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, blockquote_continuation_style, dominant_blockquote_prefix, reflow_blockquote_content,
16 split_into_sentences,
17};
18use pulldown_cmark::LinkType;
19use toml;
20
21mod block_builder;
22mod helpers;
23pub mod md013_config;
24use crate::rules::md030_list_marker_space::MD030Config;
25use crate::utils::is_template_directive_only;
26use block_builder::{Block, BlockBuilder};
27use helpers::{
28 extract_list_marker_and_content, has_hard_break, is_github_alert_marker, is_horizontal_rule, is_html_only_line,
29 is_list_item, is_standalone_link_or_image_line, is_unwrappable_line, source_list_marker, split_into_segments,
30 trim_preserving_hard_break,
31};
32pub use md013_config::MD013Config;
33use md013_config::{LengthMode, ReflowMode};
34
35#[cfg(test)]
36mod tests;
37use unicode_width::UnicodeWidthStr;
38
39#[derive(Clone, Default)]
40pub struct MD013LineLength {
41 pub(crate) config: MD013Config,
42 /// MD030 list-marker spacing, applied when reflowing list items so the rewrite
43 /// uses the configured post-marker spacing rather than a hard-coded single
44 /// space. Defaults to MD030's defaults (a single space everywhere), which
45 /// reproduces the previous behaviour exactly. See [`MD030Config::expected_spaces`].
46 pub(crate) list_spacing: MD030Config,
47}
48
49/// Blockquote paragraph line collected for reflow, with original line index for range computation.
50struct CollectedBlockquoteLine {
51 line_idx: usize,
52 data: BlockquoteLineData,
53}
54
55impl MD013LineLength {
56 pub fn new(line_length: usize, code_blocks: bool, tables: bool, headings: bool, strict: bool) -> Self {
57 Self {
58 config: MD013Config {
59 line_length: crate::types::LineLength::new(line_length),
60 code_blocks,
61 code_spans: true,
62 tables,
63 headings,
64 paragraphs: true, // Default to true for backwards compatibility
65 blockquotes: true, // Default to true for backwards compatibility
66 strict,
67 stern: false,
68 heading_line_length: None,
69 code_block_line_length: None,
70 reflow: false,
71 reflow_mode: ReflowMode::default(),
72 length_mode: LengthMode::default(),
73 abbreviations: Vec::new(),
74 require_sentence_capital: true,
75 ignore_link_urls: true,
76 atomic_spans: true,
77 },
78 list_spacing: MD030Config::default(),
79 }
80 }
81
82 pub fn from_config_struct(config: MD013Config) -> Self {
83 Self {
84 config,
85 list_spacing: MD030Config::default(),
86 }
87 }
88
89 /// Return a clone with code block checking disabled.
90 /// Used for doc comment linting where code blocks are Rust code managed by rustfmt.
91 pub fn with_code_blocks_disabled(&self) -> Self {
92 let mut clone = self.clone();
93 clone.config.code_blocks = false;
94 clone
95 }
96
97 /// Convert MD013 LengthMode to text_reflow ReflowLengthMode
98 /// Normalized set of reference labels defined in the document.
99 ///
100 /// Passed to reflow so a bare shortcut reference (`[text]`) is treated as an
101 /// atomic link only when its label is actually defined; an undefined
102 /// bracketed run reflows as literal prose.
103 fn defined_reference_labels(ctx: &crate::lint_context::LintContext) -> std::collections::HashSet<String> {
104 ctx.reference_defs
105 .iter()
106 .map(|d| crate::utils::text_reflow::normalize_reference_label(&d.id))
107 .collect()
108 }
109
110 /// Build the reflow options shared by every MD013 fix path.
111 ///
112 /// `line_length` varies per call site (some subtract a list-marker or
113 /// blockquote prefix); every other field is derived uniformly from the
114 /// effective config and the document flavor. Callers that need a different
115 /// `max_list_continuation_indent` override it via struct update.
116 fn reflow_options(
117 ctx: &crate::lint_context::LintContext,
118 config: &MD013Config,
119 line_length: usize,
120 ) -> crate::utils::text_reflow::ReflowOptions {
121 crate::utils::text_reflow::ReflowOptions {
122 line_length,
123 break_on_sentences: true,
124 preserve_breaks: false,
125 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
126 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
127 abbreviations: config.abbreviations_for_reflow(),
128 length_mode: config.reflow_length_mode(),
129 attr_lists: ctx.flavor.supports_attr_lists(),
130 myst_roles: ctx.flavor.supports_myst_roles(),
131 require_sentence_capital: config.require_sentence_capital,
132 max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
133 Some(4)
134 } else {
135 None
136 },
137 defined_references: Some(Self::defined_reference_labels(ctx)),
138 atomic_spans: config.atomic_spans,
139 }
140 }
141
142 fn should_ignore_line(
143 &self,
144 line: &str,
145 _lines: &[&str],
146 current_line: usize,
147 ctx: &crate::lint_context::LintContext,
148 ) -> bool {
149 if self.config.strict {
150 return false;
151 }
152
153 // Quick check for common patterns before expensive regex
154 let trimmed = line.trim();
155
156 // Only skip if the entire line is a URL (quick check first)
157 if (trimmed.starts_with("http://") || trimmed.starts_with("https://")) && URL_PATTERN.is_match(trimmed) {
158 return true;
159 }
160
161 // Only skip if the entire line is an image reference (quick check first)
162 if trimmed.starts_with("![") && trimmed.ends_with(']') && IMAGE_REF_PATTERN.is_match(trimmed) {
163 return true;
164 }
165
166 // Note: link reference definitions are handled as always-exempt (even in strict mode)
167 // in the main check loop, so they don't need to be checked here.
168
169 // Code blocks with long strings (only check if in code block)
170 if ctx.line_info(current_line + 1).is_some_and(|info| info.in_code_block)
171 && !trimmed.is_empty()
172 && !line.contains(' ')
173 && !line.contains('\t')
174 {
175 return true;
176 }
177
178 false
179 }
180
181 /// Check if rule should skip based on provided config (used for inline config support)
182 fn should_skip_with_config(&self, ctx: &crate::lint_context::LintContext, config: &MD013Config) -> bool {
183 // Skip if content is empty
184 if ctx.content.is_empty() {
185 return true;
186 }
187
188 // For sentence-per-line, semantic-line-breaks, or normalize mode, never skip based on line length
189 if config.reflow
190 && (config.reflow_mode == ReflowMode::SentencePerLine
191 || config.reflow_mode == ReflowMode::SemanticLineBreaks
192 || config.reflow_mode == ReflowMode::Normalize)
193 {
194 return false;
195 }
196
197 // Use the smallest applicable budget across line/heading/code-block
198 // contexts so a stricter context-specific limit doesn't get masked by
199 // the document-wide budget.
200 let min_limit = config.min_effective_line_length();
201 if min_limit.is_unlimited() {
202 return true;
203 }
204 let min_limit_bytes = min_limit.get();
205
206 // Quick check: if total content is shorter than the smallest line limit,
207 // definitely skip.
208 if ctx.content.len() <= min_limit_bytes {
209 return true;
210 }
211
212 // Skip if no line exceeds the smallest applicable limit.
213 !ctx.lines.iter().any(|line| line.byte_len > min_limit_bytes)
214 }
215
216 fn normalize_mode_needs_reflow<'a, I>(&self, lines: I, config: &MD013Config) -> bool
217 where
218 I: IntoIterator<Item = &'a str>,
219 {
220 let mut line_count = 0;
221 let check_length = !config.line_length.is_unlimited();
222
223 for line in lines {
224 line_count += 1;
225 if check_length && self.calculate_effective_length(line) > config.line_length.get() {
226 return true;
227 }
228 }
229
230 line_count > 1
231 }
232}
233
234impl Rule for MD013LineLength {
235 fn name(&self) -> &'static str {
236 "MD013"
237 }
238
239 fn description(&self) -> &'static str {
240 "Line length should not be excessive"
241 }
242
243 fn check(&self, ctx: &crate::lint_context::LintContext) -> LintResult {
244 // Use pre-parsed inline config from LintContext
245 let config_override = ctx.inline_config().get_rule_config("MD013");
246
247 // Apply configuration override if present
248 let effective_config = if let Some(json_config) = config_override {
249 if let Some(obj) = json_config.as_object() {
250 let mut config = self.config.clone();
251 if let Some(line_length) = obj.get("line_length").and_then(serde_json::Value::as_u64) {
252 config.line_length = crate::types::LineLength::new(line_length as usize);
253 }
254 if let Some(code_blocks) = obj.get("code_blocks").and_then(serde_json::Value::as_bool) {
255 config.code_blocks = code_blocks;
256 }
257 if let Some(code_spans) = obj.get("code_spans").and_then(serde_json::Value::as_bool) {
258 config.code_spans = code_spans;
259 }
260 if let Some(tables) = obj.get("tables").and_then(serde_json::Value::as_bool) {
261 config.tables = tables;
262 }
263 if let Some(headings) = obj.get("headings").and_then(serde_json::Value::as_bool) {
264 config.headings = headings;
265 }
266 if let Some(blockquotes) = obj.get("blockquotes").and_then(serde_json::Value::as_bool) {
267 config.blockquotes = blockquotes;
268 }
269 if let Some(strict) = obj.get("strict").and_then(serde_json::Value::as_bool) {
270 config.strict = strict;
271 }
272 if let Some(stern) = obj.get("stern").and_then(serde_json::Value::as_bool) {
273 config.stern = stern;
274 }
275 if let Some(v) = obj
276 .get("ignore_link_urls")
277 .or_else(|| obj.get("ignore-link-urls"))
278 .or_else(|| obj.get("semantic_link_understanding"))
279 .or_else(|| obj.get("semantic-link-understanding"))
280 .and_then(serde_json::Value::as_bool)
281 {
282 config.ignore_link_urls = v;
283 }
284 if let Some(reflow) = obj.get("reflow").and_then(serde_json::Value::as_bool) {
285 config.reflow = reflow;
286 }
287 if let Some(reflow_mode) = obj.get("reflow_mode").and_then(|v| v.as_str()) {
288 config.reflow_mode = match reflow_mode {
289 "default" => ReflowMode::Default,
290 "normalize" => ReflowMode::Normalize,
291 "sentence-per-line" => ReflowMode::SentencePerLine,
292 "semantic-line-breaks" => ReflowMode::SemanticLineBreaks,
293 _ => ReflowMode::default(),
294 };
295 }
296 config
297 } else {
298 self.config.clone()
299 }
300 } else {
301 self.config.clone()
302 };
303
304 // Fast early return using should_skip with EFFECTIVE config (after inline overrides)
305 // But don't skip if we're in reflow mode with Normalize or SentencePerLine
306 if self.should_skip_with_config(ctx, &effective_config)
307 && !(effective_config.reflow
308 && (effective_config.reflow_mode == ReflowMode::Normalize
309 || effective_config.reflow_mode == ReflowMode::SentencePerLine
310 || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
311 {
312 return Ok(Vec::new());
313 }
314
315 // Direct implementation without DocumentStructure
316 let mut warnings = Vec::new();
317
318 // Special handling: line_length = 0 means "no line length limit"
319 // Skip all line length checks, but still allow reflow if enabled
320 let skip_length_checks = effective_config.line_length.is_unlimited();
321
322 // Pre-filter lines that could be problematic to avoid processing all lines.
323 // Use the smallest applicable budget across line/heading/code-block contexts
324 // so candidates aren't dropped when a stricter context-specific budget applies.
325 let prefilter_limit = effective_config.min_effective_line_length();
326 let prefilter_skip = prefilter_limit.is_unlimited();
327 let mut candidate_lines = Vec::new();
328 if !skip_length_checks && !prefilter_skip {
329 for (line_idx, line_info) in ctx.lines.iter().enumerate() {
330 // Skip front matter - it should never be linted
331 if line_info.in_front_matter {
332 continue;
333 }
334
335 // Quick length check first
336 if line_info.byte_len > prefilter_limit.get() {
337 candidate_lines.push(line_idx);
338 }
339 }
340 }
341
342 // If no candidate lines and not in normalize or sentence-per-line mode, early return
343 if candidate_lines.is_empty()
344 && !(effective_config.reflow
345 && (effective_config.reflow_mode == ReflowMode::Normalize
346 || effective_config.reflow_mode == ReflowMode::SentencePerLine
347 || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
348 {
349 return Ok(warnings);
350 }
351
352 let lines = ctx.raw_lines();
353
354 // Whether a 1-indexed line is a heading. `LineInfo::heading` is an O(1)
355 // per-line field, so check it directly at each use site instead of
356 // materializing a full-document HashSet (an extra O(n) pass and
357 // allocation on a rule that runs on virtually every file).
358 let is_heading_line_num = |line_number: usize| -> bool {
359 line_number
360 .checked_sub(1)
361 .and_then(|idx| ctx.lines.get(idx))
362 .is_some_and(|line| line.heading.is_some())
363 };
364
365 // Use pre-computed table blocks from context
366 // We need this for both the table skip check AND the paragraphs check
367 let table_blocks = &ctx.table_blocks;
368 let mut table_lines_set = std::collections::HashSet::new();
369 for table in table_blocks {
370 table_lines_set.insert(table.header_line + 1);
371 table_lines_set.insert(table.delimiter_line + 1);
372 for &line in &table.content_lines {
373 table_lines_set.insert(line + 1);
374 }
375 }
376
377 // Process candidate lines for line length checks
378 'line_loop: for &line_idx in &candidate_lines {
379 let line_number = line_idx + 1;
380 let line = lines[line_idx];
381
382 // Calculate actual line length (used in warning messages)
383 let effective_length = self.calculate_effective_length(line);
384
385 // Pick the context-specific limit: heading > code-block > paragraph.
386 // Headings dominate over code-block context if a setext underline ever
387 // overlaps a fenced range (defensive — these are mutually exclusive in
388 // practice, but the explicit ordering documents intent).
389 let is_heading_line = is_heading_line_num(line_number);
390 let in_code_block = ctx.line_info(line_number).is_some_and(|info| info.in_code_block);
391 let line_limit = if is_heading_line {
392 effective_config.effective_heading_line_length().get()
393 } else if in_code_block {
394 effective_config.effective_code_block_line_length().get()
395 } else {
396 effective_config.line_length.get()
397 };
398
399 // A context-specific limit of 0 means "unlimited for this context".
400 if line_limit == 0 {
401 continue;
402 }
403
404 // Stern mode: like default, but the trailing-token forgiveness is
405 // disabled — a line with whitespace that exceeds the limit is a
406 // violation even if the excess is the final token. The "unwrappable"
407 // line exemption (single token, optionally prefixed by # or >) is
408 // still honored. Strict overrides stern entirely.
409 if effective_config.stern && !effective_config.strict && is_unwrappable_line(line) {
410 continue;
411 }
412
413 // Trailing-token forgiveness: only in default mode (not strict, not stern).
414 // If the line only exceeds the limit because of a long token at the end
415 // (URL, link chain, identifier), it passes. This matches markdownlint's
416 // behavior: line.replace(/\S*$/u, "#")
417 let check_length = if effective_config.strict || effective_config.stern {
418 effective_length
419 } else {
420 match line.rfind(char::is_whitespace) {
421 Some(pos) => {
422 let ws_char = line[pos..].chars().next().unwrap();
423 let prefix_end = pos + ws_char.len_utf8();
424 self.calculate_string_length(&line[..prefix_end]) + 1
425 }
426 None => 1, // No whitespace — entire line is a single token
427 }
428 };
429
430 // Skip lines where the check length is within the limit
431 if check_length <= line_limit {
432 continue;
433 }
434
435 // Ignore inline link/image URLs: suppress when excess comes entirely from inline URLs.
436 // Disabled by `strict` (all forgiveness off) and by `ignore_link_urls = false`
437 // (count link/image URLs toward the line length so such lines are flagged).
438 if !effective_config.strict && effective_config.ignore_link_urls {
439 let length_without_urls = self.length_without_inline_link_urls(effective_length, line_number, ctx);
440 if length_without_urls <= line_limit {
441 continue;
442 }
443 }
444
445 // Inline code spans cannot be wrapped, so reflow cannot shorten a line
446 // whose excess length is one. When code-span checking is disabled,
447 // suppress a violation that would fit once inline code spans are excluded.
448 if !effective_config.code_spans {
449 let code_span_width: usize = ctx
450 .code_spans()
451 .iter()
452 .filter(|span| span.line == line_number && span.end_line == line_number)
453 .map(|span| self.calculate_string_length(&ctx.content[span.byte_offset..span.byte_end]))
454 .sum();
455 if effective_length.saturating_sub(code_span_width) <= line_limit {
456 continue;
457 }
458 }
459
460 // Skip mkdocstrings and pymdown blocks (already handled by LintContext)
461 if ctx.lines[line_idx].in_mkdocstrings || ctx.lines[line_idx].in_pymdown_block {
462 continue;
463 }
464
465 // Skip MyST comments (% comment) — structural lines, not prose
466 if ctx.lines[line_idx].is_myst_comment {
467 continue;
468 }
469
470 // Link reference definitions are always exempt, even in strict mode.
