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