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::utils::mkdocs_admonitions;
6use crate::utils::mkdocs_attr_list::is_standalone_attr_list;
7use crate::utils::mkdocs_snippets::is_snippet_block_delimiter;
8use crate::utils::mkdocs_tabs;
9use crate::utils::range_utils::calculate_excess_range;
10use crate::utils::regex_cache::{IMAGE_REF_PATTERN, LINK_REF_PATTERN, URL_PATTERN};
11use crate::utils::table_utils::TableUtils;
12use crate::utils::text_reflow::{
13 BlockquoteLineData, blockquote_continuation_style, dominant_blockquote_prefix, is_self_contained_display_math_line,
14 join_soft_break_lines, reflow_blockquote_content, split_into_sentences,
15};
16use pulldown_cmark::LinkType;
17use toml;
18
19mod block_builder;
20mod helpers;
21pub mod md013_config;
22use crate::rules::md030_list_marker_space::MD030Config;
23use crate::utils::is_template_directive_only;
24use block_builder::{Block, BlockBuilder};
25use helpers::{
26 extract_list_marker_and_content, has_hard_break, is_github_alert_marker, is_horizontal_rule, is_html_only_line,
27 is_list_item, is_setext_heading_text_line, is_setext_underline_content, is_standalone_link_or_image_line,
28 is_unwrappable_line, source_list_marker, split_into_segments, standalone_link_ends_paragraph,
29 trim_preserving_hard_break,
30};
31pub use md013_config::MD013Config;
32use md013_config::{LengthMode, ReflowMode};
33
34#[cfg(test)]
35mod tests;
36use unicode_width::UnicodeWidthStr;
37
38#[derive(Clone, Default)]
39pub struct MD013LineLength {
40 pub(crate) config: MD013Config,
41 /// MD030 list-marker spacing, applied when reflowing list items so the rewrite
42 /// uses the configured post-marker spacing rather than a hard-coded single
43 /// space. Defaults to MD030's defaults (a single space everywhere), which
44 /// reproduces the previous behaviour exactly. See [`MD030Config::expected_spaces`].
45 pub(crate) list_spacing: MD030Config,
46}
47
48/// Blockquote paragraph line collected for reflow, with original line index for range computation.
49struct CollectedBlockquoteLine {
50 line_idx: usize,
51 data: BlockquoteLineData,
52}
53
54/// A deliberately conservative MDG step candidate.
55///
56/// Gherkin keywords are localized, so recognizing the word after the marker is
57/// not reliable without the complete dialect table. MDG represents steps as
58/// unordered Markdown list items; withholding reflow from all such items in the
59/// flavor is safer than splitting a step and changing its Gherkin text.
60fn is_potential_mdg_step(ctx: &crate::lint_context::LintContext, line_num: usize) -> bool {
61 let Some(item) = ctx.list_item_on_line(line_num) else {
62 return false;
63 };
64 !item.is_ordered() && matches!(item.marker_char(), Some('*' | '-' | '+'))
65}
66
67/// Whether line `line_num` (1-based) is touched on either boundary by a code
68/// span crossing more than one line: a span containing the newline that ends
69/// the line before it, or the newline that ends it.
70///
71/// A renderer reads the line break inside a code span as one space, so such a
72/// line is code however it is spelled, and a `$$...$$` expression on it is no
73/// display block. A span that begins and ends on the line itself does not
74/// matter. `flags` holds one entry per line of the document, indexed here by
75/// the 1-based line number, and comes from
76/// [`crate::utils::text_reflow::lines_touching_multiline_code_span`], which
77/// reads code spans on their own rather than alongside math delimiters: a
78/// `$$...$$` pair closes wherever the next `$$` sits, backtick or not, so a
79/// backtick between them that in fact opens a span reaching past the line
80/// needs a parser that isn't also deciding where the math closes.
81fn line_touches_multiline_code_span(flags: &[bool], line_num: usize) -> bool {
82 line_num
83 .checked_sub(1)
84 .and_then(|idx| flags.get(idx))
85 .copied()
86 .unwrap_or(false)
87}
88
89impl MD013LineLength {
90 pub fn new(line_length: usize, code_blocks: bool, tables: bool, headings: bool, strict: bool) -> Self {
91 Self {
92 config: MD013Config {
93 line_length: crate::types::LineLength::new(line_length),
94 code_blocks,
95 code_spans: true,
96 tables,
97 headings,
98 math_blocks: true,
99 bracket_display_math: false,
100 paragraphs: true, // Default to true for backwards compatibility
101 blockquotes: true, // Default to true for backwards compatibility
102 strict,
103 stern: false,
104 heading_line_length: None,
105 code_block_line_length: None,
106 reflow: false,
107 reflow_mode: ReflowMode::default(),
108 length_mode: LengthMode::default(),
109 abbreviations: Vec::new(),
110 require_sentence_capital: true,
111 ignore_link_urls: true,
112 atomic_spans: true,
113 reflow_break_link_text: false,
114 reflow_length_exemptions: false,
115 },
116 list_spacing: MD030Config::default(),
117 }
118 }
119
120 pub fn from_config_struct(config: MD013Config) -> Self {
121 Self {
122 config,
123 list_spacing: MD030Config::default(),
124 }
125 }
126
127 /// Return a clone with code block checking disabled.
128 /// Used for doc comment linting where code blocks are Rust code managed by rustfmt.
129 pub fn with_code_blocks_disabled(&self) -> Self {
130 let mut clone = self.clone();
131 clone.config.code_blocks = false;
132 clone
133 }
134
135 /// Convert MD013 LengthMode to text_reflow ReflowLengthMode
136 /// Normalized set of reference labels defined in the document.
137 ///
138 /// Passed to reflow so a bare shortcut reference (`[text]`) is treated as an
139 /// atomic link only when its label is actually defined; an undefined
140 /// bracketed run reflows as literal prose.
141 fn defined_reference_labels(ctx: &crate::lint_context::LintContext) -> std::collections::HashSet<String> {
142 ctx.reference_definitions()
143 .iter()
144 .map(|d| crate::utils::text_reflow::normalize_reference_label(&d.id))
145 .collect()
146 }
147
148 /// Build the reflow options shared by every MD013 fix path.
149 ///
150 /// `line_length` varies per call site (some subtract a list-marker or
151 /// blockquote prefix); every other field is derived uniformly from the
152 /// effective config and the document flavor. Callers that need a different
153 /// `max_list_continuation_indent` override it via struct update.
154 fn reflow_options(
155 ctx: &crate::lint_context::LintContext,
156 config: &MD013Config,
157 line_length: usize,
158 ) -> crate::utils::text_reflow::ReflowOptions {
159 crate::utils::text_reflow::ReflowOptions {
160 line_length,
161 break_on_sentences: true,
162 preserve_breaks: false,
163 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
164 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
165 abbreviations: config.abbreviations_for_reflow(),
166 length_mode: config.reflow_length_mode(),
167 attr_lists: ctx.flavor.supports_attr_lists(),
168 myst_roles: ctx.flavor.supports_myst_roles(),
169 require_sentence_capital: config.require_sentence_capital,
170 max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
171 Some(4)
172 } else {
173 None
174 },
175 defined_references: Some(Self::defined_reference_labels(ctx)),
176 atomic_spans: config.atomic_spans,
177 break_link_text: config.reflow_break_link_text,
178 length_exemptions: config.length_exemptions_for_reflow(),
179 }
180 }
181
182 fn should_ignore_line(
183 &self,
184 line: &str,
185 _lines: &[&str],
186 current_line: usize,
187 ctx: &crate::lint_context::LintContext,
188 ) -> bool {
189 if self.config.strict {
190 return false;
191 }
192
193 // Quick check for common patterns before expensive regex
194 let trimmed = line.trim();
195
196 // Only skip if the entire line is a URL (quick check first)
197 if (trimmed.starts_with("http://") || trimmed.starts_with("https://")) && URL_PATTERN.is_match(trimmed) {
198 return true;
199 }
200
201 // Only skip if the entire line is an image reference (quick check first)
202 if trimmed.starts_with("![") && trimmed.ends_with(']') && IMAGE_REF_PATTERN.is_match(trimmed) {
203 return true;
204 }
205
206 // Note: link reference definitions are handled as always-exempt (even in strict mode)
207 // in the main check loop, so they don't need to be checked here.
208
209 // Code blocks with long strings (only check if in code block)
210 if ctx.line_info(current_line + 1).is_some_and(|info| info.in_code_block)
211 && !trimmed.is_empty()
212 && !line.contains(' ')
213 && !line.contains('\t')
214 {
215 return true;
216 }
217
218 false
219 }
220
221 /// Check if rule should skip based on provided config (used for inline config support)
222 fn should_skip_with_config(&self, ctx: &crate::lint_context::LintContext, config: &MD013Config) -> bool {
223 // Skip if content is empty
224 if ctx.content.is_empty() {
225 return true;
226 }
227
228 // For sentence-per-line, semantic-line-breaks, or normalize mode, never skip based on line length
229 if config.reflow
230 && (config.reflow_mode == ReflowMode::SentencePerLine
231 || config.reflow_mode == ReflowMode::SemanticLineBreaks
232 || config.reflow_mode == ReflowMode::Normalize)
233 {
234 return false;
235 }
236
237 // Use the smallest applicable budget across line/heading/code-block
238 // contexts so a stricter context-specific limit doesn't get masked by
239 // the document-wide budget.
240 let min_limit = config.min_effective_line_length();
241 if min_limit.is_unlimited() {
242 return true;
243 }
244 let min_limit_bytes = min_limit.get();
245
246 // Quick check: if total content is shorter than the smallest line limit,
247 // definitely skip.
248 if ctx.content.len() <= min_limit_bytes {
249 return true;
250 }
251
252 // Skip if no line exceeds the smallest applicable limit.
253 !ctx.lines.iter().any(|line| line.byte_len > min_limit_bytes)
254 }
255
256 fn normalize_mode_needs_reflow<'a, I>(&self, lines: I, config: &MD013Config) -> bool
257 where
258 I: IntoIterator<Item = &'a str>,
259 {
260 let mut line_count = 0;
261 let check_length = !config.line_length.is_unlimited();
262
263 for line in lines {
264 line_count += 1;
265 if check_length && self.calculate_effective_length(line) > config.line_length.get() {
266 return true;
267 }
268 }
269
270 line_count > 1
271 }
272}
273
274impl Rule for MD013LineLength {
275 fn name(&self) -> &'static str {
276 "MD013"
277 }
278
279 fn description(&self) -> &'static str {
280 "Line length should not be excessive"
281 }
282
283 fn check(&self, ctx: &crate::lint_context::LintContext) -> LintResult {
284 // Use pre-parsed inline config from LintContext
285 let config_override = ctx.inline_config().get_rule_config("MD013");
286
287 // Apply configuration override if present
288 let effective_config = if let Some(json_config) = config_override {
289 if let Some(obj) = json_config.as_object() {
290 let mut config = self.config.clone();
291 if let Some(line_length) = obj.get("line_length").and_then(serde_json::Value::as_u64) {
292 config.line_length = crate::types::LineLength::new(line_length as usize);
293 }
294 if let Some(code_blocks) = obj.get("code_blocks").and_then(serde_json::Value::as_bool) {
295 config.code_blocks = code_blocks;
296 }
297 if let Some(code_spans) = obj.get("code_spans").and_then(serde_json::Value::as_bool) {
298 config.code_spans = code_spans;
299 }
300 if let Some(tables) = obj.get("tables").and_then(serde_json::Value::as_bool) {
301 config.tables = tables;
302 }
303 if let Some(headings) = obj.get("headings").and_then(serde_json::Value::as_bool) {
304 config.headings = headings;
305 }
306 if let Some(math_blocks) = obj
307 .get("math_blocks")
308 .or_else(|| obj.get("math-blocks"))
309 .and_then(serde_json::Value::as_bool)
310 {
311 config.math_blocks = math_blocks;
312 }
313 if let Some(enabled) = obj
314 .get("bracket_display_math")
315 .or_else(|| obj.get("bracket-display-math"))
316 .and_then(serde_json::Value::as_bool)
317 {
318 config.bracket_display_math = enabled;
319 }
320 if let Some(blockquotes) = obj.get("blockquotes").and_then(serde_json::Value::as_bool) {
321 config.blockquotes = blockquotes;
322 }
323 if let Some(strict) = obj.get("strict").and_then(serde_json::Value::as_bool) {
324 config.strict = strict;
325 }
326 if let Some(stern) = obj.get("stern").and_then(serde_json::Value::as_bool) {
327 config.stern = stern;
328 }
329 if let Some(v) = obj
330 .get("ignore_link_urls")
331 .or_else(|| obj.get("ignore-link-urls"))
332 .or_else(|| obj.get("semantic_link_understanding"))
333 .or_else(|| obj.get("semantic-link-understanding"))
334 .and_then(serde_json::Value::as_bool)
335 {
336 config.ignore_link_urls = v;
337 }
338 if let Some(reflow) = obj.get("reflow").and_then(serde_json::Value::as_bool) {
339 config.reflow = reflow;
340 }
341 if let Some(reflow_mode) = obj.get("reflow_mode").and_then(|v| v.as_str()) {
342 config.reflow_mode = match reflow_mode {
343 "default" => ReflowMode::Default,
344 "normalize" => ReflowMode::Normalize,
345 "sentence-per-line" => ReflowMode::SentencePerLine,
346 "semantic-line-breaks" => ReflowMode::SemanticLineBreaks,
347 _ => ReflowMode::default(),
348 };
349 }
350 config
351 } else {
352 self.config.clone()
353 }
354 } else {
355 self.config.clone()
356 };
357
358 // Fast early return using should_skip with EFFECTIVE config (after inline overrides)
359 // But don't skip if we're in reflow mode with Normalize or SentencePerLine
360 if self.should_skip_with_config(ctx, &effective_config)
361 && !(effective_config.reflow
362 && (effective_config.reflow_mode == ReflowMode::Normalize
363 || effective_config.reflow_mode == ReflowMode::SentencePerLine
364 || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
365 {
366 return Ok(Vec::new());
367 }
368
369 // Direct implementation without DocumentStructure
370 let mut warnings = Vec::new();
371
372 // Special handling: line_length = 0 means "no line length limit"
373 // Skip all line length checks, but still allow reflow if enabled
374 let skip_length_checks = effective_config.line_length.is_unlimited();
375
376 // Pre-filter lines that could be problematic to avoid processing all lines.
377 // Use the smallest applicable budget across line/heading/code-block contexts
378 // so candidates aren't dropped when a stricter context-specific budget applies.
379 let prefilter_limit = effective_config.min_effective_line_length();
380 let prefilter_skip = prefilter_limit.is_unlimited();
381 let mut candidate_lines = Vec::new();
382 if !skip_length_checks && !prefilter_skip {
383 for (line_idx, line_info) in ctx.lines.iter().enumerate() {
384 // Skip front matter - it should never be linted
385 if line_info.in_front_matter {
386 continue;
387 }
388
389 // Quick length check first
390 if line_info.byte_len > prefilter_limit.get() {
391 candidate_lines.push(line_idx);
392 }
393 }
394 }
395
396 // If no candidate lines and not in normalize or sentence-per-line mode, early return
397 if candidate_lines.is_empty()
398 && !(effective_config.reflow
399 && (effective_config.reflow_mode == ReflowMode::Normalize
400 || effective_config.reflow_mode == ReflowMode::SentencePerLine
401 || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
402 {
403 return Ok(warnings);
404 }
405
406 let lines = ctx.raw_lines();
407
408 // Whether a 1-indexed line is part of a heading. A setext heading's text
409 // is the whole paragraph its underline ends, and the parser records the
410 // heading itself only on the last of those lines, so the earlier ones
411 // are recognized through `is_setext_heading_text`. Both are O(1)
412 // per-line fields, so check them directly at each use site instead of
413 // materializing a full-document HashSet (an extra O(n) pass and
414 // allocation on a rule that runs on virtually every file).
415 let is_heading_line_num = |line_number: usize| -> bool {
416 line_number
417 .checked_sub(1)
418 .and_then(|idx| ctx.lines.get(idx))
419 .is_some_and(|line| line.heading.is_some() || line.is_setext_heading_text)
420 };
421
422 // Use pre-computed table blocks from context
423 // We need this for both the table skip check AND the paragraphs check
424 let table_blocks = &ctx.table_blocks;
425 let mut table_lines_set = std::collections::HashSet::new();
426 for table in table_blocks {
427 table_lines_set.insert(table.header_line + 1);
428 table_lines_set.insert(table.delimiter_line + 1);
429 for &line in &table.content_lines {
430 table_lines_set.insert(line + 1);
431 }
432 }
433
434 let defined_references = Self::defined_reference_labels(ctx);
435
436 // Process candidate lines for line length checks
437 'line_loop: for &line_idx in &candidate_lines {
438 let line_number = line_idx + 1;
439 let line = lines[line_idx];
440
441 // Calculate actual line length (used in warning messages)
442 let effective_length = self.calculate_effective_length(line);
443
444 // Pick the context-specific limit: heading > code-block > paragraph.
445 // Headings dominate over code-block context if a setext underline ever
446 // overlaps a fenced range (defensive — these are mutually exclusive in
447 // practice, but the explicit ordering documents intent).
448 let is_heading_line = is_heading_line_num(line_number);
449 let in_code_block = ctx.line_info(line_number).is_some_and(|info| info.in_code_block);
450 let line_limit = if is_heading_line {
451 effective_config.effective_heading_line_length().get()
452 } else if in_code_block {
453 effective_config.effective_code_block_line_length().get()
454 } else {
455 effective_config.line_length.get()
456 };
457
458 // A context-specific limit of 0 means "unlimited for this context".
459 if line_limit == 0 {
460 continue;
461 }
462
463 // Stern mode: like default, but the trailing-token forgiveness is
464 // disabled — a line with whitespace that exceeds the limit is a
465 // violation even if the excess is the final token. The "unwrappable"
466 // line exemption (single token, optionally prefixed by # or >) is
467 // still honored. Strict overrides stern entirely.
468 if effective_config.stern && !effective_config.strict && is_unwrappable_line(line) {
469 continue;
470 }
471
472 // Trailing-token forgiveness: only in default mode (not strict, not stern).
473 // If the line only exceeds the limit because of a long token at the end
474 // (URL, link chain, identifier), it passes. This matches markdownlint's
475 // behavior: line.replace(/\S*$/u, "#")
476 let check_length = if effective_config.strict || effective_config.stern {
477 effective_length
478 } else {
479 match line.rfind(char::is_whitespace) {
480 Some(pos) => {
481 let ws_char = line[pos..].chars().next().unwrap();
482 let prefix_end = pos + ws_char.len_utf8();
483 self.calculate_string_length(&line[..prefix_end]) + 1
484 }
485 None => 1, // No whitespace — entire line is a single token
486 }
487 };
488
489 // Skip lines where the check length is within the limit
490 if check_length <= line_limit {
491 continue;
492 }
493
494 // Ignore inline link/image URLs: suppress when excess comes entirely from inline URLs.
495 // Disabled by `strict` (all forgiveness off) and by `ignore_link_urls = false`
496 // (count link/image URLs toward the line length so such lines are flagged).
497 if !effective_config.strict && effective_config.ignore_link_urls {
498 let length_without_urls = self.length_without_inline_link_urls(effective_length, line_number, ctx);
499 if length_without_urls <= line_limit {
500 continue;
501 }
502 }
503
504 // Inline code spans cannot be wrapped, so reflow cannot shorten a line
505 // whose excess length is one. When code-span checking is disabled,
506 // suppress a violation that would fit once inline code spans are excluded.