471 // There's no way to shorten them without breaking the URL.
472 // Also check after stripping list markers, since list items may
473 // contain link ref defs as their content.
474 {
475 let trimmed = line.trim();
476 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
477 continue;
478 }
479 if is_list_item(trimmed) {
480 let (_, content) = extract_list_marker_and_content(trimmed);
481 let content_trimmed = content.trim();
482 if content_trimmed.starts_with('[')
483 && content_trimmed.contains("]:")
484 && LINK_REF_PATTERN.is_match(content_trimmed)
485 {
486 continue;
487 }
488 }
489 }
490
491 // Skip various block types efficiently
492 if !effective_config.strict {
493 // Lines whose only content is a link/image are exempt.
494 // After stripping list markers, blockquote markers, and emphasis,
495 // if only a link or image remains, there is no way to shorten it.
496 if is_standalone_link_or_image_line(line) {
497 continue;
498 }
499
500 // Lines consisting entirely of HTML tags are exempt.
501 // Badge lines, images with attributes, and similar inline HTML
502 // are long due to URLs in attributes and can't be meaningfully shortened.
503 if is_html_only_line(line) {
504 continue;
505 }
506
507 // Skip setext heading underlines
508 if !line.trim().is_empty() && line.trim().chars().all(|c| c == '=' || c == '-') {
509 continue;
510 }
511
512 // Skip block elements according to config flags
513 // The flags mean: true = check these elements, false = skip these elements
514 // So we skip when the flag is FALSE and the line is in that element type
515 if (!effective_config.headings && is_heading_line_num(line_number))
516 || (!effective_config.code_blocks
517 && ctx.line_info(line_number).is_some_and(|info| info.in_code_block))
518 || (!effective_config.tables && table_lines_set.contains(&line_number))
519 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
520 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
521 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
522 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
523 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
524 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
525 || ctx.line_info(line_number).is_some_and(|info| info.in_pymdown_block)
526 {
527 continue;
528 }
529
530 // Check if this is a paragraph/regular text line
531 // If paragraphs = false, skip lines that are NOT in special blocks
532 // Blockquote content is treated as paragraph text, so it's not
533 // included in the special blocks list here.
534 if !effective_config.paragraphs {
535 let is_special_block = is_heading_line_num(line_number)
536 || ctx.line_info(line_number).is_some_and(|info| info.in_code_block)
537 || table_lines_set.contains(&line_number)
538 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
539 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
540 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
541 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
542 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
543 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
544 || ctx
545 .line_info(line_number)
546 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container);
547
548 // Skip regular paragraph text when paragraphs = false
549 if !is_special_block {
550 continue;
551 }
552 }
553
554 // Skip blockquote lines when blockquotes = false.
555 // Also skip lazy continuation lines that belong to a blockquote
556 // (lines without `>` prefix that follow a blockquote line).
557 if !effective_config.blockquotes {
558 if ctx.lines[line_number - 1].blockquote.is_some() {
559 continue;
560 }
561 // Check for lazy continuation: scan backwards through
562 // non-blank lines to find if this paragraph started with
563 // a blockquote marker
564 if !line.trim().is_empty() {
565 let mut scan = line_number.saturating_sub(2);
566 loop {
567 if ctx.lines[scan].blockquote.is_some() {
568 // Found a blockquote ancestor — this is a lazy continuation
569 continue 'line_loop;
570 }
571 if lines[scan].trim().is_empty() || scan == 0 {
572 break;
573 }
574 scan -= 1;
575 }
576 }
577 }
578
579 // Skip lines that are only a URL, image ref, or link ref
580 if self.should_ignore_line(line, lines, line_idx, ctx) {
581 continue;
582 }
583 }
584
585 // In sentence-per-line mode, check if this is a single long sentence
586 // If so, emit a warning without a fix (user must manually rephrase)
587 if effective_config.reflow_mode == ReflowMode::SentencePerLine {
588 let sentences = split_into_sentences(line.trim());
589 if sentences.len() == 1 {
590 // Single sentence that's too long - warn but don't auto-fix
591 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
592
593 let (start_line, start_col, end_line, end_col) =
594 calculate_excess_range(line_number, line, line_limit);
595
596 warnings.push(LintWarning {
597 rule_name: Some(self.name().to_string()),
598 message,
599 line: start_line,
600 column: start_col,
601 end_line,
602 end_column: end_col,
603 severity: Severity::Warning,
604 fix: None, // No auto-fix for long single sentences
605 });
606 continue;
607 }
608 // Multiple sentences will be handled by paragraph-based reflow
609 continue;
610 }
611
612 // In semantic-line-breaks mode, skip per-line checks —
613 // all reflow is handled at the paragraph level with cascading splits
614 if effective_config.reflow_mode == ReflowMode::SemanticLineBreaks {
615 continue;
616 }
617
618 // Don't provide fix for individual lines when reflow is enabled
619 // Paragraph-based fixes will be handled separately
620 let fix = None;
621
622 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
623
624 // Calculate precise character range for the excess portion
625 let (start_line, start_col, end_line, end_col) = calculate_excess_range(line_number, line, line_limit);
626
627 warnings.push(LintWarning {
628 rule_name: Some(self.name().to_string()),
629 message,
630 line: start_line,
631 column: start_col,
632 end_line,
633 end_column: end_col,
634 severity: Severity::Warning,
635 fix,
636 });
637 }
638
639 // If reflow is enabled, generate paragraph-based fixes
640 if effective_config.reflow {
641 let paragraph_warnings = self.generate_paragraph_fixes(ctx, &effective_config, lines);
642 // Merge paragraph warnings with line warnings, removing duplicates
643 for pw in paragraph_warnings {
644 // Remove any line warnings that overlap with this paragraph
645 warnings.retain(|w| w.line < pw.line || w.line > pw.end_line);
646 warnings.push(pw);
647 }
648 }
649
650 Ok(warnings)
651 }
652
653 fn fix(&self, ctx: &crate::lint_context::LintContext) -> Result<String, LintError> {
654 // For CLI usage, apply fixes from warnings
655 // LSP will use the warning-based fixes directly
656 let warnings = self.check(ctx)?;
657 let warnings =
658 crate::utils::fix_utils::filter_warnings_by_inline_config(warnings, ctx.inline_config(), self.name());
659
660 // If there are no fixes, return content unchanged
661 if !warnings.iter().any(|w| w.fix.is_some()) {
662 return Ok(ctx.content.to_string());
663 }
664
665 // Apply warning-based fixes
666 crate::utils::fix_utils::apply_warning_fixes(ctx.content, &warnings)
667 .map_err(|e| LintError::FixFailed(format!("Failed to apply fixes: {e}")))
668 }
669
670 fn as_any(&self) -> &dyn std::any::Any {
671 self
672 }
673
674 fn category(&self) -> RuleCategory {
675 RuleCategory::Whitespace
676 }
677
678 fn should_skip(&self, ctx: &crate::lint_context::LintContext) -> bool {
679 self.should_skip_with_config(ctx, &self.config)
680 }
681
682 fn default_config_section(&self) -> Option<(String, toml::Value)> {
683 let table = crate::rule_config_serde::config_schema_table(&MD013Config::default())?;
684 if table.is_empty() {
685 None
686 } else {
687 Some((MD013Config::RULE_NAME.to_string(), toml::Value::Table(table)))
688 }
689 }
690
691 fn config_aliases(&self) -> Option<std::collections::HashMap<String, String>> {
692 let mut aliases = std::collections::HashMap::new();
693 aliases.insert("enable_reflow".to_string(), "reflow".to_string());
694 aliases.insert("strict_sentences".to_string(), "require-sentence-capital".to_string());
695 aliases.insert("strict-sentences".to_string(), "require-sentence-capital".to_string());
696 // Kept in step with the `alias` attributes on `MD013Config::ignore_link_urls`.
697 // Serde accepts these spellings, so a config using one is honored; without
698 // them here the key validator reports a documented, working key as unknown.
699 aliases.insert(
700 "semantic-link-understanding".to_string(),
701 "ignore-link-urls".to_string(),
702 );
703 aliases.insert(
704 "semantic_link_understanding".to_string(),
705 "ignore-link-urls".to_string(),
706 );
707 Some(aliases)
708 }
709
710 fn from_config(config: &crate::config::Config) -> Box<dyn Rule>
711 where
712 Self: Sized,
713 {
714 let mut rule_config = crate::rule_config_serde::load_rule_config::<MD013Config>(config);
715 // Use global line_length if rule-specific config still has default value
716 if rule_config.line_length.get() == 80 {
717 rule_config.line_length = config.global.line_length;
718 }
719 let mut rule = Self::from_config_struct(rule_config);
720 // Pull list-marker spacing from MD030 (via the shared serde config loader)
721 // so reflow rewrites list items with the configured spacing rather than a
722 // hard-coded single space.
723 rule.list_spacing = crate::rule_config_serde::load_rule_config::<MD030Config>(config);
724 Box::new(rule)
725 }
726}
727
728impl MD013LineLength {
729 /// True when `line_num` (1-based) sits inside a structure whose lines must be
730 /// preserved verbatim (code block, front matter, HTML/JSX/MDX block, MkDocs
731 /// container, div marker, ...). Used to keep blockquote reflow from touching
732 /// quoted-looking text embedded in such structures.
733 fn line_in_verbatim_context(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
734 ctx.line_info(line_num).is_some_and(|info| {
735 info.in_code_block
736 || info.in_front_matter
737 || info.in_html_block
738 || info.in_html_comment
739 || info.in_esm_block
740 || info.in_jsx_expression
741 || info.in_jsx_block
742 || info.in_mdx_comment
743 || info.in_mkdocstrings
744 || info.in_pymdown_block
745 || info.in_mkdocs_container()
746 || info.is_div_marker
747 })
748 }
749
750 fn is_blockquote_content_boundary(
751 &self,
752 content: &str,
753 line_num: usize,
754 ctx: &crate::lint_context::LintContext,
755 strict: bool,
756 ) -> bool {
757 let trimmed = content.trim();
758
759 trimmed.is_empty()
760 || self.line_in_verbatim_context(line_num, ctx)
761 || trimmed.starts_with('#')
762 || trimmed.starts_with("```")
763 || trimmed.starts_with("~~~")
764 || trimmed.starts_with('>')
765 || TableUtils::is_potential_table_row(content)
766 || is_list_item(trimmed)
767 || is_horizontal_rule(content)
768 || (trimmed.starts_with('[') && content.contains("]:"))
769 || is_template_directive_only(content)
770 || is_standalone_attr_list(content)
771 || is_snippet_block_delimiter(content)
772 || is_github_alert_marker(trimmed)
773 || is_html_only_line(content)
774 // A standalone link/image line is exempt from MD013 (non-strict mode),
775 // so it must end the blockquote paragraph rather than be absorbed into
776 // it, mirroring the top-level paragraph reflow boundary.
777 || (!strict && is_standalone_link_or_image_line(content))
778 }
779
780 #[allow(clippy::too_many_arguments)]
781 fn generate_blockquote_paragraph_fix(
782 &self,
783 ctx: &crate::lint_context::LintContext,
784 config: &MD013Config,
785 lines: &[&str],
786 line_index: &LineIndex,
787 start_idx: usize,
788 line_ending: &str,
789 // Extra indent (spaces) to prepend to the emitted `>` prefix so a blockquote
790 // nested in a list item moves with its parent's widened marker. Zero unless a
791 // non-default MD030 widened an ancestor list item.
792 ancestor_shift: isize,
793 ) -> (Option<LintWarning>, usize) {
794 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
795 return (None, start_idx + 1);
796 };
797 let target_level = start_bq.nesting_level;
798
799 let mut collected: Vec<CollectedBlockquoteLine> = Vec::new();
800 let mut i = start_idx;
801
802 while i < lines.len() {
803 if !collected.is_empty() && has_hard_break(&collected[collected.len() - 1].data.content) {
804 break;
805 }
806
807 let line_num = i + 1;
808 if line_num > ctx.lines.len() {
809 break;
810 }
811
812 if lines[i].trim().is_empty() {
813 break;
814 }
815
816 let line_bq = ctx.lines[i].blockquote.as_deref();
817 if let Some(bq) = line_bq {
818 if bq.nesting_level != target_level {
819 break;
820 }
821
822 if self.is_blockquote_content_boundary(&bq.content, line_num, ctx, config.strict) {
823 break;
824 }
825
826 collected.push(CollectedBlockquoteLine {
827 line_idx: i,
828 data: BlockquoteLineData::explicit(trim_preserving_hard_break(&bq.content), bq.prefix.clone()),
829 });
830 i += 1;
831 continue;
832 }
833
834 let lazy_content = lines[i].trim_start();
835 if self.is_blockquote_content_boundary(lazy_content, line_num, ctx, config.strict) {
836 break;
837 }
838
839 collected.push(CollectedBlockquoteLine {
840 line_idx: i,
841 data: BlockquoteLineData::lazy(trim_preserving_hard_break(lazy_content)),
842 });
843 i += 1;
844 }
845
846 if collected.is_empty() {
847 return (None, start_idx + 1);
848 }
849
850 let next_idx = i;
851 let paragraph_start = collected[0].line_idx;
852 let end_line = collected[collected.len() - 1].line_idx;
853 let line_data: Vec<BlockquoteLineData> = collected.iter().map(|l| l.data.clone()).collect();
854 let paragraph_text = line_data
855 .iter()
856 .map(|d| d.content.as_str())
857 .collect::<Vec<_>>()
858 .join(" ");
859
860 let contains_definition_list = line_data
861 .iter()
862 .any(|d| crate::utils::is_definition_list_item(&d.content));
863 if contains_definition_list {
864 return (None, next_idx);
865 }
866
867 let contains_snippets = line_data.iter().any(|d| is_snippet_block_delimiter(&d.content));
868 if contains_snippets {
869 return (None, next_idx);
870 }
871
872 let needs_reflow = match config.reflow_mode {
873 ReflowMode::Normalize => {
874 self.normalize_mode_needs_reflow(line_data.iter().map(|d| d.content.as_str()), config)
875 }
876 ReflowMode::SentencePerLine => {
877 let sentences = split_into_sentences(¶graph_text);
878 sentences.len() > 1 || line_data.len() > 1
879 }
880 ReflowMode::SemanticLineBreaks => {
881 let sentences = split_into_sentences(¶graph_text);
882 sentences.len() > 1
883 || line_data.len() > 1
884 || collected
885 .iter()
886 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get())
887 }
888 ReflowMode::Default => collected
889 .iter()
890 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get()),
891 };
892
893 if !needs_reflow {
894 return (None, next_idx);
895 }
896
897 let fallback_prefix = start_bq.prefix.clone();
898 let explicit_prefix = dominant_blockquote_prefix(&line_data, &fallback_prefix);
899 // Shift the whole quote right to track a widened parent list item's content
900 // column (only widening matters for nesting; a narrowed parent leaves the quote
901 // harmlessly over-indented, which MD027/MD030 tidy).
902 let explicit_prefix = if ancestor_shift > 0 {
903 format!("{}{explicit_prefix}", " ".repeat(ancestor_shift as usize))
904 } else {
905 explicit_prefix
906 };
907 let continuation_style = blockquote_continuation_style(&line_data);
908
909 let reflow_line_length = if config.line_length.is_unlimited() {
910 usize::MAX
911 } else {
912 config
913 .line_length
914 .get()
915 .saturating_sub(self.calculate_string_length(&explicit_prefix))
916 .max(1)
917 };
918
919 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
920
921 let reflowed_with_style =
922 reflow_blockquote_content(&line_data, &explicit_prefix, continuation_style, &reflow_options);
923
924 if reflowed_with_style.is_empty() {
925 return (None, next_idx);
926 }
927
928 let reflowed_text = reflowed_with_style.join(line_ending);
929
930 let start_range = line_index.whole_line_range(paragraph_start + 1);
931 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
932 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
933 } else {
934 line_index.whole_line_range(end_line + 1)
935 };
936 let byte_range = start_range.start..end_range.end;
937
938 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
939 format!("{reflowed_text}{line_ending}")
940 } else {
941 reflowed_text
942 };
943
944 let original_text = &ctx.content[byte_range.clone()];
945 if original_text == replacement {
946 return (None, next_idx);
947 }
948
949 let (warning_line, warning_end_line) = match config.reflow_mode {
950 ReflowMode::Normalize => (paragraph_start + 1, end_line + 1),
951 ReflowMode::SentencePerLine | ReflowMode::SemanticLineBreaks => (paragraph_start + 1, end_line + 1),
952 ReflowMode::Default => {
953 let violating_line = collected
954 .iter()
955 .find(|line| self.calculate_effective_length(lines[line.line_idx]) > config.line_length.get())
956 .map_or(paragraph_start + 1, |line| line.line_idx + 1);
957 (violating_line, violating_line)
958 }
959 };
960
961 let warning = LintWarning {
962 rule_name: Some(self.name().to_string()),
963 message: match config.reflow_mode {
964 ReflowMode::Normalize => format!(
965 "Paragraph could be normalized to use line length of {} characters",
966 config.line_length.get()
967 ),
968 ReflowMode::SentencePerLine => {
969 let num_sentences = split_into_sentences(¶graph_text).len();
970 if line_data.len() == 1 {
971 format!("Line contains {num_sentences} sentences (one sentence per line required)")
972 } else {
973 let num_lines = line_data.len();
974 format!(
975 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
976 )
977 }
978 }
979 ReflowMode::SemanticLineBreaks => {
980 let num_sentences = split_into_sentences(¶graph_text).len();
981 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
982 }
983 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
984 },
985 line: warning_line,
986 column: 1,
987 end_line: warning_end_line,
988 end_column: lines[warning_end_line.saturating_sub(1)].chars().count() + 1,
989 severity: Severity::Warning,
990 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
991 };
992
993 (Some(warning), next_idx)
994 }
995
996 /// Reflow a single list item that lives inside a blockquote.
