if !needs_reflow {
1770 continue;
1771 }
1772
1773 // Calculate byte range for this container paragraph
1774 let start_range = line_index.whole_line_range(container_start + 1);
1775 let end_line = container_start + container_lines.len() - 1;
1776 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
1777 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
1778 } else {
1779 line_index.whole_line_range(end_line + 1)
1780 };
1781 let byte_range = start_range.start..end_range.end;
1782
1783 // Reflow with adjusted line length (accounting for the 4-space indent)
1784 let reflow_line_length = if config.line_length.is_unlimited() {
1785 usize::MAX
1786 } else {
1787 config.line_length.get().saturating_sub(base_indent_len).max(1)
1788 };
1789 let reflow_options = crate::utils::text_reflow::ReflowOptions {
1790 line_length: reflow_line_length,
1791 break_on_sentences: true,
1792 preserve_breaks: false,
1793 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
1794 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
1795 abbreviations: config.abbreviations_for_reflow(),
1796 length_mode: self.reflow_length_mode(),
1797 attr_lists: ctx.flavor.supports_attr_lists(),
1798 myst_roles: ctx.flavor.supports_myst_roles(),
1799 require_sentence_capital: config.require_sentence_capital,
1800 max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
1801 Some(4)
1802 } else {
1803 None
1804 },
1805 };
1806 let reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
1807
1808 // Re-add the 4-space indent to each reflowed line
1809 let reflowed_with_indent: Vec<String> =
1810 reflowed.iter().map(|line| format!("{base_indent}{line}")).collect();
1811 let reflowed_text = reflowed_with_indent.join(line_ending);
1812
1813 // Preserve trailing newline
1814 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
1815 format!("{reflowed_text}{line_ending}")
1816 } else {
1817 reflowed_text
1818 };
1819
1820 // Only generate a warning if the replacement is different
1821 let original_text = &ctx.content[byte_range.clone()];
1822 if original_text != replacement {
1823 warnings.push(LintWarning {
1824 rule_name: Some(self.name().to_string()),
1825 message: format!(
1826 "Line length {} exceeds {} characters (in MkDocs container)",
1827 container_lines.iter().map(|l| l.len()).max().unwrap_or(0),
1828 config.line_length.get()
1829 ),
1830 line: container_start + 1,
1831 column: 1,
1832 end_line: end_line + 1,
1833 end_column: lines[end_line].chars().count() + 1,
1834 severity: Severity::Warning,
1835 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1836 });
1837 }
1838 continue;
1839 }
1840
1841 // Helper function to detect semantic line markers
1842 let is_semantic_line = |content: &str| -> bool {
1843 let trimmed = content.trim_start();
1844 let semantic_markers = [
1845 "NOTE:",
1846 "WARNING:",
1847 "IMPORTANT:",
1848 "CAUTION:",
1849 "TIP:",
1850 "DANGER:",
1851 "HINT:",
1852 "INFO:",
1853 ];
1854 semantic_markers.iter().any(|marker| trimmed.starts_with(marker))
1855 };
1856
1857 // Helper function to detect fence markers (opening or closing)
1858 let is_fence_marker = |content: &str| -> bool {
1859 let trimmed = content.trim_start();
1860 trimmed.starts_with("```") || trimmed.starts_with("~~~")
1861 };
1862
1863 // Check if this is a list item - handle it specially
1864 let trimmed = lines[i].trim();
1865 if is_list_item(trimmed) {
1866 // Collect the entire list item including continuation lines
1867 let list_start = i;
1868 let (marker, first_content) = extract_list_marker_and_content(lines[i]);
1869 let marker_len = marker.len();
1870
1871 // Checkbox ([ ]/[x]/[X]) is inline content, not part of the list marker.
1872 // Use the base bullet/number marker width for continuation recognition
1873 // so that continuation lines at 2+ spaces are collected for "- [ ] " items.
1874 let base_marker_len = if marker.contains("[ ] ") || marker.contains("[x] ") || marker.contains("[X] ") {
1875 marker.find('[').unwrap_or(marker_len)
1876 } else {
1877 marker_len
1878 };
1879
1880 // MkDocs flavor requires at least 4 spaces for list continuation
1881 // after a blank line (multi-paragraph list items). For non-blank
1882 // continuation (lines directly following the marker line), use
1883 // the natural marker width so that 2-space indent is recognized.
1884 let item_indent = ctx.lines[i].indent;
1885 let min_continuation_indent = if ctx.flavor.requires_strict_list_indent() {
1886 // Use 4-space relative indent from the list item's nesting level
1887 item_indent + (base_marker_len - item_indent).max(4)
1888 } else {
1889 marker_len
1890 };
1891 let content_continuation_indent = base_marker_len;
1892
1893 // Track lines and their types (content, code block, fence, nested list)
1894 #[derive(Clone)]
1895 enum LineType {
1896 Content(String),
1897 CodeBlock(String, usize), // content and original indent
1898 SemanticLine(String), // Lines starting with NOTE:, WARNING:, etc that should stay separate
1899 SnippetLine(String), // MkDocs Snippets delimiters (-8<-) that must stay on their own line
1900 DivMarker(String), // Quarto/Pandoc div markers (::: opening or closing)
1901 AdmonitionHeader(String, usize), // header text (e.g. "!!! note") and original indent
1902 AdmonitionContent(String, usize), // body content text and original indent
1903 Table(String, usize), // GFM table row, preserved verbatim with original indent
1904 Empty,
1905 }
1906
1907 let mut list_item_lines: Vec<LineType> = vec![LineType::Content(first_content)];
1908 i += 1;
1909
1910 // Collect continuation lines using ctx.lines for metadata
1911 while i < lines.len() {
1912 let line_info = &ctx.lines[i];
1913
1914 // Use pre-computed is_blank from ctx
1915 if line_info.is_blank {
1916 // Empty line - check if next line is indented (part of list item)
1917 if i + 1 < lines.len() {
1918 let next_info = &ctx.lines[i + 1];
1919
1920 // Check if next line is indented enough to be continuation
1921 if !next_info.is_blank && next_info.indent >= min_continuation_indent {
1922 // This blank line is between paragraphs/blocks in the list item
1923 list_item_lines.push(LineType::Empty);
1924 i += 1;
1925 continue;
1926 }
1927 }
1928 // No indented line after blank, end of list item
1929 break;
1930 }
1931
1932 // Use pre-computed indent from ctx
1933 let indent = line_info.indent;
1934
1935 // Valid continuation must be indented at least content_continuation_indent.
1936 // For non-blank continuation, use marker_len (e.g. 2 for "- ").
1937 // MkDocs strict 4-space requirement applies only after blank lines.
1938 if indent >= content_continuation_indent {
1939 let trimmed = line_info.content(ctx.content).trim();
1940
1941 // Check for MkDocs admonition lines inside list items BEFORE
1942 // checking in_code_block. Lines inside code blocks within
1943 // admonitions have both in_admonition and in_code_block set;
1944 // admonition membership takes priority so the entire admonition
1945 // structure (including embedded code blocks) is preserved.
