rumdl_lib/rules/md013_line_length.rs
1/// Rule MD013: Line length
2///
3/// See [docs/md013.md](../../docs/md013.md) for full documentation, configuration, and examples.
4use crate::rule::{LintError, LintResult, LintWarning, Rule, RuleCategory, Severity};
5use crate::rule_config_serde::RuleConfig;
6use crate::utils::mkdocs_admonitions;
7use crate::utils::mkdocs_attr_list::is_standalone_attr_list;
8use crate::utils::mkdocs_snippets::is_snippet_block_delimiter;
9use crate::utils::mkdocs_tabs;
10use crate::utils::range_utils::LineIndex;
11use crate::utils::range_utils::calculate_excess_range;
12use crate::utils::regex_cache::{IMAGE_REF_PATTERN, LINK_REF_PATTERN, URL_PATTERN};
13use crate::utils::table_utils::TableUtils;
14use crate::utils::text_reflow::{
15 BlockquoteLineData, blockquote_continuation_style, dominant_blockquote_prefix, reflow_blockquote_content,
16 split_into_sentences,
17};
18use pulldown_cmark::LinkType;
19use toml;
20
21mod block_builder;
22mod helpers;
23pub mod md013_config;
24use crate::rules::md030_list_marker_space::MD030Config;
25use crate::utils::is_template_directive_only;
26use block_builder::{Block, BlockBuilder};
27use helpers::{
28 extract_list_marker_and_content, has_hard_break, is_github_alert_marker, is_horizontal_rule, is_html_only_line,
29 is_list_item, is_standalone_link_or_image_line, is_unwrappable_line, source_list_marker, split_into_segments,
30 trim_preserving_hard_break,
31};
32pub use md013_config::MD013Config;
33use md013_config::{LengthMode, ReflowMode};
34
35#[cfg(test)]
36mod tests;
37use unicode_width::UnicodeWidthStr;
38
39#[derive(Clone, Default)]
40pub struct MD013LineLength {
41 pub(crate) config: MD013Config,
42 /// MD030 list-marker spacing, applied when reflowing list items so the rewrite
43 /// uses the configured post-marker spacing rather than a hard-coded single
44 /// space. Defaults to MD030's defaults (a single space everywhere), which
45 /// reproduces the previous behaviour exactly. See [`MD030Config::expected_spaces`].
46 pub(crate) list_spacing: MD030Config,
47}
48
49/// Blockquote paragraph line collected for reflow, with original line index for range computation.
50struct CollectedBlockquoteLine {
51 line_idx: usize,
52 data: BlockquoteLineData,
53}
54
55impl MD013LineLength {
56 pub fn new(line_length: usize, code_blocks: bool, tables: bool, headings: bool, strict: bool) -> Self {
57 Self {
58 config: MD013Config {
59 line_length: crate::types::LineLength::new(line_length),
60 code_blocks,
61 code_spans: true,
62 tables,
63 headings,
64 math_blocks: true,
65 paragraphs: true, // Default to true for backwards compatibility
66 blockquotes: true, // Default to true for backwards compatibility
67 strict,
68 stern: false,
69 heading_line_length: None,
70 code_block_line_length: None,
71 reflow: false,
72 reflow_mode: ReflowMode::default(),
73 length_mode: LengthMode::default(),
74 abbreviations: Vec::new(),
75 require_sentence_capital: true,
76 ignore_link_urls: true,
77 atomic_spans: true,
78 },
79 list_spacing: MD030Config::default(),
80 }
81 }
82
83 pub fn from_config_struct(config: MD013Config) -> Self {
84 Self {
85 config,
86 list_spacing: MD030Config::default(),
87 }
88 }
89
90 /// Return a clone with code block checking disabled.
91 /// Used for doc comment linting where code blocks are Rust code managed by rustfmt.
92 pub fn with_code_blocks_disabled(&self) -> Self {
93 let mut clone = self.clone();
94 clone.config.code_blocks = false;
95 clone
96 }
97
98 /// Convert MD013 LengthMode to text_reflow ReflowLengthMode
99 /// Normalized set of reference labels defined in the document.
100 ///
101 /// Passed to reflow so a bare shortcut reference (`[text]`) is treated as an
102 /// atomic link only when its label is actually defined; an undefined
103 /// bracketed run reflows as literal prose.
104 fn defined_reference_labels(ctx: &crate::lint_context::LintContext) -> std::collections::HashSet<String> {
105 ctx.reference_defs
106 .iter()
107 .map(|d| crate::utils::text_reflow::normalize_reference_label(&d.id))
108 .collect()
109 }
110
111 /// Build the reflow options shared by every MD013 fix path.
112 ///
113 /// `line_length` varies per call site (some subtract a list-marker or
114 /// blockquote prefix); every other field is derived uniformly from the
115 /// effective config and the document flavor. Callers that need a different
116 /// `max_list_continuation_indent` override it via struct update.
117 fn reflow_options(
118 ctx: &crate::lint_context::LintContext,
119 config: &MD013Config,
120 line_length: usize,
121 ) -> crate::utils::text_reflow::ReflowOptions {
122 crate::utils::text_reflow::ReflowOptions {
123 line_length,
124 break_on_sentences: true,
125 preserve_breaks: false,
126 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
127 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
128 abbreviations: config.abbreviations_for_reflow(),
129 length_mode: config.reflow_length_mode(),
130 attr_lists: ctx.flavor.supports_attr_lists(),
131 myst_roles: ctx.flavor.supports_myst_roles(),
132 require_sentence_capital: config.require_sentence_capital,
133 max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
134 Some(4)
135 } else {
136 None
137 },
138 defined_references: Some(Self::defined_reference_labels(ctx)),
139 atomic_spans: config.atomic_spans,
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 // Process candidate lines for line length checks
386 'line_loop: for &line_idx in &candidate_lines {
387 let line_number = line_idx + 1;
388 let line = lines[line_idx];
389
390 // Calculate actual line length (used in warning messages)
391 let effective_length = self.calculate_effective_length(line);
392
393 // Pick the context-specific limit: heading > code-block > paragraph.