997 ///
998 /// The blockquote paragraph reflow treats a list marker as a content boundary,
999 /// so `> - long item ...` is never wrapped even though the identical top-level
1000 /// item is. This handles that case: it collects one tight, prose-only list item
1001 /// at the starting blockquote level, reflows the item body to the configured
1002 /// width, and re-emits it with the blockquote prefix preserved and continuation
1003 /// lines aligned under the list content.
1004 ///
1005 /// Sibling and nested list items end collection and are reflowed independently
1006 /// by the caller's outer loop (a nested item carries its indent folded into the
1007 /// blockquote prefix, so it reflows correctly on its own). Items that are not
1008 /// simple tight prose - those embedding a code block, table, fence, or hard
1009 /// break - are left untouched (`None`) and the cursor still advances past the
1010 /// whole item so its inner lines are never reprocessed as loose prose.
1011 #[allow(clippy::too_many_arguments)]
1012 fn generate_blockquote_list_item_fix(
1013 &self,
1014 ctx: &crate::lint_context::LintContext,
1015 config: &MD013Config,
1016 lines: &[&str],
1017 line_index: &LineIndex,
1018 start_idx: usize,
1019 line_ending: &str,
1020 // Extra indent (spaces) to prepend to the emitted `>` prefix when this quoted
1021 // list lives inside a list item whose marker widened. Zero unless a non-default
1022 // MD030 widened an ancestor list item.
1023 ancestor_shift: isize,
1024 ) -> (Option<LintWarning>, usize) {
1025 use crate::utils::blockquote::effective_indent_in_blockquote;
1026
1027 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
1028 return (None, start_idx + 1);
1029 };
1030
1031 // A `>`-prefixed line can be marked as a blockquote even inside a fenced code
1032 // block (or other verbatim structure); such content must never be reflowed.
1033 if self.line_in_verbatim_context(start_idx + 1, ctx) {
1034 return (None, start_idx + 1);
1035 }
1036
1037 let target_level = start_bq.nesting_level;
1038
1039 // The marker line carries the canonical blockquote prefix: its content begins
1040 // with the list marker, so no list indent has been folded into the prefix.
1041 // Track a widened parent list item's content column so the quote stays nested
1042 // (only widening can detach it; narrowing just over-indents).
1043 let bq_prefix = if ancestor_shift > 0 {
1044 format!("{}{}", " ".repeat(ancestor_shift as usize), start_bq.prefix)
1045 } else {
1046 start_bq.prefix.clone()
1047 };
1048
1049 // A thematic break opens with what looks like a bullet marker (`- - -`).
1050 // It is not a list item, and reflowing it as prose destroys the break.
1051 // The top-level reflow path applies the same exemption.
1052 if is_horizontal_rule(&start_bq.content) {
1053 return (None, start_idx + 1);
1054 }
1055
1056 let (marker, first_body) = extract_list_marker_and_content(&start_bq.content);
1057 if marker.is_empty() {
1058 return (None, start_idx + 1);
1059 }
1060 let marker_width = marker.chars().count();
1061
1062 // Continuation lines of a checkbox item align under the bullet+checkbox, but
1063 // are recognized from the bullet width, matching the top-level list reflow.
1064 let base_marker_width = ["[ ] ", "[x] ", "[X] "]
1065 .iter()
1066 .find_map(|cb| marker.find(*cb))
1067 .unwrap_or(marker_width);
1068
1069 // Collect the item: the marker line plus its tight prose continuation lines.
1070 // `end_idx` always tracks the last consumed line so the cursor advances past
1071 // the entire item, even when it turns out to be too complex to reflow safely.
1072 let first_piece = trim_preserving_hard_break(&first_body);
1073 let mut simple = !has_hard_break(&first_piece);
1074 let mut body_pieces: Vec<String> = vec![first_piece];
1075 let mut end_idx = start_idx;
1076 let mut i = start_idx + 1;
1077
1078 while i < lines.len() {
1079 let Some(bq) = ctx.lines[i].blockquote.as_deref() else {
1080 // A blank line ends the item.
1081 if lines[i].trim().is_empty() {
1082 break;
1083 }
1084 // A lazy continuation (no `>` marker) is too ambiguous to reflow
1085 // safely. Consume the whole lazy run into this item's span and leave
1086 // the item untouched, so the caller does not reflow the continuation
1087 // in isolation and leave the marker line partially fixed.
1088 simple = false;
1089 while i < lines.len() && ctx.lines[i].blockquote.is_none() && !lines[i].trim().is_empty() {
1090 end_idx = i;
1091 i += 1;
1092 }
1093 break;
1094 };
1095 if bq.nesting_level != target_level {
1096 break;
1097 }
1098
1099 let content = bq.content.as_str();
1100 if content.trim().is_empty() {
1101 // Blank quoted line ends the tight paragraph. A following indented
1102 // paragraph (loose item) is reflowed on its own by the prose path.
1103 break;
1104 }
1105
1106 let eff_indent = effective_indent_in_blockquote(lines[i], target_level, 0);
1107 if eff_indent < base_marker_width {
1108 // Dedented: a sibling list item or text outside this item. Stop here
1109 // and let the outer loop classify it.
1110 break;
1111 }
1112 if is_list_item(content) {
1113 // A nested list item: its own item, handled independently.
1114 break;
1115 }
1116
1117 // An embedded structure (code block, table, fence, nested quote, ...)
1118 // means the item is not simple prose: keep consuming so the cursor clears
1119 // the whole structure, but do not produce a fix.
1120 if self.is_blockquote_content_boundary(content, i + 1, ctx, config.strict) {
1121 simple = false;
1122 }
1123
1124 let piece = trim_preserving_hard_break(content);
1125 if has_hard_break(&piece) {
1126 simple = false;
1127 }
1128 body_pieces.push(piece);
1129 end_idx = i;
1130 i += 1;
1131 }
1132
1133 let next_idx = end_idx + 1;
1134
1135 if !simple {
1136 return (None, next_idx);
1137 }
1138
1139 let exceeds_limit =
1140 || (start_idx..=end_idx).any(|idx| self.calculate_effective_length(lines[idx]) > config.line_length.get());
1141 let body_text = body_pieces.join(" ");
1142 let body_text = body_text.trim();
1143
1144 // Some bodies cannot be shortened and must stay verbatim, matching the
1145 // exemptions the top-level list reflow applies: link reference definitions
1146 // always, and (in non-strict mode) standalone links/images and HTML-only
1147 // lines. Reflowing a link reference definition would split it after the
1148 // colon/URL and break the definition.
1149 let is_link_ref_def =
1150 body_text.starts_with('[') && body_text.contains("]:") && LINK_REF_PATTERN.is_match(body_text);
1151 let raw_marker_line = lines[start_idx];
1152 let body_is_unwrappable = is_link_ref_def
1153 || (!config.strict && is_standalone_link_or_image_line(raw_marker_line))
1154 || (!config.strict && is_html_only_line(raw_marker_line));
1155 if body_is_unwrappable {
1156 return (None, next_idx);
1157 }
1158
1159 let needs_reflow = match config.reflow_mode {
1160 ReflowMode::Normalize => body_pieces.len() > 1 || exceeds_limit(),
1161 ReflowMode::Default => exceeds_limit(),
1162 ReflowMode::SentencePerLine => split_into_sentences(body_text).len() > 1 || body_pieces.len() > 1,
1163 ReflowMode::SemanticLineBreaks => split_into_sentences(body_text).len() > 1 || exceeds_limit(),
1164 };
1165 if !needs_reflow {
1166 return (None, next_idx);
1167 }
1168
1169 // Apply MD030 list-marker spacing in the spacing-normalizing modes, mirroring
1170 // the top-level list reflow: derive the post-marker spacing from MD030 and let
1171 // the continuation indent follow the resulting content width. Default MD030 (a
1172 // single space) leaves the marker unchanged. MkDocs keeps its rigid indent.
1173 let (marker, marker_width) = if matches!(config.reflow_mode, ReflowMode::Default | ReflowMode::Normalize)
1174 && !ctx.flavor.requires_strict_list_indent()
1175 {
1176 let is_ordered = marker.starts_with(|c: char| c.is_ascii_digit());
1177 // Bullet/number portion only (e.g. `-`, `1.`); the checkbox is content.
1178 let bullet = marker.split(' ').next().unwrap_or("");
1179 let bullet_len = bullet.chars().count();
1180 let checkbox_tail = &marker[base_marker_width..];
1181 // Decide single- vs multi-line spacing from the rewritten shape. This path
1182 // only handles a single tight prose paragraph (structural or multi-paragraph
1183 // items set `simple = false` and bail out above), so the emitted item stays
1184 // multi-line solely when the joined body wraps past one line. A multi-line
1185 // *source* that collapses onto the marker line must use MD030's single-line
1186 // spacing, matching the top-level reflow. The wrap test measures at the
1187 // single-line content column so it does not depend on the spacing chosen here.
1188 let single_col = self.calculate_string_length(&bq_prefix)
1189 + bullet_len
1190 + self.list_spacing.expected_spaces(is_ordered, false, bullet_len)
1191 + checkbox_tail.chars().count();
1192 let is_multi = !body_text.is_empty()
1193 && self.calculate_effective_length(&format!("{}{body_text}", " ".repeat(single_col)))
1194 > config.line_length.effective_limit();
1195 let spaces = self.list_spacing.expected_spaces(is_ordered, is_multi, bullet_len);
1196 let new_marker = format!("{bullet}{}{checkbox_tail}", " ".repeat(spaces));
1197 let width = new_marker.chars().count();
1198 (new_marker, width)
1199 } else {
1200 (marker, marker_width)
1201 };
1202
1203 let prefix_width = self.calculate_string_length(&bq_prefix) + self.calculate_string_length(&marker);
1204 let reflow_line_length = if config.line_length.is_unlimited() {
1205 usize::MAX
1206 } else {
1207 config.line_length.get().saturating_sub(prefix_width).max(1)
1208 };
1209
1210 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
1211
1212 let reflowed = crate::utils::text_reflow::reflow_line(body_text, &reflow_options);
1213 if reflowed.is_empty() {
1214 return (None, next_idx);
1215 }
1216
1217 let continuation_indent = " ".repeat(marker_width);
1218 let reflowed_text = reflowed
1219 .iter()
1220 .enumerate()
1221 .map(|(idx, line)| {
1222 if idx == 0 {
1223 format!("{bq_prefix}{marker}{line}")
1224 } else {
1225 format!("{bq_prefix}{continuation_indent}{line}")
1226 }
1227 })
1228 .collect::<Vec<_>>()
1229 .join(line_ending);
1230
1231 let start_range = line_index.whole_line_range(start_idx + 1);
1232 let end_range = if end_idx == lines.len() - 1 && !ctx.content.ends_with('\n') {
1233 line_index.line_text_range(end_idx + 1, 1, lines[end_idx].len() + 1)
1234 } else {
1235 line_index.whole_line_range(end_idx + 1)
1236 };
1237 let byte_range = start_range.start..end_range.end;
1238
1239 let replacement = if end_idx < lines.len() - 1 || ctx.content.ends_with('\n') {
1240 format!("{reflowed_text}{line_ending}")
1241 } else {
1242 reflowed_text
1243 };
1244
1245 let original_text = &ctx.content[byte_range.clone()];
1246 if original_text == replacement {
1247 return (None, next_idx);
1248 }
1249
1250 let message = match config.reflow_mode {
1251 ReflowMode::Normalize => format!(
1252 "Paragraph could be normalized to use line length of {} characters",
1253 config.line_length.get()
1254 ),
1255 ReflowMode::SentencePerLine => {
1256 let num_sentences = split_into_sentences(body_text).len();
1257 format!("List item should have one sentence per line (found {num_sentences} sentences)")
1258 }
1259 ReflowMode::SemanticLineBreaks => {
1260 let num_sentences = split_into_sentences(body_text).len();
1261 format!("List item should use semantic line breaks ({num_sentences} sentences)")
1262 }
1263 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
1264 };
1265
1266 let warning = LintWarning {
1267 rule_name: Some(self.name().to_string()),
1268 message,
1269 line: start_idx + 1,
1270 column: 1,
1271 end_line: end_idx + 1,
1272 end_column: lines[end_idx].chars().count() + 1,
1273 severity: Severity::Warning,
1274 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1275 };
1276
1277 (Some(warning), next_idx)
1278 }
1279
1280 /// Generate paragraph-based fixes
1281 fn generate_paragraph_fixes(
1282 &self,
1283 ctx: &crate::lint_context::LintContext,
1284 config: &MD013Config,
1285 lines: &[&str],
1286 ) -> Vec<LintWarning> {
1287 let mut warnings = Vec::new();
1288 let line_index = LineIndex::new(ctx.content);
1289
1290 // Detect the content's line ending style to preserve it in replacements.
1291 // The LSP receives content from editors which may use CRLF (Windows).
1292 // Replacements must match the original line endings to avoid false positives.
1293 let line_ending = crate::utils::line_ending::detect_line_ending(ctx.content);
1294
1295 // Ancestor list-item indent shifts, innermost last. When a reflowed parent's
1296 // marker widens under a non-default MD030 (e.g. ul-multi = 3 moves the parent's
1297 // content from column 2 to 4), its nested list/blockquote children are reflowed
1298 // independently and would otherwise keep their original indent — leaving them
1299 // under the parent's new content column, where a CommonMark parser reparses them
1300 // as siblings rather than children. Each frame is (normalized marker width,
1301 // cumulative shift applied to that item's content column); a descendant adds its
1302 // innermost open ancestor's shift to its own indent so the whole subtree moves
1303 // together. With a default MD030 and no marker padding every shift is 0, so this
1304 // is inert and the output is byte-identical.
1305 let mut list_shift_stack: Vec<(usize, isize)> = Vec::new();
1306
1307 let mut i = 0;
1308 while i < lines.len() {
1309 let line_num = i + 1;
1310
1311 // Close ancestor frames whose list item has ended at this line: a non-blank
1312 // line indented less than the frame's source content column is no longer
1313 // inside that item. Blank lines alone don't close a (loose) list item.
1314 if !list_shift_stack.is_empty()
1315 && let Some(info) = ctx.lines.get(i)
1316 && !info.is_blank
1317 {
1318 while let Some(&(content_column, _)) = list_shift_stack.last() {
1319 if info.indent < content_column {
1320 list_shift_stack.pop();
1321 } else {
1322 break;
1323 }
1324 }
1325 }
1326
1327 // Handle blockquote paragraphs with style-preserving reflow.
1328 // Skip blockquotes when blockquotes=false or paragraphs=false
1329 if line_num > 0 && line_num <= ctx.lines.len() && ctx.lines[line_num - 1].blockquote.is_some() {
1330 if !config.blockquotes || !config.paragraphs {
1331 // Skip past all blockquote lines (explicit and lazy continuations).
1332 // A lazy continuation is a non-blank line without `>` that follows
1333 // a blockquote line and isn't a structural element.
1334 let mut saw_explicit_bq = false;
1335 while i < lines.len() && i < ctx.lines.len() {
1336 if ctx.lines[i].blockquote.is_some() {
1337 saw_explicit_bq = true;
1338 i += 1;
1339 } else if saw_explicit_bq
1340 && !lines[i].trim().is_empty()
1341 && !lines[i].trim_start().starts_with('#')
1342 && !lines[i].trim_start().starts_with('>')
1343 {
1344 // Lazy continuation of preceding blockquote
1345 i += 1;
1346 } else {
1347 break;
1348 }
1349 }
1350 continue;
1351 }
1352 // A blockquote nested in a list item moves with its parent when the
1353 // parent's marker widens (see `list_shift_stack`); pass that shift so the
1354 // emitted `>` prefix lands under the parent's new content column instead
1355 // of detaching into a sibling.