1946 if line_info.in_admonition {
1947 let raw_content = line_info.content(ctx.content);
1948 if mkdocs_admonitions::is_admonition_start(raw_content) {
1949 let header_text = raw_content[indent..].trim_end().to_string();
1950 list_item_lines.push(LineType::AdmonitionHeader(header_text, indent));
1951 } else {
1952 let body_text = raw_content[indent..].trim_end().to_string();
1953 list_item_lines.push(LineType::AdmonitionContent(body_text, indent));
1954 }
1955 i += 1;
1956 continue;
1957 }
1958
1959 // Use pre-computed in_code_block from ctx
1960 if line_info.in_code_block {
1961 list_item_lines.push(LineType::CodeBlock(
1962 line_info.content(ctx.content)[indent..].to_string(),
1963 indent,
1964 ));
1965 i += 1;
1966 continue;
1967 }
1968
1969 // A blockquote nested inside the list item is reflowed by the
1970 // blockquote-aware path (it preserves the `>` prefix, including the
1971 // list indent), not as list-item prose. Collecting it as Content
1972 // would strip the markers and reflow `>` as words, collapsing the
1973 // blank `>` line and dropping `>` from wrapped continuations. End
1974 // the item here so the outer loop routes the blockquote line to
1975 // generate_blockquote_paragraph_fix. Uncollect a pending blank so
1976 // the separator between the list prose and the blockquote survives.
1977 if line_info.blockquote.is_some() {
1978 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
1979 list_item_lines.pop();
1980 i -= 1;
1981 }
1982 break;
1983 }
1984
1985 // Check if this is a SIBLING list item (breaks parent)
1986 // Nested lists are indented >= marker_len and are PART of the parent item
1987 // Siblings are at indent < marker_len (at or before parent marker)
1988 if is_list_item(trimmed) && indent < marker_len {
1989 // This is a sibling item at same or higher level - end parent item
1990 break;
1991 }
1992
1993 // Nested list items are always processed independently
1994 // by the outer loop, so break when we encounter one.
1995 // If a blank line was collected before this, uncollect it
1996 // so the outer loop preserves the blank between parent and nested.
1997 if is_list_item(trimmed) && indent >= marker_len {
1998 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
1999 list_item_lines.pop();
2000 i -= 1;
2001 }
2002 break;
2003 }
2004
2005 // Normal continuation vs indented code block.
2006 // Use min_continuation_indent for the threshold since
2007 // code blocks start 4 spaces beyond the expected content
2008 // level (which is min_continuation_indent for MkDocs).
2009 if indent <= min_continuation_indent + 3 {
2010 // Extract content (remove indentation and trailing whitespace)
2011 // Preserve hard breaks (2 trailing spaces) while removing excessive whitespace
2012 // See: https://github.com/rvben/rumdl/issues/76
2013 let content = trim_preserving_hard_break(&line_info.content(ctx.content)[indent..]);
2014
2015 // Check if this is a div marker (::: opening or closing)
2016 // These must be preserved on their own line, not merged into paragraphs
2017 if line_info.is_div_marker {
2018 list_item_lines.push(LineType::DivMarker(content));
2019 }
2020 // Check if this is a fence marker (opening or closing)
2021 // These should be treated as code block lines, not paragraph content
2022 else if is_fence_marker(&content) {
2023 list_item_lines.push(LineType::CodeBlock(content, indent));
2024 }
2025 // Check if this is a semantic line (NOTE:, WARNING:, etc.)
2026 else if is_semantic_line(&content) {
2027 list_item_lines.push(LineType::SemanticLine(content));
2028 }
2029 // Check if this is a snippet block delimiter (-8<- or --8<--)
2030 // These must be preserved on their own lines for MkDocs Snippets extension
2031 else if is_snippet_block_delimiter(&content) {
2032 list_item_lines.push(LineType::SnippetLine(content));
2033 }
2034 // Check if this is a GFM table row. Tables nested inside list
2035 // items must be preserved verbatim — joining them with prose
2036 // breaks the column structure.
2037 //
2038 // `is_potential_table_row` is intentionally permissive at the
2039 // row level: any line with `|` and 2+ cells qualifies. To avoid
2040 // misclassifying prose continuation lines that contain a literal
2041 // pipe (e.g. "use grep | sort to ..."), require one of:
2042 // - the row is pipe-bordered (`| ... |`), the canonical form
2043 // for tables nested in lists; or
2044 // - the next line is a delimiter row (this is a header); or
2045 // - the previous classified line was already a Table (this is
2046 // a continuation row).
2047 else if TableUtils::is_potential_table_row(&content) && {
2048 let pipe_bordered = content.trim().starts_with('|') && content.trim().ends_with('|');
2049 let next_is_delim = ctx
2050 .lines
2051 .get(i + 1)
2052 .is_some_and(|next| TableUtils::is_delimiter_row(next.content(ctx.content)));
2053 let prev_was_table = matches!(list_item_lines.last(), Some(LineType::Table(..)));
2054 pipe_bordered || next_is_delim || prev_was_table
2055 } {
2056 list_item_lines.push(LineType::Table(content, indent));
2057 } else {
2058 list_item_lines.push(LineType::Content(content));
2059 }
2060 i += 1;
2061 } else {
2062 // indent >= min_continuation_indent + 4: indented code block
2063 list_item_lines.push(LineType::CodeBlock(
2064 line_info.content(ctx.content)[indent..].to_string(),
2065 indent,
2066 ));
2067 i += 1;
2068 }
2069 } else {
2070 // Not indented enough, end of list item
2071 break;
2072 }
2073 }
2074
2075 // Determine the output continuation indent.
2076 // Normalize/Default modes canonicalize to min_continuation_indent
2077 // (fixing over-indented continuation). Semantic/SentencePerLine
2078 // modes preserve the user's actual indent since they only fix
2079 // line breaking, not indentation.
2080 let indent_size = match config.reflow_mode {
2081 ReflowMode::SemanticLineBreaks | ReflowMode::SentencePerLine => {
2082 // Find indent of the first plain text continuation line,
2083 // skipping the marker line (index 0), nested list items,
2084 // code blocks, and blank lines.
2085 list_item_lines
2086 .iter()
2087 .enumerate()
2088 .skip(1)
2089 .find_map(|(k, lt)| {
2090 if matches!(lt, LineType::Content(_)) {
2091 Some(ctx.lines[list_start + k].indent)
2092 } else {
2093 None
2094 }
2095 })
2096 .unwrap_or(min_continuation_indent)
2097 }
2098 _ => min_continuation_indent,
2099 };
2100 // For checkbox items in mkdocs flavor, enforce minimum indent so
2101 // continuation lines use the structural list indent (4), not the
2102 // content-aligned indent (6) which Python-Markdown doesn't support
2103 let has_checkbox = base_marker_len < marker_len;
2104 let indent_size = if has_checkbox && ctx.flavor.requires_strict_list_indent() {
2105 indent_size.max(min_continuation_indent)
2106 } else {
2107 indent_size
2108 };
2109 let expected_indent = " ".repeat(indent_size);
2110
2111 // Split list_item_lines into blocks (paragraphs, code blocks, nested lists, semantic lines, and HTML blocks)
2112 let mut builder = BlockBuilder::new();
2113 for line in &list_item_lines {
2114 match line {
2115 LineType::Empty => builder.feed_blank_line(),
2116 LineType::Content(content) => builder.feed_content(content),
2117 LineType::CodeBlock(content, indent) => builder.feed_code_line(content, *indent),
2118 LineType::SemanticLine(content) => builder.feed_semantic_line(content),
2119 LineType::SnippetLine(content) => builder.feed_snippet_line(content),
2120 LineType::DivMarker(content) => builder.feed_div_marker(content),
2121 LineType::AdmonitionHeader(header_text, indent) => {
2122 builder.feed_admonition_header(header_text, *indent)
2123 }
2124 LineType::AdmonitionContent(content, indent) => {
2125 builder.feed_admonition_content(content, *indent)
2126 }
2127 LineType::Table(content, indent) => builder.feed_table_line(content, *indent),
2128 }
2129 }
2130 let blocks = builder.finalize();
2131
2132 // Helper: check if a line (raw source or stripped content) is exempt
2133 // from line-length checks. Link reference definitions are always exempt;
2134 // standalone link/image lines are exempt when strict mode is off.