507 if !effective_config.code_spans {
508 let code_span_width: usize = ctx
509 .code_spans()
510 .iter()
511 .filter(|span| span.line == line_number && span.end_line == line_number)
512 .map(|span| self.calculate_string_length(&ctx.content[span.byte_offset..span.byte_end]))
513 .sum();
514 if effective_length.saturating_sub(code_span_width) <= line_limit {
515 continue;
516 }
517 }
518
519 // Skip mkdocstrings and pymdown blocks (already handled by LintContext)
520 if ctx.lines[line_idx].in_mkdocstrings || ctx.lines[line_idx].in_pymdown_block {
521 continue;
522 }
523
524 // Skip MyST comments (% comment) — structural lines, not prose
525 if ctx.lines[line_idx].is_myst_comment {
526 continue;
527 }
528
529 // Link reference definitions are always exempt, even in strict mode.
530 // There's no way to shorten them without breaking the URL.
531 // Also check after stripping list markers, since list items may
532 // contain link ref defs as their content.
533 {
534 let trimmed = line.trim();
535 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
536 continue;
537 }
538 if is_list_item(trimmed) {
539 let (_, content) = extract_list_marker_and_content(trimmed);
540 let content_trimmed = content.trim();
541 if content_trimmed.starts_with('[')
542 && content_trimmed.contains("]:")
543 && LINK_REF_PATTERN.is_match(content_trimmed)
544 {
545 continue;
546 }
547 }
548 }
549
550 // Skip various block types efficiently
551 if !effective_config.strict {
552 // Lines whose only content is a link/image are exempt.
553 // After stripping list markers, blockquote markers, and emphasis,
554 // if only a link or image remains, there is no way to shorten it.
555 if is_standalone_link_or_image_line(ctx, line_number) {
556 continue;
557 }
558
559 // Lines consisting entirely of HTML tags are exempt.
560 // Badge lines, images with attributes, and similar inline HTML
561 // are long due to URLs in attributes and can't be meaningfully shortened.
562 if is_html_only_line(line) {
563 continue;
564 }
565
566 // Skip setext heading underlines
567 if !line.trim().is_empty() && line.trim().chars().all(|c| c == '=' || c == '-') {
568 continue;
569 }
570
571 // Skip block elements according to config flags
572 // The flags mean: true = check these elements, false = skip these elements
573 // So we skip when the flag is FALSE and the line is in that element type
574 if (!effective_config.headings && is_heading_line_num(line_number))
575 || (!effective_config.code_blocks
576 && ctx.line_info(line_number).is_some_and(|info| info.in_code_block))
577 || (!effective_config.tables && table_lines_set.contains(&line_number))
578 || (!effective_config.math_blocks && self.line_is_display_math(line_number, ctx, &effective_config))
579 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
580 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
581 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
582 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
583 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
584 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
585 || ctx.line_info(line_number).is_some_and(|info| info.in_pymdown_block)
586 {
587 continue;
588 }
589
590 // Check if this is a paragraph/regular text line
591 // If paragraphs = false, skip lines that are NOT in special blocks
592 // Blockquote content is treated as paragraph text, so it's not
593 // included in the special blocks list here.
594 if !effective_config.paragraphs {
595 let is_special_block = is_heading_line_num(line_number)
596 || ctx.line_info(line_number).is_some_and(|info| info.in_code_block)
597 || table_lines_set.contains(&line_number)
598 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
599 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
600 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
601 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
602 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
603 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
604 || ctx
605 .line_info(line_number)
606 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container);
607
608 // Skip regular paragraph text when paragraphs = false
609 if !is_special_block {
610 continue;
611 }
612 }
613
614 // Skip blockquote lines when blockquotes = false.
615 // Also skip lazy continuation lines that belong to a blockquote
616 // (lines without `>` prefix that follow a blockquote line).
617 if !effective_config.blockquotes {
618 if ctx.lines[line_number - 1].blockquote.is_some() {
619 continue;
620 }
621 // Check for lazy continuation: scan backwards through
622 // non-blank lines to find if this paragraph started with
623 // a blockquote marker
624 if !line.trim().is_empty() {
625 let mut scan = line_number.saturating_sub(2);
626 loop {
627 if ctx.lines[scan].blockquote.is_some() {
628 // Found a blockquote ancestor — this is a lazy continuation
629 continue 'line_loop;
630 }
631 if lines[scan].trim().is_empty() || scan == 0 {
632 break;
633 }
634 scan -= 1;
635 }
636 }
637 }
638
639 // Skip lines that are only a URL, image ref, or link ref
640 if self.should_ignore_line(line, lines, line_idx, ctx) {
641 continue;
642 }
643 }
644
645 // In sentence-per-line mode, check if this is a single long sentence
646 // If so, emit a warning without a fix (user must manually rephrase)
647 if effective_config.reflow_mode == ReflowMode::SentencePerLine {
648 let sentences = split_into_sentences(
649 line.trim(),
650 Some(&defined_references),
651 effective_config.require_sentence_capital,
652 );
653 if sentences.len() == 1 {
654 // Single sentence that's too long - warn but don't auto-fix
655 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
656
657 let (start_line, start_col, end_line, end_col) =
658 calculate_excess_range(line_number, line, line_limit);
659
660 warnings.push(LintWarning {
661 rule_name: Some(self.name().to_string()),
662 message,
663 line: start_line,
664 column: start_col,
665 end_line,
666 end_column: end_col,
667 severity: Severity::Warning,
668 fix: None, // No auto-fix for long single sentences
669 });
670 continue;
671 }
672 // Multiple sentences will be handled by paragraph-based reflow
673 continue;
674 }
675
676 // In semantic-line-breaks mode, skip per-line checks —
677 // all reflow is handled at the paragraph level with cascading splits
678 if effective_config.reflow_mode == ReflowMode::SemanticLineBreaks {
679 continue;
680 }
681
682 // Don't provide fix for individual lines when reflow is enabled
683 // Paragraph-based fixes will be handled separately
684 let fix = None;
685
686 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
687
688 // Calculate precise character range for the excess portion
689 let (start_line, start_col, end_line, end_col) = calculate_excess_range(line_number, line, line_limit);
690
691 warnings.push(LintWarning {
692 rule_name: Some(self.name().to_string()),
693 message,
694 line: start_line,
695 column: start_col,
696 end_line,
697 end_column: end_col,
698 severity: Severity::Warning,
699 fix,
700 });
701 }
702
703 // If reflow is enabled, generate paragraph-based fixes
704 if effective_config.reflow {
705 let paragraph_warnings = self.generate_paragraph_fixes(ctx, &effective_config, lines);
706 // Merge paragraph warnings with line warnings, removing duplicates
707 for mut pw in paragraph_warnings {
708 if ctx.flavor == crate::config::MarkdownFlavor::MDG
709 && (pw.line..=pw.end_line).any(|line_number| is_potential_mdg_step(ctx, line_number))
710 {
711 // Keep reporting the excessive line, but do not offer a fix
712 // whose replacement would split a Gherkin step across lines.
713 pw.fix = None;
714 }
715 // Remove any line warnings that overlap with this paragraph
716 warnings.retain(|w| w.line < pw.line || w.line > pw.end_line);
717 warnings.push(pw);
718 }
719 }
720
721 Ok(warnings)
722 }
723
724 fn fix(&self, ctx: &crate::lint_context::LintContext) -> Result<String, LintError> {
725 // For CLI usage, apply fixes from warnings
726 // LSP will use the warning-based fixes directly
727 let warnings = self.check(ctx)?;
728 let warnings =
729 crate::utils::fix_utils::filter_warnings_by_inline_config(warnings, ctx.inline_config(), self.name());
730
731 // If there are no fixes, return content unchanged
732 if !warnings.iter().any(|w| w.fix.is_some()) {
733 return Ok(ctx.content.to_string());
734 }
735
736 // Apply warning-based fixes
737 crate::utils::fix_utils::apply_warning_fixes(ctx.content, &warnings)
738 .map_err(|e| LintError::FixFailed(format!("Failed to apply fixes: {e}")))
739 }
740
741 fn as_any(&self) -> &dyn std::any::Any {
742 self
743 }
744
745 fn category(&self) -> RuleCategory {
746 RuleCategory::Whitespace
747 }
748
749 fn should_skip(&self, ctx: &crate::lint_context::LintContext) -> bool {
750 self.should_skip_with_config(ctx, &self.config)
751 }
752
753 crate::impl_rule_config_sections!(MD013Config);
754
755 fn config_aliases(&self) -> Option<std::collections::HashMap<String, String>> {
756 let mut aliases = std::collections::HashMap::new();
757 aliases.insert("enable_reflow".to_string(), "reflow".to_string());
758 aliases.insert("strict_sentences".to_string(), "require-sentence-capital".to_string());
759 aliases.insert("strict-sentences".to_string(), "require-sentence-capital".to_string());
760 // Kept in step with the `alias` attributes on `MD013Config::ignore_link_urls`.
761 // Serde accepts these spellings, so a config using one is honored; without
762 // them here the key validator reports a documented, working key as unknown.
763 aliases.insert(
764 "semantic-link-understanding".to_string(),
765 "ignore-link-urls".to_string(),
766 );
767 aliases.insert(
768 "semantic_link_understanding".to_string(),
769 "ignore-link-urls".to_string(),
770 );
771 Some(aliases)
772 }
773
774 fn from_config(config: &crate::config::Config) -> Box<dyn Rule>
775 where
776 Self: Sized,
777 {
778 let rule_config = MD013Config::from_document_config(config);
779 let mut rule = Self::from_config_struct(rule_config);
780 // Pull list-marker spacing from MD030 (via the shared serde config loader)
781 // so reflow rewrites list items with the configured spacing rather than a
782 // hard-coded single space.
783 rule.list_spacing = crate::rule_config_serde::load_rule_config::<MD030Config>(config);
784 Box::new(rule)
785 }
786}
787
788impl MD013LineLength {
789 /// True when `line_num` (1-indexed) sits inside a `$$` span covering more
790 /// than one line, as seen by the byte-level math parser.
791 fn line_in_multiline_math_span(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
792 ctx.math_spans()
793 .iter()
794 .any(|span| span.is_display && span.end_line > span.line && (span.line..=span.end_line).contains(&line_num))
795 }
796
797 /// True when `line_num` (1-indexed) holds nothing but display math, whether
798 /// that is one line of a multi-line block, a delimiter line, or a whole line
799 /// that is a single complete `$$...$$` span. This is what `math-blocks =
800 /// false` exempts from the length check.
801 ///
802 /// rumdl models math twice and the two models miss different containers, so
803 /// this consults both. `math_spans()` is byte-level and sees a block opened on
804 /// a list marker line (`- $$`), which the line-level map cannot because that
805 /// line does not begin with `$$`. `LineInfo::in_math_block` is line-level and
806 /// sees a four-space-indented block inside a footnote, which the byte-level
807 /// parser reads as an indented code block. Taking the union only ever adds
808 /// coverage: neither signal fires on ordinary prose, and an unmatched `$$`
809 /// opener is flagged by neither, so a stray delimiter cannot exempt the rest
810 /// of the document.
811 fn line_is_display_math(
812 &self,
813 line_num: usize,
814 ctx: &crate::lint_context::LintContext,
815 config: &MD013Config,
816 ) -> bool {
817 self.line_holds_only_multiline_math(line_num, ctx)
818 || ctx.line_info(line_num).is_some_and(|info| info.in_math_block)
819 || (config.bracket_display_math
820 && ctx
821 .bracket_display_math_lines()
822 .multiline
823 .get(line_num.saturating_sub(1))
824 .copied()
825 .unwrap_or(false))
826 || (config.bracket_display_math
827 && ctx
828 .bracket_display_math_lines()
829 .standalone
830 .get(line_num.saturating_sub(1))
831 .copied()
832 .unwrap_or(false))
833 }
834
835 /// True when a multi-line `$$` span covers `line_num` (1-indexed) and the line
836 /// holds nothing besides that math.
837 ///
838 /// A delimiter line can carry Markdown outside the delimiter: `$$ trailing
839 /// prose` closes a block and then continues in prose, and `leading prose $$`
840 /// opens one at the end of a sentence. That prose is ordinary text and counts
841 /// toward the line's length like any other, so exempting the whole line would
842 /// hide arbitrarily long prose behind a delimiter. The line-level map already
843 /// leaves such mixed lines unflagged; this is the byte-level half of the same
844 /// judgement.
845 fn line_holds_only_multiline_math(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
846 let Some(info) = ctx.line_info(line_num) else {
847 return false;
848 };
849 let line = info.content(ctx.content);
850
851 ctx.math_spans().iter().any(|span| {
852 if !span.is_display || span.end_line <= span.line || !(span.line..=span.end_line).contains(&line_num) {
853 return false;
854 }
855 if line_num == span.line {
856 let before = line.get(..span.byte_offset.saturating_sub(info.byte_offset));
857 if !before.is_none_or(|before| Self::only_structure_precedes_math(before, info)) {
858 return false;
859 }
860 }
861 if line_num == span.end_line {
862 let after = line.get(span.byte_end.saturating_sub(info.byte_offset)..);
863 if !after.is_none_or(|after| after.trim().is_empty()) {
864 return false;
865 }
866 }
867 true
868 })
869 }
870
871 /// True when the text before a block's opening delimiter is only the structure
872 /// the block sits in: indentation, a blockquote marker, or the list marker
873 /// introducing it. Such a block still owns its whole line.
874 fn only_structure_precedes_math(before: &str, info: &crate::lint_context::LineInfo) -> bool {
875 let after_marker =
876 crate::utils::blockquote::parse_blockquote_prefix(before).map_or(before, |prefix| prefix.content);
877 if after_marker.trim().is_empty() {
878 return true;
879 }
880 info.list_item.as_ref().is_some_and(|item| {
881 let mut chars = before.chars();
882 chars.by_ref().take(item.content_column).count() == item.content_column && chars.as_str().trim().is_empty()
883 })
884 }
885
886 /// True when `line_num` (1-indexed) falls inside a display-math block that
887 /// spans more than one line.
888 ///
889 /// Line breaks carry meaning inside such a block: a TeX `%` comment runs to
890 /// the end of its line, so joining the lines pulls whatever followed on later
891 /// lines into the comment and drops it from the rendered equation, which can
892 /// also leave an environment unclosed. Reflow therefore leaves these blocks
893 /// alone regardless of the `math_blocks` setting, which governs only whether
894 /// their length is reported.
895 ///
896 /// This is `line_is_display_math` minus the case of a whole line that is one
897 /// complete `$$...$$` span, which
898 /// [`crate::utils::text_reflow::is_self_contained_display_math_line`]
899 /// recognizes. Such a line is a block of its own: it has no internal line
900 /// breaks to lose, and reflow keeps it on the line it was written on
901 /// through that recognizer rather than through this one.
902 fn line_in_multiline_math_block(
903 &self,
904 line_num: usize,
905 ctx: &crate::lint_context::LintContext,
906 config: &MD013Config,
907 ) -> bool {
908 self.line_in_multiline_math_span(line_num, ctx)
909 || ctx.line_info(line_num).is_some_and(|info| {
910 info.in_math_block && !is_self_contained_display_math_line(info.content(ctx.content))
911 })
912 || (config.bracket_display_math
913 && ctx
914 .bracket_display_math_lines()
915 .multiline
916 .get(line_num.saturating_sub(1))
917 .copied()
918 .unwrap_or(false))
919 }
920
921 fn line_is_standalone_bracket_math(
922 &self,
923 line_num: usize,
924 ctx: &crate::lint_context::LintContext,
925 config: &MD013Config,
926 ) -> bool {
927 config.bracket_display_math
928 && ctx
929 .bracket_display_math_lines()
930 .standalone
931 .get(line_num.saturating_sub(1))
932 .copied()
933 .unwrap_or(false)
934 }
935
936 /// True when `line_num` (1-based) sits inside a structure whose lines must be
937 /// preserved verbatim (code block, front matter, HTML/JSX/MDX block, MkDocs
938 /// container, div marker, multi-line math block, ...). Used to keep blockquote
939 /// reflow from touching quoted-looking text embedded in such structures.
940 fn line_in_verbatim_context(
941 &self,
942 line_num: usize,
943 ctx: &crate::lint_context::LintContext,
944 config: &MD013Config,
945 ) -> bool {
946 if self.line_in_multiline_math_block(line_num, ctx, config) {
947 return true;
948 }
949 ctx.line_info(line_num).is_some_and(|info| {
950 info.in_code_block
951 || info.in_front_matter
952 || info.in_html_block
953 || info.in_html_comment
954 || info.in_esm_block
955 || info.in_jsx_expression
956 || info.in_jsx_block
957 || info.in_mdx_comment
958 || info.in_mkdocstrings
959 || info.in_pymdown_block
960 || info.in_mkdocs_container()
961 || info.is_div_marker
962 })
963 }
964
965 fn is_blockquote_content_boundary(
966 &self,
967 content: &str,
968 line_num: usize,
969 ctx: &crate::lint_context::LintContext,
970 config: &MD013Config,
971 ) -> bool {
972 let trimmed = content.trim();
973
974 trimmed.is_empty()
975 || self.line_in_verbatim_context(line_num, ctx, config)
976 || trimmed.starts_with('#')
977 || trimmed.starts_with("```")
978 || trimmed.starts_with("~~~")
979 || trimmed.starts_with('>')
980 || TableUtils::is_potential_table_row_with_flavor(content, ctx.flavor)
981 || is_list_item(trimmed)
982 || is_horizontal_rule(content)
983 // A setext underline ends the quoted paragraph. The text decides
984 // it: under quoted paragraph text the run is an underline, a dash
985 // run anywhere else is a thematic break, and stopping at an equals
986 // run that is neither only leaves text unreflowed.
987 || is_setext_underline_content(content)
988 || (trimmed.starts_with('[') && content.contains("]:"))
989 || is_template_directive_only(content)
990 || is_standalone_attr_list(content)
991 || is_snippet_block_delimiter(content)
992 || is_github_alert_marker(trimmed)
993 || is_html_only_line(content)
994 || self.line_is_standalone_bracket_math(line_num, ctx, config)
995 || standalone_link_ends_paragraph(ctx, line_num, config)
996 }
997
998 #[allow(clippy::too_many_arguments)]
999 fn generate_blockquote_paragraph_fix(
1000 &self,
1001 ctx: &crate::lint_context::LintContext,
1002 config: &MD013Config,
1003 lines: &[&str],
1004 start_idx: usize,
1005 line_ending: &str,
1006 // Extra indent (spaces) to prepend to the emitted `>` prefix so a blockquote
1007 // nested in a list item moves with its parent's widened marker. Zero unless a
1008 // non-default MD030 widened an ancestor list item.