394 // Headings dominate over code-block context if a setext underline ever
395 // overlaps a fenced range (defensive — these are mutually exclusive in
396 // practice, but the explicit ordering documents intent).
397 let is_heading_line = is_heading_line_num(line_number);
398 let in_code_block = ctx.line_info(line_number).is_some_and(|info| info.in_code_block);
399 let line_limit = if is_heading_line {
400 effective_config.effective_heading_line_length().get()
401 } else if in_code_block {
402 effective_config.effective_code_block_line_length().get()
403 } else {
404 effective_config.line_length.get()
405 };
406
407 // A context-specific limit of 0 means "unlimited for this context".
408 if line_limit == 0 {
409 continue;
410 }
411
412 // Stern mode: like default, but the trailing-token forgiveness is
413 // disabled — a line with whitespace that exceeds the limit is a
414 // violation even if the excess is the final token. The "unwrappable"
415 // line exemption (single token, optionally prefixed by # or >) is
416 // still honored. Strict overrides stern entirely.
417 if effective_config.stern && !effective_config.strict && is_unwrappable_line(line) {
418 continue;
419 }
420
421 // Trailing-token forgiveness: only in default mode (not strict, not stern).
422 // If the line only exceeds the limit because of a long token at the end
423 // (URL, link chain, identifier), it passes. This matches markdownlint's
424 // behavior: line.replace(/\S*$/u, "#")
425 let check_length = if effective_config.strict || effective_config.stern {
426 effective_length
427 } else {
428 match line.rfind(char::is_whitespace) {
429 Some(pos) => {
430 let ws_char = line[pos..].chars().next().unwrap();
431 let prefix_end = pos + ws_char.len_utf8();
432 self.calculate_string_length(&line[..prefix_end]) + 1
433 }
434 None => 1, // No whitespace — entire line is a single token
435 }
436 };
437
438 // Skip lines where the check length is within the limit
439 if check_length <= line_limit {
440 continue;
441 }
442
443 // Ignore inline link/image URLs: suppress when excess comes entirely from inline URLs.
444 // Disabled by `strict` (all forgiveness off) and by `ignore_link_urls = false`
445 // (count link/image URLs toward the line length so such lines are flagged).
446 if !effective_config.strict && effective_config.ignore_link_urls {
447 let length_without_urls = self.length_without_inline_link_urls(effective_length, line_number, ctx);
448 if length_without_urls <= line_limit {
449 continue;
450 }
451 }
452
453 // Inline code spans cannot be wrapped, so reflow cannot shorten a line
454 // whose excess length is one. When code-span checking is disabled,
455 // suppress a violation that would fit once inline code spans are excluded.
456 if !effective_config.code_spans {
457 let code_span_width: usize = ctx
458 .code_spans()
459 .iter()
460 .filter(|span| span.line == line_number && span.end_line == line_number)
461 .map(|span| self.calculate_string_length(&ctx.content[span.byte_offset..span.byte_end]))
462 .sum();
463 if effective_length.saturating_sub(code_span_width) <= line_limit {
464 continue;
465 }
466 }
467
468 // Skip mkdocstrings and pymdown blocks (already handled by LintContext)
469 if ctx.lines[line_idx].in_mkdocstrings || ctx.lines[line_idx].in_pymdown_block {
470 continue;
471 }
472
473 // Skip MyST comments (% comment) — structural lines, not prose
474 if ctx.lines[line_idx].is_myst_comment {
475 continue;
476 }
477
478 // Link reference definitions are always exempt, even in strict mode.
479 // There's no way to shorten them without breaking the URL.
480 // Also check after stripping list markers, since list items may
481 // contain link ref defs as their content.
482 {
483 let trimmed = line.trim();
484 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
485 continue;
486 }
487 if is_list_item(trimmed) {
488 let (_, content) = extract_list_marker_and_content(trimmed);
489 let content_trimmed = content.trim();
490 if content_trimmed.starts_with('[')
491 && content_trimmed.contains("]:")
492 && LINK_REF_PATTERN.is_match(content_trimmed)
493 {
494 continue;
495 }
496 }
497 }
498
499 // Skip various block types efficiently
500 if !effective_config.strict {
501 // Lines whose only content is a link/image are exempt.
502 // After stripping list markers, blockquote markers, and emphasis,
503 // if only a link or image remains, there is no way to shorten it.
504 if is_standalone_link_or_image_line(line) {
505 continue;
506 }
507
508 // Lines consisting entirely of HTML tags are exempt.
509 // Badge lines, images with attributes, and similar inline HTML
510 // are long due to URLs in attributes and can't be meaningfully shortened.
511 if is_html_only_line(line) {
512 continue;
513 }
514
515 // Skip setext heading underlines
516 if !line.trim().is_empty() && line.trim().chars().all(|c| c == '=' || c == '-') {
517 continue;
518 }
519
520 // Skip block elements according to config flags
521 // The flags mean: true = check these elements, false = skip these elements
522 // So we skip when the flag is FALSE and the line is in that element type
523 if (!effective_config.headings && is_heading_line_num(line_number))
524 || (!effective_config.code_blocks
525 && ctx.line_info(line_number).is_some_and(|info| info.in_code_block))
526 || (!effective_config.tables && table_lines_set.contains(&line_number))
527 || (!effective_config.math_blocks && self.line_is_display_math(line_number, ctx))
528 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
529 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
530 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
531 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
532 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
533 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
534 || ctx.line_info(line_number).is_some_and(|info| info.in_pymdown_block)
535 {
536 continue;
537 }
538
539 // Check if this is a paragraph/regular text line
540 // If paragraphs = false, skip lines that are NOT in special blocks
541 // Blockquote content is treated as paragraph text, so it's not
542 // included in the special blocks list here.