2135 // Also checks content after stripping list markers, since list item
2136 // continuation lines may contain link ref defs.
2137 let is_exempt_line = |raw_line: &str| -> bool {
2138 let trimmed = raw_line.trim();
2139 // Link reference definitions: always exempt
2140 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
2141 return true;
2142 }
2143 // Also check after stripping list markers (for list item content)
2144 if is_list_item(trimmed) {
2145 let (_, content) = extract_list_marker_and_content(trimmed);
2146 let content_trimmed = content.trim();
2147 if content_trimmed.starts_with('[')
2148 && content_trimmed.contains("]:")
2149 && LINK_REF_PATTERN.is_match(content_trimmed)
2150 {
2151 return true;
2152 }
2153 }
2154 // Standalone link/image lines: exempt when not strict
2155 if !config.strict && is_standalone_link_or_image_line(raw_line) {
2156 return true;
2157 }
2158 // HTML-only lines: exempt when not strict
2159 if !config.strict && is_html_only_line(raw_line) {
2160 return true;
2161 }
2162 false
2163 };
2164
2165 // Check if reflowing is needed (only for content paragraphs, not code blocks or nested lists)
2166 // Exclude link reference definitions and standalone link lines from content
2167 // so they don't pollute combined_content or trigger false reflow.
2168 let content_lines: Vec<String> = list_item_lines
2169 .iter()
2170 .filter_map(|line| {
2171 if let LineType::Content(s) = line {
2172 if is_exempt_line(s) {
2173 return None;
2174 }
2175 Some(s.clone())
2176 } else {
2177 None
2178 }
2179 })
2180 .collect();
2181
2182 // Check if we need to reflow this list item
2183 // We check the combined content to see if it exceeds length limits
2184 let combined_content = content_lines.join(" ").trim().to_string();
2185
2186 // Helper to check if we should reflow in normalize mode
2187 let should_normalize = || {
2188 // Don't normalize if the list item only contains nested lists, code blocks, or semantic lines
2189 // DO normalize if it has plain text content that spans multiple lines
2190 let has_code_blocks = blocks.iter().any(|b| matches!(b, Block::Code { .. }));
2191 let has_semantic_lines = blocks.iter().any(|b| matches!(b, Block::SemanticLine(_)));
2192 let has_snippet_lines = blocks.iter().any(|b| matches!(b, Block::SnippetLine(_)));
2193 let has_div_markers = blocks.iter().any(|b| matches!(b, Block::DivMarker(_)));
2194 let has_admonitions = blocks.iter().any(|b| matches!(b, Block::Admonition { .. }));
2195 let has_tables = blocks.iter().any(|b| matches!(b, Block::Table { .. }));
2196 let has_paragraphs = blocks.iter().any(|b| matches!(b, Block::Paragraph(_)));
2197
2198 // If we have structural blocks but no paragraphs, don't normalize
2199 if (has_code_blocks
2200 || has_semantic_lines
2201 || has_snippet_lines
2202 || has_div_markers
2203 || has_admonitions
2204 || has_tables)
2205 && !has_paragraphs
2206 {
2207 return false;
2208 }
2209
2210 // If we have paragraphs, check if they span multiple lines or there are multiple blocks
2211 if has_paragraphs {
2212 // Count only paragraphs that contain at least one non-exempt line.
2213 // Paragraphs consisting entirely of link ref defs or standalone links
2214 // should not trigger normalization.
2215 let paragraph_count = blocks
2216 .iter()
2217 .filter(|b| {
2218 if let Block::Paragraph(para_lines) = b {
2219 !para_lines.iter().all(|line| is_exempt_line(line))
2220 } else {
2221 false
2222 }
2223 })
2224 .count();
2225 if paragraph_count > 1 {
2226 // Multiple non-exempt paragraph blocks should be normalized
2227 return true;
2228 }
2229
2230 // Single paragraph block: normalize if it has multiple content lines
2231 if content_lines.len() > 1 {
2232 return true;
2233 }
2234 }
2235
2236 false
2237 };
2238
2239 let needs_reflow = match config.reflow_mode {
2240 ReflowMode::Normalize => {
2241 // Only reflow if:
2242 // 1. Any non-exempt paragraph, when joined, exceeds the limit, OR
2243 // 2. Any admonition content line exceeds the limit, OR
2244 // 3. The list item should be normalized (has multi-line plain text)
2245 let any_paragraph_exceeds = blocks.iter().any(|block| match block {
2246 Block::Paragraph(para_lines) => {
2247 if para_lines.iter().all(|line| is_exempt_line(line)) {
2248 return false;
2249 }
2250 let joined = para_lines.join(" ");
2251 let with_marker = format!("{}{}", " ".repeat(indent_size), joined.trim());
2252 self.calculate_effective_length(&with_marker) > config.line_length.get()
2253 }
2254 Block::Admonition {
2255 content_lines,
2256 header_indent,
2257 ..
2258 } => content_lines.iter().any(|(content, indent)| {
2259 if content.is_empty() {
2260 return false;
2261 }
2262 let with_indent = format!("{}{}", " ".repeat(*indent.max(header_indent)), content);
2263 self.calculate_effective_length(&with_indent) > config.line_length.get()
2264 }),
2265 _ => false,
2266 });
2267 if any_paragraph_exceeds {
2268 true
2269 } else {
2270 should_normalize()
2271 }
2272 }
2273 ReflowMode::SentencePerLine => {
2274 // Check if list item has multiple sentences
2275 let sentences = split_into_sentences(&combined_content);
2276 sentences.len() > 1
2277 }
2278 ReflowMode::SemanticLineBreaks => {
2279 let sentences = split_into_sentences(&combined_content);
2280 sentences.len() > 1
2281 || (list_start..i).any(|line_idx| {
2282 let line = lines[line_idx];
2283 let trimmed = line.trim();
2284 if trimmed.is_empty() || is_exempt_line(line) {
2285 return false;
2286 }
2287 self.calculate_effective_length(line) > config.line_length.get()
2288 })
2289 }
2290 ReflowMode::Default => {
2291 // In default mode, only reflow if any individual non-exempt line exceeds limit
2292 (list_start..i).any(|line_idx| {
2293 let line = lines[line_idx];
2294 let trimmed = line.trim();
2295 // Skip blank lines and exempt lines
2296 if trimmed.is_empty() || is_exempt_line(line) {
2297 return false;
2298 }
2299 self.calculate_effective_length(line) > config.line_length.get()
2300 })
2301 }
2302 };
2303
2304 if needs_reflow {
2305 let start_range = line_index.whole_line_range(list_start + 1);
2306 let end_line = i - 1;
2307 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
2308 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
2309 } else {
2310 line_index.whole_line_range(end_line + 1)
2311 };
2312 let byte_range = start_range.start..end_range.end;
2313
2314 // Reflow each block (paragraphs only, preserve code blocks)
2315 // When line_length = 0 (no limit), use a very large value for reflow
2316 let reflow_line_length = if config.line_length.is_unlimited() {
2317 usize::MAX
2318 } else {
2319 config.line_length.get().saturating_sub(indent_size).max(1)
2320 };
2321 let reflow_options = crate::utils::text_reflow::ReflowOptions {
2322 line_length: reflow_line_length,
2323 break_on_sentences: true,
2324 preserve_breaks: false,
2325 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
2326 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
2327 abbreviations: config.abbreviations_for_reflow(),
2328 length_mode: self.reflow_length_mode(),
2329 attr_lists: ctx.flavor.supports_attr_lists(),
2330 myst_roles: ctx.flavor.supports_myst_roles(),
2331 require_sentence_capital: config.require_sentence_capital,
2332 max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
2333 Some(4)
2334 } else {
2335 None
2336 },
2337 };
2338
2339 let mut result: Vec<String> = Vec::new();
2340 let mut is_first_block = true;
2341
2342 for (block_idx, block) in blocks.iter().enumerate() {
2343 match block {
2344 Block::Paragraph(para_lines) => {
2345 // If every line in this paragraph is exempt (link ref defs,
2346 // standalone links), preserve the paragraph verbatim instead
2347 // of reflowing it. Reflowing would corrupt link ref defs.