1009 ancestor_shift: isize,
1010 ) -> (Option<LintWarning>, usize) {
1011 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
1012 return (None, start_idx + 1);
1013 };
1014 let target_level = start_bq.nesting_level;
1015 let defined_references = Self::defined_reference_labels(ctx);
1016
1017 let mut collected: Vec<CollectedBlockquoteLine> = Vec::new();
1018 let mut i = start_idx;
1019
1020 while i < lines.len() {
1021 if !collected.is_empty() && has_hard_break(&collected[collected.len() - 1].data.content) {
1022 break;
1023 }
1024
1025 let line_num = i + 1;
1026 if line_num > ctx.lines.len() {
1027 break;
1028 }
1029
1030 if lines[i].trim().is_empty() {
1031 break;
1032 }
1033
1034 let line_bq = ctx.lines[i].blockquote.as_deref();
1035 if let Some(bq) = line_bq {
1036 if bq.nesting_level != target_level {
1037 break;
1038 }
1039
1040 if self.is_blockquote_content_boundary(&bq.content, line_num, ctx, config) {
1041 break;
1042 }
1043
1044 collected.push(CollectedBlockquoteLine {
1045 line_idx: i,
1046 data: BlockquoteLineData::explicit(trim_preserving_hard_break(&bq.content), bq.prefix.clone()),
1047 });
1048 i += 1;
1049 continue;
1050 }
1051
1052 let lazy_content = lines[i].trim_start();
1053 if self.is_blockquote_content_boundary(lazy_content, line_num, ctx, config) {
1054 break;
1055 }
1056
1057 collected.push(CollectedBlockquoteLine {
1058 line_idx: i,
1059 data: BlockquoteLineData::lazy(trim_preserving_hard_break(lazy_content)),
1060 });
1061 i += 1;
1062 }
1063
1064 if collected.is_empty() {
1065 return (None, start_idx + 1);
1066 }
1067
1068 let next_idx = i;
1069 let paragraph_start = collected[0].line_idx;
1070 let end_line = collected[collected.len() - 1].line_idx;
1071 let line_data: Vec<BlockquoteLineData> = collected.iter().map(|l| l.data.clone()).collect();
1072 let paragraph_text = line_data
1073 .iter()
1074 .map(|d| d.content.as_str())
1075 .collect::<Vec<_>>()
1076 .join(" ");
1077
1078 // A colon-led line with a line of the paragraph before it opens a
1079 // definition, and joining the lines would flatten the definition list
1080 // into prose. The paragraph's first line is prose whatever it starts
1081 // with, since a definition needs a term on the line before it.
1082 let contains_definition_list = line_data
1083 .iter()
1084 .skip(1)
1085 .any(|d| crate::utils::text_reflow::is_definition_list_marker(&d.content));
1086 if contains_definition_list {
1087 return (None, next_idx);
1088 }
1089
1090 let contains_snippets = line_data.iter().any(|d| is_snippet_block_delimiter(&d.content));
1091 if contains_snippets {
1092 return (None, next_idx);
1093 }
1094
1095 let needs_reflow = match config.reflow_mode {
1096 ReflowMode::Normalize => {
1097 self.normalize_mode_needs_reflow(line_data.iter().map(|d| d.content.as_str()), config)
1098 }
1099 ReflowMode::SentencePerLine => {
1100 let sentences = split_into_sentences(
1101 ¶graph_text,
1102 Some(&defined_references),
1103 config.require_sentence_capital,
1104 );
1105 sentences.len() > 1 || line_data.len() > 1
1106 }
1107 ReflowMode::SemanticLineBreaks => {
1108 let sentences = split_into_sentences(
1109 ¶graph_text,
1110 Some(&defined_references),
1111 config.require_sentence_capital,
1112 );
1113 sentences.len() > 1
1114 || line_data.len() > 1
1115 || collected
1116 .iter()
1117 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get())
1118 }
1119 ReflowMode::Default => collected
1120 .iter()
1121 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get()),
1122 };
1123
1124 if !needs_reflow {
1125 return (None, next_idx);
1126 }
1127
1128 let fallback_prefix = start_bq.prefix.clone();
1129 let explicit_prefix = dominant_blockquote_prefix(&line_data, &fallback_prefix);
1130 // Shift the whole quote right to track a widened parent list item's content
1131 // column (only widening matters for nesting; a narrowed parent leaves the quote
1132 // harmlessly over-indented, which MD027/MD030 tidy).
1133 let explicit_prefix = if ancestor_shift > 0 {
1134 format!("{}{explicit_prefix}", " ".repeat(ancestor_shift as usize))
1135 } else {
1136 explicit_prefix
1137 };
1138 let continuation_style = blockquote_continuation_style(&line_data);
1139
1140 let reflow_line_length = if config.line_length.is_unlimited() {
1141 usize::MAX
1142 } else {
1143 config
1144 .line_length
1145 .get()
1146 .saturating_sub(self.calculate_string_length(&explicit_prefix))
1147 .max(1)
1148 };
1149
1150 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
1151
1152 let reflowed_with_style =
1153 reflow_blockquote_content(&line_data, &explicit_prefix, continuation_style, &reflow_options);
1154
1155 if reflowed_with_style.is_empty() {
1156 return (None, next_idx);
1157 }
1158
1159 let reflowed_text = reflowed_with_style.join(line_ending);
1160
1161 let start_range = ctx.whole_line_byte_range(paragraph_start + 1);
1162 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
1163 ctx.line_text_byte_range(end_line + 1, 1, lines[end_line].len() + 1)
1164 } else {
1165 ctx.whole_line_byte_range(end_line + 1)
1166 };
1167 let byte_range = start_range.start..end_range.end;
1168
1169 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
1170 format!("{reflowed_text}{line_ending}")
1171 } else {
1172 reflowed_text
1173 };
1174
1175 let original_text = &ctx.content[byte_range.clone()];
1176 if original_text == replacement {
1177 return (None, next_idx);
1178 }
1179
1180 let (warning_line, warning_end_line) = match config.reflow_mode {
1181 ReflowMode::Normalize => (paragraph_start + 1, end_line + 1),
1182 ReflowMode::SentencePerLine | ReflowMode::SemanticLineBreaks => (paragraph_start + 1, end_line + 1),
1183 ReflowMode::Default => {
1184 let violating_line = collected
1185 .iter()
1186 .find(|line| self.calculate_effective_length(lines[line.line_idx]) > config.line_length.get())
1187 .map_or(paragraph_start + 1, |line| line.line_idx + 1);
1188 (violating_line, violating_line)
1189 }
1190 };
1191
1192 let warning = LintWarning {
1193 rule_name: Some(self.name().to_string()),
1194 message: match config.reflow_mode {
1195 ReflowMode::Normalize => format!(
1196 "Paragraph could be normalized to use line length of {} characters",
1197 config.line_length.get()
1198 ),
1199 ReflowMode::SentencePerLine => {
1200 let num_sentences = split_into_sentences(
1201 ¶graph_text,
1202 Some(&defined_references),
1203 config.require_sentence_capital,
1204 )
1205 .len();
1206 if line_data.len() == 1 {
1207 format!("Line contains {num_sentences} sentences (one sentence per line required)")
1208 } else {
1209 let num_lines = line_data.len();
1210 format!(
1211 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
1212 )
1213 }
1214 }
1215 ReflowMode::SemanticLineBreaks => {
1216 let num_sentences = split_into_sentences(
1217 ¶graph_text,
1218 Some(&defined_references),
1219 config.require_sentence_capital,
1220 )
1221 .len();
1222 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
1223 }
1224 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
1225 },
1226 line: warning_line,
1227 column: 1,
1228 end_line: warning_end_line,
1229 end_column: lines[warning_end_line.saturating_sub(1)].chars().count() + 1,
1230 severity: Severity::Warning,
1231 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1232 };
1233
1234 (Some(warning), next_idx)
1235 }
1236
1237 /// Reflow a single list item that lives inside a blockquote.
1238 ///
1239 /// The blockquote paragraph reflow treats a list marker as a content boundary,
1240 /// so `> - long item ...` is never wrapped even though the identical top-level
1241 /// item is. This handles that case: it collects one tight, prose-only list item
1242 /// at the starting blockquote level, reflows the item body to the configured
1243 /// width, and re-emits it with the blockquote prefix preserved and continuation
1244 /// lines aligned under the list content.
1245 ///
1246 /// Sibling and nested list items end collection and are reflowed independently
1247 /// by the caller's outer loop (a nested item carries its indent folded into the
1248 /// blockquote prefix, so it reflows correctly on its own). Items that are not
1249 /// simple tight prose - those embedding a code block, table, fence, or hard
1250 /// break - are left untouched (`None`) and the cursor still advances past the
1251 /// whole item so its inner lines are never reprocessed as loose prose.
1252 #[allow(clippy::too_many_arguments)]
1253 fn generate_blockquote_list_item_fix(
1254 &self,
1255 ctx: &crate::lint_context::LintContext,
1256 config: &MD013Config,
1257 lines: &[&str],
1258 start_idx: usize,
1259 line_ending: &str,
1260 // Extra indent (spaces) to prepend to the emitted `>` prefix when this quoted
1261 // list lives inside a list item whose marker widened. Zero unless a non-default
1262 // MD030 widened an ancestor list item.
1263 ancestor_shift: isize,
1264 // One entry per document line, true where a code span crosses one of the
1265 // line's boundaries. The caller reads them off one pass over the document
1266 // and shares them across every item, so an item costs no pass of its own.
1267 code_span_touches: &[bool],
1268 ) -> (Option<LintWarning>, usize) {
1269 use crate::utils::blockquote::effective_indent_in_blockquote;
1270
1271 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
1272 return (None, start_idx + 1);
1273 };
1274
1275 let defined_references = Self::defined_reference_labels(ctx);
1276
1277 // A `>`-prefixed line can be marked as a blockquote even inside a fenced code
1278 // block (or other verbatim structure); such content must never be reflowed.
1279 if self.line_in_verbatim_context(start_idx + 1, ctx, config) {
1280 return (None, start_idx + 1);
1281 }
1282
1283 let target_level = start_bq.nesting_level;
1284
1285 // The marker line carries the canonical blockquote prefix: its content begins
1286 // with the list marker, so no list indent has been folded into the prefix.
1287 // Track a widened parent list item's content column so the quote stays nested
1288 // (only widening can detach it; narrowing just over-indents).
1289 let bq_prefix = if ancestor_shift > 0 {
1290 format!("{}{}", " ".repeat(ancestor_shift as usize), start_bq.prefix)
1291 } else {
1292 start_bq.prefix.clone()
1293 };
1294
1295 // A thematic break opens with what looks like a bullet marker (`- - -`).
1296 // It is not a list item, and reflowing it as prose destroys the break.
1297 // The top-level reflow path applies the same exemption.
1298 if is_horizontal_rule(&start_bq.content) {
1299 return (None, start_idx + 1);
1300 }
1301
1302 let (marker, first_body) = extract_list_marker_and_content(&start_bq.content);
1303 if marker.is_empty() {
1304 return (None, start_idx + 1);
1305 }
1306 let marker_width = marker.chars().count();
1307
1308 // Continuation lines of a checkbox item align under the bullet+checkbox, but
1309 // are recognized from the bullet width, matching the top-level list reflow.
1310 let base_marker_width = ["[ ] ", "[x] ", "[X] "]
1311 .iter()
1312 .find_map(|cb| marker.find(*cb))
1313 .unwrap_or(marker_width);
1314
1315 // Collect the item: the marker line plus its tight prose continuation lines.
1316 // `end_idx` always tracks the last consumed line so the cursor advances past
1317 // the entire item, even when it turns out to be too complex to reflow safely.
1318 let first_piece = trim_preserving_hard_break(&first_body);
1319 let mut simple = !has_hard_break(&first_piece);
1320 let mut body_pieces: Vec<String> = vec![first_piece];
1321 let mut end_idx = start_idx;
1322 let mut i = start_idx + 1;
1323
1324 while i < lines.len() {
1325 let Some(bq) = ctx.lines[i].blockquote.as_deref() else {
1326 // A blank line ends the item.
1327 if lines[i].trim().is_empty() {
1328 break;
1329 }
1330 // A lazy continuation (no `>` marker) is too ambiguous to reflow
1331 // safely. Consume the whole lazy run into this item's span and leave
1332 // the item untouched, so the caller does not reflow the continuation
1333 // in isolation and leave the marker line partially fixed.
1334 simple = false;
1335 while i < lines.len() && ctx.lines[i].blockquote.is_none() && !lines[i].trim().is_empty() {
1336 end_idx = i;
1337 i += 1;
1338 }
1339 break;
1340 };
1341 if bq.nesting_level != target_level {
1342 break;
1343 }
1344
1345 let content = bq.content.as_str();
1346 if content.trim().is_empty() {
1347 // Blank quoted line ends the tight paragraph. A following indented
1348 // paragraph (loose item) is reflowed on its own by the prose path.
1349 break;
1350 }
1351
1352 let eff_indent = effective_indent_in_blockquote(lines[i], target_level, 0);
1353 if eff_indent < base_marker_width {
1354 // Dedented: a sibling list item or text outside this item. Stop here
1355 // and let the outer loop classify it.
1356 break;
1357 }
1358 if is_list_item(content) {
1359 // A nested list item: its own item, handled independently.
1360 break;
1361 }
1362
1363 // An embedded structure (code block, table, fence, nested quote, ...)
1364 // means the item is not simple prose: keep consuming so the cursor clears
1365 // the whole structure, but do not produce a fix.
1366 if self.is_blockquote_content_boundary(content, i + 1, ctx, config) {
1367 simple = false;
1368 }
1369
1370 let piece = trim_preserving_hard_break(content);
1371 if has_hard_break(&piece) {
1372 simple = false;
1373 }
1374 body_pieces.push(piece);
1375 end_idx = i;
1376 i += 1;
1377 }
1378
1379 let next_idx = end_idx + 1;
1380
1381 if !simple {
1382 return (None, next_idx);
1383 }
1384
1385 let exceeds_limit =
1386 || (start_idx..=end_idx).any(|idx| self.calculate_effective_length(lines[idx]) > config.line_length.get());
1387 let body_text = body_pieces.join(" ");
1388 let body_text = body_text.trim();
1389
1390 // A body line that is one whole `$$...$$` expression renders as a display
1391 // block, so it holds a line of its own and the prose on either side of it
1392 // is reflowed separately. Each segment carries whether it is that line.
1393 // A line touched by a code span crossing one of its boundaries is code,
1394 // not such a block. The pieces are the consecutive lines from
1395 // `start_idx`, one piece each.
1396 let body_segments: Vec<(bool, Vec<&str>)> = {
1397 let mut segments: Vec<(bool, Vec<&str>)> = Vec::new();
1398 let mut current: Vec<&str> = Vec::new();
1399 for (offset, piece) in body_pieces.iter().enumerate() {
1400 if (is_self_contained_display_math_line(piece)
1401 || self.line_is_standalone_bracket_math(start_idx + offset + 1, ctx, config))
1402 && !line_touches_multiline_code_span(code_span_touches, start_idx + offset + 1)
1403 {
1404 if !current.is_empty() {
1405 segments.push((false, std::mem::take(&mut current)));
1406 }
1407 segments.push((true, vec![piece.trim_start()]));
1408 } else {
1409 current.push(piece.as_str());
1410 }
1411 }
1412 if !current.is_empty() {
1413 segments.push((false, current));
1414 }
1415 segments
1416 };
1417 let holds_display_math = body_segments.iter().any(|(is_math, _)| *is_math);
1418
1419 // Some bodies cannot be shortened and must stay verbatim, matching the
1420 // exemptions the top-level list reflow applies: link reference definitions
1421 // always, and (in non-strict mode) standalone links/images and HTML-only
1422 // lines. Reflowing a link reference definition would split it after the
1423 // colon/URL and break the definition.
1424 let is_link_ref_def =
1425 body_text.starts_with('[') && body_text.contains("]:") && LINK_REF_PATTERN.is_match(body_text);
1426 let raw_marker_line = lines[start_idx];
1427 let body_is_unwrappable = is_link_ref_def
1428 || standalone_link_ends_paragraph(ctx, start_idx + 1, config)
1429 || (!config.strict && is_html_only_line(raw_marker_line));
1430 if body_is_unwrappable {
1431 return (None, next_idx);
1432 }
1433
1434 let needs_reflow = match config.reflow_mode {
1435 ReflowMode::Normalize => body_pieces.len() > 1 || exceeds_limit(),
1436 ReflowMode::Default => exceeds_limit(),
1437 ReflowMode::SentencePerLine => {
1438 split_into_sentences(body_text, Some(&defined_references), config.require_sentence_capital).len() > 1
1439 || body_pieces.len() > 1
1440 }
1441 ReflowMode::SemanticLineBreaks => {
1442 split_into_sentences(body_text, Some(&defined_references), config.require_sentence_capital).len() > 1
1443 || exceeds_limit()
1444 }
1445 };
1446 if !needs_reflow {
1447 return (None, next_idx);
1448 }
1449
1450 // Apply MD030 list-marker spacing in the spacing-normalizing modes, mirroring
1451 // the top-level list reflow: derive the post-marker spacing from MD030 and let
1452 // the continuation indent follow the resulting content width. Default MD030 (a
1453 // single space) leaves the marker unchanged. MkDocs keeps its rigid indent.
1454 let (marker, marker_width) = if matches!(config.reflow_mode, ReflowMode::Default | ReflowMode::Normalize)
1455 && !ctx.flavor.requires_strict_list_indent()
1456 {
1457 let is_ordered = marker.starts_with(|c: char| c.is_ascii_digit());
1458 // Bullet/number portion only (e.g. `-`, `1.`); the checkbox is content.
1459 let bullet = marker.split(' ').next().unwrap_or("");
1460 let bullet_len = bullet.chars().count();
1461 let checkbox_tail = &marker[base_marker_width..];
1462 // Decide single- vs multi-line spacing from the rewritten shape. This path
1463 // only handles a single tight prose paragraph (structural or multi-paragraph
1464 // items set `simple = false` and bail out above), so the emitted item stays
1465 // multi-line solely when the joined body wraps past one line. A multi-line
1466 // *source* that collapses onto the marker line must use MD030's single-line
1467 // spacing, matching the top-level reflow. The wrap test measures at the
1468 // single-line content column so it does not depend on the spacing chosen here.
1469 let single_col = self.calculate_string_length(&bq_prefix)
1470 + bullet_len
1471 + self.list_spacing.expected_spaces(is_ordered, false, bullet_len)
1472 + checkbox_tail.chars().count();
1473 // A display-math line always holds a line of its own, so an item that
1474 // carries one spans several lines whatever the joined body measures.
1475 let is_multi = holds_display_math
1476 || (!body_text.is_empty()
1477 && self.calculate_effective_length(&format!("{}{body_text}", " ".repeat(single_col)))
1478 > config.line_length.effective_limit());
1479 let spaces = self.list_spacing.expected_spaces(is_ordered, is_multi, bullet_len);
1480 let new_marker = format!("{bullet}{}{checkbox_tail}", " ".repeat(spaces));
1481 let width = new_marker.chars().count();
1482 (new_marker, width)
1483 } else {
1484 (marker, marker_width)
1485 };
1486
1487 let prefix_width = self.calculate_string_length(&bq_prefix) + self.calculate_string_length(&marker);
1488 let reflow_line_length = if config.line_length.is_unlimited() {
1489 usize::MAX
1490 } else {
1491 config.line_length.get().saturating_sub(prefix_width).max(1)
1492 };
1493
1494 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
1495
1496 // A display-math segment is emitted as written; a prose segment is joined
1497 // and reflowed on its own, so the prose above and below the expression
1498 // wraps within its own paragraph.