543 if !effective_config.paragraphs {
544 let is_special_block = is_heading_line_num(line_number)
545 || ctx.line_info(line_number).is_some_and(|info| info.in_code_block)
546 || table_lines_set.contains(&line_number)
547 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
548 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
549 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
550 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
551 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
552 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
553 || ctx
554 .line_info(line_number)
555 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container);
556
557 // Skip regular paragraph text when paragraphs = false
558 if !is_special_block {
559 continue;
560 }
561 }
562
563 // Skip blockquote lines when blockquotes = false.
564 // Also skip lazy continuation lines that belong to a blockquote
565 // (lines without `>` prefix that follow a blockquote line).
566 if !effective_config.blockquotes {
567 if ctx.lines[line_number - 1].blockquote.is_some() {
568 continue;
569 }
570 // Check for lazy continuation: scan backwards through
571 // non-blank lines to find if this paragraph started with
572 // a blockquote marker
573 if !line.trim().is_empty() {
574 let mut scan = line_number.saturating_sub(2);
575 loop {
576 if ctx.lines[scan].blockquote.is_some() {
577 // Found a blockquote ancestor — this is a lazy continuation
578 continue 'line_loop;
579 }
580 if lines[scan].trim().is_empty() || scan == 0 {
581 break;
582 }
583 scan -= 1;
584 }
585 }
586 }
587
588 // Skip lines that are only a URL, image ref, or link ref
589 if self.should_ignore_line(line, lines, line_idx, ctx) {
590 continue;
591 }
592 }
593
594 // In sentence-per-line mode, check if this is a single long sentence
595 // If so, emit a warning without a fix (user must manually rephrase)
596 if effective_config.reflow_mode == ReflowMode::SentencePerLine {
597 let sentences = split_into_sentences(line.trim());
598 if sentences.len() == 1 {
599 // Single sentence that's too long - warn but don't auto-fix
600 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
601
602 let (start_line, start_col, end_line, end_col) =
603 calculate_excess_range(line_number, line, line_limit);
604
605 warnings.push(LintWarning {
606 rule_name: Some(self.name().to_string()),
607 message,
608 line: start_line,
609 column: start_col,
610 end_line,
611 end_column: end_col,
612 severity: Severity::Warning,
613 fix: None, // No auto-fix for long single sentences
614 });
615 continue;
616 }
617 // Multiple sentences will be handled by paragraph-based reflow
618 continue;
619 }
620
621 // In semantic-line-breaks mode, skip per-line checks —
622 // all reflow is handled at the paragraph level with cascading splits
623 if effective_config.reflow_mode == ReflowMode::SemanticLineBreaks {
624 continue;
625 }
626
627 // Don't provide fix for individual lines when reflow is enabled
628 // Paragraph-based fixes will be handled separately
629 let fix = None;
630
631 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
632
633 // Calculate precise character range for the excess portion
634 let (start_line, start_col, end_line, end_col) = calculate_excess_range(line_number, line, line_limit);
635
636 warnings.push(LintWarning {
637 rule_name: Some(self.name().to_string()),
638 message,
639 line: start_line,
640 column: start_col,
641 end_line,
642 end_column: end_col,
643 severity: Severity::Warning,
644 fix,
645 });
646 }
647
648 // If reflow is enabled, generate paragraph-based fixes
649 if effective_config.reflow {
650 let paragraph_warnings = self.generate_paragraph_fixes(ctx, &effective_config, lines);
651 // Merge paragraph warnings with line warnings, removing duplicates
652 for pw in paragraph_warnings {
653 // Remove any line warnings that overlap with this paragraph
654 warnings.retain(|w| w.line < pw.line || w.line > pw.end_line);
655 warnings.push(pw);
656 }
657 }
658
659 Ok(warnings)
660 }
661
662 fn fix(&self, ctx: &crate::lint_context::LintContext) -> Result<String, LintError> {
663 // For CLI usage, apply fixes from warnings
664 // LSP will use the warning-based fixes directly
665 let warnings = self.check(ctx)?;
666 let warnings =
667 crate::utils::fix_utils::filter_warnings_by_inline_config(warnings, ctx.inline_config(), self.name());
668
669 // If there are no fixes, return content unchanged
670 if !warnings.iter().any(|w| w.fix.is_some()) {
671 return Ok(ctx.content.to_string());
672 }
673
674 // Apply warning-based fixes
675 crate::utils::fix_utils::apply_warning_fixes(ctx.content, &warnings)
676 .map_err(|e| LintError::FixFailed(format!("Failed to apply fixes: {e}")))
677 }
678
679 fn as_any(&self) -> &dyn std::any::Any {
680 self
681 }
682
683 fn category(&self) -> RuleCategory {
684 RuleCategory::Whitespace
685 }
686
687 fn should_skip(&self, ctx: &crate::lint_context::LintContext) -> bool {
688 self.should_skip_with_config(ctx, &self.config)
689 }
690
691 fn default_config_section(&self) -> Option<(String, toml::Value)> {
692 let table = crate::rule_config_serde::config_schema_table(&MD013Config::default())?;
693 if table.is_empty() {
694 None
695 } else {
696 Some((MD013Config::RULE_NAME.to_string(), toml::Value::Table(table)))
697 }
698 }
699
700 fn config_aliases(&self) -> Option<std::collections::HashMap<String, String>> {
701 let mut aliases = std::collections::HashMap::new();
702 aliases.insert("enable_reflow".to_string(), "reflow".to_string());
703 aliases.insert("strict_sentences".to_string(), "require-sentence-capital".to_string());
704 aliases.insert("strict-sentences".to_string(), "require-sentence-capital".to_string());
705 // Kept in step with the `alias` attributes on `MD013Config::ignore_link_urls`.