2348 let all_exempt = para_lines.iter().all(|line| is_exempt_line(line));
2349
2350 if all_exempt {
2351 for (idx, line) in para_lines.iter().enumerate() {
2352 if is_first_block && idx == 0 {
2353 result.push(format!("{marker}{line}"));
2354 is_first_block = false;
2355 } else {
2356 result.push(format!("{expected_indent}{line}"));
2357 }
2358 }
2359 } else {
2360 // Split the paragraph into segments at hard break boundaries
2361 // Each segment can be reflowed independently
2362 let segments = split_into_segments(para_lines);
2363
2364 for (segment_idx, segment) in segments.iter().enumerate() {
2365 // Check if this segment ends with a hard break and what type
2366 let hard_break_type = segment.last().and_then(|line| {
2367 let line = line.strip_suffix('\r').unwrap_or(line);
2368 if line.ends_with('\\') {
2369 Some("\\")
2370 } else if line.ends_with(" ") {
2371 Some(" ")
2372 } else {
2373 None
2374 }
2375 });
2376
2377 // Join and reflow the segment (removing the hard break marker for processing)
2378 let segment_for_reflow: Vec<String> = segment
2379 .iter()
2380 .map(|line| {
2381 // Strip hard break marker (2 spaces or backslash) for reflow processing
2382 if line.ends_with('\\') {
2383 line[..line.len() - 1].trim_end().to_string()
2384 } else if line.ends_with(" ") {
2385 line[..line.len() - 2].trim_end().to_string()
2386 } else {
2387 line.clone()
2388 }
2389 })
2390 .collect();
2391
2392 let segment_text = segment_for_reflow.join(" ").trim().to_string();
2393 if !segment_text.is_empty() {
2394 let reflowed =
2395 crate::utils::text_reflow::reflow_line(&segment_text, &reflow_options);
2396
2397 if is_first_block && segment_idx == 0 {
2398 // First segment of first block starts with marker
2399 result.push(format!("{marker}{}", reflowed[0]));
2400 for line in reflowed.iter().skip(1) {
2401 result.push(format!("{expected_indent}{line}"));
2402 }
2403 is_first_block = false;
2404 } else {
2405 // Subsequent segments
2406 for line in reflowed {
2407 result.push(format!("{expected_indent}{line}"));
2408 }
2409 }
2410
2411 // If this segment had a hard break, add it back to the last line
2412 // Preserve the original hard break format (backslash or two spaces)
2413 if let Some(break_marker) = hard_break_type
2414 && let Some(last_line) = result.last_mut()
2415 {
2416 last_line.push_str(break_marker);
2417 }
2418 }
2419 }
2420 }
2421
2422 // Add blank line after paragraph block if there's a next block.
2423 // Check if next block is a code block that doesn't want a preceding blank.
2424 // Also don't add blank lines before snippet lines (they should stay tight).
2425 // Only add if not already ending with one (avoids double blanks).
2426 if block_idx < blocks.len() - 1 {
2427 let next_block = &blocks[block_idx + 1];
2428 let should_add_blank = match next_block {
2429 Block::Code {
2430 has_preceding_blank, ..
2431 } => *has_preceding_blank,
2432 Block::Table {
2433 has_preceding_blank, ..
2434 } => *has_preceding_blank,
2435 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2436 _ => true, // For all other blocks, add blank line
2437 };
2438 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2439 result.push(String::new());
2440 }
2441 }
2442 }
2443 Block::Code {
2444 lines: code_lines,
2445 has_preceding_blank: _,
2446 } => {
2447 // Preserve code blocks as-is with original indentation
2448 // NOTE: Blank line before code block is handled by the previous block
2449 // (see paragraph block's logic above)
2450
2451 for (idx, (content, orig_indent)) in code_lines.iter().enumerate() {
2452 if is_first_block && idx == 0 {
2453 // First line of first block gets marker
2454 result.push(format!(
2455 "{marker}{}",
2456 " ".repeat(orig_indent - marker_len) + content
2457 ));
2458 is_first_block = false;
2459 } else if content.is_empty() {
2460 result.push(String::new());
2461 } else {
2462 result.push(format!("{}{}", " ".repeat(*orig_indent), content));
2463 }
2464 }
2465 }
2466 Block::SemanticLine(content) => {
2467 // Preserve semantic lines (NOTE:, WARNING:, etc.) as-is on their own line.
2468 // Only add blank before if not already ending with one.
2469 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
2470 result.push(String::new());
2471 }
2472
2473 if is_first_block {
2474 // First block starts with marker
2475 result.push(format!("{marker}{content}"));
2476 is_first_block = false;
2477 } else {
2478 // Subsequent blocks use expected indent
2479 result.push(format!("{expected_indent}{content}"));
2480 }
2481
2482 // Add blank line after semantic line if there's a next block.
2483 // Only add if not already ending with one.
2484 if block_idx < blocks.len() - 1 {
2485 let next_block = &blocks[block_idx + 1];
2486 let should_add_blank = match next_block {
2487 Block::Code {
2488 has_preceding_blank, ..
2489 } => *has_preceding_blank,
2490 Block::Table {
2491 has_preceding_blank, ..
2492 } => *has_preceding_blank,
2493 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2494 _ => true, // For all other blocks, add blank line
2495 };
2496 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2497 result.push(String::new());
2498 }
2499 }
2500 }
2501 Block::SnippetLine(content) => {
2502 // Preserve snippet delimiters (-8<-) as-is on their own line
2503 // Unlike semantic lines, snippet lines don't add extra blank lines
2504 if is_first_block {
2505 // First block starts with marker
2506 result.push(format!("{marker}{content}"));
2507 is_first_block = false;
2508 } else {
2509 // Subsequent blocks use expected indent
2510 result.push(format!("{expected_indent}{content}"));
2511 }
2512 // No blank lines added before or after snippet delimiters
2513 }
2514 Block::DivMarker(content) => {
2515 // Preserve div markers (::: opening or closing) as-is on their own line
2516 if is_first_block {
2517 result.push(format!("{marker}{content}"));
2518 is_first_block = false;
2519 } else {
2520 result.push(format!("{expected_indent}{content}"));
2521 }
2522 }
2523 Block::Html {
2524 lines: html_lines,
2525 has_preceding_blank: _,
2526 } => {
2527 // Preserve HTML blocks exactly as-is with original indentation
2528 // NOTE: Blank line before HTML block is handled by the previous block
2529
2530 for (idx, line) in html_lines.iter().enumerate() {
2531 if is_first_block && idx == 0 {
2532 // First line of first block gets marker
2533 result.push(format!("{marker}{line}"));
2534 is_first_block = false;
2535 } else if line.is_empty() {
2536 // Preserve blank lines inside HTML blocks
2537 result.push(String::new());
2538 } else {
2539 // Preserve lines with their original content (already includes indentation)
2540 result.push(format!("{expected_indent}{line}"));
2541 }
2542 }
2543
2544 // Add blank line after HTML block if there's a next block.
2545 // Only add if not already ending with one (avoids double blanks
2546 // when the HTML block itself contained a trailing blank line).