1356 let ancestor_shift = list_shift_stack.last().map_or(0isize, |&(_, shift)| shift);
1357 // A list item inside the blockquote needs list-aware reflow (marker +
1358 // continuation indent); plain prose goes through the paragraph path.
1359 let is_bq_list_item = ctx.lines[i]
1360 .blockquote
1361 .as_deref()
1362 .is_some_and(|bq| is_list_item(&bq.content));
1363 let (warning, next_idx) = if is_bq_list_item {
1364 self.generate_blockquote_list_item_fix(
1365 ctx,
1366 config,
1367 lines,
1368 &line_index,
1369 i,
1370 line_ending,
1371 ancestor_shift,
1372 )
1373 } else {
1374 self.generate_blockquote_paragraph_fix(
1375 ctx,
1376 config,
1377 lines,
1378 &line_index,
1379 i,
1380 line_ending,
1381 ancestor_shift,
1382 )
1383 };
1384 if let Some(warning) = warning {
1385 warnings.push(warning);
1386 }
1387 i = next_idx;
1388 continue;
1389 }
1390
1391 // Skip special structures (but NOT MkDocs containers - those get special handling)
1392 let should_skip_due_to_line_info = ctx.line_info(line_num).is_some_and(|info| {
1393 info.in_code_block
1394 || info.in_front_matter
1395 || info.in_html_block
1396 || info.in_html_comment
1397 || info.in_esm_block
1398 || info.in_jsx_expression
1399 || info.in_jsx_block
1400 || info.in_mdx_comment
1401 || info.in_mkdocstrings
1402 || info.in_pymdown_block
1403 });
1404
1405 // Skip link reference definitions but NOT footnote definitions.
1406 // Footnote definitions (`[^id]: prose`) contain reflowable text,
1407 // while link reference definitions (`[ref]: URL`) contain URLs
1408 // that cannot be shortened.
1409 let is_link_ref_def =
1410 lines[i].trim().starts_with('[') && !lines[i].trim().starts_with("[^") && lines[i].contains("]:");
1411
1412 if should_skip_due_to_line_info
1413 || lines[i].trim().starts_with('#')
1414 || TableUtils::is_potential_table_row(lines[i])
1415 || lines[i].trim().is_empty()
1416 || is_horizontal_rule(lines[i])
1417 || is_template_directive_only(lines[i])
1418 || is_link_ref_def
1419 || ctx.line_info(line_num).is_some_and(|info| info.is_div_marker)
1420 || is_html_only_line(lines[i])
1421 || (!config.strict && is_standalone_link_or_image_line(lines[i]))
1422 {
1423 i += 1;
1424 continue;
1425 }
1426
1427 // Handle footnote definitions: `[^id]: prose text that can be reflowed`
1428 // Supports multi-paragraph footnotes with code blocks, blockquotes,
1429 // tables, and lists preserved verbatim.
1430 // Validate structure: must start with `[^`, contain `]:`, and the ID
1431 // must not contain `[` or `]` (prevents false matches on nested brackets)
1432 if lines[i].trim().starts_with("[^") && lines[i].contains("]:") && {
1433 let after_caret = &lines[i].trim()[2..];
1434 after_caret
1435 .find("]:")
1436 .is_some_and(|pos| pos > 0 && !after_caret[..pos].contains(['[', ']']))
1437 } {
1438 let footnote_start = i;
1439 let line = lines[i];
1440
1441 // Extract the prefix `[^id]:`
1442 let Some(colon_pos) = line.find("]:") else {
1443 i += 1;
1444 continue;
1445 };
1446 let prefix_end = colon_pos + 2;
1447 let prefix = &line[..prefix_end];
1448
1449 // Content starts after `]: ` (with optional space)
1450 let content_start = if line[prefix_end..].starts_with(' ') {
1451 prefix_end + 1
1452 } else {
1453 prefix_end
1454 };
1455 let first_content = &line[content_start..];
1456
1457 // CommonMark footnotes use 4-space continuation indent
1458 const FN_INDENT: usize = 4;
1459
1460 // --- Line classification for footnote content ---
1461 #[derive(Debug, Clone)]
1462 enum FnLineType {
1463 Content(String),
1464 Verbatim(String, usize), // preserved text, original indent
1465 Empty,
1466 }
1467
1468 // Helper: compute visual indent (tabs = 4 spaces)
1469 let visual_indent = |s: &str| -> usize {
1470 s.chars()
1471 .take_while(|c| c.is_whitespace())
1472 .map(|c| if c == '\t' { 4 } else { 1 })
1473 .sum::<usize>()
1474 };
1475
1476 // Helper: check if a trimmed line is a fence marker (homogeneous chars)
1477 let is_fence = |s: &str| -> bool {
1478 let t = s.trim();
1479 let fence_char = t.chars().next();
1480 matches!(fence_char, Some('`') | Some('~'))
1481 && t.chars().take_while(|&c| c == fence_char.unwrap()).count() >= 3
1482 };
1483
1484 // Helper: check if a trimmed line is a setext underline
1485 let is_setext_underline = |s: &str| -> bool {
1486 let t = s.trim();
1487 !t.is_empty()
1488 && (t.chars().all(|c| c == '=' || c == ' ') || t.chars().all(|c| c == '-' || c == ' '))
1489 && t.contains(['=', '-'])
1490 };
1491
1492 // Deferred body: `[^id]:\n content` — first line has no content,
1493 // actual content starts on the next indented line
1494 let deferred_body = first_content.trim().is_empty();
1495
1496 // Collect all lines belonging to this footnote definition
1497 let mut fn_lines: Vec<FnLineType> = Vec::new();
1498 if !deferred_body {
1499 fn_lines.push(FnLineType::Content(first_content.to_string()));
1500 }
1501 let mut last_consumed = i;
1502 i += 1;
1503
1504 // Strip only the footnote continuation indent, preserving
1505 // internal indentation (e.g., code block body indent)
1506 let strip_fn_indent = |s: &str| -> String {
1507 let mut chars = s.chars();
1508 let mut stripped = 0;
1509 while stripped < FN_INDENT {
1510 match chars.next() {
1511 Some('\t') => stripped += 4,
1512 Some(c) if c.is_whitespace() => stripped += 1,
1513 _ => break,
1514 }
1515 }
1516 chars.as_str().to_string()
1517 };
1518
1519 let mut in_fenced_code = false;
1520 let mut consecutive_blanks = 0u32;
1521
1522 while i < lines.len() {
1523 let next = lines[i];
1524 let next_trimmed = next.trim();
1525
1526 // Blank line handling
1527 if next_trimmed.is_empty() {
1528 consecutive_blanks += 1;
1529 // 2+ consecutive blanks terminate the footnote
1530 if consecutive_blanks >= 2 {
1531 break;
1532 }
1533
1534 // Inside a fenced code block, blank lines are part of the code
1535 if in_fenced_code {
1536 consecutive_blanks = 0; // Don't count blanks inside code blocks
1537 fn_lines.push(FnLineType::Verbatim(String::new(), 0));
1538 last_consumed = i;
1539 i += 1;
1540 continue;
1541 }
1542
1543 // Peek ahead: if next non-blank line is indented >= FN_INDENT,
1544 // this blank is an internal paragraph separator
1545 if i + 1 < lines.len() {
1546 let peek = lines[i + 1];
1547 let peek_indent = visual_indent(peek);
1548 if !peek.trim().is_empty() && peek_indent >= FN_INDENT {
1549 fn_lines.push(FnLineType::Empty);
1550 last_consumed = i;
1551 i += 1;
1552 continue;
1553 }
1554 }
1555 // No valid continuation after blank — end of footnote
1556 break;
1557 }
1558
1559 consecutive_blanks = 0;
1560 let indent = visual_indent(next);
1561
1562 // Not indented enough — end of footnote
1563 if indent < FN_INDENT {
1564 break;
1565 }
1566
1567 // Inside a fenced code block: everything is verbatim until closing fence
1568 if in_fenced_code {
1569 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1570 if is_fence(next_trimmed) {
1571 in_fenced_code = false;
1572 }
1573 last_consumed = i;
1574 i += 1;
1575 continue;
1576 }
1577
1578 // Fence opener — start verbatim code block
1579 if is_fence(next_trimmed) {
1580 in_fenced_code = true;
1581 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1582 last_consumed = i;
1583 i += 1;
1584 continue;
1585 }
1586
1587 // Indented code block: indent >= FN_INDENT + 4 (= 8 spaces)
1588 if indent >= FN_INDENT + 4 {
1589 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1590 last_consumed = i;
1591 i += 1;
1592 continue;
1593 }
1594
1595 // Structural content that must be preserved verbatim
1596 if next_trimmed.starts_with('#')
1597 || is_list_item(next_trimmed)
1598 || next_trimmed.starts_with('>')
1599 || TableUtils::is_potential_table_row(next_trimmed)
1600 || is_setext_underline(next_trimmed)
1601 || is_horizontal_rule(next_trimmed)
1602 || crate::utils::mkdocs_footnotes::is_footnote_definition(next_trimmed)
1603 {
1604 // Preserve verbatim: blockquotes, tables, lists, setext
1605 // underlines, and horizontal rules inside the footnote
1606 if next_trimmed.starts_with('>')
1607 || TableUtils::is_potential_table_row(next_trimmed)
1608 || is_list_item(next_trimmed)
1609 || is_setext_underline(next_trimmed)
1610 || is_horizontal_rule(next_trimmed)
1611 {
1612 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1613 last_consumed = i;
1614 i += 1;
1615 continue;
1616 }
1617 // Headings, new footnote defs, link refs — end the footnote
1618 break;
1619 }
1620
1621 // Link reference definitions inside footnotes are not reflowable
1622 if next_trimmed.starts_with('[')
1623 && !next_trimmed.starts_with("[^")
1624 && next_trimmed.contains("]:")
1625 && LINK_REF_PATTERN.is_match(next_trimmed)
1626 {
1627 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1628 last_consumed = i;
1629 i += 1;
1630 continue;
1631 }
1632
1633 // HTML-only lines inside footnotes are not reflowable
1634 if is_html_only_line(next_trimmed) {
1635 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1636 last_consumed = i;
1637 i += 1;
1638 continue;
1639 }
1640
1641 // Regular prose content
1642 fn_lines.push(FnLineType::Content(next_trimmed.to_string()));
1643 last_consumed = i;
1644 i += 1;
1645 }
1646
1647 // Nothing collected or only empty lines
1648 if fn_lines.iter().all(|l| matches!(l, FnLineType::Empty)) || fn_lines.is_empty() {
1649 continue;
1650 }
1651
1652 // --- Group into blocks ---
1653 #[derive(Debug)]
1654 enum FnBlock {
1655 Paragraph(Vec<String>),
1656 Verbatim(Vec<(String, usize)>), // (content, indent) preserved as-is
1657 }
1658
1659 let mut blocks: Vec<FnBlock> = Vec::new();
1660 let mut current_para: Vec<String> = Vec::new();
1661 let mut current_verbatim: Vec<(String, usize)> = Vec::new();
1662
1663 for fl in &fn_lines {
1664 match fl {
1665 FnLineType::Content(s) => {
1666 if !current_verbatim.is_empty() {
1667 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
1668 }
1669 current_para.push(s.clone());
1670 }
1671 FnLineType::Verbatim(s, indent) => {
1672 if !current_para.is_empty() {
1673 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
1674 }
1675 current_verbatim.push((s.clone(), *indent));
1676 }
1677 FnLineType::Empty => {
1678 if !current_para.is_empty() {
1679 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
1680 }
1681 if !current_verbatim.is_empty() {
1682 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
1683 }
1684 }
1685 }
1686 }
1687 if !current_para.is_empty() {
1688 blocks.push(FnBlock::Paragraph(current_para));
1689 }
1690 if !current_verbatim.is_empty() {
1691 blocks.push(FnBlock::Verbatim(current_verbatim));
1692 }
1693
1694 // --- Reflow paragraphs and reconstruct ---
1695 let prefix_display_width = prefix.chars().count() + 1; // +1 for space
1696 let reflow_line_length = if config.line_length.is_unlimited() {
1697 usize::MAX
1698 } else {
1699 config
1700 .line_length
1701 .get()
1702 .saturating_sub(FN_INDENT.max(prefix_display_width))
1703 .max(20)
1704 };
1705 // Footnote continuation uses a fixed 4-space indent, so list
1706 // continuation capping does not apply here.
1707 let reflow_options = crate::utils::text_reflow::ReflowOptions {
1708 max_list_continuation_indent: None,
1709 ..Self::reflow_options(ctx, config, reflow_line_length)
1710 };
1711
1712 let indent_str = " ".repeat(FN_INDENT);
1713 let mut result_lines: Vec<String> = Vec::new();
1714 let mut is_first_block = true;
1715
1716 for block in &blocks {
1717 match block {
1718 FnBlock::Paragraph(para_lines) => {
1719 let paragraph_text = para_lines.join(" ");
1720 let paragraph_text = paragraph_text.trim();
1721 if paragraph_text.is_empty() {
1722 continue;
1723 }
1724
1725 let reflowed = crate::utils::text_reflow::reflow_line(paragraph_text, &reflow_options);
1726 if reflowed.is_empty() {
1727 continue;
1728 }
1729
1730 // Blank line separator between blocks
1731 if !result_lines.is_empty() {
1732 result_lines.push(String::new());
1733 }
1734
1735 for (idx, rline) in reflowed.iter().enumerate() {
1736 if is_first_block && idx == 0 {
1737 result_lines.push(format!("{prefix} {rline}"));
1738 } else {
1739 result_lines.push(format!("{indent_str}{rline}"));
1740 }
1741 }
1742 is_first_block = false;
1743 }
1744 FnBlock::Verbatim(verb_lines) => {
1745 // Blank line separator between blocks
1746 if !result_lines.is_empty() {
1747 result_lines.push(String::new());
1748 }
1749
1750 if is_first_block {
1751 // Verbatim as first block in a deferred-body footnote
1752 if deferred_body {
1753 result_lines.push(prefix.to_string());
1754 }
1755 is_first_block = false;
1756 }
1757 for (content, _orig_indent) in verb_lines {
1758 result_lines.push(format!("{indent_str}{content}"));
1759 }
1760 }
1761 }
1762 }
1763
1764 // If nothing was produced, skip
1765 if result_lines.is_empty() {
1766 continue;
1767 }
1768
1769 let reflowed_text = result_lines.join(line_ending);
1770
1771 // Calculate byte range using last_consumed
1772 let start_range = line_index.whole_line_range(footnote_start + 1);
1773 let end_range = if last_consumed == lines.len() - 1 && !ctx.content.ends_with('\n') {
1774 line_index.line_text_range(last_consumed + 1, 1, lines[last_consumed].len() + 1)
1775 } else {
1776 line_index.whole_line_range(last_consumed + 1)
1777 };
1778 let byte_range = start_range.start..end_range.end;
1779
1780 let replacement = if last_consumed < lines.len() - 1 || ctx.content.ends_with('\n') {
1781 format!("{reflowed_text}{line_ending}")
1782 } else {
1783 reflowed_text
1784 };
1785
1786 let original_text = &ctx.content[byte_range.clone()];
1787 let max_length = (footnote_start..=last_consumed)
1788 .map(|idx| self.calculate_effective_length(lines[idx]))
1789 .max()
1790 .unwrap_or(0);
1791 let line_limit = if config.line_length.is_unlimited() {
1792 usize::MAX
1793 } else {
1794 config.line_length.get()
1795 };
1796 if original_text != replacement && max_length > line_limit {
1797 warnings.push(LintWarning {
1798 rule_name: Some(self.name().to_string()),
1799 message: format!(
1800 "Line length {} exceeds {} characters",
1801 max_length,
1802 config.line_length.get()
1803 ),
1804 line: footnote_start + 1,
1805 column: 1,
1806 end_line: last_consumed + 1,
1807 end_column: lines[last_consumed].chars().count() + 1,
1808 severity: Severity::Warning,
1809 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1810 });
1811 }
1812 continue;
1813 }
1814
1815 // Handle MkDocs container content (admonitions and tabs) with indent-preserving reflow
1816 if ctx
1817 .line_info(line_num)
1818 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
1819 {
1820 // Skip admonition/tab marker lines — only reflow their indented content
1821 let current_line = lines[i];
1822 if mkdocs_admonitions::is_admonition_start(current_line) || mkdocs_tabs::is_tab_marker(current_line) {
1823 i += 1;
1824 continue;
1825 }
1826
1827 let container_start = i;
1828
1829 // Detect the actual indent level from the first content line
1830 // (supports nested admonitions with 8+ spaces)
1831 let first_line = lines[i];
1832 let base_indent_len = first_line.len() - first_line.trim_start().len();
1833 let base_indent: String = " ".repeat(base_indent_len);
1834
1835 // Collect consecutive MkDocs container paragraph lines
1836 let mut container_lines: Vec<&str> = Vec::new();
1837 while i < lines.len() {
1838 let current_line_num = i + 1;
1839 let line_info = ctx.line_info(current_line_num);
1840
1841 // Stop if we leave the MkDocs container
1842 if !line_info.is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container) {
1843 break;
1844 }
1845
1846 let line = lines[i];
1847
1848 // Stop at paragraph boundaries within the container
1849 if line.trim().is_empty() {
1850 break;
1851 }
1852
1853 // Skip list items, code blocks, headings, HTML-only lines within containers
1854 if is_list_item(line.trim())
1855 || line.trim().starts_with("```")
1856 || line.trim().starts_with("~~~")
1857 || line.trim().starts_with('#')
1858 || is_html_only_line(line)
1859 {
1860 break;
1861 }
1862
1863 container_lines.push(line);
1864 i += 1;
1865 }
1866
1867 if container_lines.is_empty() {
1868 // Must advance i to avoid infinite loop when we encounter
1869 // non-paragraph content (code block, list, heading, empty line)
1870 // at the start of an MkDocs container
1871 i += 1;
1872 continue;
1873 }
1874
1875 // Strip the base indent from each line and join for reflow
1876 let stripped_lines: Vec<&str> = container_lines
1877 .iter()
1878 .map(|line| {
1879 if line.starts_with(&base_indent) {
1880 &line[base_indent_len..]
