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