1499 let mut reflowed: Vec<String> = Vec::new();
1500 for (is_math, segment) in &body_segments {
1501 if *is_math {
1502 reflowed.push(segment[0].to_string());
1503 continue;
1504 }
1505 let segment_text = segment.join(" ");
1506 let segment_text = segment_text.trim();
1507 if segment_text.is_empty() {
1508 continue;
1509 }
1510 reflowed.extend(crate::utils::text_reflow::reflow_line(segment_text, &reflow_options));
1511 }
1512 if reflowed.is_empty() {
1513 return (None, next_idx);
1514 }
1515
1516 let continuation_indent = " ".repeat(marker_width);
1517 let reflowed_text = reflowed
1518 .iter()
1519 .enumerate()
1520 .map(|(idx, line)| {
1521 if idx == 0 {
1522 format!("{bq_prefix}{marker}{line}")
1523 } else {
1524 format!("{bq_prefix}{continuation_indent}{line}")
1525 }
1526 })
1527 .collect::<Vec<_>>()
1528 .join(line_ending);
1529
1530 let start_range = ctx.whole_line_byte_range(start_idx + 1);
1531 let end_range = if end_idx == lines.len() - 1 && !ctx.content.ends_with('\n') {
1532 ctx.line_text_byte_range(end_idx + 1, 1, lines[end_idx].len() + 1)
1533 } else {
1534 ctx.whole_line_byte_range(end_idx + 1)
1535 };
1536 let byte_range = start_range.start..end_range.end;
1537
1538 let replacement = if end_idx < lines.len() - 1 || ctx.content.ends_with('\n') {
1539 format!("{reflowed_text}{line_ending}")
1540 } else {
1541 reflowed_text
1542 };
1543
1544 let original_text = &ctx.content[byte_range.clone()];
1545 if original_text == replacement {
1546 return (None, next_idx);
1547 }
1548
1549 let message = match config.reflow_mode {
1550 ReflowMode::Normalize => format!(
1551 "Paragraph could be normalized to use line length of {} characters",
1552 config.line_length.get()
1553 ),
1554 ReflowMode::SentencePerLine => {
1555 let num_sentences =
1556 split_into_sentences(body_text, Some(&defined_references), config.require_sentence_capital).len();
1557 format!("List item should have one sentence per line (found {num_sentences} sentences)")
1558 }
1559 ReflowMode::SemanticLineBreaks => {
1560 let num_sentences =
1561 split_into_sentences(body_text, Some(&defined_references), config.require_sentence_capital).len();
1562 format!("List item should use semantic line breaks ({num_sentences} sentences)")
1563 }
1564 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
1565 };
1566
1567 let warning = LintWarning {
1568 rule_name: Some(self.name().to_string()),
1569 message,
1570 line: start_idx + 1,
1571 column: 1,
1572 end_line: end_idx + 1,
1573 end_column: lines[end_idx].chars().count() + 1,
1574 severity: Severity::Warning,
1575 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1576 };
1577
1578 (Some(warning), next_idx)
1579 }
1580
1581 /// Generate paragraph-based fixes
1582 fn generate_paragraph_fixes(
1583 &self,
1584 ctx: &crate::lint_context::LintContext,
1585 config: &MD013Config,
1586 lines: &[&str],
1587 ) -> Vec<LintWarning> {
1588 let mut warnings = Vec::new();
1589 let defined_references = Self::defined_reference_labels(ctx);
1590 // A line touched by a code span crossing one of its boundaries is code
1591 // however it is spelled, so a `$$...$$` expression on one is no display
1592 // block.
1593 let code_span_touches = crate::utils::text_reflow::lines_touching_multiline_code_span(ctx.content);
1594
1595 // Detect the content's line ending style to preserve it in replacements.
1596 // The LSP receives content from editors which may use CRLF (Windows).
1597 // Replacements must match the original line endings to avoid false positives.
1598 let line_ending = crate::utils::line_ending::detect_line_ending(ctx.content);
1599
1600 // Ancestor list-item indent shifts, innermost last. When a reflowed parent's
1601 // marker widens under a non-default MD030 (e.g. ul-multi = 3 moves the parent's
1602 // content from column 2 to 4), its nested list/blockquote children are reflowed
1603 // independently and would otherwise keep their original indent — leaving them
1604 // under the parent's new content column, where a CommonMark parser reparses them
1605 // as siblings rather than children. Each frame is (normalized marker width,
1606 // cumulative shift applied to that item's content column); a descendant adds its
1607 // innermost open ancestor's shift to its own indent so the whole subtree moves
1608 // together. With a default MD030 and no marker padding every shift is 0, so this
1609 // is inert and the output is byte-identical.
1610 let mut list_shift_stack: Vec<(usize, isize)> = Vec::new();
1611
1612 let mut i = 0;
1613 while i < lines.len() {
1614 let line_num = i + 1;
1615
1616 // Close ancestor frames whose list item has ended at this line: a non-blank
1617 // line indented less than the frame's source content column is no longer
1618 // inside that item. Blank lines alone don't close a (loose) list item.
1619 if !list_shift_stack.is_empty()
1620 && let Some(info) = ctx.lines.get(i)
1621 && !info.is_blank
1622 {
1623 while let Some(&(content_column, _)) = list_shift_stack.last() {
1624 if info.indent < content_column {
1625 list_shift_stack.pop();
1626 } else {
1627 break;
1628 }
1629 }
1630 }
1631
1632 // Handle blockquote paragraphs with style-preserving reflow.
1633 // Skip blockquotes when blockquotes=false or paragraphs=false
1634 if line_num > 0 && line_num <= ctx.lines.len() && ctx.lines[line_num - 1].blockquote.is_some() {
1635 if !config.blockquotes || !config.paragraphs {
1636 // Skip past all blockquote lines (explicit and lazy continuations).
1637 // A lazy continuation is a non-blank line without `>` that follows
1638 // a blockquote line and isn't a structural element.
1639 let mut saw_explicit_bq = false;
1640 while i < lines.len() && i < ctx.lines.len() {
1641 if ctx.lines[i].blockquote.is_some() {
1642 saw_explicit_bq = true;
1643 i += 1;
1644 } else if saw_explicit_bq
1645 && !lines[i].trim().is_empty()
1646 && !lines[i].trim_start().starts_with('#')
1647 && !lines[i].trim_start().starts_with('>')
1648 {
1649 // Lazy continuation of preceding blockquote
1650 i += 1;
1651 } else {
1652 break;
1653 }
1654 }
1655 continue;
1656 }
1657 // A blockquote nested in a list item moves with its parent when the
1658 // parent's marker widens (see `list_shift_stack`); pass that shift so the
1659 // emitted `>` prefix lands under the parent's new content column instead
1660 // of detaching into a sibling.
1661 let ancestor_shift = list_shift_stack.last().map_or(0isize, |&(_, shift)| shift);
1662 // A list item inside the blockquote needs list-aware reflow (marker +
1663 // continuation indent); plain prose goes through the paragraph path.
1664 let is_bq_list_item = ctx.lines[i]
1665 .blockquote
1666 .as_deref()
1667 .is_some_and(|bq| is_list_item(&bq.content));
1668 let (warning, next_idx) = if is_bq_list_item {
1669 self.generate_blockquote_list_item_fix(
1670 ctx,
1671 config,
1672 lines,
1673 i,
1674 line_ending,
1675 ancestor_shift,
1676 &code_span_touches,
1677 )
1678 } else {
1679 self.generate_blockquote_paragraph_fix(ctx, config, lines, i, line_ending, ancestor_shift)
1680 };
1681 if let Some(warning) = warning {
1682 warnings.push(warning);
1683 }
1684 i = next_idx;
1685 continue;
1686 }
1687
1688 // Skip special structures (but NOT MkDocs containers - those get special handling)
1689 let should_skip_due_to_line_info = ctx.line_info(line_num).is_some_and(|info| {
1690 info.in_code_block
1691 || info.in_front_matter
1692 || info.in_html_block
1693 || info.in_html_comment
1694 || info.in_esm_block
1695 || info.in_jsx_expression
1696 || info.in_jsx_block
1697 || info.in_mdx_comment
1698 || info.in_mkdocstrings
1699 || info.in_pymdown_block
1700 });
1701
1702 // Skip link reference definitions but NOT footnote definitions.
1703 // Footnote definitions (`[^id]: prose`) contain reflowable text,
1704 // while link reference definitions (`[ref]: URL`) contain URLs
1705 // that cannot be shortened.
1706 let is_link_ref_def =
1707 lines[i].trim().starts_with('[') && !lines[i].trim().starts_with("[^") && lines[i].contains("]:");
1708
1709 // A setext heading is a heading, not a paragraph: skip every line of
1710 // its text and its underline together, the way an ATX heading is
1711 // skipped just below. Reflowing any part rewrites the document's
1712 // structure - joining the underline onto the text demotes the
1713 // heading to prose, and rewrapping the text moves words across the
1714 // heading boundary. The text spans the whole paragraph the underline
1715 // ends, so walk the flag to the end of it; the line after the last
1716 // text line is the underline.
1717 if is_setext_heading_text_line(ctx, line_num) {
1718 while i < lines.len() && is_setext_heading_text_line(ctx, i + 1) {
1719 i += 1;
1720 }
1721 i += 1;
1722 continue;
1723 }
1724
1725 if should_skip_due_to_line_info
1726 || lines[i].trim().starts_with('#')
1727 || TableUtils::is_potential_table_row_with_flavor(lines[i], ctx.flavor)
1728 || lines[i].trim().is_empty()
1729 || is_horizontal_rule(lines[i])
1730 || is_template_directive_only(lines[i])
1731 || is_link_ref_def
1732 || ctx.line_info(line_num).is_some_and(|info| info.is_div_marker)
1733 || is_html_only_line(lines[i])
1734 || standalone_link_ends_paragraph(ctx, line_num, config)
1735 {
1736 i += 1;
1737 continue;
1738 }
1739
1740 // Handle footnote definitions: `[^id]: prose text that can be reflowed`
1741 // Supports multi-paragraph footnotes with code blocks, blockquotes,
1742 // tables, and lists preserved verbatim.
1743 // Validate structure: must start with `[^`, contain `]:`, and the ID
1744 // must not contain `[` or `]` (prevents false matches on nested brackets)
1745 if lines[i].trim().starts_with("[^") && lines[i].contains("]:") && {
1746 let after_caret = &lines[i].trim()[2..];
1747 after_caret
1748 .find("]:")
1749 .is_some_and(|pos| pos > 0 && !after_caret[..pos].contains(['[', ']']))
1750 } {
1751 let footnote_start = i;
1752 let line = lines[i];
1753
1754 // Extract the prefix `[^id]:`
1755 let Some(colon_pos) = line.find("]:") else {
1756 i += 1;
1757 continue;
1758 };
1759 let prefix_end = colon_pos + 2;
1760 let prefix = &line[..prefix_end];
1761
1762 // Content starts after `]: ` (with optional space)
1763 let content_start = if line[prefix_end..].starts_with(' ') {
1764 prefix_end + 1
1765 } else {
1766 prefix_end
1767 };
1768 let first_content = &line[content_start..];
1769
1770 // CommonMark footnotes use 4-space continuation indent
1771 const FN_INDENT: usize = 4;
1772
1773 // --- Line classification for footnote content ---
1774 #[derive(Debug, Clone)]
1775 enum FnLineType {
1776 Content(String),
1777 Verbatim(String, usize), // preserved text, original indent
1778 Empty,
1779 }
1780
1781 // Helper: compute visual indent (tabs = 4 spaces)
1782 let visual_indent = |s: &str| -> usize {
1783 s.chars()
1784 .take_while(|c| c.is_whitespace())
1785 .map(|c| if c == '\t' { 4 } else { 1 })
1786 .sum::<usize>()
1787 };
1788
1789 // Helper: check if a trimmed line is a fence marker (homogeneous chars)
1790 let is_fence = |s: &str| -> bool {
1791 let t = s.trim();
1792 let fence_char = t.chars().next();
1793 matches!(fence_char, Some('`') | Some('~'))
1794 && t.chars().take_while(|&c| c == fence_char.unwrap()).count() >= 3
1795 };
1796
1797 // Helper: check if a trimmed line is a setext underline
1798 let is_setext_underline = |s: &str| -> bool {
1799 let t = s.trim();
1800 !t.is_empty()
1801 && (t.chars().all(|c| c == '=' || c == ' ') || t.chars().all(|c| c == '-' || c == ' '))
1802 && t.contains(['=', '-'])
1803 };
1804
1805 // Deferred body: `[^id]:\n content` — first line has no content,
1806 // actual content starts on the next indented line
1807 let deferred_body = first_content.trim().is_empty();
1808
1809 // Collect all lines belonging to this footnote definition
1810 let mut fn_lines: Vec<FnLineType> = Vec::new();
1811 if !deferred_body {
1812 fn_lines.push(FnLineType::Content(first_content.to_string()));
1813 }
1814 let mut last_consumed = i;
1815 i += 1;
1816
1817 // Strip only the footnote continuation indent, preserving
1818 // internal indentation (e.g., code block body indent)
1819 let strip_fn_indent = |s: &str| -> String {
1820 let mut chars = s.chars();
1821 let mut stripped = 0;
1822 while stripped < FN_INDENT {
1823 match chars.next() {
1824 Some('\t') => stripped += 4,
1825 Some(c) if c.is_whitespace() => stripped += 1,
1826 _ => break,
1827 }
1828 }
1829 chars.as_str().to_string()
1830 };
1831
1832 let mut in_fenced_code = false;
1833 let mut consecutive_blanks = 0u32;
1834
1835 while i < lines.len() {
1836 let next = lines[i];
1837 let next_trimmed = next.trim();
1838
1839 // Blank line handling
1840 if next_trimmed.is_empty() {
1841 consecutive_blanks += 1;
1842 // 2+ consecutive blanks terminate the footnote
1843 if consecutive_blanks >= 2 {
1844 break;
1845 }
1846
1847 // Inside a fenced code block, blank lines are part of the code
1848 if in_fenced_code {
1849 consecutive_blanks = 0; // Don't count blanks inside code blocks
1850 fn_lines.push(FnLineType::Verbatim(String::new(), 0));
1851 last_consumed = i;
1852 i += 1;
1853 continue;
1854 }
1855
1856 // Peek ahead: if next non-blank line is indented >= FN_INDENT,
1857 // this blank is an internal paragraph separator
1858 if i + 1 < lines.len() {
1859 let peek = lines[i + 1];
1860 let peek_indent = visual_indent(peek);
1861 if !peek.trim().is_empty() && peek_indent >= FN_INDENT {
1862 fn_lines.push(FnLineType::Empty);
1863 last_consumed = i;
1864 i += 1;
1865 continue;
1866 }
1867 }
1868 // No valid continuation after blank — end of footnote
1869 break;
1870 }
1871
1872 consecutive_blanks = 0;
1873 let indent = visual_indent(next);
1874
1875 // Not indented enough — end of footnote
1876 if indent < FN_INDENT {
1877 break;
1878 }
1879
1880 // Inside a fenced code block: everything is verbatim until closing fence
1881 if in_fenced_code {
1882 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1883 if is_fence(next_trimmed) {
1884 in_fenced_code = false;
1885 }
1886 last_consumed = i;
1887 i += 1;
1888 continue;
1889 }
1890
1891 // Fence opener — start verbatim code block
1892 if is_fence(next_trimmed) {
1893 in_fenced_code = true;
1894 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1895 last_consumed = i;
1896 i += 1;
1897 continue;
1898 }
1899
1900 // A multi-line display-math block is verbatim: its line breaks
1901 // carry meaning (see `line_in_multiline_math_block`).
1902 if self.line_in_multiline_math_block(i + 1, ctx, config) {
1903 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1904 last_consumed = i;
1905 i += 1;
1906 continue;
1907 }
1908
1909 // Indented code block: indent >= FN_INDENT + 4 (= 8 spaces)
1910 if indent >= FN_INDENT + 4 {
1911 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1912 last_consumed = i;
1913 i += 1;
1914 continue;
1915 }
1916
1917 // Structural content that must be preserved verbatim
1918 if next_trimmed.starts_with('#')
1919 || is_list_item(next_trimmed)
1920 || next_trimmed.starts_with('>')
1921 || TableUtils::is_potential_table_row_with_flavor(next_trimmed, ctx.flavor)
1922 || is_setext_underline(next_trimmed)
1923 || is_horizontal_rule(next_trimmed)
1924 || crate::utils::mkdocs_footnotes::is_footnote_definition(next_trimmed)
1925 {
1926 // Preserve verbatim: blockquotes, tables, lists, setext
1927 // underlines, and horizontal rules inside the footnote
1928 if next_trimmed.starts_with('>')
1929 || TableUtils::is_potential_table_row_with_flavor(next_trimmed, ctx.flavor)
1930 || is_list_item(next_trimmed)
1931 || is_setext_underline(next_trimmed)
1932 || is_horizontal_rule(next_trimmed)
1933 {
1934 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1935 last_consumed = i;
1936 i += 1;
1937 continue;
1938 }
1939 // Headings, new footnote defs, link refs — end the footnote
1940 break;
1941 }
1942
1943 // Link reference definitions inside footnotes are not reflowable
1944 if next_trimmed.starts_with('[')
1945 && !next_trimmed.starts_with("[^")
1946 && next_trimmed.contains("]:")
1947 && LINK_REF_PATTERN.is_match(next_trimmed)
1948 {
1949 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1950 last_consumed = i;
1951 i += 1;
1952 continue;
1953 }
1954
1955 // HTML-only lines inside footnotes are not reflowable
1956 if is_html_only_line(next_trimmed) {
1957 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1958 last_consumed = i;
1959 i += 1;
1960 continue;
1961 }
1962
1963 // Regular prose content
1964 fn_lines.push(FnLineType::Content(next_trimmed.to_string()));
1965 last_consumed = i;
1966 i += 1;
1967 }
1968
1969 // Nothing collected or only empty lines
1970 if fn_lines.iter().all(|l| matches!(l, FnLineType::Empty)) || fn_lines.is_empty() {
1971 continue;
1972 }
1973
1974 // --- Group into blocks ---
1975 #[derive(Debug)]
1976 enum FnBlock {
1977 Paragraph(Vec<String>),
1978 Verbatim(Vec<(String, usize)>), // (content, indent) preserved as-is
1979 }
1980
1981 let mut blocks: Vec<FnBlock> = Vec::new();
1982 let mut current_para: Vec<String> = Vec::new();
1983 let mut current_verbatim: Vec<(String, usize)> = Vec::new();
1984
1985 for fl in &fn_lines {
1986 match fl {
1987 FnLineType::Content(s) => {
1988 if !current_verbatim.is_empty() {
1989 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
1990 }
1991 current_para.push(s.clone());
1992 }
1993 FnLineType::Verbatim(s, indent) => {
1994 if !current_para.is_empty() {
1995 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
1996 }
1997 current_verbatim.push((s.clone(), *indent));
1998 }
1999 FnLineType::Empty => {
2000 if !current_para.is_empty() {
2001 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
2002 }
2003 if !current_verbatim.is_empty() {
2004 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
2005 }
2006 }
2007 }
2008 }
2009 if !current_para.is_empty() {
2010 blocks.push(FnBlock::Paragraph(current_para));
2011 }
2012 if !current_verbatim.is_empty() {
2013 blocks.push(FnBlock::Verbatim(current_verbatim));
2014 }
2015
2016 // --- Reflow paragraphs and reconstruct ---
2017 let prefix_display_width = prefix.chars().count() + 1; // +1 for space
2018 let reflow_line_length = if config.line_length.is_unlimited() {
2019 usize::MAX
2020 } else {
2021 config
2022 .line_length
2023 .get()
2024 .saturating_sub(FN_INDENT.max(prefix_display_width))
2025 .max(20)
2026 };
2027 // Footnote continuation uses a fixed 4-space indent, so list
2028 // continuation capping does not apply here.