2547 if block_idx < blocks.len() - 1 {
2548 let next_block = &blocks[block_idx + 1];
2549 let should_add_blank = match next_block {
2550 Block::Code {
2551 has_preceding_blank, ..
2552 } => *has_preceding_blank,
2553 Block::Html {
2554 has_preceding_blank, ..
2555 } => *has_preceding_blank,
2556 Block::Table {
2557 has_preceding_blank, ..
2558 } => *has_preceding_blank,
2559 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2560 _ => true, // For all other blocks, add blank line
2561 };
2562 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2563 result.push(String::new());
2564 }
2565 }
2566 }
2567 Block::Table {
2568 lines: table_lines,
2569 has_preceding_blank: _,
2570 } => {
2571 // Preserve table rows verbatim with their original indentation.
2572 // Reflowing rows would corrupt column alignment and inject `|`
2573 // characters mid-paragraph (issue #590).
2574 // The leading blank line is emitted by the previous block.
2575 for (idx, (content, orig_indent)) in table_lines.iter().enumerate() {
2576 if is_first_block && idx == 0 {
2577 // First line of first block gets the list marker
2578 result.push(format!(
2579 "{marker}{}",
2580 " ".repeat(orig_indent.saturating_sub(marker_len)) + content
2581 ));
2582 is_first_block = false;
2583 } else {
2584 result.push(format!("{}{}", " ".repeat(*orig_indent), content));
2585 }
2586 }
2587
2588 // Add blank line after table block if there's a next block.
2589 if block_idx < blocks.len() - 1 {
2590 let next_block = &blocks[block_idx + 1];
2591 let should_add_blank = match next_block {
2592 Block::Code {
2593 has_preceding_blank, ..
2594 } => *has_preceding_blank,
2595 Block::Html {
2596 has_preceding_blank, ..
2597 } => *has_preceding_blank,
2598 Block::Table {
2599 has_preceding_blank, ..
2600 } => *has_preceding_blank,
2601 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2602 _ => true,
2603 };
2604 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2605 result.push(String::new());
2606 }
2607 }
2608 }
2609 Block::Admonition {
2610 header,
2611 header_indent,
2612 content_lines: admon_lines,
2613 } => {
2614 // Reconstruct admonition block with header at original indent
2615 // and body content reflowed to fit within the line length limit
2616
2617 // Add blank line before admonition if not first block
2618 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
2619 result.push(String::new());
2620 }
2621
2622 // Output the header at its original indent
2623 let header_indent_str = " ".repeat(*header_indent);
2624 if is_first_block {
2625 result.push(format!(
2626 "{marker}{}",
2627 " ".repeat(header_indent.saturating_sub(marker_len)) + header
2628 ));
2629 is_first_block = false;
2630 } else {
2631 result.push(format!("{header_indent_str}{header}"));
2632 }
2633
2634 // Derive body indent from the first non-empty content line's
2635 // stored indent, falling back to header_indent + 4 for
2636 // empty-body admonitions
2637 let body_indent = admon_lines
2638 .iter()
2639 .find(|(content, _)| !content.is_empty())
2640 .map_or(header_indent + 4, |(_, indent)| *indent);
2641 let body_indent_str = " ".repeat(body_indent);
2642
2643 // Segment body content into code blocks (verbatim) and
2644 // text paragraphs (reflowable), separated by blank lines.
2645 // Code lines store (content, orig_indent) to reconstruct
2646 // internal indentation relative to body_indent.
2647 enum AdmonSegment {
2648 Text(Vec<String>),
2649 Code(Vec<(String, usize)>),
2650 }
2651
2652 let mut segments: Vec<AdmonSegment> = Vec::new();
2653 let mut current_text: Vec<String> = Vec::new();
2654 let mut current_code: Vec<(String, usize)> = Vec::new();
2655 let mut in_admon_code = false;
2656 // Track the opening fence character so closing fences
2657 // must match (backticks close backticks, tildes close tildes)
2658 let mut fence_char: char = '`';
2659
2660 // Opening fences: ``` or ~~~ followed by optional info string
2661 let get_opening_fence = |s: &str| -> Option<(char, usize)> {
2662 let t = s.trim_start();
2663 if t.starts_with("```") {
2664 Some(('`', t.bytes().take_while(|&b| b == b'`').count()))
2665 } else if t.starts_with("~~~") {
2666 Some(('~', t.bytes().take_while(|&b| b == b'~').count()))
2667 } else {
2668 None
2669 }
2670 };
2671 // Closing fences: ONLY fence chars + optional trailing spaces
2672 let get_closing_fence = |s: &str| -> Option<(char, usize)> {
2673 let t = s.trim();
2674 if t.starts_with("```") && t.bytes().all(|b| b == b'`') {
2675 Some(('`', t.len()))
2676 } else if t.starts_with("~~~") && t.bytes().all(|b| b == b'~') {
2677 Some(('~', t.len()))
2678 } else {
2679 None
2680 }
2681 };
2682 let mut fence_len: usize = 3;
2683
2684 for (content, orig_indent) in admon_lines {
2685 if in_admon_code {
2686 // Closing fence must use the same character, be
2687 // at least as long, and have no info string
2688 if let Some((ch, len)) = get_closing_fence(content)
2689 && ch == fence_char
2690 && len >= fence_len
2691 {
2692 current_code.push((content.clone(), *orig_indent));
2693 in_admon_code = false;
2694 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
2695 continue;
2696 }
2697 current_code.push((content.clone(), *orig_indent));
2698 } else if let Some((ch, len)) = get_opening_fence(content) {
2699 if !current_text.is_empty() {
2700 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
2701 }
2702 in_admon_code = true;
2703 fence_char = ch;
2704 fence_len = len;
2705 current_code.push((content.clone(), *orig_indent));
2706 } else if content.is_empty() {
2707 if !current_text.is_empty() {
2708 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
2709 }
2710 } else {
2711 current_text.push(content.clone());
2712 }
2713 }
2714 if in_admon_code && !current_code.is_empty() {
2715 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
2716 }
2717 if !current_text.is_empty() {
2718 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
2719 }
2720
2721 // Build reflow options once for all text segments
2722 let admon_reflow_length = if config.line_length.is_unlimited() {
2723 usize::MAX
2724 } else {
2725 config.line_length.get().saturating_sub(body_indent).max(1)
2726 };
2727
2728 let admon_reflow_options = crate::utils::text_reflow::ReflowOptions {
2729 line_length: admon_reflow_length,
2730 break_on_sentences: true,
2731 preserve_breaks: false,
2732 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
2733 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
2734 abbreviations: config.abbreviations_for_reflow(),
2735 length_mode: self.reflow_length_mode(),
2736 attr_lists: ctx.flavor.supports_attr_lists(),
2737 myst_roles: ctx.flavor.supports_myst_roles(),
2738 require_sentence_capital: config.require_sentence_capital,
2739 max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
2740 Some(4)
2741 } else {
2742 None
2743 },
2744 };
2745
2746 // Output each segment
2747 for segment in &segments {
2748 // Blank line before each segment (after the header or previous segment)
2749 result.push(String::new());
2750
2751 match segment {
2752 AdmonSegment::Code(lines) => {
2753 for (line, orig_indent) in lines {
2754 if line.is_empty() {
2755 // Preserve blank lines inside code blocks
2756 result.push(String::new());
2757 } else {
2758 // Reconstruct with body_indent + any extra
2759 // indentation the line had beyond body_indent
2760 let extra = orig_indent.saturating_sub(body_indent);
2761 let indent_str = " ".repeat(body_indent + extra);
2762 result.push(format!("{indent_str}{line}"));
2763 }
2764 }
2765 }
2766 AdmonSegment::Text(lines) => {
2767 let paragraph_text = lines.join(" ").trim().to_string();
2768 if paragraph_text.is_empty() {
2769 continue;
2770 }
2771 let reflowed = crate::utils::text_reflow::reflow_line(
2772 ¶graph_text,
2773 &admon_reflow_options,
2774 );
2775 for line in &reflowed {
2776 result.push(format!("{body_indent_str}{line}"));
2777 }
2778 }
2779 }
2780 }
2781
2782 // Add blank line after admonition if there's a next block
2783 if block_idx < blocks.len() - 1 {
2784 let next_block = &blocks[block_idx + 1];
2785 let should_add_blank = match next_block {
2786 Block::Code {
2787 has_preceding_blank, ..