706 // Serde accepts these spellings, so a config using one is honored; without
707 // them here the key validator reports a documented, working key as unknown.
708 aliases.insert(
709 "semantic-link-understanding".to_string(),
710 "ignore-link-urls".to_string(),
711 );
712 aliases.insert(
713 "semantic_link_understanding".to_string(),
714 "ignore-link-urls".to_string(),
715 );
716 Some(aliases)
717 }
718
719 fn from_config(config: &crate::config::Config) -> Box<dyn Rule>
720 where
721 Self: Sized,
722 {
723 let mut rule_config = crate::rule_config_serde::load_rule_config::<MD013Config>(config);
724 // Use global line_length if rule-specific config still has default value
725 if rule_config.line_length.get() == 80 {
726 rule_config.line_length = config.global.line_length;
727 }
728 let mut rule = Self::from_config_struct(rule_config);
729 // Pull list-marker spacing from MD030 (via the shared serde config loader)
730 // so reflow rewrites list items with the configured spacing rather than a
731 // hard-coded single space.
732 rule.list_spacing = crate::rule_config_serde::load_rule_config::<MD030Config>(config);
733 Box::new(rule)
734 }
735}
736
737impl MD013LineLength {
738 /// True when `line_num` (1-indexed) sits inside a `$$` span covering more
739 /// than one line, as seen by the byte-level math parser.
740 fn line_in_multiline_math_span(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
741 ctx.math_spans()
742 .iter()
743 .any(|span| span.is_display && span.end_line > span.line && (span.line..=span.end_line).contains(&line_num))
744 }
745
746 /// True when `line_num` (1-indexed) holds nothing but display math, whether
747 /// that is one line of a multi-line block, a delimiter line, or a whole line
748 /// that is a single complete `$$...$$` span. This is what `math-blocks =
749 /// false` exempts from the length check.
750 ///
751 /// rumdl models math twice and the two models miss different containers, so
752 /// this consults both. `math_spans()` is byte-level and sees a block opened on
753 /// a list marker line (`- $$`), which the line-level map cannot because that
754 /// line does not begin with `$$`. `LineInfo::in_math_block` is line-level and
755 /// sees a four-space-indented block inside a footnote, which the byte-level
756 /// parser reads as an indented code block. Taking the union only ever adds
757 /// coverage: neither signal fires on ordinary prose, and an unmatched `$$`
758 /// opener is flagged by neither, so a stray delimiter cannot exempt the rest
759 /// of the document.
760 fn line_is_display_math(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
761 self.line_holds_only_multiline_math(line_num, ctx)
762 || ctx.line_info(line_num).is_some_and(|info| info.in_math_block)
763 }
764
765 /// True when a multi-line `$$` span covers `line_num` (1-indexed) and the line
766 /// holds nothing besides that math.
767 ///
768 /// A delimiter line can carry Markdown outside the delimiter: `$$ trailing
769 /// prose` closes a block and then continues in prose, and `leading prose $$`
770 /// opens one at the end of a sentence. That prose is ordinary text and counts
771 /// toward the line's length like any other, so exempting the whole line would
772 /// hide arbitrarily long prose behind a delimiter. The line-level map already
773 /// leaves such mixed lines unflagged; this is the byte-level half of the same
774 /// judgement.
775 fn line_holds_only_multiline_math(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
776 let Some(info) = ctx.line_info(line_num) else {
777 return false;
778 };
779 let line = info.content(ctx.content);
780
781 ctx.math_spans().iter().any(|span| {
782 if !span.is_display || span.end_line <= span.line || !(span.line..=span.end_line).contains(&line_num) {
783 return false;
784 }
785 if line_num == span.line {
786 let before = line.get(..span.byte_offset.saturating_sub(info.byte_offset));
787 if !before.is_none_or(|before| Self::only_structure_precedes_math(before, info)) {
788 return false;
789 }
790 }
791 if line_num == span.end_line {
792 let after = line.get(span.byte_end.saturating_sub(info.byte_offset)..);
793 if !after.is_none_or(|after| after.trim().is_empty()) {
794 return false;
795 }
796 }
797 true
798 })
799 }
800
801 /// True when the text before a block's opening delimiter is only the structure
802 /// the block sits in: indentation, a blockquote marker, or the list marker
803 /// introducing it. Such a block still owns its whole line.
804 fn only_structure_precedes_math(before: &str, info: &crate::lint_context::LineInfo) -> bool {
805 let after_marker =
806 crate::utils::blockquote::parse_blockquote_prefix(before).map_or(before, |prefix| prefix.content);
807 if after_marker.trim().is_empty() {
808 return true;
809 }
810 info.list_item.as_ref().is_some_and(|item| {
811 let mut chars = before.chars();
812 chars.by_ref().take(item.content_column).count() == item.content_column && chars.as_str().trim().is_empty()
813 })
814 }
815
816 /// True when `line_num` (1-indexed) falls inside a display-math block that
817 /// spans more than one line.
818 ///
819 /// Line breaks carry meaning inside such a block: a TeX `%` comment runs to
820 /// the end of its line, so joining the lines pulls whatever followed on later
821 /// lines into the comment and drops it from the rendered equation, which can
822 /// also leave an environment unclosed. Reflow therefore leaves these blocks
823 /// alone regardless of the `math_blocks` setting, which governs only whether
824 /// their length is reported.