2029 let reflow_options = crate::utils::text_reflow::ReflowOptions {
2030 max_list_continuation_indent: None,
2031 ..Self::reflow_options(ctx, config, reflow_line_length)
2032 };
2033
2034 let indent_str = " ".repeat(FN_INDENT);
2035 let mut result_lines: Vec<String> = Vec::new();
2036 let mut is_first_block = true;
2037
2038 for block in &blocks {
2039 match block {
2040 FnBlock::Paragraph(para_lines) => {
2041 let paragraph_text = para_lines.join(" ");
2042 let paragraph_text = paragraph_text.trim();
2043 if paragraph_text.is_empty() {
2044 continue;
2045 }
2046
2047 let reflowed = crate::utils::text_reflow::reflow_line(paragraph_text, &reflow_options);
2048 if reflowed.is_empty() {
2049 continue;
2050 }
2051
2052 // Blank line separator between blocks
2053 if !result_lines.is_empty() {
2054 result_lines.push(String::new());
2055 }
2056
2057 for (idx, rline) in reflowed.iter().enumerate() {
2058 if is_first_block && idx == 0 {
2059 result_lines.push(format!("{prefix} {rline}"));
2060 } else {
2061 result_lines.push(format!("{indent_str}{rline}"));
2062 }
2063 }
2064 is_first_block = false;
2065 }
2066 FnBlock::Verbatim(verb_lines) => {
2067 // Blank line separator between blocks
2068 if !result_lines.is_empty() {
2069 result_lines.push(String::new());
2070 }
2071
2072 if is_first_block {
2073 // Verbatim as first block in a deferred-body footnote
2074 if deferred_body {
2075 result_lines.push(prefix.to_string());
2076 }
2077 is_first_block = false;
2078 }
2079 for (content, _orig_indent) in verb_lines {
2080 result_lines.push(format!("{indent_str}{content}"));
2081 }
2082 }
2083 }
2084 }
2085
2086 // If nothing was produced, skip
2087 if result_lines.is_empty() {
2088 continue;
2089 }
2090
2091 let reflowed_text = result_lines.join(line_ending);
2092
2093 // Calculate byte range using last_consumed
2094 let start_range = ctx.whole_line_byte_range(footnote_start + 1);
2095 let end_range = if last_consumed == lines.len() - 1 && !ctx.content.ends_with('\n') {
2096 ctx.line_text_byte_range(last_consumed + 1, 1, lines[last_consumed].len() + 1)
2097 } else {
2098 ctx.whole_line_byte_range(last_consumed + 1)
2099 };
2100 let byte_range = start_range.start..end_range.end;
2101
2102 let replacement = if last_consumed < lines.len() - 1 || ctx.content.ends_with('\n') {
2103 format!("{reflowed_text}{line_ending}")
2104 } else {
2105 reflowed_text
2106 };
2107
2108 let original_text = &ctx.content[byte_range.clone()];
2109 let max_length = (footnote_start..=last_consumed)
2110 .map(|idx| self.calculate_effective_length(lines[idx]))
2111 .max()
2112 .unwrap_or(0);
2113 let line_limit = if config.line_length.is_unlimited() {
2114 usize::MAX
2115 } else {
2116 config.line_length.get()
2117 };
2118 if original_text != replacement && max_length > line_limit {
2119 warnings.push(LintWarning {
2120 rule_name: Some(self.name().to_string()),
2121 message: format!(
2122 "Line length {} exceeds {} characters",
2123 max_length,
2124 config.line_length.get()
2125 ),
2126 line: footnote_start + 1,
2127 column: 1,
2128 end_line: last_consumed + 1,
2129 end_column: lines[last_consumed].chars().count() + 1,
2130 severity: Severity::Warning,
2131 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
2132 });
2133 }
2134 continue;
2135 }
2136
2137 // Handle MkDocs container content (admonitions and tabs) with indent-preserving reflow
2138 if ctx
2139 .line_info(line_num)
2140 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
2141 {
2142 // Skip admonition/tab marker lines — only reflow their indented content
2143 let current_line = lines[i];
2144 if mkdocs_admonitions::is_admonition_start(current_line) || mkdocs_tabs::is_tab_marker(current_line) {
2145 i += 1;
2146 continue;
2147 }
2148
2149 let container_start = i;
2150
2151 // Detect the actual indent level from the first content line
2152 // (supports nested admonitions with 8+ spaces)
2153 let first_line = lines[i];
2154 let base_indent_len = first_line.len() - first_line.trim_start().len();
2155 let base_indent: String = " ".repeat(base_indent_len);
2156
2157 // Collect consecutive MkDocs container paragraph lines
2158 let mut container_lines: Vec<&str> = Vec::new();
2159 while i < lines.len() {
2160 let current_line_num = i + 1;
2161 let line_info = ctx.line_info(current_line_num);
2162
2163 // Stop if we leave the MkDocs container
2164 if !line_info.is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container) {
2165 break;
2166 }
2167
2168 let line = lines[i];
2169
2170 // Stop at paragraph boundaries within the container
2171 if line.trim().is_empty() {
2172 break;
2173 }
2174
2175 // Skip list items, code blocks, headings, HTML-only lines within containers
2176 if is_list_item(line.trim())
2177 || line.trim().starts_with("```")
2178 || line.trim().starts_with("~~~")
2179 || line.trim().starts_with('#')
2180 || is_html_only_line(line)
2181 {
2182 break;
2183 }
2184
2185 container_lines.push(line);
2186 i += 1;
2187 }
2188
2189 if container_lines.is_empty() {
2190 // Must advance i to avoid infinite loop when we encounter
2191 // non-paragraph content (code block, list, heading, empty line)
2192 // at the start of an MkDocs container
2193 i += 1;
2194 continue;
2195 }
2196
2197 // Strip the base indent from each line and join for reflow
2198 let stripped_lines: Vec<&str> = container_lines
2199 .iter()
2200 .map(|line| {
2201 if line.starts_with(&base_indent) {
2202 &line[base_indent_len..]
2203 } else {
2204 line.trim_start()
2205 }
2206 })
2207 .collect();
2208 let paragraph_text = stripped_lines.join(" ");
2209
2210 // Check if reflow is needed
2211 let needs_reflow = match config.reflow_mode {
2212 ReflowMode::Normalize => self.normalize_mode_needs_reflow(container_lines.iter().copied(), config),
2213 ReflowMode::SentencePerLine => {
2214 let sentences = split_into_sentences(
2215 ¶graph_text,
2216 Some(&defined_references),
2217 config.require_sentence_capital,
2218 );
2219 sentences.len() > 1 || container_lines.len() > 1
2220 }
2221 ReflowMode::SemanticLineBreaks => {
2222 let sentences = split_into_sentences(
2223 ¶graph_text,
2224 Some(&defined_references),
2225 config.require_sentence_capital,
2226 );
2227 sentences.len() > 1
2228 || container_lines.len() > 1
2229 || container_lines
2230 .iter()
2231 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
2232 }
2233 ReflowMode::Default => container_lines
2234 .iter()
2235 .any(|line| self.calculate_effective_length(line) > config.line_length.get()),
2236 };
2237
2238 if !needs_reflow {
2239 continue;
2240 }
2241
2242 // Calculate byte range for this container paragraph
2243 let start_range = ctx.whole_line_byte_range(container_start + 1);
2244 let end_line = container_start + container_lines.len() - 1;
2245 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
2246 ctx.line_text_byte_range(end_line + 1, 1, lines[end_line].len() + 1)
2247 } else {
2248 ctx.whole_line_byte_range(end_line + 1)
2249 };
2250 let byte_range = start_range.start..end_range.end;
2251
2252 // Reflow with adjusted line length (accounting for the 4-space indent)
2253 let reflow_line_length = if config.line_length.is_unlimited() {
2254 usize::MAX
2255 } else {
2256 config.line_length.get().saturating_sub(base_indent_len).max(1)
2257 };
2258 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
2259 let reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
2260
2261 // Re-add the 4-space indent to each reflowed line
2262 let reflowed_with_indent: Vec<String> =
2263 reflowed.iter().map(|line| format!("{base_indent}{line}")).collect();
2264 let reflowed_text = reflowed_with_indent.join(line_ending);
2265
2266 // Preserve trailing newline
2267 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
2268 format!("{reflowed_text}{line_ending}")
2269 } else {
2270 reflowed_text
2271 };
2272
2273 // Only generate a warning if the replacement is different
2274 let original_text = &ctx.content[byte_range.clone()];
2275 if original_text != replacement {
2276 warnings.push(LintWarning {
2277 rule_name: Some(self.name().to_string()),
2278 message: format!(
2279 "Line length {} exceeds {} characters (in MkDocs container)",
2280 container_lines.iter().map(|l| l.len()).max().unwrap_or(0),
2281 config.line_length.get()
2282 ),
2283 line: container_start + 1,
2284 column: 1,
2285 end_line: end_line + 1,
2286 end_column: lines[end_line].chars().count() + 1,
2287 severity: Severity::Warning,
2288 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
2289 });
2290 }
2291 continue;
2292 }
2293
2294 // Helper function to detect semantic line markers
2295 let is_semantic_line = |content: &str| -> bool {
2296 let trimmed = content.trim_start();
2297 let semantic_markers = [
2298 "NOTE:",
2299 "WARNING:",
2300 "IMPORTANT:",
2301 "CAUTION:",
2302 "TIP:",
2303 "DANGER:",
2304 "HINT:",
2305 "INFO:",
2306 ];
2307 semantic_markers.iter().any(|marker| trimmed.starts_with(marker))
2308 };
2309
2310 // Helper function to detect fence markers (opening or closing)
2311 let is_fence_marker = |content: &str| -> bool {
2312 let trimmed = content.trim_start();
2313 trimmed.starts_with("```") || trimmed.starts_with("~~~")
2314 };
2315
2316 // Check if this is a list item - handle it specially
2317 let trimmed = lines[i].trim();
2318 if is_list_item(trimmed) {
2319 // Collect the entire list item including continuation lines
2320 let list_start = i;
2321 let (marker, first_content) = extract_list_marker_and_content(lines[i]);
2322 let marker_len = marker.len();
2323 // The normalized marker above is what gets re-emitted; the source marker
2324 // is where the item's content actually starts. Nested blocks move by the
2325 // difference between the two content columns, so the shift must be
2326 // measured against the source, not against the normalized width.
2327 let source_marker = source_list_marker(lines[i]);
2328 let source_content_col = source_marker.as_ref().map_or(marker_len, |m| m.content_col);
2329
2330 // Checkbox ([ ]/[x]/[X]) is inline content, not part of the list marker.
2331 // Use the base bullet/number marker width for continuation recognition
2332 // so that continuation lines at 2+ spaces are collected for "- [ ] " items.
2333 let base_marker_len = if marker.contains("[ ] ") || marker.contains("[x] ") || marker.contains("[X] ") {
2334 marker.find('[').unwrap_or(marker_len)
2335 } else {
2336 marker_len
2337 };
2338
2339 // MkDocs flavor requires at least 4 spaces for list continuation
2340 // after a blank line (multi-paragraph list items). For non-blank
2341 // continuation (lines directly following the marker line), use
2342 // the natural marker width so that 2-space indent is recognized.
2343 let item_indent = ctx.lines[i].indent;
2344 let min_continuation_indent = if ctx.flavor.requires_strict_list_indent() {
2345 // Use 4-space relative indent from the list item's nesting level
2346 item_indent + (base_marker_len - item_indent).max(4)
2347 } else {
2348 marker_len
2349 };
2350 let content_continuation_indent = base_marker_len;
2351
2352 // Track lines and their types (content, code block, fence, nested list)
2353 #[derive(Clone)]
2354 enum LineType {
2355 Content(String, usize), // content and 1-indexed line number
2356 CodeBlock(String, usize), // content and original indent
2357 SemanticLine(String), // Lines starting with NOTE:, WARNING:, etc that should stay separate
2358 SnippetLine(String), // MkDocs Snippets delimiters (-8<-) that must stay on their own line
2359 DivMarker(String), // Quarto/Pandoc div markers (::: opening or closing)
2360 AdmonitionHeader(String, usize), // header text (e.g. "!!! note") and original indent
2361 AdmonitionContent(String, usize), // body content text and original indent
2362 Table(String, usize), // GFM table row, preserved verbatim with original indent
2363 Empty,
2364 }
2365
2366 let start_idx = i;
2367 // A marker line whose content is one whole `$$...$$` expression
2368 // renders as a display block, so it keeps the line it was written
2369 // on and the prose under it starts a paragraph of its own. The
2370 // code-block carrier re-emits it unchanged after the marker.
2371 //
2372 // The marker keeps one padding space and leaves the rest at the
2373 // head of the content, and the carrier writes the marker as the
2374 // author spelled it, padding included. The carrier therefore
2375 // holds the content with its leading whitespace off, so the
2376 // padding is written once.
2377 //
2378 // A marker that cannot interrupt a paragraph, such as an
2379 // ordered one not starting at one, can sit inside a code span
2380 // the paragraph above opened, and its content is code then.
2381 // The marker line can just as well be the one that opens the
2382 // span, closing on a line still to come.
2383 let mut list_item_lines: Vec<LineType> = if (is_self_contained_display_math_line(&first_content)
2384 || self.line_is_standalone_bracket_math(i + 1, ctx, config))
2385 && !line_touches_multiline_code_span(&code_span_touches, i + 1)
2386 {
2387 vec![LineType::CodeBlock(first_content.trim_start().to_string(), marker_len)]
2388 } else {
2389 vec![LineType::Content(first_content, i + 1)]
2390 };
2391 // Set when collection stops at a nested list item or a nested
2392 // blockquote that belongs to this item. Such structure is reflowed
2393 // independently and is therefore absent from `list_item_lines`/`blocks`,
2394 // but it still keeps the emitted item spanning multiple physical lines,
2395 // which the MD030 multi-line spacing decision below must account for.
2396 let mut has_trailing_nested_structure = false;
2397 i += 1;
2398
2399 // Collect continuation lines using ctx.lines for metadata
2400 while i < lines.len() {
2401 let line_info = &ctx.lines[i];
2402
2403 // Use pre-computed is_blank from ctx
2404 if line_info.is_blank {
2405 // Empty line - check if next line is indented (part of list item)
2406 if i + 1 < lines.len() {
2407 let next_info = &ctx.lines[i + 1];
2408
2409 // Check if next line is indented enough to be continuation
2410 if !next_info.is_blank && next_info.indent >= min_continuation_indent {
2411 // This blank line is between paragraphs/blocks in the list item
2412 list_item_lines.push(LineType::Empty);
2413 i += 1;
2414 continue;
2415 }
2416 }
2417 // No indented line after blank, end of list item
2418 break;
2419 }
2420
2421 // Use pre-computed indent from ctx
2422 let indent = line_info.indent;
2423
2424 // Valid continuation must be indented at least content_continuation_indent.
2425 // For non-blank continuation, use marker_len (e.g. 2 for "- ").
2426 // MkDocs strict 4-space requirement applies only after blank lines.
2427 if indent >= content_continuation_indent {
2428 let trimmed = line_info.content(ctx.content).trim();
2429
2430 // Check for MkDocs admonition lines inside list items BEFORE
2431 // checking in_code_block. Lines inside code blocks within
2432 // admonitions have both in_admonition and in_code_block set;
2433 // admonition membership takes priority so the entire admonition
2434 // structure (including embedded code blocks) is preserved.
2435 if line_info.in_admonition {
2436 let raw_content = line_info.content(ctx.content);
2437 if mkdocs_admonitions::is_admonition_start(raw_content) {
2438 let header_text = raw_content[indent..].trim_end().to_string();
2439 list_item_lines.push(LineType::AdmonitionHeader(header_text, indent));
2440 } else {
2441 let body_text = raw_content[indent..].trim_end().to_string();
2442 list_item_lines.push(LineType::AdmonitionContent(body_text, indent));
2443 }
2444 i += 1;
2445 continue;
2446 }
2447
2448 // Use pre-computed in_code_block from ctx
2449 if line_info.in_code_block {
2450 list_item_lines.push(LineType::CodeBlock(
2451 line_info.content(ctx.content)[indent..].to_string(),
2452 indent,
2453 ));
2454 i += 1;
2455 continue;
2456 }
2457
2458 // A multi-line display-math block inside the item is verbatim:
2459 // its line breaks carry meaning (see
2460 // `line_in_multiline_math_block`), so reuse the code-block
2461 // carrier to re-emit it unchanged.
2462 if self.line_in_multiline_math_block(i + 1, ctx, config) {
2463 list_item_lines.push(LineType::CodeBlock(
2464 line_info.content(ctx.content)[indent..].to_string(),
2465 indent,
2466 ));
2467 i += 1;
2468 continue;
2469 }
2470
2471 // A blockquote nested inside the list item is reflowed by the
2472 // blockquote-aware path (it preserves the `>` prefix, including the
2473 // list indent), not as list-item prose. Collecting it as Content
2474 // would strip the markers and reflow `>` as words, collapsing the
2475 // blank `>` line and dropping `>` from wrapped continuations. End
2476 // the item here so the outer loop routes the blockquote line to
2477 // generate_blockquote_paragraph_fix. Uncollect a pending blank so
2478 // the separator between the list prose and the blockquote survives.
2479 if line_info.blockquote.is_some() {
2480 has_trailing_nested_structure = true;
2481 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
2482 list_item_lines.pop();
2483 i -= 1;
2484 }
2485 break;
2486 }
2487
2488 // Check if this is a SIBLING list item (breaks parent)
2489 // Nested lists are indented >= marker_len and are PART of the parent item
2490 // Siblings are at indent < marker_len (at or before parent marker)
2491 if is_list_item(trimmed) && indent < marker_len {
2492 // This is a sibling item at same or higher level - end parent item
2493 break;
2494 }
2495
2496 // Nested list items are always processed independently
2497 // by the outer loop, so break when we encounter one.
2498 // If a blank line was collected before this, uncollect it
2499 // so the outer loop preserves the blank between parent and nested.
2500 if is_list_item(trimmed) && indent >= marker_len {
2501 has_trailing_nested_structure = true;
2502 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
2503 list_item_lines.pop();
2504 i -= 1;
2505 }
2506 break;
2507 }
2508
2509 // Normal continuation vs indented code block.
2510 // Use min_continuation_indent for the threshold since
2511 // code blocks start 4 spaces beyond the expected content
2512 // level (which is min_continuation_indent for MkDocs).
2513 if indent <= min_continuation_indent + 3 {
2514 // Extract content (remove indentation and trailing whitespace)
2515 // Preserve hard breaks (2 trailing spaces) while removing excessive whitespace
2516 // See: https://github.com/rvben/rumdl/issues/76
2517 let content = trim_preserving_hard_break(&line_info.content(ctx.content)[indent..]);
2518
2519 // Check if this is a div marker (::: opening or closing)
2520 // These must be preserved on their own line, not merged into paragraphs
2521 if line_info.is_div_marker {
2522 list_item_lines.push(LineType::DivMarker(content));
2523 }
2524 // A fence marker opens or closes a code block, and a line
2525 // that is one whole `$$...$$` expression renders as a
2526 // display block. Both keep the line they were written on,
2527 // so the code-block carrier re-emits them unchanged
2528 // between the prose above and below. A line touched by a
2529 // code span crossing one of its boundaries is code, not
2530 // such a block.
2531 else if is_fence_marker(&content)
2532 || ((is_self_contained_display_math_line(&content)
2533 || self.line_is_standalone_bracket_math(i + 1, ctx, config))
2534 && !line_touches_multiline_code_span(&code_span_touches, i + 1))
2535 {
2536 list_item_lines.push(LineType::CodeBlock(content, indent));
2537 }
2538 // Check if this is a semantic line (NOTE:, WARNING:, etc.)