2788 } => *has_preceding_blank,
2789 Block::Table {
2790 has_preceding_blank, ..
2791 } => *has_preceding_blank,
2792 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2793 _ => true,
2794 };
2795 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2796 result.push(String::new());
2797 }
2798 }
2799 }
2800 }
2801 }
2802
2803 let reflowed_text = result.join(line_ending);
2804
2805 // Preserve trailing newline
2806 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
2807 format!("{reflowed_text}{line_ending}")
2808 } else {
2809 reflowed_text
2810 };
2811
2812 // Get the original text to compare
2813 let original_text = &ctx.content[byte_range.clone()];
2814
2815 // Physical-line-length scan, shared by the Normalize-mode gate and its
2816 // message. The list-item reflow preserves code blocks, HTML blocks,
2817 // admonition headers, fence markers, semantic markers, and snippet/div
2818 // markers verbatim; only paragraph content and admonition bodies are
2819 // restructured. Only those lines drive the length warning, so that
2820 // preserved-but-overlong content does not keep the paragraph-level
2821 // warning alive when the reflow would not fix that line.
2822 let should_count_for_length = |line_idx: usize| -> bool {
2823 let line = lines[line_idx];
2824 let trimmed = line.trim();
2825 if trimmed.is_empty() || is_exempt_line(line) {
2826 return false;
2827 }
2828 let info = &ctx.lines[line_idx];
2829 if info.in_code_block || info.in_html_block {
2830 return false;
2831 }
2832 if info.in_admonition && mkdocs_admonitions::is_admonition_start(line) {
2833 return false;
2834 }
2835 if is_fence_marker(line) || is_semantic_line(line) {
2836 return false;
2837 }
2838 if is_snippet_block_delimiter(line) {
2839 return false;
2840 }
2841 if line.trim_start().starts_with(":::") {
2842 return false;
2843 }
2844 true
2845 };
2846 let max_physical_length = (list_start..i)
2847 .filter(|&idx| should_count_for_length(idx))
2848 .map(|idx| self.calculate_effective_length(lines[idx]))
2849 .max()
2850 .unwrap_or(0);
2851 // `line-length = 0` means "no limit", so no physical line can be
2852 // "over"; the message below then describes a structural join rather
2853 // than a length violation.
2854 let any_paragraph_line_over =
2855 !config.line_length.is_unlimited() && max_physical_length > config.line_length.get();
2856
2857 // Normalize mode reflows list-item prose just like paragraphs:
2858 // joining continuation lines and re-wrapping to `line-length`.
2859 // `prose_changed` is true only when the reflow alters the words or
2860 // line breaks, not when it would merely re-indent continuation
2861 // lines or trim trailing whitespace. Comparing the texts with each
2862 // line's leading and trailing whitespace removed isolates "did the
2863 // words/line breaks change" from "did the surrounding whitespace
2864 // change". Continuation indentation is MD077's responsibility and
2865 // trailing whitespace is MD009's; an MD013 warning for either would
2866 // both duplicate those rules and resurface a persistent advisory on
2867 // already-fitting items that users disable MD013 fixing to avoid.
2868 let prose_changed = {
2869 let stripped = |text: &str| text.lines().map(str::trim).collect::<Vec<_>>().join("\n");
2870 stripped(original_text) != stripped(&replacement)
2871 };
2872 // Warn when the reflow rewraps prose (the normalize feature for
2873 // list items), or when a physical line genuinely exceeds the limit
2874 // and the reflow can change something (a true length violation,
2875 // even if all that changes is the continuation indent). A line that
2876 // is already optimal in both respects produces no warning.
2877 let gate_ok = prose_changed || (any_paragraph_line_over && original_text != replacement);
2878 if gate_ok {
2879 // Generate an appropriate message based on why reflow is needed
2880 let message = match config.reflow_mode {
2881 ReflowMode::SentencePerLine => {
2882 let num_sentences = split_into_sentences(&combined_content).len();
2883 let num_lines = content_lines.len();
2884 if num_lines == 1 {
2885 // Single line with multiple sentences
2886 format!("Line contains {num_sentences} sentences (one sentence per line required)")
2887 } else {
2888 // Multiple lines - could be split sentences or mixed
2889 format!(
2890 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
2891 )
2892 }
2893 }
2894 ReflowMode::SemanticLineBreaks => {
2895 let num_sentences = split_into_sentences(&combined_content).len();
2896 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
2897 }
2898 ReflowMode::Normalize => {
2899 // When a physical line genuinely exceeds the limit, report
2900 // it as a length violation. Otherwise the reflow is a
2901 // structural normalization (joining/re-wrapping multi-line
2902 // content that already fits), mirroring the paragraph path.
2903 if any_paragraph_line_over {
2904 format!(
2905 "Line length {} exceeds {} characters",
2906 max_physical_length,
2907 config.line_length.get()
2908 )
2909 } else {
2910 format!(
2911 "List item could be normalized to use line length of {} characters",
2912 config.line_length.get()
2913 )
2914 }
2915 }
2916 ReflowMode::Default => {
2917 // Report the actual longest non-exempt line, not the combined content
2918 let max_length = (list_start..i)
2919 .filter(|&line_idx| {
2920 let line = lines[line_idx];
2921 let trimmed = line.trim();
2922 !trimmed.is_empty() && !is_exempt_line(line)
2923 })
2924 .map(|line_idx| self.calculate_effective_length(lines[line_idx]))
2925 .max()
2926 .unwrap_or(0);
2927 format!(
2928 "Line length {} exceeds {} characters",
2929 max_length,
2930 config.line_length.get()
2931 )
2932 }
2933 };
2934
2935 warnings.push(LintWarning {
2936 rule_name: Some(self.name().to_string()),
2937 message,
2938 line: list_start + 1,
2939 column: 1,
2940 end_line: end_line + 1,
2941 end_column: lines[end_line].chars().count() + 1,
2942 severity: Severity::Warning,
2943 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
2944 });
2945 }
2946 }
2947 continue;
2948 }
2949
2950 // Found start of a paragraph - collect all lines in it
2951 let paragraph_start = i;
2952 let mut paragraph_lines = vec![lines[i]];
2953 i += 1;
2954
2955 while i < lines.len() {
2956 let next_line = lines[i];
2957 let next_line_num = i + 1;
2958 let next_trimmed = next_line.trim();
2959
2960 // Stop at paragraph boundaries
2961 if next_trimmed.is_empty()
2962 || ctx.line_info(next_line_num).is_some_and(|info| info.in_code_block)
2963 || ctx.line_info(next_line_num).is_some_and(|info| info.in_front_matter)
2964 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_block)
2965 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_comment)
2966 || ctx.line_info(next_line_num).is_some_and(|info| info.in_esm_block)
2967 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_expression)
2968 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_block)
2969 || ctx.line_info(next_line_num).is_some_and(|info| info.in_mdx_comment)
2970 || ctx
2971 .line_info(next_line_num)
2972 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
2973 || (next_line_num > 0
2974 && next_line_num <= ctx.lines.len()
2975 && ctx.lines[next_line_num - 1].blockquote.is_some())
2976 || next_trimmed.starts_with('#')
2977 || TableUtils::is_potential_table_row(next_line)
2978 || is_list_item(next_trimmed)
2979 || is_horizontal_rule(next_line)
2980 || (next_trimmed.starts_with('[') && next_line.contains("]:"))
2981 || is_template_directive_only(next_line)
2982 || is_standalone_attr_list(next_line)
2983 || is_snippet_block_delimiter(next_line)
2984 || ctx.line_info(next_line_num).is_some_and(|info| info.is_div_marker)
2985 || is_html_only_line(next_line)
2986 || (!config.strict && is_standalone_link_or_image_line(next_line))
2987 {
2988 break;
2989 }
2990
2991 // Check if the previous line ends with a hard break (2+ spaces or backslash)
2992 if i > 0 && has_hard_break(lines[i - 1]) {
2993 // Don't include lines after hard breaks in the same paragraph
2994 break;
2995 }
2996
2997 paragraph_lines.push(next_line);
2998 i += 1;
2999 }
3000
3001 // Compute the common leading indent of all non-empty paragraph lines,
3002 // but only when those lines are structurally inside a list block.