825 ///
826 /// This is `line_is_display_math` minus the case of a whole line that is one
827 /// complete `$$...$$` span: such a line is a single atomic element reflow can
828 /// move around freely, and only a multi-line block has meaningful internal
829 /// line breaks.
830 fn line_in_multiline_math_block(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
831 self.line_in_multiline_math_span(line_num, ctx)
832 || ctx.line_info(line_num).is_some_and(|info| {
833 info.in_math_block && !Self::is_self_contained_display_math_line(info.content(ctx.content))
834 })
835 }
836
837 /// True when `line` is a whole line holding exactly one closed `$$...$$` span.
838 fn is_self_contained_display_math_line(line: &str) -> bool {
839 let trimmed = line.trim();
840 let inner = crate::utils::blockquote::parse_blockquote_prefix(trimmed).map_or(trimmed, |p| p.content.trim());
841 inner.strip_prefix("$$").is_some_and(|rest| rest.contains("$$"))
842 }
843
844 /// True when `line_num` (1-based) sits inside a structure whose lines must be
845 /// preserved verbatim (code block, front matter, HTML/JSX/MDX block, MkDocs
846 /// container, div marker, multi-line math block, ...). Used to keep blockquote
847 /// reflow from touching quoted-looking text embedded in such structures.
848 fn line_in_verbatim_context(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
849 if self.line_in_multiline_math_block(line_num, ctx) {
850 return true;
851 }
852 ctx.line_info(line_num).is_some_and(|info| {
853 info.in_code_block
854 || info.in_front_matter
855 || info.in_html_block
856 || info.in_html_comment
857 || info.in_esm_block
858 || info.in_jsx_expression
859 || info.in_jsx_block
860 || info.in_mdx_comment
861 || info.in_mkdocstrings
862 || info.in_pymdown_block
863 || info.in_mkdocs_container()
864 || info.is_div_marker
865 })
866 }
867
868 fn is_blockquote_content_boundary(
869 &self,
870 content: &str,
871 line_num: usize,
872 ctx: &crate::lint_context::LintContext,
873 strict: bool,
874 ) -> bool {
875 let trimmed = content.trim();
876
877 trimmed.is_empty()
878 || self.line_in_verbatim_context(line_num, ctx)
879 || trimmed.starts_with('#')
880 || trimmed.starts_with("```")
881 || trimmed.starts_with("~~~")
882 || trimmed.starts_with('>')
883 || TableUtils::is_potential_table_row(content)
884 || is_list_item(trimmed)
885 || is_horizontal_rule(content)
886 || (trimmed.starts_with('[') && content.contains("]:"))
887 || is_template_directive_only(content)
888 || is_standalone_attr_list(content)
889 || is_snippet_block_delimiter(content)
890 || is_github_alert_marker(trimmed)
891 || is_html_only_line(content)
892 // A standalone link/image line is exempt from MD013 (non-strict mode),
893 // so it must end the blockquote paragraph rather than be absorbed into
894 // it, mirroring the top-level paragraph reflow boundary.
895 || (!strict && is_standalone_link_or_image_line(content))
896 }
897
898 #[allow(clippy::too_many_arguments)]
899 fn generate_blockquote_paragraph_fix(
900 &self,
901 ctx: &crate::lint_context::LintContext,
902 config: &MD013Config,
903 lines: &[&str],
904 line_index: &LineIndex,
905 start_idx: usize,
906 line_ending: &str,
907 // Extra indent (spaces) to prepend to the emitted `>` prefix so a blockquote
908 // nested in a list item moves with its parent's widened marker. Zero unless a
909 // non-default MD030 widened an ancestor list item.
910 ancestor_shift: isize,
911 ) -> (Option<LintWarning>, usize) {
912 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
913 return (None, start_idx + 1);
914 };
915 let target_level = start_bq.nesting_level;
916
917 let mut collected: Vec<CollectedBlockquoteLine> = Vec::new();
918 let mut i = start_idx;
919
920 while i < lines.len() {
921 if !collected.is_empty() && has_hard_break(&collected[collected.len() - 1].data.content) {
922 break;
923 }
924
925 let line_num = i + 1;
926 if line_num > ctx.lines.len() {
927 break;
928 }
929
930 if lines[i].trim().is_empty() {
931 break;
932 }
933
934 let line_bq = ctx.lines[i].blockquote.as_deref();
935 if let Some(bq) = line_bq {
936 if bq.nesting_level != target_level {
937 break;
938 }
939
940 if self.is_blockquote_content_boundary(&bq.content, line_num, ctx, config.strict) {
941 break;
942 }
943
944 collected.push(CollectedBlockquoteLine {
945 line_idx: i,
946 data: BlockquoteLineData::explicit(trim_preserving_hard_break(&bq.content), bq.prefix.clone()),
947 });
948 i += 1;
949 continue;
950 }
951
952 let lazy_content = lines[i].trim_start();
953 if self.is_blockquote_content_boundary(lazy_content, line_num, ctx, config.strict) {
954 break;
955 }
956
957 collected.push(CollectedBlockquoteLine {
958 line_idx: i,
959 data: BlockquoteLineData::lazy(trim_preserving_hard_break(lazy_content)),
960 });
961 i += 1;
962 }
963
964 if collected.is_empty() {
965 return (None, start_idx + 1);
966 }
967
968 let next_idx = i;
969 let paragraph_start = collected[0].line_idx;
970 let end_line = collected[collected.len() - 1].line_idx;
971 let line_data: Vec<BlockquoteLineData> = collected.iter().map(|l| l.data.clone()).collect();
972 let paragraph_text = line_data
973 .iter()
974 .map(|d| d.content.as_str())
975 .collect::<Vec<_>>()
976 .join(" ");
977
978 let contains_definition_list = line_data
979 .iter()
980 .any(|d| crate::utils::is_definition_list_item(&d.content));
981 if contains_definition_list {
982 return (None, next_idx);
983 }
984
985 let contains_snippets = line_data.iter().any(|d| is_snippet_block_delimiter(&d.content));
986 if contains_snippets {
987 return (None, next_idx);
988 }
989
990 let needs_reflow = match config.reflow_mode {
991 ReflowMode::Normalize => {
992 self.normalize_mode_needs_reflow(line_data.iter().map(|d| d.content.as_str()), config)
993 }
994 ReflowMode::SentencePerLine => {
995 let sentences = split_into_sentences(¶graph_text);
996 sentences.len() > 1 || line_data.len() > 1
997 }
998 ReflowMode::SemanticLineBreaks => {
999 let sentences = split_into_sentences(¶graph_text);
1000 sentences.len() > 1
1001 || line_data.len() > 1
1002 || collected
1003 .iter()
1004 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get())
1005 }
1006 ReflowMode::Default => collected
1007 .iter()
1008 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get()),
1009 };
1010
1011 if !needs_reflow {
1012 return (None, next_idx);
1013 }
1014
1015 let fallback_prefix = start_bq.prefix.clone();
1016 let explicit_prefix = dominant_blockquote_prefix(&line_data, &fallback_prefix);
1017 // Shift the whole quote right to track a widened parent list item's content
1018 // column (only widening matters for nesting; a narrowed parent leaves the quote
1019 // harmlessly over-indented, which MD027/MD030 tidy).