2539 else if is_semantic_line(&content) {
2540 list_item_lines.push(LineType::SemanticLine(content));
2541 }
2542 // Check if this is a snippet block delimiter (-8<- or --8<--)
2543 // These must be preserved on their own lines for MkDocs Snippets extension
2544 else if is_snippet_block_delimiter(&content) {
2545 list_item_lines.push(LineType::SnippetLine(content));
2546 }
2547 // Check if this is a GFM table row. Tables nested inside list
2548 // items must be preserved verbatim — joining them with prose
2549 // breaks the column structure.
2550 //
2551 // `is_potential_table_row` is intentionally permissive at the
2552 // row level: any line with `|` and 2+ cells qualifies. To avoid
2553 // misclassifying prose continuation lines that contain a literal
2554 // pipe (e.g. "use grep | sort to ..."), require one of:
2555 // - the row is pipe-bordered (`| ... |`), the canonical form
2556 // for tables nested in lists; or
2557 // - the next line is a delimiter row (this is a header); or
2558 // - the previous classified line was already a Table (this is
2559 // a continuation row).
2560 else if TableUtils::is_potential_table_row_with_flavor(&content, ctx.flavor) && {
2561 let pipe_bordered = content.trim().starts_with('|') && content.trim().ends_with('|');
2562 let next_is_delim = ctx
2563 .lines
2564 .get(i + 1)
2565 .is_some_and(|next| TableUtils::is_delimiter_row(next.content(ctx.content)));
2566 let prev_was_table = matches!(list_item_lines.last(), Some(LineType::Table(..)));
2567 pipe_bordered || next_is_delim || prev_was_table
2568 } {
2569 list_item_lines.push(LineType::Table(content, indent));
2570 } else {
2571 list_item_lines.push(LineType::Content(content, i + 1));
2572 }
2573 i += 1;
2574 } else {
2575 // indent >= min_continuation_indent + 4: indented code block
2576 list_item_lines.push(LineType::CodeBlock(
2577 line_info.content(ctx.content)[indent..].to_string(),
2578 indent,
2579 ));
2580 i += 1;
2581 }
2582 } else {
2583 // Not indented enough, end of list item
2584 break;
2585 }
2586 }
2587
2588 // Determine the output continuation indent.
2589 // Normalize/Default modes canonicalize to min_continuation_indent
2590 // (fixing over-indented continuation). Semantic/SentencePerLine
2591 // modes preserve the user's actual indent since they only fix
2592 // line breaking, not indentation.
2593 let indent_size = match config.reflow_mode {
2594 ReflowMode::SemanticLineBreaks | ReflowMode::SentencePerLine => {
2595 // Find indent of the first plain text continuation line,
2596 // skipping the marker line (index 0), nested list items,
2597 // code blocks, and blank lines.
2598 list_item_lines
2599 .iter()
2600 .enumerate()
2601 .skip(1)
2602 .find_map(|(k, lt)| {
2603 if matches!(lt, LineType::Content(..)) {
2604 Some(ctx.lines[list_start + k].indent)
2605 } else {
2606 None
2607 }
2608 })
2609 .unwrap_or(min_continuation_indent)
2610 }
2611 _ => min_continuation_indent,
2612 };
2613 // For checkbox items in mkdocs flavor, enforce minimum indent so
2614 // continuation lines use the structural list indent (4), not the
2615 // content-aligned indent (6) which Python-Markdown doesn't support
2616 let has_checkbox = base_marker_len < marker_len;
2617 let indent_size = if has_checkbox && ctx.flavor.requires_strict_list_indent() {
2618 indent_size.max(min_continuation_indent)
2619 } else {
2620 indent_size
2621 };
2622
2623 // Split list_item_lines into blocks (paragraphs, code blocks, nested lists, semantic lines, and HTML blocks)
2624 let mut builder = BlockBuilder::new_with_start_line(start_idx + 1);
2625 for line in &list_item_lines {
2626 match line {
2627 LineType::Empty => builder.feed_blank_line(),
2628 LineType::Content(content, _) => builder.feed_content(content),
2629 LineType::CodeBlock(content, indent) => builder.feed_code_line(content, *indent),
2630 LineType::SemanticLine(content) => builder.feed_semantic_line(content),
2631 LineType::SnippetLine(content) => builder.feed_snippet_line(content),
2632 LineType::DivMarker(content) => builder.feed_div_marker(content),
2633 LineType::AdmonitionHeader(header_text, indent) => {
2634 builder.feed_admonition_header(header_text, *indent)
2635 }
2636 LineType::AdmonitionContent(content, indent) => {
2637 builder.feed_admonition_content(content, *indent)
2638 }
2639 LineType::Table(content, indent) => builder.feed_table_line(content, *indent),
2640 }
2641 }
2642 let blocks = builder.finalize();
2643
2644 // Helper: check if a line (raw source or stripped content) is exempt
2645 // from line-length checks. Link reference definitions are always exempt;
2646 // standalone link/image lines are exempt when strict mode is off.
2647 // Also checks content after stripping list markers, since list item
2648 // continuation lines may contain link ref defs.
2649 let is_exempt_line = |raw_line: &str, line_num: usize| -> bool {
2650 let trimmed = raw_line.trim();
2651 // Link reference definitions: always exempt
2652 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
2653 return true;
2654 }
2655 // Also check after stripping list markers (for list item content)
2656 if is_list_item(trimmed) {
2657 let (_, content) = extract_list_marker_and_content(trimmed);
2658 let content_trimmed = content.trim();
2659 if content_trimmed.starts_with('[')
2660 && content_trimmed.contains("]:")
2661 && LINK_REF_PATTERN.is_match(content_trimmed)
2662 {
2663 return true;
2664 }
2665 }
2666 // Standalone link/image lines: exempt when not strict
2667 if standalone_link_ends_paragraph(ctx, line_num, config) {
2668 return true;
2669 }
2670 // HTML-only lines: exempt when not strict
2671 if !config.strict && is_html_only_line(raw_line) {
2672 return true;
2673 }
2674 false
2675 };
2676
2677 // Check if reflowing is needed (only for content paragraphs, not code blocks or nested lists)
2678 // Exclude link reference definitions and standalone link lines from content
2679 // so they don't pollute combined_content or trigger false reflow.
2680 let content_lines: Vec<String> = list_item_lines
2681 .iter()
2682 .filter_map(|line| {
2683 if let LineType::Content(s, line_num) = line {
2684 if is_exempt_line(s, *line_num) {
2685 return None;
2686 }
2687 Some(s.clone())
2688 } else {
2689 None
2690 }
2691 })
2692 .collect();
2693
2694 // Check if we need to reflow this list item
2695 // We check the combined content to see if it exceeds length limits
2696 let combined_content = content_lines.join(" ").trim().to_string();
2697
2698 // Helper to check if we should reflow in normalize mode
2699 let should_normalize = || {
2700 // Don't normalize if the list item only contains nested lists, code blocks, or semantic lines
2701 // DO normalize if it has plain text content that spans multiple lines
2702 let has_code_blocks = blocks.iter().any(|b| matches!(b, Block::Code { .. }));
2703 let has_semantic_lines = blocks.iter().any(|b| matches!(b, Block::SemanticLine(_)));
2704 let has_snippet_lines = blocks.iter().any(|b| matches!(b, Block::SnippetLine(_)));
2705 let has_div_markers = blocks.iter().any(|b| matches!(b, Block::DivMarker(_)));
2706 let has_admonitions = blocks.iter().any(|b| matches!(b, Block::Admonition { .. }));
2707 let has_tables = blocks.iter().any(|b| matches!(b, Block::Table { .. }));
2708 let has_paragraphs = blocks.iter().any(|b| matches!(b, Block::Paragraph(_)));
2709
2710 // If we have structural blocks but no paragraphs, don't normalize
2711 if (has_code_blocks
2712 || has_semantic_lines
2713 || has_snippet_lines
2714 || has_div_markers
2715 || has_admonitions
2716 || has_tables)
2717 && !has_paragraphs
2718 {
2719 return false;
2720 }
2721
2722 // If we have paragraphs, check if they span multiple lines or there are multiple blocks
2723 if has_paragraphs {
2724 // Count only paragraphs that contain at least one non-exempt line.
2725 // Paragraphs consisting entirely of link ref defs or standalone links
2726 // should not trigger normalization.
2727 let paragraph_count = blocks
2728 .iter()
2729 .filter(|b| {
2730 if let Block::Paragraph(para_lines) = b {
2731 !para_lines
2732 .iter()
2733 .all(|(line, line_num)| is_exempt_line(line, *line_num))
2734 } else {
2735 false
2736 }
2737 })
2738 .count();
2739 if paragraph_count > 1 {
2740 // Multiple non-exempt paragraph blocks should be normalized
2741 return true;
2742 }
2743
2744 // Single paragraph block: normalize if it has multiple content lines
2745 if content_lines.len() > 1 {
2746 return true;
2747 }
2748 }
2749
2750 false
2751 };
2752
2753 // Integrate MD030 list-marker spacing (and MD007's text-aligned
2754 // continuation). In Default/Normalize modes — the modes that already
2755 // canonicalize spacing/indent — derive the number of spaces after the
2756 // marker from the configured MD030 values instead of forcing a single
2757 // space, then align continuation lines to the resulting content column.
2758 // Sentence/Semantic modes only adjust line breaks, so they keep the
2759 // marker spacing and indentation already present in the source.
2760 //
2761 // With default MD030 (a single space everywhere) the rebuilt marker is
2762 // byte-identical to the source marker, so this is a no-op and existing
2763 // behaviour is preserved; only a non-default MD030 changes the output.
2764 // The MkDocs flavor enforces a rigid structural indent (4 spaces,
2765 // capped via max_list_continuation_indent) that Python-Markdown
2766 // requires; leave its specialized handling untouched.
2767 let (marker, indent_size, code_indent_shift) =
2768 if matches!(config.reflow_mode, ReflowMode::Default | ReflowMode::Normalize)
2769 && !ctx.flavor.requires_strict_list_indent()
2770 && let Some(li) = ctx.lines[list_start].list_item.as_deref()
2771 {
2772 let bullet_len = li.marker.len();
2773 // The checkbox (e.g. `[ ] `) is content, not part of the list
2774 // marker MD030 governs; carry it over verbatim after the spacing.
2775 let checkbox_tail = marker[base_marker_len..].to_string();
2776 // Shift this item right by its ancestors' cumulative marker
2777 // widening so a nested item stays under its parent's (widened)
2778 // content column. Zero for top-level items and for the whole
2779 // tree under default MD030, where the source indent is preserved
2780 // verbatim (byte-identical output).
2781 let ancestor_shift = list_shift_stack.last().map_or(0isize, |&(_, shift)| shift);
2782 let shifted_indent = (item_indent as isize + ancestor_shift).max(0) as usize;
2783 let indent_prefix = if ancestor_shift == 0 {
2784 marker[..item_indent].to_string()
2785 } else {
2786 " ".repeat(shifted_indent)
2787 };
2788
2789 // Decide single- vs multi-line spacing from the *rewritten* shape,
2790 // not the source. A multi-line source is not enough: plain prose
2791 // continuation collapses onto the marker line during reflow, so a
2792 // two-line bullet that fits becomes a single physical line and must
2793 // use MD030's single-line spacing (otherwise MD013 emits a result
2794 // that MD030 immediately rewrites). The emitted item stays
2795 // multi-line only when reflow cannot collapse it:
2796 // - the prose wraps past the line length, or
2797 // - a structural block remains (code, table, admonition, semantic
2798 // line, snippet, div marker, HTML) that is not joinable prose, or
2799 // - more than one paragraph remains (blank-separated), or
2800 // - a nested list/blockquote follows (reflowed independently, so it
2801 // is absent from `blocks` but still keeps the item multi-line).
2802 // The wrap test uses the single-line content column so it is
2803 // independent of the spacing we are about to choose (avoiding a
2804 // circular result). `ol-align-column` ignores this flag entirely in
2805 // expected_spaces(), so ordered lists are unaffected.
2806 //
2807 // This is the rewritten-shape counterpart of MD030's
2808 // `is_multi_line_list_item` (which keys off the *source*). The two
2809 // are related but technically distinct and intentionally separate;
2810 // if the notion of "multi-line" changes in one, revisit the other.
2811 let single_col = shifted_indent
2812 + bullet_len
2813 + self.list_spacing.expected_spaces(li.is_ordered, false, bullet_len)
2814 + checkbox_tail.len();
2815 let prose_wraps = !combined_content.is_empty()
2816 && self
2817 .calculate_effective_length(&format!("{}{combined_content}", " ".repeat(single_col)))
2818 > config.line_length.effective_limit();
2819 let has_structural_block = blocks.iter().any(|b| !matches!(b, Block::Paragraph(_)));
2820 let multiple_paragraphs =
2821 blocks.iter().filter(|b| matches!(b, Block::Paragraph(_))).count() > 1;
2822 let is_multi =
2823 prose_wraps || has_structural_block || multiple_paragraphs || has_trailing_nested_structure;
2824
2825 let spaces = self.list_spacing.expected_spaces(li.is_ordered, is_multi, bullet_len);
2826 let new_marker = format!("{indent_prefix}{}{}{checkbox_tail}", li.marker, " ".repeat(spaces));
2827 let new_col = new_marker.chars().count();
2828 let shift = new_col as isize - source_content_col as isize;
2829 (new_marker, new_col, shift)
2830 } else {
2831 // MkDocs enforces a rigid structural indent, so the item is emitted
2832 // exactly as written and its nested blocks never move. Re-emitting
2833 // the normalized marker here would narrow the content column while
2834 // leaving those blocks behind.
2835 let marker = source_marker.map_or(marker, |m| m.text);
2836 (marker, indent_size, 0isize)
2837 };
2838 let expected_indent = " ".repeat(indent_size);
2839
2840 // A colon-led line with a line of its paragraph before it opens a
2841 // definition and makes the item a definition list. Joining the
2842 // lines flattens that into prose, so the item is left as the
2843 // author wrote it. The first line of each paragraph is prose
2844 // whatever it starts with, since a definition needs a term on
2845 // the line before it. Content lines arrive with the item's own
2846 // indentation already off, which is what the marker's
2847 // indentation is counted from.
2848 let contains_definition_list = blocks.iter().any(|block| match block {
2849 Block::Paragraph(para_lines) => para_lines
2850 .iter()
2851 .skip(1)
2852 .any(|(line, _)| crate::utils::text_reflow::is_definition_list_marker(line)),
2853 _ => false,
2854 });
2855
2856 let needs_reflow = !contains_definition_list
2857 && match config.reflow_mode {
2858 ReflowMode::Normalize => {
2859 // Only reflow if:
2860 // 1. Any non-exempt paragraph, when joined, exceeds the limit, OR
2861 // 2. Any admonition content line exceeds the limit, OR
2862 // 3. The list item should be normalized (has multi-line plain text)
2863 let any_paragraph_exceeds = blocks.iter().any(|block| match block {
2864 Block::Paragraph(para_lines) => {
2865 if para_lines
2866 .iter()
2867 .all(|(line, line_num)| is_exempt_line(line, *line_num))
2868 {
2869 return false;
2870 }
2871 let joined =
2872 para_lines.iter().map(|(l, _)| l.as_str()).collect::<Vec<_>>().join(" ");
2873 let with_marker = format!("{}{}", " ".repeat(indent_size), joined.trim());
2874 self.calculate_effective_length(&with_marker) > config.line_length.get()
2875 }
2876 Block::Admonition {
2877 content_lines,
2878 header_indent,
2879 ..
2880 } => content_lines.iter().any(|(content, indent)| {
2881 if content.is_empty() {
2882 return false;
2883 }
2884 let with_indent = format!("{}{}", " ".repeat(*indent.max(header_indent)), content);
2885 self.calculate_effective_length(&with_indent) > config.line_length.get()
2886 }),
2887 _ => false,
2888 });
2889 if any_paragraph_exceeds {
2890 true
2891 } else {
2892 should_normalize()
2893 }
2894 }
2895 ReflowMode::SentencePerLine => {
2896 // Check if list item has multiple sentences
2897 let sentences = split_into_sentences(
2898 &combined_content,
2899 Some(&defined_references),
2900 config.require_sentence_capital,
2901 );
2902 sentences.len() > 1
2903 }
2904 ReflowMode::SemanticLineBreaks => {
2905 let sentences = split_into_sentences(
2906 &combined_content,
2907 Some(&defined_references),
2908 config.require_sentence_capital,
2909 );
2910 sentences.len() > 1
2911 || (list_start..i).any(|line_idx| {
2912 let line = lines[line_idx];
2913 let trimmed = line.trim();
2914 if trimmed.is_empty() || is_exempt_line(line, line_idx + 1) {
2915 return false;
2916 }
2917 self.calculate_effective_length(line) > config.line_length.get()
2918 })
2919 }
2920 ReflowMode::Default => {
2921 // In default mode, only reflow if any individual non-exempt line exceeds limit
2922 (list_start..i).any(|line_idx| {
2923 let line = lines[line_idx];
2924 let trimmed = line.trim();
2925 // Skip blank lines and exempt lines
2926 if trimmed.is_empty() || is_exempt_line(line, line_idx + 1) {
2927 return false;
2928 }
2929 self.calculate_effective_length(line) > config.line_length.get()
2930 })
2931 }
2932 };
2933
2934 // Record this item's frame so its nested children inherit the shift.
2935 // Only a reflowed item's marker actually moves; an unreflowed one keeps
2936 // its source position and so contributes no shift to its children. The
2937 // threshold that decides which following lines are inside this item is
2938 // the normalized marker width, NOT `source_content_col`: the collection
2939 // loop above gathers continuations by that same width, so the frame
2940 // boundary must match it or the two would disagree about ownership of
2941 // lines indented between the normalized and the source content column.
2942 list_shift_stack.push((marker_len, if needs_reflow { code_indent_shift } else { 0 }));
2943
2944 if needs_reflow {
2945 let start_range = ctx.whole_line_byte_range(list_start + 1);
2946 let end_line = i - 1;
2947 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
2948 ctx.line_text_byte_range(end_line + 1, 1, lines[end_line].len() + 1)
2949 } else {
2950 ctx.whole_line_byte_range(end_line + 1)
2951 };
2952 let byte_range = start_range.start..end_range.end;
2953
2954 // Reflow each block (paragraphs only, preserve code blocks)
2955 // When line_length = 0 (no limit), use a very large value for reflow
2956 let reflow_line_length = if config.line_length.is_unlimited() {
2957 usize::MAX
2958 } else {
2959 config.line_length.get().saturating_sub(indent_size).max(1)
2960 };
2961 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
2962
2963 let mut result: Vec<String> = Vec::new();
2964 let mut is_first_block = true;
2965
2966 for (block_idx, block) in blocks.iter().enumerate() {
2967 match block {
2968 Block::Paragraph(para_lines) => {
2969 // If every line in this paragraph is exempt (link ref defs,
2970 // standalone links), preserve the paragraph verbatim instead
2971 // of reflowing it. Reflowing would corrupt link ref defs.