3003 // Indented continuation lines that follow a nested list arrive here
3004 // with their structural indentation intact (e.g. 2 spaces for a
3005 // top-level list item). Stripping the indent before reflow and
3006 // re-applying it afterward prevents the fixer from moving those
3007 // lines to column 0.
3008 //
3009 // The list-block guard is essential: top-level paragraphs that happen
3010 // to start with spaces (insignificant in Markdown) must NOT have those
3011 // spaces preserved or injected by the fixer.
3012 let common_indent: String = if ctx.is_in_list_block(paragraph_start + 1) {
3013 let min_len = paragraph_lines
3014 .iter()
3015 .filter(|l| !l.trim().is_empty())
3016 .map(|l| l.len() - l.trim_start().len())
3017 .min()
3018 .unwrap_or(0);
3019 paragraph_lines
3020 .iter()
3021 .find(|l| !l.trim().is_empty())
3022 .map(|l| l[..min_len].to_string())
3023 .unwrap_or_default()
3024 } else {
3025 String::new()
3026 };
3027
3028 // Combine paragraph lines into a single string for processing.
3029 // This must be done BEFORE the needs_reflow check for sentence-per-line mode.
3030 let paragraph_text = if common_indent.is_empty() {
3031 paragraph_lines.join(" ")
3032 } else {
3033 paragraph_lines
3034 .iter()
3035 .map(|l| {
3036 if l.starts_with(common_indent.as_str()) {
3037 &l[common_indent.len()..]
3038 } else {
3039 l.trim_start()
3040 }
3041 })
3042 .collect::<Vec<_>>()
3043 .join(" ")
3044 };
3045
3046 // Skip reflowing if this paragraph contains definition list items
3047 // Definition lists are multi-line structures that should not be joined
3048 let contains_definition_list = paragraph_lines
3049 .iter()
3050 .any(|line| crate::utils::is_definition_list_item(line));
3051
3052 if contains_definition_list {
3053 // Don't reflow definition lists - skip this paragraph
3054 i = paragraph_start + paragraph_lines.len();
3055 continue;
3056 }
3057
3058 // Skip reflowing if this paragraph contains MkDocs Snippets markers
3059 // Snippets blocks (-8<- ... -8<-) should be preserved exactly
3060 let contains_snippets = paragraph_lines.iter().any(|line| is_snippet_block_delimiter(line));
3061
3062 if contains_snippets {
3063 // Don't reflow Snippets blocks - skip this paragraph
3064 i = paragraph_start + paragraph_lines.len();
3065 continue;
3066 }
3067
3068 // Check if this paragraph needs reflowing
3069 let needs_reflow = match config.reflow_mode {
3070 ReflowMode::Normalize => self.normalize_mode_needs_reflow(paragraph_lines.iter().copied(), config),
3071 ReflowMode::SentencePerLine => {
3072 // In sentence-per-line mode, check if the JOINED paragraph has multiple sentences
3073 // Note: we check the joined text because sentences can span multiple lines
3074 let sentences = split_into_sentences(¶graph_text);
3075
3076 // Always reflow if multiple sentences on one line
3077 if sentences.len() > 1 {
3078 true
3079 } else if paragraph_lines.len() > 1 {
3080 // For single-sentence paragraphs spanning multiple lines:
3081 // Reflow if they COULD fit on one line (respecting line-length constraint)
3082 if config.line_length.is_unlimited() {
3083 // No line-length constraint - always join single sentences
3084 true
3085 } else {
3086 // Only join if it fits within line-length.
3087 // paragraph_text has the common indent stripped, so add it
3088 // back to get the true output length before comparing.
3089 let effective_length =
3090 self.calculate_effective_length(¶graph_text) + common_indent.len();
3091 effective_length <= config.line_length.get()
3092 }
3093 } else {
3094 false
3095 }
3096 }
3097 ReflowMode::SemanticLineBreaks => {
3098 let sentences = split_into_sentences(¶graph_text);
3099 // Reflow if multiple sentences, multiple lines, or any line exceeds limit
3100 sentences.len() > 1
3101 || paragraph_lines.len() > 1
3102 || paragraph_lines
3103 .iter()
3104 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3105 }
3106 ReflowMode::Default => {
3107 // In default mode, only reflow if lines exceed limit
3108 paragraph_lines
3109 .iter()
3110 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3111 }
3112 };
3113
3114 if needs_reflow {
3115 // Calculate byte range for this paragraph
3116 // Use whole_line_range for each line and combine
3117 let start_range = line_index.whole_line_range(paragraph_start + 1);
3118 let end_line = paragraph_start + paragraph_lines.len() - 1;
3119
3120 // For the last line, we want to preserve any trailing newline
3121 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
3122 // Last line without trailing newline - use line_text_range
3123 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
3124 } else {
3125 // Not the last line or has trailing newline - use whole_line_range
3126 line_index.whole_line_range(end_line + 1)
3127 };
3128
3129 let byte_range = start_range.start..end_range.end;
3130
3131 // Check if the paragraph ends with a hard break and what type
3132 let hard_break_type = paragraph_lines.last().and_then(|line| {
3133 let line = line.strip_suffix('\r').unwrap_or(line);
3134 if line.ends_with('\\') {
3135 Some("\\")
3136 } else if line.ends_with(" ") {
3137 Some(" ")
3138 } else {
3139 None
3140 }
3141 });
3142
3143 // Reflow the paragraph
3144 // When line_length = 0 (no limit), use a very large value for reflow
3145 let reflow_line_length = if config.line_length.is_unlimited() {
3146 usize::MAX
3147 } else {
3148 config.line_length.get()
3149 };
3150 let reflow_options = crate::utils::text_reflow::ReflowOptions {
3151 line_length: reflow_line_length,
3152 break_on_sentences: true,
3153 preserve_breaks: false,
3154 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
3155 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
3156 abbreviations: config.abbreviations_for_reflow(),
3157 length_mode: self.reflow_length_mode(),
3158 attr_lists: ctx.flavor.supports_attr_lists(),
3159 myst_roles: ctx.flavor.supports_myst_roles(),
3160 require_sentence_capital: config.require_sentence_capital,
3161 max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
3162 Some(4)
3163 } else {
3164 None
3165 },
3166 };
3167 let mut reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
3168
3169 // Re-apply the common indent to each non-empty reflowed line so
3170 // that the replacement preserves the original structural indentation.