1881 } else {
1882 line.trim_start()
1883 }
1884 })
1885 .collect();
1886 let paragraph_text = stripped_lines.join(" ");
1887
1888 // Check if reflow is needed
1889 let needs_reflow = match config.reflow_mode {
1890 ReflowMode::Normalize => self.normalize_mode_needs_reflow(container_lines.iter().copied(), config),
1891 ReflowMode::SentencePerLine => {
1892 let sentences = split_into_sentences(¶graph_text);
1893 sentences.len() > 1 || container_lines.len() > 1
1894 }
1895 ReflowMode::SemanticLineBreaks => {
1896 let sentences = split_into_sentences(¶graph_text);
1897 sentences.len() > 1
1898 || container_lines.len() > 1
1899 || container_lines
1900 .iter()
1901 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
1902 }
1903 ReflowMode::Default => container_lines
1904 .iter()
1905 .any(|line| self.calculate_effective_length(line) > config.line_length.get()),
1906 };
1907
1908 if !needs_reflow {
1909 continue;
1910 }
1911
1912 // Calculate byte range for this container paragraph
1913 let start_range = line_index.whole_line_range(container_start + 1);
1914 let end_line = container_start + container_lines.len() - 1;
1915 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
1916 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
1917 } else {
1918 line_index.whole_line_range(end_line + 1)
1919 };
1920 let byte_range = start_range.start..end_range.end;
1921
1922 // Reflow with adjusted line length (accounting for the 4-space indent)
1923 let reflow_line_length = if config.line_length.is_unlimited() {
1924 usize::MAX
1925 } else {
1926 config.line_length.get().saturating_sub(base_indent_len).max(1)
1927 };
1928 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
1929 let reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
1930
1931 // Re-add the 4-space indent to each reflowed line
1932 let reflowed_with_indent: Vec<String> =
1933 reflowed.iter().map(|line| format!("{base_indent}{line}")).collect();
1934 let reflowed_text = reflowed_with_indent.join(line_ending);
1935
1936 // Preserve trailing newline
1937 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
1938 format!("{reflowed_text}{line_ending}")
1939 } else {
1940 reflowed_text
1941 };
1942
1943 // Only generate a warning if the replacement is different
1944 let original_text = &ctx.content[byte_range.clone()];
1945 if original_text != replacement {
1946 warnings.push(LintWarning {
1947 rule_name: Some(self.name().to_string()),
1948 message: format!(
1949 "Line length {} exceeds {} characters (in MkDocs container)",
1950 container_lines.iter().map(|l| l.len()).max().unwrap_or(0),
1951 config.line_length.get()
1952 ),
1953 line: container_start + 1,
1954 column: 1,
1955 end_line: end_line + 1,
1956 end_column: lines[end_line].chars().count() + 1,
1957 severity: Severity::Warning,
1958 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1959 });
1960 }
1961 continue;
1962 }
1963
1964 // Helper function to detect semantic line markers
1965 let is_semantic_line = |content: &str| -> bool {
1966 let trimmed = content.trim_start();
1967 let semantic_markers = [
1968 "NOTE:",
1969 "WARNING:",
1970 "IMPORTANT:",
1971 "CAUTION:",
1972 "TIP:",
1973 "DANGER:",
1974 "HINT:",
1975 "INFO:",
1976 ];
1977 semantic_markers.iter().any(|marker| trimmed.starts_with(marker))
1978 };
1979
1980 // Helper function to detect fence markers (opening or closing)
1981 let is_fence_marker = |content: &str| -> bool {
1982 let trimmed = content.trim_start();
1983 trimmed.starts_with("```") || trimmed.starts_with("~~~")
1984 };
1985
1986 // Check if this is a list item - handle it specially
1987 let trimmed = lines[i].trim();
1988 if is_list_item(trimmed) {
1989 // Collect the entire list item including continuation lines
1990 let list_start = i;
1991 let (marker, first_content) = extract_list_marker_and_content(lines[i]);
1992 let marker_len = marker.len();
1993 // The normalized marker above is what gets re-emitted; the source marker
1994 // is where the item's content actually starts. Nested blocks move by the
1995 // difference between the two content columns, so the shift must be
1996 // measured against the source, not against the normalized width.
1997 let source_marker = source_list_marker(lines[i]);
1998 let source_content_col = source_marker.as_ref().map_or(marker_len, |m| m.content_col);
1999
2000 // Checkbox ([ ]/[x]/[X]) is inline content, not part of the list marker.
2001 // Use the base bullet/number marker width for continuation recognition
2002 // so that continuation lines at 2+ spaces are collected for "- [ ] " items.
2003 let base_marker_len = if marker.contains("[ ] ") || marker.contains("[x] ") || marker.contains("[X] ") {
2004 marker.find('[').unwrap_or(marker_len)
2005 } else {
2006 marker_len
2007 };
2008
2009 // MkDocs flavor requires at least 4 spaces for list continuation
2010 // after a blank line (multi-paragraph list items). For non-blank
2011 // continuation (lines directly following the marker line), use
2012 // the natural marker width so that 2-space indent is recognized.
2013 let item_indent = ctx.lines[i].indent;
2014 let min_continuation_indent = if ctx.flavor.requires_strict_list_indent() {
2015 // Use 4-space relative indent from the list item's nesting level
2016 item_indent + (base_marker_len - item_indent).max(4)
2017 } else {
2018 marker_len
2019 };
2020 let content_continuation_indent = base_marker_len;
2021
2022 // Track lines and their types (content, code block, fence, nested list)
2023 #[derive(Clone)]
2024 enum LineType {
2025 Content(String),
2026 CodeBlock(String, usize), // content and original indent
2027 SemanticLine(String), // Lines starting with NOTE:, WARNING:, etc that should stay separate
2028 SnippetLine(String), // MkDocs Snippets delimiters (-8<-) that must stay on their own line
2029 DivMarker(String), // Quarto/Pandoc div markers (::: opening or closing)
2030 AdmonitionHeader(String, usize), // header text (e.g. "!!! note") and original indent
2031 AdmonitionContent(String, usize), // body content text and original indent
2032 Table(String, usize), // GFM table row, preserved verbatim with original indent
2033 Empty,
2034 }
2035
2036 let mut list_item_lines: Vec<LineType> = vec![LineType::Content(first_content)];
2037 // Set when collection stops at a nested list item or a nested
2038 // blockquote that belongs to this item. Such structure is reflowed
2039 // independently and is therefore absent from `list_item_lines`/`blocks`,
2040 // but it still keeps the emitted item spanning multiple physical lines,
2041 // which the MD030 multi-line spacing decision below must account for.
2042 let mut has_trailing_nested_structure = false;
2043 i += 1;
2044
2045 // Collect continuation lines using ctx.lines for metadata
2046 while i < lines.len() {
2047 let line_info = &ctx.lines[i];
2048
2049 // Use pre-computed is_blank from ctx
2050 if line_info.is_blank {
2051 // Empty line - check if next line is indented (part of list item)
2052 if i + 1 < lines.len() {
2053 let next_info = &ctx.lines[i + 1];
2054
2055 // Check if next line is indented enough to be continuation
2056 if !next_info.is_blank && next_info.indent >= min_continuation_indent {
2057 // This blank line is between paragraphs/blocks in the list item
2058 list_item_lines.push(LineType::Empty);
2059 i += 1;
2060 continue;
2061 }
2062 }
2063 // No indented line after blank, end of list item
2064 break;
2065 }
2066
2067 // Use pre-computed indent from ctx
2068 let indent = line_info.indent;
2069
2070 // Valid continuation must be indented at least content_continuation_indent.
2071 // For non-blank continuation, use marker_len (e.g. 2 for "- ").
2072 // MkDocs strict 4-space requirement applies only after blank lines.
2073 if indent >= content_continuation_indent {
2074 let trimmed = line_info.content(ctx.content).trim();
2075
2076 // Check for MkDocs admonition lines inside list items BEFORE
2077 // checking in_code_block. Lines inside code blocks within
2078 // admonitions have both in_admonition and in_code_block set;
2079 // admonition membership takes priority so the entire admonition
2080 // structure (including embedded code blocks) is preserved.
2081 if line_info.in_admonition {
2082 let raw_content = line_info.content(ctx.content);
2083 if mkdocs_admonitions::is_admonition_start(raw_content) {
2084 let header_text = raw_content[indent..].trim_end().to_string();
2085 list_item_lines.push(LineType::AdmonitionHeader(header_text, indent));
2086 } else {
2087 let body_text = raw_content[indent..].trim_end().to_string();
2088 list_item_lines.push(LineType::AdmonitionContent(body_text, indent));
2089 }
2090 i += 1;
2091 continue;
2092 }
2093
2094 // Use pre-computed in_code_block from ctx
2095 if line_info.in_code_block {
2096 list_item_lines.push(LineType::CodeBlock(
2097 line_info.content(ctx.content)[indent..].to_string(),
2098 indent,
2099 ));
2100 i += 1;
2101 continue;
2102 }
2103
2104 // A blockquote nested inside the list item is reflowed by the
2105 // blockquote-aware path (it preserves the `>` prefix, including the
2106 // list indent), not as list-item prose. Collecting it as Content
2107 // would strip the markers and reflow `>` as words, collapsing the
2108 // blank `>` line and dropping `>` from wrapped continuations. End
2109 // the item here so the outer loop routes the blockquote line to
2110 // generate_blockquote_paragraph_fix. Uncollect a pending blank so
2111 // the separator between the list prose and the blockquote survives.
2112 if line_info.blockquote.is_some() {
2113 has_trailing_nested_structure = true;
2114 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
2115 list_item_lines.pop();
2116 i -= 1;
2117 }
2118 break;
2119 }
2120
2121 // Check if this is a SIBLING list item (breaks parent)
2122 // Nested lists are indented >= marker_len and are PART of the parent item
2123 // Siblings are at indent < marker_len (at or before parent marker)
2124 if is_list_item(trimmed) && indent < marker_len {
2125 // This is a sibling item at same or higher level - end parent item
2126 break;
2127 }
2128
2129 // Nested list items are always processed independently
2130 // by the outer loop, so break when we encounter one.
2131 // If a blank line was collected before this, uncollect it
2132 // so the outer loop preserves the blank between parent and nested.
2133 if is_list_item(trimmed) && indent >= marker_len {
2134 has_trailing_nested_structure = true;
2135 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
2136 list_item_lines.pop();
2137 i -= 1;
2138 }
2139 break;
2140 }
2141
2142 // Normal continuation vs indented code block.
2143 // Use min_continuation_indent for the threshold since
2144 // code blocks start 4 spaces beyond the expected content
2145 // level (which is min_continuation_indent for MkDocs).
2146 if indent <= min_continuation_indent + 3 {
2147 // Extract content (remove indentation and trailing whitespace)
2148 // Preserve hard breaks (2 trailing spaces) while removing excessive whitespace
2149 // See: https://github.com/rvben/rumdl/issues/76
2150 let content = trim_preserving_hard_break(&line_info.content(ctx.content)[indent..]);
2151
2152 // Check if this is a div marker (::: opening or closing)
2153 // These must be preserved on their own line, not merged into paragraphs
2154 if line_info.is_div_marker {
2155 list_item_lines.push(LineType::DivMarker(content));
2156 }
2157 // Check if this is a fence marker (opening or closing)
2158 // These should be treated as code block lines, not paragraph content
2159 else if is_fence_marker(&content) {
2160 list_item_lines.push(LineType::CodeBlock(content, indent));
2161 }
2162 // Check if this is a semantic line (NOTE:, WARNING:, etc.)
2163 else if is_semantic_line(&content) {
2164 list_item_lines.push(LineType::SemanticLine(content));
2165 }
2166 // Check if this is a snippet block delimiter (-8<- or --8<--)
2167 // These must be preserved on their own lines for MkDocs Snippets extension
2168 else if is_snippet_block_delimiter(&content) {
2169 list_item_lines.push(LineType::SnippetLine(content));
2170 }
2171 // Check if this is a GFM table row. Tables nested inside list
2172 // items must be preserved verbatim — joining them with prose
2173 // breaks the column structure.
2174 //
2175 // `is_potential_table_row` is intentionally permissive at the
2176 // row level: any line with `|` and 2+ cells qualifies. To avoid
2177 // misclassifying prose continuation lines that contain a literal
2178 // pipe (e.g. "use grep | sort to ..."), require one of:
2179 // - the row is pipe-bordered (`| ... |`), the canonical form
2180 // for tables nested in lists; or
2181 // - the next line is a delimiter row (this is a header); or
2182 // - the previous classified line was already a Table (this is
2183 // a continuation row).
2184 else if TableUtils::is_potential_table_row(&content) && {
2185 let pipe_bordered = content.trim().starts_with('|') && content.trim().ends_with('|');
2186 let next_is_delim = ctx
2187 .lines
2188 .get(i + 1)
2189 .is_some_and(|next| TableUtils::is_delimiter_row(next.content(ctx.content)));
2190 let prev_was_table = matches!(list_item_lines.last(), Some(LineType::Table(..)));
2191 pipe_bordered || next_is_delim || prev_was_table
2192 } {
2193 list_item_lines.push(LineType::Table(content, indent));
2194 } else {
2195 list_item_lines.push(LineType::Content(content));
2196 }
2197 i += 1;
2198 } else {
2199 // indent >= min_continuation_indent + 4: indented code block
2200 list_item_lines.push(LineType::CodeBlock(
2201 line_info.content(ctx.content)[indent..].to_string(),
2202 indent,
2203 ));
2204 i += 1;
2205 }
2206 } else {
2207 // Not indented enough, end of list item
2208 break;
2209 }
2210 }
2211
2212 // Determine the output continuation indent.
2213 // Normalize/Default modes canonicalize to min_continuation_indent
2214 // (fixing over-indented continuation). Semantic/SentencePerLine
2215 // modes preserve the user's actual indent since they only fix
2216 // line breaking, not indentation.
2217 let indent_size = match config.reflow_mode {
2218 ReflowMode::SemanticLineBreaks | ReflowMode::SentencePerLine => {
2219 // Find indent of the first plain text continuation line,
2220 // skipping the marker line (index 0), nested list items,
2221 // code blocks, and blank lines.
