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