3171 if !common_indent.is_empty() {
3172 for line in &mut reflowed {
3173 if !line.is_empty() {
3174 *line = format!("{common_indent}{line}");
3175 }
3176 }
3177 }
3178
3179 // If the original paragraph ended with a hard break, preserve it
3180 // Preserve the original hard break format (backslash or two spaces)
3181 if let Some(break_marker) = hard_break_type
3182 && !reflowed.is_empty()
3183 {
3184 let last_idx = reflowed.len() - 1;
3185 if !has_hard_break(&reflowed[last_idx]) {
3186 reflowed[last_idx].push_str(break_marker);
3187 }
3188 }
3189
3190 let reflowed_text = reflowed.join(line_ending);
3191
3192 // Preserve trailing newline if the original paragraph had one
3193 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
3194 format!("{reflowed_text}{line_ending}")
3195 } else {
3196 reflowed_text
3197 };
3198
3199 // Get the original text to compare
3200 let original_text = &ctx.content[byte_range.clone()];
3201
3202 // Only generate a warning if the replacement is different from the original
3203 if original_text != replacement {
3204 // Determine which line ranges and messages to report based on the reflow mode.
3205 let warnings_to_report: Vec<(usize, usize, String)> = match config.reflow_mode {
3206 ReflowMode::Default => {
3207 // In default mode, report a warning for *every* line in the paragraph
3208 // that exceeds the limit. Each warning will carry the same paragraph-level
3209 // fix, making all of them auto-fixable.
3210 paragraph_lines
3211 .iter()
3212 .enumerate()
3213 .filter(|(_, line)| self.calculate_effective_length(line) > config.line_length.get())
3214 .map(|(idx, _)| {
3215 let violating_line = paragraph_start + idx + 1;
3216 (
3217 violating_line,
3218 violating_line,
3219 format!("Line length exceeds {} characters", config.line_length.get()),
3220 )
3221 })
3222 .collect()
3223 }
3224 ReflowMode::Normalize => {
3225 // In normalize mode, report the whole paragraph as needing normalization.
3226 vec![(
3227 paragraph_start + 1,
3228 end_line + 1,
3229 format!(
3230 "Paragraph could be normalized to use line length of {} characters",
3231 config.line_length.get()
3232 ),
3233 )]
3234 }
3235 ReflowMode::SentencePerLine => {
3236 // In sentence-per-line mode, highlight the entire paragraph that needs reformatting.
3237 let num_sentences = split_into_sentences(¶graph_text).len();
3238 let message = if paragraph_lines.len() == 1 {
3239 // Single line with multiple sentences
3240 format!("Line contains {num_sentences} sentences (one sentence per line required)")
3241 } else {
3242 // Multiple lines - could be split sentences or mixed
3243 let num_lines = paragraph_lines.len();
3244 format!(
3245 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
3246 )
3247 };
3248 vec![(paragraph_start + 1, paragraph_start + paragraph_lines.len(), message)]
3249 }
3250 ReflowMode::SemanticLineBreaks => {
3251 // In semantic-line-breaks mode, highlight the entire paragraph.
3252 let num_sentences = split_into_sentences(¶graph_text).len();
3253 vec![(
3254 paragraph_start + 1,
3255 paragraph_start + paragraph_lines.len(),
3256 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)"),
3257 )]
3258 }
3259 };
3260
3261 // Generate the actual lint warnings. All warnings for this paragraph
3262 // share the same paragraph-level fix.
3263 for (w_start, w_end, msg) in warnings_to_report {
3264 warnings.push(LintWarning {
3265 rule_name: Some(self.name().to_string()),
3266 message: msg,
3267 line: w_start,
3268 column: 1,
3269 end_line: w_end,
3270 end_column: lines[w_end.saturating_sub(1)].chars().count() + 1,
3271 severity: Severity::Warning,
3272 fix: Some(crate::rule::Fix::new(byte_range.clone(), replacement.clone())),
3273 });
3274 }
3275 }
3276 }
3277 }
3278
3279 warnings
3280 }
3281
3282 /// Calculate string length based on the configured length mode
3283 fn calculate_string_length(&self, s: &str) -> usize {
3284 match self.config.length_mode {
3285 LengthMode::Chars => s.chars().count(),
3286 LengthMode::Visual => s.width(),
3287 LengthMode::Bytes => s.len(),
3288 }
3289 }
3290
3291 /// Calculate effective line length
3292 ///
3293 /// Returns the actual display length of the line using the configured length mode.
3294 fn calculate_effective_length(&self, line: &str) -> usize {
3295 self.calculate_string_length(line)
3296 }
3297
3298 /// Calculate line length with inline link/image URLs removed.
3299 ///
3300 /// For each inline link `[text](url)` or image `` on the line,
3301 /// computes the "savings" from removing the URL portion (keeping only `[text]`
3302 /// or `![alt]`). Returns `effective_length - total_savings`.
3303 ///
3304 /// Handles nested constructs (e.g., `[](url)`) by only counting the
3305 /// outermost construct to avoid double-counting.
3306 fn calculate_text_only_length(
3307 &self,
3308 effective_length: usize,
3309 line_number: usize,
3310 ctx: &crate::lint_context::LintContext,
3311 ) -> usize {
3312 let line_range = ctx.line_index.line_content_range(line_number);
3313 let line_byte_end = line_range.end;
3314
3315 // Collect inline links/images on this line: (byte_offset, byte_end, text_only_display_len)
3316 let mut constructs: Vec<(usize, usize, usize)> = Vec::new();
3317
3318 // Binary search: links are sorted by byte_offset, so link.line is non-decreasing
3319 let link_start = ctx.links.partition_point(|l| l.line < line_number);
3320 for link in &ctx.links[link_start..] {
3321 if link.line != line_number {
3322 break;
3323 }
3324 if link.is_reference {
3325 continue;
3326 }
3327 if !matches!(link.link_type, LinkType::Inline) {
3328 continue;
3329 }
3330 if link.byte_end > line_byte_end {
3331 continue;
3332 }
3333 let text_only_len = 2 + self.calculate_string_length(&link.text);
3334 constructs.push((link.byte_offset, link.byte_end, text_only_len));
3335 }
3336
3337 let img_start = ctx.images.partition_point(|i| i.line < line_number);
3338 for image in &ctx.images[img_start..] {
3339 if image.line != line_number {
3340 break;
3341 }
3342 if image.is_reference {
3343 continue;
3344 }
3345 if !matches!(image.link_type, LinkType::Inline) {
3346 continue;
3347 }
3348 if image.byte_end > line_byte_end {
3349 continue;
3350 }
3351 let text_only_len = 3 + self.calculate_string_length(&image.alt_text);
3352 constructs.push((image.byte_offset, image.byte_end, text_only_len));
3353 }
3354
3355 if constructs.is_empty() {
3356 return effective_length;
3357 }
3358
3359 // Sort by byte offset to handle overlapping/nested constructs
3360 constructs.sort_by_key(|&(start, _, _)| start);
3361
3362 let mut total_savings: usize = 0;
3363 let mut last_end: usize = 0;
3364
3365 for (start, end, text_only_len) in &constructs {
3366 // Skip constructs nested inside a previously counted one
3367 if *start < last_end {
3368 continue;
3369 }
3370 // Full construct length in configured length mode
3371 let full_source = &ctx.content[*start..*end];
3372 let full_len = self.calculate_string_length(full_source);
3373 total_savings += full_len.saturating_sub(*text_only_len);
3374 last_end = *end;
3375 }
3376
3377 effective_length.saturating_sub(total_savings)
3378 }
3379}