2222 list_item_lines
2223 .iter()
2224 .enumerate()
2225 .skip(1)
2226 .find_map(|(k, lt)| {
2227 if matches!(lt, LineType::Content(_)) {
2228 Some(ctx.lines[list_start + k].indent)
2229 } else {
2230 None
2231 }
2232 })
2233 .unwrap_or(min_continuation_indent)
2234 }
2235 _ => min_continuation_indent,
2236 };
2237 // For checkbox items in mkdocs flavor, enforce minimum indent so
2238 // continuation lines use the structural list indent (4), not the
2239 // content-aligned indent (6) which Python-Markdown doesn't support
2240 let has_checkbox = base_marker_len < marker_len;
2241 let indent_size = if has_checkbox && ctx.flavor.requires_strict_list_indent() {
2242 indent_size.max(min_continuation_indent)
2243 } else {
2244 indent_size
2245 };
2246
2247 // Split list_item_lines into blocks (paragraphs, code blocks, nested lists, semantic lines, and HTML blocks)
2248 let mut builder = BlockBuilder::new();
2249 for line in &list_item_lines {
2250 match line {
2251 LineType::Empty => builder.feed_blank_line(),
2252 LineType::Content(content) => builder.feed_content(content),
2253 LineType::CodeBlock(content, indent) => builder.feed_code_line(content, *indent),
2254 LineType::SemanticLine(content) => builder.feed_semantic_line(content),
2255 LineType::SnippetLine(content) => builder.feed_snippet_line(content),
2256 LineType::DivMarker(content) => builder.feed_div_marker(content),
2257 LineType::AdmonitionHeader(header_text, indent) => {
2258 builder.feed_admonition_header(header_text, *indent)
2259 }
2260 LineType::AdmonitionContent(content, indent) => {
2261 builder.feed_admonition_content(content, *indent)
2262 }
2263 LineType::Table(content, indent) => builder.feed_table_line(content, *indent),
2264 }
2265 }
2266 let blocks = builder.finalize();
2267
2268 // Helper: check if a line (raw source or stripped content) is exempt
2269 // from line-length checks. Link reference definitions are always exempt;
2270 // standalone link/image lines are exempt when strict mode is off.
2271 // Also checks content after stripping list markers, since list item
2272 // continuation lines may contain link ref defs.
2273 let is_exempt_line = |raw_line: &str| -> bool {
2274 let trimmed = raw_line.trim();
2275 // Link reference definitions: always exempt
2276 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
2277 return true;
2278 }
2279 // Also check after stripping list markers (for list item content)
2280 if is_list_item(trimmed) {
2281 let (_, content) = extract_list_marker_and_content(trimmed);
2282 let content_trimmed = content.trim();
2283 if content_trimmed.starts_with('[')
2284 && content_trimmed.contains("]:")
2285 && LINK_REF_PATTERN.is_match(content_trimmed)
2286 {
2287 return true;
2288 }
2289 }
2290 // Standalone link/image lines: exempt when not strict
2291 if !config.strict && is_standalone_link_or_image_line(raw_line) {
2292 return true;
2293 }
2294 // HTML-only lines: exempt when not strict
2295 if !config.strict && is_html_only_line(raw_line) {
2296 return true;
2297 }
2298 false
2299 };
2300
2301 // Check if reflowing is needed (only for content paragraphs, not code blocks or nested lists)
2302 // Exclude link reference definitions and standalone link lines from content
2303 // so they don't pollute combined_content or trigger false reflow.
2304 let content_lines: Vec<String> = list_item_lines
2305 .iter()
2306 .filter_map(|line| {
2307 if let LineType::Content(s) = line {
2308 if is_exempt_line(s) {
2309 return None;
2310 }
2311 Some(s.clone())
2312 } else {
2313 None
2314 }
2315 })
2316 .collect();
2317
2318 // Check if we need to reflow this list item
2319 // We check the combined content to see if it exceeds length limits
2320 let combined_content = content_lines.join(" ").trim().to_string();
2321
2322 // Helper to check if we should reflow in normalize mode
2323 let should_normalize = || {
2324 // Don't normalize if the list item only contains nested lists, code blocks, or semantic lines
2325 // DO normalize if it has plain text content that spans multiple lines
2326 let has_code_blocks = blocks.iter().any(|b| matches!(b, Block::Code { .. }));
2327 let has_semantic_lines = blocks.iter().any(|b| matches!(b, Block::SemanticLine(_)));
2328 let has_snippet_lines = blocks.iter().any(|b| matches!(b, Block::SnippetLine(_)));
2329 let has_div_markers = blocks.iter().any(|b| matches!(b, Block::DivMarker(_)));
2330 let has_admonitions = blocks.iter().any(|b| matches!(b, Block::Admonition { .. }));
2331 let has_tables = blocks.iter().any(|b| matches!(b, Block::Table { .. }));
2332 let has_paragraphs = blocks.iter().any(|b| matches!(b, Block::Paragraph(_)));
2333
2334 // If we have structural blocks but no paragraphs, don't normalize
2335 if (has_code_blocks
2336 || has_semantic_lines
2337 || has_snippet_lines
2338 || has_div_markers
2339 || has_admonitions
2340 || has_tables)
2341 && !has_paragraphs
2342 {
2343 return false;
2344 }
2345
2346 // If we have paragraphs, check if they span multiple lines or there are multiple blocks
2347 if has_paragraphs {
2348 // Count only paragraphs that contain at least one non-exempt line.
2349 // Paragraphs consisting entirely of link ref defs or standalone links
2350 // should not trigger normalization.
2351 let paragraph_count = blocks
2352 .iter()
2353 .filter(|b| {
2354 if let Block::Paragraph(para_lines) = b {
2355 !para_lines.iter().all(|line| is_exempt_line(line))
2356 } else {
2357 false
2358 }
2359 })
2360 .count();
2361 if paragraph_count > 1 {
2362 // Multiple non-exempt paragraph blocks should be normalized
2363 return true;
2364 }
2365
2366 // Single paragraph block: normalize if it has multiple content lines
2367 if content_lines.len() > 1 {
2368 return true;
2369 }
2370 }
2371
2372 false
2373 };
2374
2375 // Integrate MD030 list-marker spacing (and MD007's text-aligned
2376 // continuation). In Default/Normalize modes — the modes that already
2377 // canonicalize spacing/indent — derive the number of spaces after the
2378 // marker from the configured MD030 values instead of forcing a single
2379 // space, then align continuation lines to the resulting content column.
2380 // Sentence/Semantic modes only adjust line breaks, so they keep the
2381 // marker spacing and indentation already present in the source.
2382 //
2383 // With default MD030 (a single space everywhere) the rebuilt marker is
2384 // byte-identical to the source marker, so this is a no-op and existing
2385 // behaviour is preserved; only a non-default MD030 changes the output.
2386 // The MkDocs flavor enforces a rigid structural indent (4 spaces,
2387 // capped via max_list_continuation_indent) that Python-Markdown
2388 // requires; leave its specialized handling untouched.
2389 let (marker, indent_size, code_indent_shift) =
2390 if matches!(config.reflow_mode, ReflowMode::Default | ReflowMode::Normalize)
2391 && !ctx.flavor.requires_strict_list_indent()
2392 && let Some(li) = ctx.lines[list_start].list_item.as_deref()
2393 {
2394 let bullet_len = li.marker.len();
2395 // The checkbox (e.g. `[ ] `) is content, not part of the list
2396 // marker MD030 governs; carry it over verbatim after the spacing.
2397 let checkbox_tail = marker[base_marker_len..].to_string();
2398 // Shift this item right by its ancestors' cumulative marker
2399 // widening so a nested item stays under its parent's (widened)
2400 // content column. Zero for top-level items and for the whole
2401 // tree under default MD030, where the source indent is preserved
2402 // verbatim (byte-identical output).
2403 let ancestor_shift = list_shift_stack.last().map_or(0isize, |&(_, shift)| shift);
2404 let shifted_indent = (item_indent as isize + ancestor_shift).max(0) as usize;
2405 let indent_prefix = if ancestor_shift == 0 {
2406 marker[..item_indent].to_string()
2407 } else {
2408 " ".repeat(shifted_indent)
2409 };
2410
2411 // Decide single- vs multi-line spacing from the *rewritten* shape,
2412 // not the source. A multi-line source is not enough: plain prose
2413 // continuation collapses onto the marker line during reflow, so a
2414 // two-line bullet that fits becomes a single physical line and must
2415 // use MD030's single-line spacing (otherwise MD013 emits a result
2416 // that MD030 immediately rewrites). The emitted item stays
2417 // multi-line only when reflow cannot collapse it:
2418 // - the prose wraps past the line length, or
2419 // - a structural block remains (code, table, admonition, semantic
2420 // line, snippet, div marker, HTML) that is not joinable prose, or
2421 // - more than one paragraph remains (blank-separated), or
2422 // - a nested list/blockquote follows (reflowed independently, so it
2423 // is absent from `blocks` but still keeps the item multi-line).
2424 // The wrap test uses the single-line content column so it is
2425 // independent of the spacing we are about to choose (avoiding a
2426 // circular result). `ol-align-column` ignores this flag entirely in
2427 // expected_spaces(), so ordered lists are unaffected.
2428 //
2429 // This is the rewritten-shape counterpart of MD030's
2430 // `is_multi_line_list_item` (which keys off the *source*). The two
2431 // are related but technically distinct and intentionally separate;
2432 // if the notion of "multi-line" changes in one, revisit the other.
2433 let single_col = shifted_indent
2434 + bullet_len
2435 + self.list_spacing.expected_spaces(li.is_ordered, false, bullet_len)
2436 + checkbox_tail.len();
2437 let prose_wraps = !combined_content.is_empty()
2438 && self
2439 .calculate_effective_length(&format!("{}{combined_content}", " ".repeat(single_col)))
2440 > config.line_length.effective_limit();
2441 let has_structural_block = blocks.iter().any(|b| !matches!(b, Block::Paragraph(_)));
2442 let multiple_paragraphs =
2443 blocks.iter().filter(|b| matches!(b, Block::Paragraph(_))).count() > 1;
2444 let is_multi =
2445 prose_wraps || has_structural_block || multiple_paragraphs || has_trailing_nested_structure;
2446
2447 let spaces = self.list_spacing.expected_spaces(li.is_ordered, is_multi, bullet_len);
2448 let new_marker = format!("{indent_prefix}{}{}{checkbox_tail}", li.marker, " ".repeat(spaces));
2449 let new_col = new_marker.chars().count();
2450 let shift = new_col as isize - source_content_col as isize;
2451 (new_marker, new_col, shift)
2452 } else {
2453 // MkDocs enforces a rigid structural indent, so the item is emitted
2454 // exactly as written and its nested blocks never move. Re-emitting
2455 // the normalized marker here would narrow the content column while
2456 // leaving those blocks behind.
2457 let marker = source_marker.map_or(marker, |m| m.text);
2458 (marker, indent_size, 0isize)
2459 };
2460 let expected_indent = " ".repeat(indent_size);
2461
2462 let needs_reflow = match config.reflow_mode {
2463 ReflowMode::Normalize => {
2464 // Only reflow if:
2465 // 1. Any non-exempt paragraph, when joined, exceeds the limit, OR
2466 // 2. Any admonition content line exceeds the limit, OR
2467 // 3. The list item should be normalized (has multi-line plain text)
2468 let any_paragraph_exceeds = blocks.iter().any(|block| match block {
2469 Block::Paragraph(para_lines) => {
2470 if para_lines.iter().all(|line| is_exempt_line(line)) {
2471 return false;
2472 }
2473 let joined = para_lines.join(" ");
2474 let with_marker = format!("{}{}", " ".repeat(indent_size), joined.trim());
2475 self.calculate_effective_length(&with_marker) > config.line_length.get()
2476 }
2477 Block::Admonition {
2478 content_lines,
2479 header_indent,
2480 ..
2481 } => content_lines.iter().any(|(content, indent)| {
2482 if content.is_empty() {
2483 return false;
2484 }
2485 let with_indent = format!("{}{}", " ".repeat(*indent.max(header_indent)), content);
2486 self.calculate_effective_length(&with_indent) > config.line_length.get()
2487 }),
2488 _ => false,
2489 });
2490 if any_paragraph_exceeds {
2491 true
2492 } else {
2493 should_normalize()
2494 }
2495 }
2496 ReflowMode::SentencePerLine => {
2497 // Check if list item has multiple sentences
2498 let sentences = split_into_sentences(&combined_content);
2499 sentences.len() > 1
2500 }
2501 ReflowMode::SemanticLineBreaks => {
2502 let sentences = split_into_sentences(&combined_content);
2503 sentences.len() > 1
2504 || (list_start..i).any(|line_idx| {
2505 let line = lines[line_idx];
2506 let trimmed = line.trim();
2507 if trimmed.is_empty() || is_exempt_line(line) {
2508 return false;
2509 }
2510 self.calculate_effective_length(line) > config.line_length.get()
2511 })
2512 }
2513 ReflowMode::Default => {
2514 // In default mode, only reflow if any individual non-exempt line exceeds limit
2515 (list_start..i).any(|line_idx| {
2516 let line = lines[line_idx];
2517 let trimmed = line.trim();
2518 // Skip blank lines and exempt lines
2519 if trimmed.is_empty() || is_exempt_line(line) {
2520 return false;
2521 }
2522 self.calculate_effective_length(line) > config.line_length.get()
2523 })
2524 }
2525 };
2526
2527 // Record this item's frame so its nested children inherit the shift.
2528 // Only a reflowed item's marker actually moves; an unreflowed one keeps
2529 // its source position and so contributes no shift to its children. The
2530 // threshold that decides which following lines are inside this item is
2531 // the normalized marker width, NOT `source_content_col`: the collection
2532 // loop above gathers continuations by that same width, so the frame
2533 // boundary must match it or the two would disagree about ownership of
2534 // lines indented between the normalized and the source content column.
2535 list_shift_stack.push((marker_len, if needs_reflow { code_indent_shift } else { 0 }));
2536
2537 if needs_reflow {
2538 let start_range = line_index.whole_line_range(list_start + 1);
2539 let end_line = i - 1;
2540 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
2541 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
2542 } else {
2543 line_index.whole_line_range(end_line + 1)
2544 };
2545 let byte_range = start_range.start..end_range.end;
2546
2547 // Reflow each block (paragraphs only, preserve code blocks)
2548 // When line_length = 0 (no limit), use a very large value for reflow
2549 let reflow_line_length = if config.line_length.is_unlimited() {
2550 usize::MAX
2551 } else {
2552 config.line_length.get().saturating_sub(indent_size).max(1)
2553 };
2554 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
2555
2556 let mut result: Vec<String> = Vec::new();
2557 let mut is_first_block = true;
2558
2559 for (block_idx, block) in blocks.iter().enumerate() {
2560 match block {
2561 Block::Paragraph(para_lines) => {
2562 // If every line in this paragraph is exempt (link ref defs,
2563 // standalone links), preserve the paragraph verbatim instead
2564 // of reflowing it. Reflowing would corrupt link ref defs.
2565 let all_exempt = para_lines.iter().all(|line| is_exempt_line(line));
2566
2567 if all_exempt {
2568 for (idx, line) in para_lines.iter().enumerate() {
2569 if is_first_block && idx == 0 {
2570 result.push(format!("{marker}{line}"));
2571 is_first_block = false;
2572 } else {
2573 result.push(format!("{expected_indent}{line}"));
2574 }
2575 }
2576 } else {
2577 // Split the paragraph into segments at hard break boundaries
2578 // Each segment can be reflowed independently
2579 let segments = split_into_segments(para_lines);
2580
2581 for (segment_idx, segment) in segments.iter().enumerate() {
2582 // Check if this segment ends with a hard break and what type
2583 let hard_break_type = segment.last().and_then(|line| {
2584 let line = line.strip_suffix('\r').unwrap_or(line);
2585 if line.ends_with('\\') {
2586 Some("\\")
2587 } else if line.ends_with(" ") {
2588 Some(" ")
2589 } else {
2590 None
2591 }
2592 });
2593
2594 // Join and reflow the segment (removing the hard break marker for processing)
2595 let segment_for_reflow: Vec<String> = segment
2596 .iter()
2597 .map(|line| {
2598 // Strip hard break marker (2 spaces or backslash) for reflow processing
2599 if line.ends_with('\\') {
2600 line[..line.len() - 1].trim_end().to_string()
2601 } else if line.ends_with(" ") {
2602 line[..line.len() - 2].trim_end().to_string()
2603 } else {
2604 line.clone()
2605 }
2606 })
2607 .collect();
2608
2609 let segment_text = segment_for_reflow.join(" ").trim().to_string();
2610 if !segment_text.is_empty() {
2611 let reflowed =
2612 crate::utils::text_reflow::reflow_line(&segment_text, &reflow_options);
2613
2614 if is_first_block && segment_idx == 0 {
2615 // First segment of first block starts with marker
2616 result.push(format!("{marker}{}", reflowed[0]));
2617 for line in reflowed.iter().skip(1) {
2618 result.push(format!("{expected_indent}{line}"));
2619 }
2620 is_first_block = false;
2621 } else {
2622 // Subsequent segments
2623 for line in reflowed {
2624 result.push(format!("{expected_indent}{line}"));
2625 }
2626 }
2627
2628 // If this segment had a hard break, add it back to the last line
2629 // Preserve the original hard break format (backslash or two spaces)
2630 if let Some(break_marker) = hard_break_type
2631 && let Some(last_line) = result.last_mut()
2632 {
2633 last_line.push_str(break_marker);
2634 }
2635 }
2636 }
2637 }
2638
2639 // Add blank line after paragraph block if there's a next block.