2972 let all_exempt = para_lines
2973 .iter()
2974 .all(|(line, line_num)| is_exempt_line(line, *line_num));
2975
2976 if all_exempt {
2977 for (idx, (line, _)) in para_lines.iter().enumerate() {
2978 if is_first_block && idx == 0 {
2979 result.push(format!("{marker}{line}"));
2980 is_first_block = false;
2981 } else {
2982 result.push(format!("{expected_indent}{line}"));
2983 }
2984 }
2985 } else {
2986 // Split the paragraph into segments at hard break boundaries
2987 // Each segment can be reflowed independently
2988 let segments = split_into_segments(para_lines);
2989
2990 for (segment_idx, segment) in segments.iter().enumerate() {
2991 // Check if this segment ends with a hard break and what type
2992 let hard_break_type = segment.last().and_then(|(line, _)| {
2993 let line = line.strip_suffix('\r').unwrap_or(line);
2994 if line.ends_with('\\') {
2995 Some("\\")
2996 } else if line.ends_with(" ") {
2997 Some(" ")
2998 } else {
2999 None
3000 }
3001 });
3002
3003 // Join and reflow the segment (removing the hard break marker for processing)
3004 let segment_for_reflow: Vec<String> = segment
3005 .iter()
3006 .map(|(line, _)| {
3007 // Strip hard break marker (2 spaces or backslash) for reflow processing
3008 if line.ends_with('\\') {
3009 line[..line.len() - 1].trim_end().to_string()
3010 } else if line.ends_with(" ") {
3011 line[..line.len() - 2].trim_end().to_string()
3012 } else {
3013 line.clone()
3014 }
3015 })
3016 .collect();
3017
3018 let segment_text = segment_for_reflow.join(" ").trim().to_string();
3019 if !segment_text.is_empty() {
3020 let reflowed =
3021 crate::utils::text_reflow::reflow_line(&segment_text, &reflow_options);
3022
3023 if is_first_block && segment_idx == 0 {
3024 // First segment of first block starts with marker
3025 result.push(format!("{marker}{}", reflowed[0]));
3026 for line in reflowed.iter().skip(1) {
3027 result.push(format!("{expected_indent}{line}"));
3028 }
3029 is_first_block = false;
3030 } else {
3031 // Subsequent segments
3032 for line in reflowed {
3033 result.push(format!("{expected_indent}{line}"));
3034 }
3035 }
3036
3037 // If this segment had a hard break, add it back to the last line
3038 // Preserve the original hard break format (backslash or two spaces)
3039 if let Some(break_marker) = hard_break_type
3040 && let Some(last_line) = result.last_mut()
3041 {
3042 last_line.push_str(break_marker);
3043 }
3044 }
3045 }
3046 }
3047
3048 // Add blank line after paragraph block if there's a next block.
3049 // Check if next block is a code block that doesn't want a preceding blank.
3050 // Also don't add blank lines before snippet lines (they should stay tight).
3051 // Only add if not already ending with one (avoids double blanks).
3052 if block_idx < blocks.len() - 1 {
3053 let next_block = &blocks[block_idx + 1];
3054 let should_add_blank = match next_block {
3055 Block::Code {
3056 has_preceding_blank, ..
3057 } => *has_preceding_blank,
3058 Block::Table {
3059 has_preceding_blank, ..
3060 } => *has_preceding_blank,
3061 Block::SnippetLine(_) | Block::DivMarker(_) => false,
3062 _ => true, // For all other blocks, add blank line
3063 };
3064 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
3065 result.push(String::new());
3066 }
3067 }
3068 }
3069 Block::Code {
3070 lines: code_lines,
3071 has_preceding_blank: _,
3072 } => {
3073 // Preserve code blocks as-is with original indentation
3074 // NOTE: Blank line before code block is handled by the previous block
3075 // (see paragraph block's logic above)
3076
3077 for (idx, (content, orig_indent)) in code_lines.iter().enumerate() {
3078 if is_first_block && idx == 0 {
3079 // First line of first block gets marker
3080 result.push(format!(
3081 "{marker}{}",
3082 " ".repeat(orig_indent - marker_len) + content.as_str()
3083 ));
3084 is_first_block = false;
3085 } else if content.is_empty() {
3086 result.push(String::new());
3087 } else {
3088 // Shift nested code with the marker so it stays
3089 // aligned under content when MD030 widens spacing.
3090 result.push(format!(
3091 "{}{}",
3092 " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
3093 content
3094 ));
3095 }
3096 }
3097 }
3098 Block::SemanticLine(content) => {
3099 // Preserve semantic lines (NOTE:, WARNING:, etc.) as-is on their own line.
3100 // Only add blank before if not already ending with one.
3101 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
3102 result.push(String::new());
3103 }
3104
3105 if is_first_block {
3106 // First block starts with marker
3107 result.push(format!("{marker}{content}"));
3108 is_first_block = false;
3109 } else {
3110 // Subsequent blocks use expected indent
3111 result.push(format!("{expected_indent}{content}"));
3112 }
3113
3114 // Add blank line after semantic line if there's a next block.
3115 // Only add if not already ending with one.
3116 if block_idx < blocks.len() - 1 {
3117 let next_block = &blocks[block_idx + 1];
3118 let should_add_blank = match next_block {
3119 Block::Code {
3120 has_preceding_blank, ..
3121 } => *has_preceding_blank,
3122 Block::Table {
3123 has_preceding_blank, ..
3124 } => *has_preceding_blank,
3125 Block::SnippetLine(_) | Block::DivMarker(_) => false,
3126 _ => true, // For all other blocks, add blank line
3127 };
3128 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
3129 result.push(String::new());
3130 }
3131 }
3132 }
3133 Block::SnippetLine(content) => {
3134 // Preserve snippet delimiters (-8<-) as-is on their own line
3135 // Unlike semantic lines, snippet lines don't add extra blank lines
3136 if is_first_block {
3137 // First block starts with marker
3138 result.push(format!("{marker}{content}"));
3139 is_first_block = false;
3140 } else {
3141 // Subsequent blocks use expected indent
3142 result.push(format!("{expected_indent}{content}"));
3143 }
3144 // No blank lines added before or after snippet delimiters
3145 }
3146 Block::DivMarker(content) => {
3147 // Preserve div markers (::: opening or closing) as-is on their own line
3148 if is_first_block {
3149 result.push(format!("{marker}{content}"));
3150 is_first_block = false;
3151 } else {
3152 result.push(format!("{expected_indent}{content}"));
3153 }
3154 }
3155 Block::Html {
3156 lines: html_lines,
3157 has_preceding_blank: _,
3158 } => {
3159 // Preserve HTML blocks exactly as-is with original indentation
3160 // NOTE: Blank line before HTML block is handled by the previous block
3161
3162 for (idx, line) in html_lines.iter().enumerate() {
3163 if is_first_block && idx == 0 {
3164 // First line of first block gets marker
3165 result.push(format!("{marker}{line}"));
3166 is_first_block = false;
3167 } else if line.is_empty() {
3168 // Preserve blank lines inside HTML blocks
3169 result.push(String::new());
3170 } else {
3171 // Preserve lines with their original content (already includes indentation)
3172 result.push(format!("{expected_indent}{line}"));
3173 }
3174 }
3175
3176 // Add blank line after HTML block if there's a next block.
3177 // Only add if not already ending with one (avoids double blanks
3178 // when the HTML block itself contained a trailing blank line).
3179 if block_idx < blocks.len() - 1 {
3180 let next_block = &blocks[block_idx + 1];
3181 let should_add_blank = match next_block {
3182 Block::Code {
3183 has_preceding_blank, ..
3184 } => *has_preceding_blank,
3185 Block::Html {
3186 has_preceding_blank, ..
3187 } => *has_preceding_blank,
3188 Block::Table {
3189 has_preceding_blank, ..
3190 } => *has_preceding_blank,
3191 Block::SnippetLine(_) | Block::DivMarker(_) => false,
3192 _ => true, // For all other blocks, add blank line
3193 };
3194 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
3195 result.push(String::new());
3196 }
3197 }
3198 }
3199 Block::Table {
3200 lines: table_lines,
3201 has_preceding_blank: _,
3202 } => {
3203 // Preserve table rows verbatim with their original indentation.
3204 // Reflowing rows would corrupt column alignment and inject `|`
3205 // characters mid-paragraph (issue #590).
3206 // The leading blank line is emitted by the previous block.
3207 for (idx, (content, orig_indent)) in table_lines.iter().enumerate() {
3208 if is_first_block && idx == 0 {
3209 // First line of first block gets the list marker
3210 result.push(format!(
3211 "{marker}{}",
3212 " ".repeat(orig_indent.saturating_sub(marker_len)) + content.as_str()
3213 ));
3214 is_first_block = false;
3215 } else {
3216 // Shift nested table rows with the marker so they
3217 // stay aligned when MD030 widens marker spacing.
3218 result.push(format!(
3219 "{}{}",
3220 " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
3221 content
3222 ));
3223 }
3224 }
3225
3226 // Add blank line after table block if there's a next block.
3227 if block_idx < blocks.len() - 1 {
3228 let next_block = &blocks[block_idx + 1];
3229 let should_add_blank = match next_block {
3230 Block::Code {
3231 has_preceding_blank, ..
3232 } => *has_preceding_blank,
3233 Block::Html {
3234 has_preceding_blank, ..
3235 } => *has_preceding_blank,
3236 Block::Table {
3237 has_preceding_blank, ..
3238 } => *has_preceding_blank,
3239 Block::SnippetLine(_) | Block::DivMarker(_) => false,
3240 _ => true,
3241 };
3242 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
3243 result.push(String::new());
3244 }
3245 }
3246 }
3247 Block::Admonition {
3248 header,
3249 header_indent,
3250 content_lines: admon_lines,
3251 } => {
3252 // Reconstruct admonition block with header at original indent
3253 // and body content reflowed to fit within the line length limit
3254
3255 // Add blank line before admonition if not first block
3256 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
3257 result.push(String::new());
3258 }
3259
3260 // Output the header at its original indent
3261 let header_indent_str = " ".repeat(*header_indent);
3262 if is_first_block {
3263 result.push(format!(
3264 "{marker}{}",
3265 " ".repeat(header_indent.saturating_sub(marker_len)) + header.as_str()
3266 ));
3267 is_first_block = false;
3268 } else {
3269 result.push(format!("{header_indent_str}{header}"));
3270 }
3271
3272 // Derive body indent from the first non-empty content line's
3273 // stored indent, falling back to header_indent + 4 for
3274 // empty-body admonitions
3275 let body_indent = admon_lines
3276 .iter()
3277 .find(|(content, _)| !content.is_empty())
3278 .map_or(header_indent + 4, |(_, indent)| *indent);
3279 let body_indent_str = " ".repeat(body_indent);
3280
3281 // Segment body content into code blocks (verbatim) and
3282 // text paragraphs (reflowable), separated by blank lines.
3283 // Code lines store (content, orig_indent) to reconstruct
3284 // internal indentation relative to body_indent.
3285 enum AdmonSegment {
3286 Text(Vec<String>),
3287 Code(Vec<(String, usize)>),
3288 }
3289
3290 let mut segments: Vec<AdmonSegment> = Vec::new();
3291 let mut current_text: Vec<String> = Vec::new();
3292 let mut current_code: Vec<(String, usize)> = Vec::new();
3293 let mut in_admon_code = false;
3294 // Track the opening fence character so closing fences
3295 // must match (backticks close backticks, tildes close tildes)
3296 let mut fence_char: char = '`';
3297
3298 // Opening fences: ``` or ~~~ followed by optional info string
3299 let get_opening_fence = |s: &str| -> Option<(char, usize)> {
3300 let t = s.trim_start();
3301 if t.starts_with("```") {
3302 Some(('`', t.bytes().take_while(|&b| b == b'`').count()))
3303 } else if t.starts_with("~~~") {
3304 Some(('~', t.bytes().take_while(|&b| b == b'~').count()))
3305 } else {
3306 None
3307 }
3308 };
3309 // Closing fences: ONLY fence chars + optional trailing spaces
3310 let get_closing_fence = |s: &str| -> Option<(char, usize)> {
3311 let t = s.trim();
3312 if t.starts_with("```") && t.bytes().all(|b| b == b'`') {
3313 Some(('`', t.len()))
3314 } else if t.starts_with("~~~") && t.bytes().all(|b| b == b'~') {
3315 Some(('~', t.len()))
3316 } else {
3317 None
3318 }
3319 };
3320 let mut fence_len: usize = 3;
3321
3322 for (content, orig_indent) in admon_lines {
3323 if in_admon_code {
3324 // Closing fence must use the same character, be
3325 // at least as long, and have no info string
3326 if let Some((ch, len)) = get_closing_fence(content)
3327 && ch == fence_char
3328 && len >= fence_len
3329 {
3330 current_code.push((content.clone(), *orig_indent));
3331 in_admon_code = false;
3332 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
3333 continue;
3334 }
3335 current_code.push((content.clone(), *orig_indent));
3336 } else if let Some((ch, len)) = get_opening_fence(content) {
3337 if !current_text.is_empty() {
3338 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
3339 }
3340 in_admon_code = true;
3341 fence_char = ch;
3342 fence_len = len;
3343 current_code.push((content.clone(), *orig_indent));
3344 } else if content.is_empty() {
3345 if !current_text.is_empty() {
3346 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
3347 }
3348 } else {
3349 current_text.push(content.clone());
3350 }
3351 }
3352 if in_admon_code && !current_code.is_empty() {
3353 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
3354 }
3355 if !current_text.is_empty() {
3356 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
3357 }
3358
3359 // Build reflow options once for all text segments
3360 let admon_reflow_length = if config.line_length.is_unlimited() {
3361 usize::MAX
3362 } else {
3363 config.line_length.get().saturating_sub(body_indent).max(1)
3364 };
3365
3366 let admon_reflow_options = Self::reflow_options(ctx, config, admon_reflow_length);
3367
3368 // Output each segment
3369 for segment in &segments {
3370 // Blank line before each segment (after the header or previous segment)
3371 result.push(String::new());
3372
3373 match segment {
3374 AdmonSegment::Code(lines) => {
3375 for (line, orig_indent) in lines {
3376 if line.is_empty() {
3377 // Preserve blank lines inside code blocks
3378 result.push(String::new());
3379 } else {
3380 // Reconstruct with body_indent + any extra
3381 // indentation the line had beyond body_indent
3382 let extra = orig_indent.saturating_sub(body_indent);
3383 let indent_str = " ".repeat(body_indent + extra);
3384 result.push(format!("{indent_str}{line}"));
3385 }
3386 }
3387 }
3388 AdmonSegment::Text(lines) => {
3389 let paragraph_text = lines.join(" ").trim().to_string();
3390 if paragraph_text.is_empty() {
3391 continue;
3392 }
3393 let reflowed = crate::utils::text_reflow::reflow_line(
3394 ¶graph_text,
3395 &admon_reflow_options,
3396 );
3397 for line in &reflowed {
3398 result.push(format!("{body_indent_str}{line}"));
3399 }
3400 }
3401 }
3402 }
3403
3404 // Add blank line after admonition if there's a next block
3405 if block_idx < blocks.len() - 1 {
3406 let next_block = &blocks[block_idx + 1];
3407 let should_add_blank = match next_block {
3408 Block::Code {
3409 has_preceding_blank, ..
3410 } => *has_preceding_blank,
3411 Block::Table {
3412 has_preceding_blank, ..
3413 } => *has_preceding_blank,
3414 Block::SnippetLine(_) | Block::DivMarker(_) => false,
3415 _ => true,
3416 };
3417 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
3418 result.push(String::new());
3419 }
3420 }
3421 }
3422 }
3423 }
3424
3425 let reflowed_text = result.join(line_ending);
3426
3427 // Preserve trailing newline
3428 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
3429 format!("{reflowed_text}{line_ending}")
3430 } else {
3431 reflowed_text
3432 };
3433
3434 // Get the original text to compare
3435 let original_text = &ctx.content[byte_range.clone()];
3436
3437 // Physical-line-length scan, shared by the Normalize-mode gate and its
3438 // message. The list-item reflow preserves code blocks, HTML blocks,
3439 // admonition headers, fence markers, semantic markers, and snippet/div
3440 // markers verbatim; only paragraph content and admonition bodies are
3441 // restructured. Only those lines drive the length warning, so that
3442 // preserved-but-overlong content does not keep the paragraph-level
3443 // warning alive when the reflow would not fix that line.
3444 let should_count_for_length = |line_idx: usize| -> bool {
3445 let line = lines[line_idx];
3446 let trimmed = line.trim();
3447 if trimmed.is_empty() || is_exempt_line(line, line_idx + 1) {
3448 return false;
3449 }
3450 let info = &ctx.lines[line_idx];
3451 if info.in_code_block || info.in_html_block {
3452 return false;
3453 }
3454 if info.in_admonition && mkdocs_admonitions::is_admonition_start(line) {
3455 return false;
3456 }
3457 if is_fence_marker(line) || is_semantic_line(line) {
3458 return false;
3459 }
3460 if is_snippet_block_delimiter(line) {
3461 return false;
3462 }
3463 if line.trim_start().starts_with(":::") {
3464 return false;
3465 }
3466 true
3467 };
3468 let max_physical_length = (list_start..i)
3469 .filter(|&idx| should_count_for_length(idx))
3470 .map(|idx| self.calculate_effective_length(lines[idx]))
3471 .max()
3472 .unwrap_or(0);
3473 // `line-length = 0` means "no limit", so no physical line can be
3474 // "over"; the message below then describes a structural join rather
3475 // than a length violation.
3476 let any_paragraph_line_over =
3477 !config.line_length.is_unlimited() && max_physical_length > config.line_length.get();
3478
3479 // Normalize mode reflows list-item prose just like paragraphs:
3480 // joining continuation lines and re-wrapping to `line-length`.
3481 // `prose_changed` is true only when the reflow alters the words or
3482 // line breaks, not when it would merely re-indent continuation
3483 // lines or trim trailing whitespace. Comparing the texts with each
3484 // line's leading and trailing whitespace removed isolates "did the
3485 // words/line breaks change" from "did the surrounding whitespace
3486 // change". Continuation indentation is MD077's responsibility and
3487 // trailing whitespace is MD009's; an MD013 warning for either would
3488 // both duplicate those rules and resurface a persistent advisory on
3489 // already-fitting items that users disable MD013 fixing to avoid.
3490 let prose_changed = {
3491 let stripped = |text: &str| text.lines().map(str::trim).collect::<Vec<_>>().join("\n");
3492 stripped(original_text) != stripped(&replacement)
3493 };
3494 // Warn when the reflow rewraps prose (the normalize feature for
3495 // list items), or when a physical line genuinely exceeds the limit
3496 // and the reflow can change something (a true length violation,
3497 // even if all that changes is the continuation indent). A line that
3498 // is already optimal in both respects produces no warning.
3499 let gate_ok = prose_changed || (any_paragraph_line_over && original_text != replacement);
3500 if gate_ok {
3501 // Generate an appropriate message based on why reflow is needed
3502 let message = match config.reflow_mode {
3503 ReflowMode::SentencePerLine => {
3504 let num_sentences = split_into_sentences(
3505 &combined_content,
3506 Some(&defined_references),
3507 config.require_sentence_capital,
3508 )
3509 .len();
3510 let num_lines = content_lines.len();
3511 if num_lines == 1 {
3512 // Single line with multiple sentences
3513 format!("Line contains {num_sentences} sentences (one sentence per line required)")
3514 } else {
3515 // Multiple lines - could be split sentences or mixed
3516 format!(
3517 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
3518 )
3519 }
3520 }
3521 ReflowMode::SemanticLineBreaks => {
3522 let num_sentences = split_into_sentences(
3523 &combined_content,
3524 Some(&defined_references),
3525 config.require_sentence_capital,
3526 )
3527 .len();
3528 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
3529 }
3530 ReflowMode::Normalize => {
3531 // When a physical line genuinely exceeds the limit, report
3532 // it as a length violation. Otherwise the reflow is a
3533 // structural normalization (joining/re-wrapping multi-line
3534 // content that already fits), mirroring the paragraph path.