1020 let explicit_prefix = if ancestor_shift > 0 {
1021 format!("{}{explicit_prefix}", " ".repeat(ancestor_shift as usize))
1022 } else {
1023 explicit_prefix
1024 };
1025 let continuation_style = blockquote_continuation_style(&line_data);
1026
1027 let reflow_line_length = if config.line_length.is_unlimited() {
1028 usize::MAX
1029 } else {
1030 config
1031 .line_length
1032 .get()
1033 .saturating_sub(self.calculate_string_length(&explicit_prefix))
1034 .max(1)
1035 };
1036
1037 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
1038
1039 let reflowed_with_style =
1040 reflow_blockquote_content(&line_data, &explicit_prefix, continuation_style, &reflow_options);
1041
1042 if reflowed_with_style.is_empty() {
1043 return (None, next_idx);
1044 }
1045
1046 let reflowed_text = reflowed_with_style.join(line_ending);
1047
1048 let start_range = line_index.whole_line_range(paragraph_start + 1);
1049 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
1050 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
1051 } else {
1052 line_index.whole_line_range(end_line + 1)
1053 };
1054 let byte_range = start_range.start..end_range.end;
1055
1056 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
1057 format!("{reflowed_text}{line_ending}")
1058 } else {
1059 reflowed_text
1060 };
1061
1062 let original_text = &ctx.content[byte_range.clone()];
1063 if original_text == replacement {
1064 return (None, next_idx);
1065 }
1066
1067 let (warning_line, warning_end_line) = match config.reflow_mode {
1068 ReflowMode::Normalize => (paragraph_start + 1, end_line + 1),
1069 ReflowMode::SentencePerLine | ReflowMode::SemanticLineBreaks => (paragraph_start + 1, end_line + 1),
1070 ReflowMode::Default => {
1071 let violating_line = collected
1072 .iter()
1073 .find(|line| self.calculate_effective_length(lines[line.line_idx]) > config.line_length.get())
1074 .map_or(paragraph_start + 1, |line| line.line_idx + 1);
1075 (violating_line, violating_line)
1076 }
1077 };
1078
1079 let warning = LintWarning {
1080 rule_name: Some(self.name().to_string()),
1081 message: match config.reflow_mode {
1082 ReflowMode::Normalize => format!(
1083 "Paragraph could be normalized to use line length of {} characters",
1084 config.line_length.get()
1085 ),
1086 ReflowMode::SentencePerLine => {
1087 let num_sentences = split_into_sentences(¶graph_text).len();
1088 if line_data.len() == 1 {
1089 format!("Line contains {num_sentences} sentences (one sentence per line required)")
1090 } else {
1091 let num_lines = line_data.len();
1092 format!(
1093 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
1094 )
1095 }
1096 }
1097 ReflowMode::SemanticLineBreaks => {
1098 let num_sentences = split_into_sentences(¶graph_text).len();
1099 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
1100 }
1101 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
1102 },
1103 line: warning_line,
1104 column: 1,
1105 end_line: warning_end_line,
1106 end_column: lines[warning_end_line.saturating_sub(1)].chars().count() + 1,
1107 severity: Severity::Warning,
1108 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1109 };
1110
1111 (Some(warning), next_idx)
1112 }
1113
1114 /// Reflow a single list item that lives inside a blockquote.
1115 ///
1116 /// The blockquote paragraph reflow treats a list marker as a content boundary,
1117 /// so `> - long item ...` is never wrapped even though the identical top-level
1118 /// item is. This handles that case: it collects one tight, prose-only list item
1119 /// at the starting blockquote level, reflows the item body to the configured
1120 /// width, and re-emits it with the blockquote prefix preserved and continuation
1121 /// lines aligned under the list content.
1122 ///
1123 /// Sibling and nested list items end collection and are reflowed independently
1124 /// by the caller's outer loop (a nested item carries its indent folded into the
1125 /// blockquote prefix, so it reflows correctly on its own). Items that are not
1126 /// simple tight prose - those embedding a code block, table, fence, or hard
1127 /// break - are left untouched (`None`) and the cursor still advances past the
1128 /// whole item so its inner lines are never reprocessed as loose prose.