2640 // Check if next block is a code block that doesn't want a preceding blank.
2641 // Also don't add blank lines before snippet lines (they should stay tight).
2642 // Only add if not already ending with one (avoids double blanks).
2643 if block_idx < blocks.len() - 1 {
2644 let next_block = &blocks[block_idx + 1];
2645 let should_add_blank = match next_block {
2646 Block::Code {
2647 has_preceding_blank, ..
2648 } => *has_preceding_blank,
2649 Block::Table {
2650 has_preceding_blank, ..
2651 } => *has_preceding_blank,
2652 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2653 _ => true, // For all other blocks, add blank line
2654 };
2655 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2656 result.push(String::new());
2657 }
2658 }
2659 }
2660 Block::Code {
2661 lines: code_lines,
2662 has_preceding_blank: _,
2663 } => {
2664 // Preserve code blocks as-is with original indentation
2665 // NOTE: Blank line before code block is handled by the previous block
2666 // (see paragraph block's logic above)
2667
2668 for (idx, (content, orig_indent)) in code_lines.iter().enumerate() {
2669 if is_first_block && idx == 0 {
2670 // First line of first block gets marker
2671 result.push(format!(
2672 "{marker}{}",
2673 " ".repeat(orig_indent - marker_len) + content.as_str()
2674 ));
2675 is_first_block = false;
2676 } else if content.is_empty() {
2677 result.push(String::new());
2678 } else {
2679 // Shift nested code with the marker so it stays
2680 // aligned under content when MD030 widens spacing.
2681 result.push(format!(
2682 "{}{}",
2683 " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
2684 content
2685 ));
2686 }
2687 }
2688 }
2689 Block::SemanticLine(content) => {
2690 // Preserve semantic lines (NOTE:, WARNING:, etc.) as-is on their own line.
2691 // Only add blank before if not already ending with one.
2692 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
2693 result.push(String::new());
2694 }
2695
2696 if is_first_block {
2697 // First block starts with marker
2698 result.push(format!("{marker}{content}"));
2699 is_first_block = false;
2700 } else {
2701 // Subsequent blocks use expected indent
2702 result.push(format!("{expected_indent}{content}"));
2703 }
2704
2705 // Add blank line after semantic line if there's a next block.
2706 // Only add if not already ending with one.
2707 if block_idx < blocks.len() - 1 {
2708 let next_block = &blocks[block_idx + 1];
2709 let should_add_blank = match next_block {
2710 Block::Code {
2711 has_preceding_blank, ..
2712 } => *has_preceding_blank,
2713 Block::Table {
2714 has_preceding_blank, ..
2715 } => *has_preceding_blank,
2716 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2717 _ => true, // For all other blocks, add blank line
2718 };
2719 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2720 result.push(String::new());
2721 }
2722 }
2723 }
2724 Block::SnippetLine(content) => {
2725 // Preserve snippet delimiters (-8<-) as-is on their own line
2726 // Unlike semantic lines, snippet lines don't add extra blank lines
2727 if is_first_block {
2728 // First block starts with marker
2729 result.push(format!("{marker}{content}"));
2730 is_first_block = false;
2731 } else {
2732 // Subsequent blocks use expected indent
2733 result.push(format!("{expected_indent}{content}"));
2734 }
2735 // No blank lines added before or after snippet delimiters
2736 }
2737 Block::DivMarker(content) => {
2738 // Preserve div markers (::: opening or closing) as-is on their own line
2739 if is_first_block {
2740 result.push(format!("{marker}{content}"));
2741 is_first_block = false;
2742 } else {
2743 result.push(format!("{expected_indent}{content}"));
2744 }
2745 }
2746 Block::Html {
2747 lines: html_lines,
2748 has_preceding_blank: _,
2749 } => {
2750 // Preserve HTML blocks exactly as-is with original indentation
2751 // NOTE: Blank line before HTML block is handled by the previous block
2752
2753 for (idx, line) in html_lines.iter().enumerate() {
2754 if is_first_block && idx == 0 {
2755 // First line of first block gets marker
2756 result.push(format!("{marker}{line}"));
2757 is_first_block = false;
2758 } else if line.is_empty() {
2759 // Preserve blank lines inside HTML blocks
2760 result.push(String::new());
2761 } else {
2762 // Preserve lines with their original content (already includes indentation)
2763 result.push(format!("{expected_indent}{line}"));
2764 }
2765 }
2766
2767 // Add blank line after HTML block if there's a next block.
2768 // Only add if not already ending with one (avoids double blanks
2769 // when the HTML block itself contained a trailing blank line).
2770 if block_idx < blocks.len() - 1 {
2771 let next_block = &blocks[block_idx + 1];
2772 let should_add_blank = match next_block {
2773 Block::Code {
2774 has_preceding_blank, ..
2775 } => *has_preceding_blank,
2776 Block::Html {
2777 has_preceding_blank, ..
2778 } => *has_preceding_blank,
2779 Block::Table {
2780 has_preceding_blank, ..
2781 } => *has_preceding_blank,
2782 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2783 _ => true, // For all other blocks, add blank line
2784 };
2785 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2786 result.push(String::new());
2787 }
2788 }
2789 }
2790 Block::Table {
2791 lines: table_lines,
2792 has_preceding_blank: _,
2793 } => {
2794 // Preserve table rows verbatim with their original indentation.
2795 // Reflowing rows would corrupt column alignment and inject `|`
2796 // characters mid-paragraph (issue #590).
2797 // The leading blank line is emitted by the previous block.
2798 for (idx, (content, orig_indent)) in table_lines.iter().enumerate() {
2799 if is_first_block && idx == 0 {
2800 // First line of first block gets the list marker
2801 result.push(format!(
2802 "{marker}{}",
2803 " ".repeat(orig_indent.saturating_sub(marker_len)) + content.as_str()
2804 ));
2805 is_first_block = false;
2806 } else {
2807 // Shift nested table rows with the marker so they
2808 // stay aligned when MD030 widens marker spacing.
2809 result.push(format!(
2810 "{}{}",
2811 " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
2812 content
2813 ));
2814 }
2815 }
2816
2817 // Add blank line after table block if there's a next block.
2818 if block_idx < blocks.len() - 1 {
2819 let next_block = &blocks[block_idx + 1];
2820 let should_add_blank = match next_block {
2821 Block::Code {
2822 has_preceding_blank, ..
2823 } => *has_preceding_blank,
2824 Block::Html {
2825 has_preceding_blank, ..
2826 } => *has_preceding_blank,
2827 Block::Table {
2828 has_preceding_blank, ..
2829 } => *has_preceding_blank,
2830 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2831 _ => true,
2832 };
2833 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2834 result.push(String::new());
2835 }
2836 }
2837 }
2838 Block::Admonition {
2839 header,
2840 header_indent,
2841 content_lines: admon_lines,
2842 } => {
2843 // Reconstruct admonition block with header at original indent
2844 // and body content reflowed to fit within the line length limit
2845
2846 // Add blank line before admonition if not first block
2847 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
2848 result.push(String::new());
2849 }
2850
2851 // Output the header at its original indent
2852 let header_indent_str = " ".repeat(*header_indent);
2853 if is_first_block {
2854 result.push(format!(
2855 "{marker}{}",
2856 " ".repeat(header_indent.saturating_sub(marker_len)) + header.as_str()
2857 ));
2858 is_first_block = false;
2859 } else {
2860 result.push(format!("{header_indent_str}{header}"));
2861 }
2862
2863 // Derive body indent from the first non-empty content line's
2864 // stored indent, falling back to header_indent + 4 for
2865 // empty-body admonitions
2866 let body_indent = admon_lines
2867 .iter()
2868 .find(|(content, _)| !content.is_empty())
2869 .map_or(header_indent + 4, |(_, indent)| *indent);
2870 let body_indent_str = " ".repeat(body_indent);
2871
2872 // Segment body content into code blocks (verbatim) and
2873 // text paragraphs (reflowable), separated by blank lines.
2874 // Code lines store (content, orig_indent) to reconstruct
2875 // internal indentation relative to body_indent.
2876 enum AdmonSegment {
2877 Text(Vec<String>),
2878 Code(Vec<(String, usize)>),
2879 }
2880
2881 let mut segments: Vec<AdmonSegment> = Vec::new();
2882 let mut current_text: Vec<String> = Vec::new();
2883 let mut current_code: Vec<(String, usize)> = Vec::new();
2884 let mut in_admon_code = false;
2885 // Track the opening fence character so closing fences
2886 // must match (backticks close backticks, tildes close tildes)
2887 let mut fence_char: char = '`';
2888
2889 // Opening fences: ``` or ~~~ followed by optional info string
2890 let get_opening_fence = |s: &str| -> Option<(char, usize)> {
2891 let t = s.trim_start();
2892 if t.starts_with("```") {
2893 Some(('`', t.bytes().take_while(|&b| b == b'`').count()))
2894 } else if t.starts_with("~~~") {
2895 Some(('~', t.bytes().take_while(|&b| b == b'~').count()))
2896 } else {
2897 None
2898 }
2899 };
2900 // Closing fences: ONLY fence chars + optional trailing spaces
2901 let get_closing_fence = |s: &str| -> Option<(char, usize)> {
2902 let t = s.trim();
2903 if t.starts_with("```") && t.bytes().all(|b| b == b'`') {
2904 Some(('`', t.len()))
2905 } else if t.starts_with("~~~") && t.bytes().all(|b| b == b'~') {
2906 Some(('~', t.len()))
2907 } else {
2908 None
2909 }
2910 };
2911 let mut fence_len: usize = 3;
2912
2913 for (content, orig_indent) in admon_lines {
2914 if in_admon_code {
2915 // Closing fence must use the same character, be
2916 // at least as long, and have no info string
2917 if let Some((ch, len)) = get_closing_fence(content)
2918 && ch == fence_char
2919 && len >= fence_len
2920 {
2921 current_code.push((content.clone(), *orig_indent));
2922 in_admon_code = false;
2923 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
2924 continue;
2925 }
2926 current_code.push((content.clone(), *orig_indent));
2927 } else if let Some((ch, len)) = get_opening_fence(content) {
2928 if !current_text.is_empty() {
2929 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
2930 }
2931 in_admon_code = true;
2932 fence_char = ch;
2933 fence_len = len;
2934 current_code.push((content.clone(), *orig_indent));
2935 } else if content.is_empty() {
2936 if !current_text.is_empty() {
2937 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
2938 }
2939 } else {
2940 current_text.push(content.clone());
2941 }
2942 }
2943 if in_admon_code && !current_code.is_empty() {
2944 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
2945 }
2946 if !current_text.is_empty() {
2947 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
2948 }
2949
2950 // Build reflow options once for all text segments
2951 let admon_reflow_length = if config.line_length.is_unlimited() {
2952 usize::MAX
2953 } else {
2954 config.line_length.get().saturating_sub(body_indent).max(1)
2955 };
2956
2957 let admon_reflow_options = Self::reflow_options(ctx, config, admon_reflow_length);
2958
2959 // Output each segment
2960 for segment in &segments {
2961 // Blank line before each segment (after the header or previous segment)
2962 result.push(String::new());
2963
2964 match segment {
2965 AdmonSegment::Code(lines) => {
2966 for (line, orig_indent) in lines {
2967 if line.is_empty() {
2968 // Preserve blank lines inside code blocks
2969 result.push(String::new());
2970 } else {
2971 // Reconstruct with body_indent + any extra
2972 // indentation the line had beyond body_indent
2973 let extra = orig_indent.saturating_sub(body_indent);
2974 let indent_str = " ".repeat(body_indent + extra);
2975 result.push(format!("{indent_str}{line}"));
2976 }
2977 }
2978 }
2979 AdmonSegment::Text(lines) => {
2980 let paragraph_text = lines.join(" ").trim().to_string();
2981 if paragraph_text.is_empty() {
2982 continue;
2983 }
2984 let reflowed = crate::utils::text_reflow::reflow_line(
2985 ¶graph_text,
2986 &admon_reflow_options,
2987 );
2988 for line in &reflowed {
2989 result.push(format!("{body_indent_str}{line}"));
2990 }
2991 }
2992 }
2993 }
2994
2995 // Add blank line after admonition if there's a next block
2996 if block_idx < blocks.len() - 1 {
2997 let next_block = &blocks[block_idx + 1];
2998 let should_add_blank = match next_block {
2999 Block::Code {
3000 has_preceding_blank, ..
3001 } => *has_preceding_blank,
3002 Block::Table {
3003 has_preceding_blank, ..
3004 } => *has_preceding_blank,
3005 Block::SnippetLine(_) | Block::DivMarker(_) => false,
3006 _ => true,
3007 };
3008 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
3009 result.push(String::new());
3010 }
3011 }
3012 }
3013 }
3014 }
3015
3016 let reflowed_text = result.join(line_ending);
3017
3018 // Preserve trailing newline
3019 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
3020 format!("{reflowed_text}{line_ending}")
3021 } else {
3022 reflowed_text
3023 };
3024
3025 // Get the original text to compare
3026 let original_text = &ctx.content[byte_range.clone()];
3027
3028 // Physical-line-length scan, shared by the Normalize-mode gate and its
3029 // message. The list-item reflow preserves code blocks, HTML blocks,
3030 // admonition headers, fence markers, semantic markers, and snippet/div
3031 // markers verbatim; only paragraph content and admonition bodies are
3032 // restructured. Only those lines drive the length warning, so that
3033 // preserved-but-overlong content does not keep the paragraph-level
3034 // warning alive when the reflow would not fix that line.
3035 let should_count_for_length = |line_idx: usize| -> bool {
3036 let line = lines[line_idx];
3037 let trimmed = line.trim();
3038 if trimmed.is_empty() || is_exempt_line(line) {
3039 return false;
3040 }
3041 let info = &ctx.lines[line_idx];
3042 if info.in_code_block || info.in_html_block {
3043 return false;
3044 }
3045 if info.in_admonition && mkdocs_admonitions::is_admonition_start(line) {
3046 return false;
3047 }
3048 if is_fence_marker(line) || is_semantic_line(line) {
3049 return false;
3050 }
3051 if is_snippet_block_delimiter(line) {
3052 return false;
3053 }
3054 if line.trim_start().starts_with(":::") {
3055 return false;
3056 }
3057 true
3058 };
3059 let max_physical_length = (list_start..i)
3060 .filter(|&idx| should_count_for_length(idx))
3061 .map(|idx| self.calculate_effective_length(lines[idx]))
3062 .max()
3063 .unwrap_or(0);
3064 // `line-length = 0` means "no limit", so no physical line can be
3065 // "over"; the message below then describes a structural join rather
3066 // than a length violation.
3067 let any_paragraph_line_over =
3068 !config.line_length.is_unlimited() && max_physical_length > config.line_length.get();
3069
3070 // Normalize mode reflows list-item prose just like paragraphs:
3071 // joining continuation lines and re-wrapping to `line-length`.
3072 // `prose_changed` is true only when the reflow alters the words or
3073 // line breaks, not when it would merely re-indent continuation
3074 // lines or trim trailing whitespace. Comparing the texts with each
3075 // line's leading and trailing whitespace removed isolates "did the
3076 // words/line breaks change" from "did the surrounding whitespace
3077 // change". Continuation indentation is MD077's responsibility and
3078 // trailing whitespace is MD009's; an MD013 warning for either would
3079 // both duplicate those rules and resurface a persistent advisory on
3080 // already-fitting items that users disable MD013 fixing to avoid.
3081 let prose_changed = {
3082 let stripped = |text: &str| text.lines().map(str::trim).collect::<Vec<_>>().join("\n");
3083 stripped(original_text) != stripped(&replacement)
3084 };
3085 // Warn when the reflow rewraps prose (the normalize feature for
3086 // list items), or when a physical line genuinely exceeds the limit
3087 // and the reflow can change something (a true length violation,
3088 // even if all that changes is the continuation indent). A line that
3089 // is already optimal in both respects produces no warning.
3090 let gate_ok = prose_changed || (any_paragraph_line_over && original_text != replacement);
3091 if gate_ok {
3092 // Generate an appropriate message based on why reflow is needed
3093 let message = match config.reflow_mode {
3094 ReflowMode::SentencePerLine => {
3095 let num_sentences = split_into_sentences(&combined_content).len();
3096 let num_lines = content_lines.len();
3097 if num_lines == 1 {
3098 // Single line with multiple sentences
3099 format!("Line contains {num_sentences} sentences (one sentence per line required)")
3100 } else {
3101 // Multiple lines - could be split sentences or mixed
3102 format!(
3103 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
3104 )
3105 }
3106 }
3107 ReflowMode::SemanticLineBreaks => {
3108 let num_sentences = split_into_sentences(&combined_content).len();
3109 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
3110 }
3111 ReflowMode::Normalize => {
3112 // When a physical line genuinely exceeds the limit, report
3113 // it as a length violation. Otherwise the reflow is a
3114 // structural normalization (joining/re-wrapping multi-line
3115 // content that already fits), mirroring the paragraph path.