3535 if any_paragraph_line_over {
3536 format!(
3537 "Line length {} exceeds {} characters",
3538 max_physical_length,
3539 config.line_length.get()
3540 )
3541 } else {
3542 format!(
3543 "List item could be normalized to use line length of {} characters",
3544 config.line_length.get()
3545 )
3546 }
3547 }
3548 ReflowMode::Default => {
3549 // Report the actual longest non-exempt line, not the combined content
3550 let max_length = (list_start..i)
3551 .filter(|&line_idx| {
3552 let line = lines[line_idx];
3553 let trimmed = line.trim();
3554 !trimmed.is_empty() && !is_exempt_line(line, line_idx + 1)
3555 })
3556 .map(|line_idx| self.calculate_effective_length(lines[line_idx]))
3557 .max()
3558 .unwrap_or(0);
3559 format!(
3560 "Line length {} exceeds {} characters",
3561 max_length,
3562 config.line_length.get()
3563 )
3564 }
3565 };
3566
3567 warnings.push(LintWarning {
3568 rule_name: Some(self.name().to_string()),
3569 message,
3570 line: list_start + 1,
3571 column: 1,
3572 end_line: end_line + 1,
3573 end_column: lines[end_line].chars().count() + 1,
3574 severity: Severity::Warning,
3575 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
3576 });
3577 }
3578 }
3579 continue;
3580 }
3581
3582 // Found start of a paragraph - collect all lines in it
3583 let paragraph_start = i;
3584 let mut paragraph_lines = vec![lines[i]];
3585 i += 1;
3586
3587 while i < lines.len() {
3588 let next_line = lines[i];
3589 let next_line_num = i + 1;
3590 let next_trimmed = next_line.trim();
3591
3592 // Stop at paragraph boundaries
3593 if next_trimmed.is_empty()
3594 || ctx.line_info(next_line_num).is_some_and(|info| info.in_code_block)
3595 || ctx.line_info(next_line_num).is_some_and(|info| info.in_front_matter)
3596 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_block)
3597 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_comment)
3598 || ctx.line_info(next_line_num).is_some_and(|info| info.in_esm_block)
3599 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_expression)
3600 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_block)
3601 || ctx.line_info(next_line_num).is_some_and(|info| info.in_mdx_comment)
3602 || ctx
3603 .line_info(next_line_num)
3604 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
3605 || (next_line_num > 0
3606 && next_line_num <= ctx.lines.len()
3607 && ctx.lines[next_line_num - 1].blockquote.is_some())
3608 || next_trimmed.starts_with('#')
3609 // A setext heading ends the paragraph before it. The flag is
3610 // set on every line of the heading's text, so the paragraph
3611 // stops at the first of them and absorbs no heading text.
3612 // That also protects the underline, which can only follow the
3613 // last text line, so absorbing the construct and joining it
3614 // into prose is unreachable from here. `is_horizontal_rule`
3615 // below catches a `---` underline only by coincidence (3+
3616 // dashes are also a thematic break); `=` and short `-` runs
3617 // have no such overlap.
3618 || is_setext_heading_text_line(ctx, next_line_num)
3619 || TableUtils::is_potential_table_row_with_flavor(next_line, ctx.flavor)
3620 || is_list_item(next_trimmed)
3621 || is_horizontal_rule(next_line)
3622 || (next_trimmed.starts_with('[') && next_line.contains("]:"))
3623 || is_template_directive_only(next_line)
3624 || is_standalone_attr_list(next_line)
3625 || is_snippet_block_delimiter(next_line)
3626 || ctx.line_info(next_line_num).is_some_and(|info| info.is_div_marker)
3627 || is_html_only_line(next_line)
3628 || self.line_in_multiline_math_block(next_line_num, ctx, config)
3629 // A line that is one whole `$$...$$` expression renders as a
3630 // display block, so it ends the paragraph above it and is
3631 // reflowed on its own. A line touched by a code span crossing
3632 // one of its boundaries is code, not such a block.
3633 || ((is_self_contained_display_math_line(next_line)
3634 || self.line_is_standalone_bracket_math(next_line_num, ctx, config))
3635 && !line_touches_multiline_code_span(&code_span_touches, next_line_num))
3636 || standalone_link_ends_paragraph(ctx, next_line_num, config)
3637 {
3638 break;
3639 }
3640
3641 // Check if the previous line ends with a hard break (2+ spaces or backslash)
3642 if i > 0 && has_hard_break(lines[i - 1]) {
3643 // Don't include lines after hard breaks in the same paragraph
3644 break;
3645 }
3646
3647 paragraph_lines.push(next_line);
3648 i += 1;
3649 }
3650
3651 // Compute the common leading indent of all non-empty paragraph lines,
3652 // but only when those lines are structurally inside a list block.
3653 // Indented continuation lines that follow a nested list arrive here
3654 // with their structural indentation intact (e.g. 2 spaces for a
3655 // top-level list item). Stripping the indent before reflow and
3656 // re-applying it afterward prevents the fixer from moving those
3657 // lines to column 0.
3658 //
3659 // The list-block guard is essential: top-level paragraphs that happen
3660 // to start with spaces (insignificant in Markdown) must NOT have those
3661 // spaces preserved or injected by the fixer.
3662 let common_indent: String = if ctx.is_in_list_block(paragraph_start + 1) {
3663 let min_len = paragraph_lines
3664 .iter()
3665 .filter(|l| !l.trim().is_empty())
3666 .map(|l| l.len() - l.trim_start().len())
3667 .min()
3668 .unwrap_or(0);
3669 paragraph_lines
3670 .iter()
3671 .find(|l| !l.trim().is_empty())
3672 .map(|l| l[..min_len].to_string())
3673 .unwrap_or_default()
3674 } else {
3675 String::new()
3676 };
3677
3678 // Combine paragraph lines into a single string for processing.
3679 // This must be done BEFORE the needs_reflow check for sentence-per-line mode.
3680 let paragraph_text = if common_indent.is_empty() {
3681 join_soft_break_lines(¶graph_lines)
3682 } else {
3683 let stripped: Vec<&str> = paragraph_lines
3684 .iter()
3685 .map(|l| {
3686 if l.starts_with(common_indent.as_str()) {
3687 &l[common_indent.len()..]
3688 } else {
3689 l.trim_start()
3690 }
3691 })
3692 .collect();
3693 join_soft_break_lines(&stripped)
3694 };
3695
3696 // A colon-led line with a line of the paragraph before it opens a
3697 // definition, and joining the lines would flatten the definition
3698 // list into prose, so the paragraph is skipped. Its first line is
3699 // prose whatever it starts with, since a definition needs a term on
3700 // the line before it. The indentation a marker is allowed is
3701 // counted from the block's own content, so a list item's
3702 // indentation comes off first.
3703 let contains_definition_list = paragraph_lines.iter().skip(1).any(|line| {
3704 let content = line.strip_prefix(common_indent.as_str()).unwrap_or(line.trim_start());
3705 crate::utils::text_reflow::is_definition_list_marker(content)
3706 });
3707
3708 if contains_definition_list {
3709 // Don't reflow definition lists - skip this paragraph
3710 i = paragraph_start + paragraph_lines.len();
3711 continue;
3712 }
3713
3714 // Skip reflowing if this paragraph contains MkDocs Snippets markers
3715 // Snippets blocks (-8<- ... -8<-) should be preserved exactly
3716 let contains_snippets = paragraph_lines.iter().any(|line| is_snippet_block_delimiter(line));
3717
3718 if contains_snippets {
3719 // Don't reflow Snippets blocks - skip this paragraph
3720 i = paragraph_start + paragraph_lines.len();
3721 continue;
3722 }
3723
3724 // Leave a line of a multi-line display-math block as it is, where
3725 // joining lines would corrupt the equation (see
3726 // `line_in_multiline_math_block`). Only the first line has to be
3727 // asked about: such a line ends the paragraph above it, so a
3728 // paragraph reaching here holds one only when it starts on one.
3729 //
3730 // Only that line is passed over, not the rest of what was collected
3731 // with it: prose written directly under the closing delimiter is an
3732 // ordinary paragraph and still reflows.
3733 if self.line_in_multiline_math_block(paragraph_start + 1, ctx, config) {
3734 i = paragraph_start + 1;
3735 continue;
3736 }
3737
3738 // A line that is one whole `$$...$$` expression is a display block:
3739 // it keeps the line it was written on, and the prose under it is an
3740 // ordinary paragraph that still reflows. The line above ended at
3741 // this one, so a paragraph reaching here holds one only when it
3742 // starts on one. A paragraph can start on a line touched by a code
3743 // span crossing one of its boundaries, under a hard break the span
3744 // holds, and its first line is code then.
3745 if (is_self_contained_display_math_line(lines[paragraph_start])
3746 || self.line_is_standalone_bracket_math(paragraph_start + 1, ctx, config))
3747 && !line_touches_multiline_code_span(&code_span_touches, paragraph_start + 1)
3748 {
3749 i = paragraph_start + 1;
3750 continue;
3751 }
3752
3753 // Check if this paragraph needs reflowing
3754 let needs_reflow = match config.reflow_mode {
3755 ReflowMode::Normalize => self.normalize_mode_needs_reflow(paragraph_lines.iter().copied(), config),
3756 ReflowMode::SentencePerLine => {
3757 // In sentence-per-line mode, check if the JOINED paragraph has multiple sentences
3758 // Note: we check the joined text because sentences can span multiple lines
3759 let sentences = split_into_sentences(
3760 ¶graph_text,
3761 Some(&defined_references),
3762 config.require_sentence_capital,
3763 );
3764
3765 // Always reflow if multiple sentences on one line
3766 if sentences.len() > 1 {
3767 true
3768 } else if paragraph_lines.len() > 1 {
3769 // For single-sentence paragraphs spanning multiple lines:
3770 // Reflow if they COULD fit on one line (respecting line-length constraint)
3771 if config.line_length.is_unlimited() {
3772 // No line-length constraint - always join single sentences
3773 true
3774 } else {
3775 // Only join if it fits within line-length.
3776 // paragraph_text has the common indent stripped, so add it
3777 // back to get the true output length before comparing.
3778 let effective_length =
3779 self.calculate_effective_length(¶graph_text) + common_indent.len();
3780 effective_length <= config.line_length.get()
3781 }
3782 } else {
3783 false
3784 }
3785 }
3786 ReflowMode::SemanticLineBreaks => {
3787 let sentences = split_into_sentences(
3788 ¶graph_text,
3789 Some(&defined_references),
3790 config.require_sentence_capital,
3791 );
3792 // Reflow if multiple sentences, multiple lines, or any line exceeds limit
3793 sentences.len() > 1
3794 || paragraph_lines.len() > 1
3795 || paragraph_lines
3796 .iter()
3797 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3798 }
3799 ReflowMode::Default => {
3800 // In default mode, only reflow if lines exceed limit
3801 paragraph_lines
3802 .iter()
3803 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3804 }
3805 };
3806
3807 if needs_reflow {
3808 // Calculate byte range for this paragraph
3809 // Use whole_line_range for each line and combine
3810 let start_range = ctx.whole_line_byte_range(paragraph_start + 1);
3811 let end_line = paragraph_start + paragraph_lines.len() - 1;
3812
3813 // For the last line, we want to preserve any trailing newline
3814 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
3815 // Last line without trailing newline - use line_text_range
3816 ctx.line_text_byte_range(end_line + 1, 1, lines[end_line].len() + 1)
3817 } else {
3818 // Not the last line or has trailing newline - use whole_line_range
3819 ctx.whole_line_byte_range(end_line + 1)
3820 };
3821
3822 let byte_range = start_range.start..end_range.end;
3823
3824 // Check if the paragraph ends with a hard break and what type
3825 let hard_break_type = paragraph_lines.last().and_then(|line| {
3826 let line = line.strip_suffix('\r').unwrap_or(line);
3827 if line.ends_with('\\') {
3828 Some("\\")
3829 } else if line.ends_with(" ") {
3830 Some(" ")
3831 } else {
3832 None
3833 }
3834 });
3835
3836 // Reflow the paragraph
3837 // When line_length = 0 (no limit), use a very large value for reflow
3838 let reflow_line_length = if config.line_length.is_unlimited() {
3839 usize::MAX
3840 } else {
3841 config.line_length.get().saturating_sub(common_indent.len()).max(1)
3842 };
3843 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
3844 let mut reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
3845
3846 // Re-apply the common indent to each non-empty reflowed line so
3847 // that the replacement preserves the original structural indentation.
3848 if !common_indent.is_empty() {
3849 for line in &mut reflowed {
3850 if !line.is_empty() {
3851 *line = format!("{common_indent}{line}");
3852 }
3853 }
3854 }
3855
3856 // If the original paragraph ended with a hard break, preserve it
3857 // Preserve the original hard break format (backslash or two spaces)
3858 if let Some(break_marker) = hard_break_type
3859 && !reflowed.is_empty()
3860 {
3861 let last_idx = reflowed.len() - 1;
3862 if !has_hard_break(&reflowed[last_idx]) {
3863 reflowed[last_idx].push_str(break_marker);
3864 }
3865 }
3866
3867 let reflowed_text = reflowed.join(line_ending);
3868
3869 // Preserve trailing newline if the original paragraph had one
3870 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
3871 format!("{reflowed_text}{line_ending}")
3872 } else {
3873 reflowed_text
3874 };
3875
3876 // Get the original text to compare
3877 let original_text = &ctx.content[byte_range.clone()];
3878
3879 // Only generate a warning if the replacement is different from the original
3880 if original_text != replacement {
3881 // Determine which line ranges and messages to report based on the reflow mode.
3882 let warnings_to_report: Vec<(usize, usize, String)> = match config.reflow_mode {
3883 ReflowMode::Default => {
3884 // In default mode, report a warning for *every* line in the paragraph
3885 // that exceeds the limit. Each warning will carry the same paragraph-level
3886 // fix, making all of them auto-fixable.
3887 paragraph_lines
3888 .iter()
3889 .enumerate()
3890 .filter(|(_, line)| self.calculate_effective_length(line) > config.line_length.get())
3891 .map(|(idx, _)| {
3892 let violating_line = paragraph_start + idx + 1;
3893 (
3894 violating_line,
3895 violating_line,
3896 format!("Line length exceeds {} characters", config.line_length.get()),
3897 )
3898 })
3899 .collect()
3900 }
3901 ReflowMode::Normalize => {
3902 // In normalize mode, report the whole paragraph as needing normalization.
3903 vec![(
3904 paragraph_start + 1,
3905 end_line + 1,
3906 format!(
3907 "Paragraph could be normalized to use line length of {} characters",
3908 config.line_length.get()
3909 ),
3910 )]
3911 }
3912 ReflowMode::SentencePerLine => {
3913 // In sentence-per-line mode, highlight the entire paragraph that needs reformatting.
3914 let num_sentences = split_into_sentences(
3915 ¶graph_text,
3916 Some(&defined_references),
3917 config.require_sentence_capital,
3918 )
3919 .len();
3920 let message = if paragraph_lines.len() == 1 {
3921 // Single line with multiple sentences
3922 format!("Line contains {num_sentences} sentences (one sentence per line required)")
3923 } else {
3924 // Multiple lines - could be split sentences or mixed
3925 let num_lines = paragraph_lines.len();
3926 format!(
3927 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
3928 )
3929 };
3930 vec![(paragraph_start + 1, paragraph_start + paragraph_lines.len(), message)]
3931 }
3932 ReflowMode::SemanticLineBreaks => {
3933 // In semantic-line-breaks mode, highlight the entire paragraph.
3934 let num_sentences = split_into_sentences(
3935 ¶graph_text,
3936 Some(&defined_references),
3937 config.require_sentence_capital,
3938 )
3939 .len();
3940 vec![(
3941 paragraph_start + 1,
3942 paragraph_start + paragraph_lines.len(),
3943 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)"),
3944 )]
3945 }
3946 };
3947
3948 // Generate the actual lint warnings. All warnings for this paragraph
3949 // share the same paragraph-level fix.
3950 for (w_start, w_end, msg) in warnings_to_report {
3951 warnings.push(LintWarning {
3952 rule_name: Some(self.name().to_string()),
3953 message: msg,
3954 line: w_start,
3955 column: 1,
3956 end_line: w_end,
3957 end_column: lines[w_end.saturating_sub(1)].chars().count() + 1,
3958 severity: Severity::Warning,
3959 fix: Some(crate::rule::Fix::new(byte_range.clone(), replacement.clone())),
3960 });
3961 }
3962 }
3963 }
3964 }
3965
3966 warnings
3967 }
3968
3969 /// Calculate string length based on the configured length mode
3970 fn calculate_string_length(&self, s: &str) -> usize {
3971 match self.config.length_mode {
3972 LengthMode::Chars => s.chars().count(),
3973 LengthMode::Visual => s.width(),
3974 LengthMode::Bytes => s.len(),
3975 }
3976 }
3977
3978 /// Calculate effective line length
3979 ///
3980 /// Returns the actual display length of the line using the configured length mode.
3981 fn calculate_effective_length(&self, line: &str) -> usize {
3982 self.calculate_string_length(line)
3983 }
3984
3985 /// Calculate line length with inline link/image URLs removed.
3986 ///
3987 /// For each inline link `[text](url)` or image `` on the line,
3988 /// computes the "savings" from removing the URL portion (keeping only `[text]`
3989 /// or `![alt]`). Returns `effective_length - total_savings`.
3990 ///
3991 /// Handles nested constructs (e.g., `[](url)`) by only counting the
3992 /// outermost construct to avoid double-counting.
3993 fn length_without_inline_link_urls(
3994 &self,
3995 effective_length: usize,
3996 line_number: usize,
3997 ctx: &crate::lint_context::LintContext,
3998 ) -> usize {
3999 let line_range = ctx.line_content_byte_range(line_number);
4000 let line_byte_end = line_range.end;
4001
4002 // Collect inline links/images on this line: (byte_offset, byte_end, text_only_display_len)
4003 let mut constructs: Vec<(usize, usize, usize)> = Vec::new();
4004
4005 for link in ctx.links_on_line(line_number) {
4006 if link.is_reference {
4007 continue;
4008 }
4009 if !matches!(link.link_type, LinkType::Inline) {
4010 continue;
4011 }
4012 if link.byte_end > line_byte_end {
4013 continue;
4014 }
4015 let text_only_len = 2 + self.calculate_string_length(&link.text);
4016 constructs.push((link.byte_offset, link.byte_end, text_only_len));
4017 }
4018
4019 for image in ctx.images_on_line(line_number) {
4020 if image.is_reference {
4021 continue;
4022 }
4023 if !matches!(image.link_type, LinkType::Inline) {
4024 continue;
4025 }
4026 if image.byte_end > line_byte_end {
4027 continue;
4028 }
4029 let text_only_len = 3 + self.calculate_string_length(&image.alt_text);
4030 constructs.push((image.byte_offset, image.byte_end, text_only_len));
4031 }
4032
4033 if constructs.is_empty() {
4034 return effective_length;
4035 }
4036
4037 // Sort by byte offset to handle overlapping/nested constructs
4038 constructs.sort_by_key(|&(start, _, _)| start);
4039
4040 let mut total_savings: usize = 0;
4041 let mut last_end: usize = 0;
4042
4043 for (start, end, text_only_len) in &constructs {
4044 // Skip constructs nested inside a previously counted one
4045 if *start < last_end {
4046 continue;
4047 }
4048 // Full construct length in configured length mode
4049 let full_source = &ctx.content[*start..*end];
4050 let full_len = self.calculate_string_length(full_source);
4051 total_savings += full_len.saturating_sub(*text_only_len);
4052 last_end = *end;
4053 }
4054
4055 effective_length.saturating_sub(total_savings)
4056 }
4057}