1129 #[allow(clippy::too_many_arguments)]
1130 fn generate_blockquote_list_item_fix(
1131 &self,
1132 ctx: &crate::lint_context::LintContext,
1133 config: &MD013Config,
1134 lines: &[&str],
1135 line_index: &LineIndex,
1136 start_idx: usize,
1137 line_ending: &str,
1138 // Extra indent (spaces) to prepend to the emitted `>` prefix when this quoted
1139 // list lives inside a list item whose marker widened. Zero unless a non-default
1140 // MD030 widened an ancestor list item.
1141 ancestor_shift: isize,
1142 ) -> (Option<LintWarning>, usize) {
1143 use crate::utils::blockquote::effective_indent_in_blockquote;
1144
1145 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
1146 return (None, start_idx + 1);
1147 };
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(raw_marker_line))
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 => split_into_sentences(body_text).len() > 1 || body_pieces.len() > 1,
1281 ReflowMode::SemanticLineBreaks => split_into_sentences(body_text).len() > 1 || exceeds_limit(),
1282 };
1283 if !needs_reflow {
1284 return (None, next_idx);
1285 }
1286
1287 // Apply MD030 list-marker spacing in the spacing-normalizing modes, mirroring
1288 // the top-level list reflow: derive the post-marker spacing from MD030 and let
1289 // the continuation indent follow the resulting content width. Default MD030 (a
1290 // single space) leaves the marker unchanged. MkDocs keeps its rigid indent.
1291 let (marker, marker_width) = if matches!(config.reflow_mode, ReflowMode::Default | ReflowMode::Normalize)
1292 && !ctx.flavor.requires_strict_list_indent()
1293 {
1294 let is_ordered = marker.starts_with(|c: char| c.is_ascii_digit());
1295 // Bullet/number portion only (e.g. `-`, `1.`); the checkbox is content.
1296 let bullet = marker.split(' ').next().unwrap_or("");
1297 let bullet_len = bullet.chars().count();
1298 let checkbox_tail = &marker[base_marker_width..];
1299 // Decide single- vs multi-line spacing from the rewritten shape. This path
1300 // only handles a single tight prose paragraph (structural or multi-paragraph
1301 // items set `simple = false` and bail out above), so the emitted item stays
1302 // multi-line solely when the joined body wraps past one line. A multi-line
1303 // *source* that collapses onto the marker line must use MD030's single-line
1304 // spacing, matching the top-level reflow. The wrap test measures at the
1305 // single-line content column so it does not depend on the spacing chosen here.
1306 let single_col = self.calculate_string_length(&bq_prefix)
1307 + bullet_len
1308 + self.list_spacing.expected_spaces(is_ordered, false, bullet_len)
1309 + checkbox_tail.chars().count();
1310 let is_multi = !body_text.is_empty()
1311 && self.calculate_effective_length(&format!("{}{body_text}", " ".repeat(single_col)))
1312 > config.line_length.effective_limit();
1313 let spaces = self.list_spacing.expected_spaces(is_ordered, is_multi, bullet_len);
1314 let new_marker = format!("{bullet}{}{checkbox_tail}", " ".repeat(spaces));
1315 let width = new_marker.chars().count();
1316 (new_marker, width)
1317 } else {
1318 (marker, marker_width)
1319 };
1320
1321 let prefix_width = self.calculate_string_length(&bq_prefix) + self.calculate_string_length(&marker);
1322 let reflow_line_length = if config.line_length.is_unlimited() {
1323 usize::MAX
1324 } else {
1325 config.line_length.get().saturating_sub(prefix_width).max(1)
1326 };
1327
1328 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
1329
1330 let reflowed = crate::utils::text_reflow::reflow_line(body_text, &reflow_options);
1331 if reflowed.is_empty() {
1332 return (None, next_idx);
1333 }
1334
1335 let continuation_indent = " ".repeat(marker_width);
1336 let reflowed_text = reflowed
1337 .iter()
1338 .enumerate()
1339 .map(|(idx, line)| {
1340 if idx == 0 {
1341 format!("{bq_prefix}{marker}{line}")
1342 } else {
1343 format!("{bq_prefix}{continuation_indent}{line}")
1344 }
1345 })
1346 .collect::<Vec<_>>()
1347 .join(line_ending);
1348
1349 let start_range = line_index.whole_line_range(start_idx + 1);
1350 let end_range = if end_idx == lines.len() - 1 && !ctx.content.ends_with('\n') {
1351 line_index.line_text_range(end_idx + 1, 1, lines[end_idx].len() + 1)
1352 } else {
1353 line_index.whole_line_range(end_idx + 1)
1354 };
1355 let byte_range = start_range.start..end_range.end;
1356
1357 let replacement = if end_idx < lines.len() - 1 || ctx.content.ends_with('\n') {
1358 format!("{reflowed_text}{line_ending}")
1359 } else {
1360 reflowed_text
1361 };
1362
1363 let original_text = &ctx.content[byte_range.clone()];
1364 if original_text == replacement {
1365 return (None, next_idx);
1366 }
1367
1368 let message = match config.reflow_mode {
1369 ReflowMode::Normalize => format!(
1370 "Paragraph could be normalized to use line length of {} characters",
1371 config.line_length.get()
1372 ),
1373 ReflowMode::SentencePerLine => {
1374 let num_sentences = split_into_sentences(body_text).len();
1375 format!("List item should have one sentence per line (found {num_sentences} sentences)")
1376 }
1377 ReflowMode::SemanticLineBreaks => {
1378 let num_sentences = split_into_sentences(body_text).len();
1379 format!("List item should use semantic line breaks ({num_sentences} sentences)")
1380 }
1381 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
1382 };
1383
1384 let warning = LintWarning {
1385 rule_name: Some(self.name().to_string()),
1386 message,
1387 line: start_idx + 1,
1388 column: 1,
1389 end_line: end_idx + 1,
1390 end_column: lines[end_idx].chars().count() + 1,
1391 severity: Severity::Warning,
1392 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1393 };
1394
1395 (Some(warning), next_idx)
1396 }
1397
1398 /// Generate paragraph-based fixes
1399 fn generate_paragraph_fixes(
1400 &self,
1401 ctx: &crate::lint_context::LintContext,
1402 config: &MD013Config,
1403 lines: &[&str],
1404 ) -> Vec<LintWarning> {
1405 let mut warnings = Vec::new();
1406 let line_index = LineIndex::new(ctx.content);
1407
1408 // Detect the content's line ending style to preserve it in replacements.