3116 if any_paragraph_line_over {
3117 format!(
3118 "Line length {} exceeds {} characters",
3119 max_physical_length,
3120 config.line_length.get()
3121 )
3122 } else {
3123 format!(
3124 "List item could be normalized to use line length of {} characters",
3125 config.line_length.get()
3126 )
3127 }
3128 }
3129 ReflowMode::Default => {
3130 // Report the actual longest non-exempt line, not the combined content
3131 let max_length = (list_start..i)
3132 .filter(|&line_idx| {
3133 let line = lines[line_idx];
3134 let trimmed = line.trim();
3135 !trimmed.is_empty() && !is_exempt_line(line)
3136 })
3137 .map(|line_idx| self.calculate_effective_length(lines[line_idx]))
3138 .max()
3139 .unwrap_or(0);
3140 format!(
3141 "Line length {} exceeds {} characters",
3142 max_length,
3143 config.line_length.get()
3144 )
3145 }
3146 };
3147
3148 warnings.push(LintWarning {
3149 rule_name: Some(self.name().to_string()),
3150 message,
3151 line: list_start + 1,
3152 column: 1,
3153 end_line: end_line + 1,
3154 end_column: lines[end_line].chars().count() + 1,
3155 severity: Severity::Warning,
3156 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
3157 });
3158 }
3159 }
3160 continue;
3161 }
3162
3163 // Found start of a paragraph - collect all lines in it
3164 let paragraph_start = i;
3165 let mut paragraph_lines = vec![lines[i]];
3166 i += 1;
3167
3168 while i < lines.len() {
3169 let next_line = lines[i];
3170 let next_line_num = i + 1;
3171 let next_trimmed = next_line.trim();
3172
3173 // Stop at paragraph boundaries
3174 if next_trimmed.is_empty()
3175 || ctx.line_info(next_line_num).is_some_and(|info| info.in_code_block)
3176 || ctx.line_info(next_line_num).is_some_and(|info| info.in_front_matter)
3177 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_block)
3178 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_comment)
3179 || ctx.line_info(next_line_num).is_some_and(|info| info.in_esm_block)
3180 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_expression)
3181 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_block)
3182 || ctx.line_info(next_line_num).is_some_and(|info| info.in_mdx_comment)
3183 || ctx
3184 .line_info(next_line_num)
3185 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
3186 || (next_line_num > 0
3187 && next_line_num <= ctx.lines.len()
3188 && ctx.lines[next_line_num - 1].blockquote.is_some())
3189 || next_trimmed.starts_with('#')
3190 || TableUtils::is_potential_table_row(next_line)
3191 || is_list_item(next_trimmed)
3192 || is_horizontal_rule(next_line)
3193 || (next_trimmed.starts_with('[') && next_line.contains("]:"))
3194 || is_template_directive_only(next_line)
3195 || is_standalone_attr_list(next_line)
3196 || is_snippet_block_delimiter(next_line)
3197 || ctx.line_info(next_line_num).is_some_and(|info| info.is_div_marker)
3198 || is_html_only_line(next_line)
3199 || (!config.strict && is_standalone_link_or_image_line(next_line))
3200 {
3201 break;
3202 }
3203
3204 // Check if the previous line ends with a hard break (2+ spaces or backslash)
3205 if i > 0 && has_hard_break(lines[i - 1]) {
3206 // Don't include lines after hard breaks in the same paragraph
3207 break;
3208 }
3209
3210 paragraph_lines.push(next_line);
3211 i += 1;
3212 }
3213
3214 // Compute the common leading indent of all non-empty paragraph lines,
3215 // but only when those lines are structurally inside a list block.
3216 // Indented continuation lines that follow a nested list arrive here
3217 // with their structural indentation intact (e.g. 2 spaces for a
3218 // top-level list item). Stripping the indent before reflow and
3219 // re-applying it afterward prevents the fixer from moving those
3220 // lines to column 0.
3221 //
3222 // The list-block guard is essential: top-level paragraphs that happen
3223 // to start with spaces (insignificant in Markdown) must NOT have those
3224 // spaces preserved or injected by the fixer.
3225 let common_indent: String = if ctx.is_in_list_block(paragraph_start + 1) {
3226 let min_len = paragraph_lines
3227 .iter()
3228 .filter(|l| !l.trim().is_empty())
3229 .map(|l| l.len() - l.trim_start().len())
3230 .min()
3231 .unwrap_or(0);
3232 paragraph_lines
3233 .iter()
3234 .find(|l| !l.trim().is_empty())
3235 .map(|l| l[..min_len].to_string())
3236 .unwrap_or_default()
3237 } else {
3238 String::new()
3239 };
3240
3241 // Combine paragraph lines into a single string for processing.
3242 // This must be done BEFORE the needs_reflow check for sentence-per-line mode.
3243 let paragraph_text = if common_indent.is_empty() {
3244 paragraph_lines.join(" ")
3245 } else {
3246 paragraph_lines
3247 .iter()
3248 .map(|l| {
3249 if l.starts_with(common_indent.as_str()) {
3250 &l[common_indent.len()..]
3251 } else {
3252 l.trim_start()
3253 }
3254 })
3255 .collect::<Vec<_>>()
3256 .join(" ")
3257 };
3258
3259 // Skip reflowing if this paragraph contains definition list items
3260 // Definition lists are multi-line structures that should not be joined
3261 let contains_definition_list = paragraph_lines
3262 .iter()
3263 .any(|line| crate::utils::is_definition_list_item(line));
3264
3265 if contains_definition_list {
3266 // Don't reflow definition lists - skip this paragraph
3267 i = paragraph_start + paragraph_lines.len();
3268 continue;
3269 }
3270
3271 // Skip reflowing if this paragraph contains MkDocs Snippets markers
3272 // Snippets blocks (-8<- ... -8<-) should be preserved exactly
3273 let contains_snippets = paragraph_lines.iter().any(|line| is_snippet_block_delimiter(line));
3274
3275 if contains_snippets {
3276 // Don't reflow Snippets blocks - skip this paragraph
3277 i = paragraph_start + paragraph_lines.len();
3278 continue;
3279 }
3280
3281 // Check if this paragraph needs reflowing
3282 let needs_reflow = match config.reflow_mode {
3283 ReflowMode::Normalize => self.normalize_mode_needs_reflow(paragraph_lines.iter().copied(), config),
3284 ReflowMode::SentencePerLine => {
3285 // In sentence-per-line mode, check if the JOINED paragraph has multiple sentences
3286 // Note: we check the joined text because sentences can span multiple lines
3287 let sentences = split_into_sentences(¶graph_text);
3288
3289 // Always reflow if multiple sentences on one line
3290 if sentences.len() > 1 {
3291 true
3292 } else if paragraph_lines.len() > 1 {
3293 // For single-sentence paragraphs spanning multiple lines:
3294 // Reflow if they COULD fit on one line (respecting line-length constraint)
3295 if config.line_length.is_unlimited() {
3296 // No line-length constraint - always join single sentences
3297 true
3298 } else {
3299 // Only join if it fits within line-length.
3300 // paragraph_text has the common indent stripped, so add it
3301 // back to get the true output length before comparing.
3302 let effective_length =
3303 self.calculate_effective_length(¶graph_text) + common_indent.len();
3304 effective_length <= config.line_length.get()
3305 }
3306 } else {
3307 false
3308 }
3309 }
3310 ReflowMode::SemanticLineBreaks => {
3311 let sentences = split_into_sentences(¶graph_text);
3312 // Reflow if multiple sentences, multiple lines, or any line exceeds limit
3313 sentences.len() > 1
3314 || paragraph_lines.len() > 1
3315 || paragraph_lines
3316 .iter()
3317 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3318 }
3319 ReflowMode::Default => {
3320 // In default mode, only reflow if lines exceed limit
3321 paragraph_lines
3322 .iter()
3323 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3324 }
3325 };
3326
3327 if needs_reflow {
3328 // Calculate byte range for this paragraph
3329 // Use whole_line_range for each line and combine
3330 let start_range = line_index.whole_line_range(paragraph_start + 1);
3331 let end_line = paragraph_start + paragraph_lines.len() - 1;
3332
3333 // For the last line, we want to preserve any trailing newline
3334 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
3335 // Last line without trailing newline - use line_text_range
3336 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
3337 } else {
3338 // Not the last line or has trailing newline - use whole_line_range
3339 line_index.whole_line_range(end_line + 1)
3340 };
3341
3342 let byte_range = start_range.start..end_range.end;
3343
3344 // Check if the paragraph ends with a hard break and what type
3345 let hard_break_type = paragraph_lines.last().and_then(|line| {
3346 let line = line.strip_suffix('\r').unwrap_or(line);
3347 if line.ends_with('\\') {
3348 Some("\\")
3349 } else if line.ends_with(" ") {
3350 Some(" ")
3351 } else {
3352 None
3353 }
3354 });
3355
3356 // Reflow the paragraph
3357 // When line_length = 0 (no limit), use a very large value for reflow
3358 let reflow_line_length = if config.line_length.is_unlimited() {
3359 usize::MAX
3360 } else {
3361 config.line_length.get()
3362 };
3363 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
3364 let mut reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
3365
3366 // Re-apply the common indent to each non-empty reflowed line so
3367 // that the replacement preserves the original structural indentation.
3368 if !common_indent.is_empty() {
3369 for line in &mut reflowed {
3370 if !line.is_empty() {
3371 *line = format!("{common_indent}{line}");
3372 }
3373 }
3374 }
3375
3376 // If the original paragraph ended with a hard break, preserve it
3377 // Preserve the original hard break format (backslash or two spaces)
3378 if let Some(break_marker) = hard_break_type
3379 && !reflowed.is_empty()
3380 {
3381 let last_idx = reflowed.len() - 1;
3382 if !has_hard_break(&reflowed[last_idx]) {
3383 reflowed[last_idx].push_str(break_marker);
3384 }
3385 }
3386
3387 let reflowed_text = reflowed.join(line_ending);
3388
3389 // Preserve trailing newline if the original paragraph had one
3390 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
3391 format!("{reflowed_text}{line_ending}")
3392 } else {
3393 reflowed_text
3394 };
3395
3396 // Get the original text to compare
3397 let original_text = &ctx.content[byte_range.clone()];
3398
3399 // Only generate a warning if the replacement is different from the original
3400 if original_text != replacement {
3401 // Determine which line ranges and messages to report based on the reflow mode.
3402 let warnings_to_report: Vec<(usize, usize, String)> = match config.reflow_mode {
3403 ReflowMode::Default => {
3404 // In default mode, report a warning for *every* line in the paragraph
3405 // that exceeds the limit. Each warning will carry the same paragraph-level
3406 // fix, making all of them auto-fixable.
3407 paragraph_lines
3408 .iter()
3409 .enumerate()
3410 .filter(|(_, line)| self.calculate_effective_length(line) > config.line_length.get())
3411 .map(|(idx, _)| {
3412 let violating_line = paragraph_start + idx + 1;
3413 (
3414 violating_line,
3415 violating_line,
3416 format!("Line length exceeds {} characters", config.line_length.get()),
3417 )
3418 })
3419 .collect()
3420 }
3421 ReflowMode::Normalize => {
3422 // In normalize mode, report the whole paragraph as needing normalization.
3423 vec![(
3424 paragraph_start + 1,
3425 end_line + 1,
3426 format!(
3427 "Paragraph could be normalized to use line length of {} characters",
3428 config.line_length.get()
3429 ),
3430 )]
3431 }
3432 ReflowMode::SentencePerLine => {
3433 // In sentence-per-line mode, highlight the entire paragraph that needs reformatting.
3434 let num_sentences = split_into_sentences(¶graph_text).len();
3435 let message = if paragraph_lines.len() == 1 {
3436 // Single line with multiple sentences
3437 format!("Line contains {num_sentences} sentences (one sentence per line required)")
3438 } else {
3439 // Multiple lines - could be split sentences or mixed
3440 let num_lines = paragraph_lines.len();
3441 format!(
3442 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
3443 )
3444 };
3445 vec![(paragraph_start + 1, paragraph_start + paragraph_lines.len(), message)]
3446 }
3447 ReflowMode::SemanticLineBreaks => {
3448 // In semantic-line-breaks mode, highlight the entire paragraph.
3449 let num_sentences = split_into_sentences(¶graph_text).len();
3450 vec![(
3451 paragraph_start + 1,
3452 paragraph_start + paragraph_lines.len(),
3453 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)"),
3454 )]
3455 }
3456 };
3457
3458 // Generate the actual lint warnings. All warnings for this paragraph
3459 // share the same paragraph-level fix.
3460 for (w_start, w_end, msg) in warnings_to_report {
3461 warnings.push(LintWarning {
3462 rule_name: Some(self.name().to_string()),
3463 message: msg,
3464 line: w_start,
3465 column: 1,
3466 end_line: w_end,
3467 end_column: lines[w_end.saturating_sub(1)].chars().count() + 1,
3468 severity: Severity::Warning,
3469 fix: Some(crate::rule::Fix::new(byte_range.clone(), replacement.clone())),
3470 });
3471 }
3472 }
3473 }
3474 }
3475
3476 warnings
3477 }
3478
3479 /// Calculate string length based on the configured length mode
3480 fn calculate_string_length(&self, s: &str) -> usize {
3481 match self.config.length_mode {
3482 LengthMode::Chars => s.chars().count(),
3483 LengthMode::Visual => s.width(),
3484 LengthMode::Bytes => s.len(),
3485 }
3486 }
3487
3488 /// Calculate effective line length
3489 ///
3490 /// Returns the actual display length of the line using the configured length mode.
3491 fn calculate_effective_length(&self, line: &str) -> usize {
3492 self.calculate_string_length(line)
3493 }
3494
3495 /// Calculate line length with inline link/image URLs removed.
3496 ///
3497 /// For each inline link `[text](url)` or image `` on the line,
3498 /// computes the "savings" from removing the URL portion (keeping only `[text]`
3499 /// or `![alt]`). Returns `effective_length - total_savings`.
3500 ///
3501 /// Handles nested constructs (e.g., `[](url)`) by only counting the
3502 /// outermost construct to avoid double-counting.
3503 fn length_without_inline_link_urls(
3504 &self,
3505 effective_length: usize,
3506 line_number: usize,
3507 ctx: &crate::lint_context::LintContext,
3508 ) -> usize {
3509 let line_range = ctx.line_index.line_content_range(line_number);
3510 let line_byte_end = line_range.end;
3511
3512 // Collect inline links/images on this line: (byte_offset, byte_end, text_only_display_len)
3513 let mut constructs: Vec<(usize, usize, usize)> = Vec::new();
3514
3515 // Binary search: links are sorted by byte_offset, so link.line is non-decreasing
3516 let link_start = ctx.links.partition_point(|l| l.line < line_number);
3517 for link in &ctx.links[link_start..] {
3518 if link.line != line_number {
3519 break;
3520 }
3521 if link.is_reference {
3522 continue;
3523 }
3524 if !matches!(link.link_type, LinkType::Inline) {
3525 continue;
3526 }
3527 if link.byte_end > line_byte_end {
3528 continue;
3529 }
3530 let text_only_len = 2 + self.calculate_string_length(&link.text);
3531 constructs.push((link.byte_offset, link.byte_end, text_only_len));
3532 }
3533
3534 let img_start = ctx.images.partition_point(|i| i.line < line_number);
3535 for image in &ctx.images[img_start..] {
3536 if image.line != line_number {
3537 break;
3538 }
3539 if image.is_reference {
3540 continue;
3541 }
3542 if !matches!(image.link_type, LinkType::Inline) {
3543 continue;
3544 }
3545 if image.byte_end > line_byte_end {
3546 continue;
3547 }
3548 let text_only_len = 3 + self.calculate_string_length(&image.alt_text);
3549 constructs.push((image.byte_offset, image.byte_end, text_only_len));
3550 }
3551
3552 if constructs.is_empty() {
3553 return effective_length;
3554 }
3555
3556 // Sort by byte offset to handle overlapping/nested constructs
3557 constructs.sort_by_key(|&(start, _, _)| start);
3558
3559 let mut total_savings: usize = 0;
3560 let mut last_end: usize = 0;
3561
3562 for (start, end, text_only_len) in &constructs {
3563 // Skip constructs nested inside a previously counted one
3564 if *start < last_end {
3565 continue;
3566 }
3567 // Full construct length in configured length mode
3568 let full_source = &ctx.content[*start..*end];
3569 let full_len = self.calculate_string_length(full_source);
3570 total_savings += full_len.saturating_sub(*text_only_len);
3571 last_end = *end;
3572 }
3573
3574 effective_length.saturating_sub(total_savings)
3575 }
3576}