1409 // The LSP receives content from editors which may use CRLF (Windows).
1410 // Replacements must match the original line endings to avoid false positives.
1411 let line_ending = crate::utils::line_ending::detect_line_ending(ctx.content);
1412
1413 // Ancestor list-item indent shifts, innermost last. When a reflowed parent's
1414 // marker widens under a non-default MD030 (e.g. ul-multi = 3 moves the parent's
1415 // content from column 2 to 4), its nested list/blockquote children are reflowed
1416 // independently and would otherwise keep their original indent — leaving them
1417 // under the parent's new content column, where a CommonMark parser reparses them
1418 // as siblings rather than children. Each frame is (normalized marker width,
1419 // cumulative shift applied to that item's content column); a descendant adds its
1420 // innermost open ancestor's shift to its own indent so the whole subtree moves
1421 // together. With a default MD030 and no marker padding every shift is 0, so this
1422 // is inert and the output is byte-identical.
1423 let mut list_shift_stack: Vec<(usize, isize)> = Vec::new();
1424
1425 let mut i = 0;
1426 while i < lines.len() {
1427 let line_num = i + 1;
1428
1429 // Close ancestor frames whose list item has ended at this line: a non-blank
1430 // line indented less than the frame's source content column is no longer
1431 // inside that item. Blank lines alone don't close a (loose) list item.
1432 if !list_shift_stack.is_empty()
1433 && let Some(info) = ctx.lines.get(i)
1434 && !info.is_blank
1435 {
1436 while let Some(&(content_column, _)) = list_shift_stack.last() {
1437 if info.indent < content_column {
1438 list_shift_stack.pop();
1439 } else {
1440 break;
1441 }
1442 }
1443 }
1444
1445 // Handle blockquote paragraphs with style-preserving reflow.
1446 // Skip blockquotes when blockquotes=false or paragraphs=false
1447 if line_num > 0 && line_num <= ctx.lines.len() && ctx.lines[line_num - 1].blockquote.is_some() {
1448 if !config.blockquotes || !config.paragraphs {
1449 // Skip past all blockquote lines (explicit and lazy continuations).
1450 // A lazy continuation is a non-blank line without `>` that follows
1451 // a blockquote line and isn't a structural element.
1452 let mut saw_explicit_bq = false;
1453 while i < lines.len() && i < ctx.lines.len() {
1454 if ctx.lines[i].blockquote.is_some() {
1455 saw_explicit_bq = true;
1456 i += 1;
1457 } else if saw_explicit_bq
1458 && !lines[i].trim().is_empty()
1459 && !lines[i].trim_start().starts_with('#')
1460 && !lines[i].trim_start().starts_with('>')
1461 {
1462 // Lazy continuation of preceding blockquote
1463 i += 1;
1464 } else {
1465 break;
1466 }
1467 }
1468 continue;
1469 }
1470 // A blockquote nested in a list item moves with its parent when the
1471 // parent's marker widens (see `list_shift_stack`); pass that shift so the
1472 // emitted `>` prefix lands under the parent's new content column instead
1473 // of detaching into a sibling.
1474 let ancestor_shift = list_shift_stack.last().map_or(0isize, |&(_, shift)| shift);
1475 // A list item inside the blockquote needs list-aware reflow (marker +
1476 // continuation indent); plain prose goes through the paragraph path.
1477 let is_bq_list_item = ctx.lines[i]
1478 .blockquote
1479 .as_deref()
1480 .is_some_and(|bq| is_list_item(&bq.content));
1481 let (warning, next_idx) = if is_bq_list_item {
1482 self.generate_blockquote_list_item_fix(
1483 ctx,
1484 config,
1485 lines,
1486 &line_index,
1487 i,
1488 line_ending,
1489 ancestor_shift,
1490 )
1491 } else {
1492 self.generate_blockquote_paragraph_fix(
1493 ctx,
1494 config,
1495 lines,
1496 &line_index,
1497 i,
1498 line_ending,
1499 ancestor_shift,
1500 )
1501 };
1502 if let Some(warning) = warning {
1503 warnings.push(warning);
1504 }
1505 i = next_idx;
1506 continue;
1507 }
1508
1509 // Skip special structures (but NOT MkDocs containers - those get special handling)
1510 let should_skip_due_to_line_info = ctx.line_info(line_num).is_some_and(|info| {
1511 info.in_code_block
1512 || info.in_front_matter
1513 || info.in_html_block
1514 || info.in_html_comment
1515 || info.in_esm_block
1516 || info.in_jsx_expression
1517 || info.in_jsx_block
1518 || info.in_mdx_comment
1519 || info.in_mkdocstrings
1520 || info.in_pymdown_block
1521 });
1522
1523 // Skip link reference definitions but NOT footnote definitions.
1524 // Footnote definitions (`[^id]: prose`) contain reflowable text,
1525 // while link reference definitions (`[ref]: URL`) contain URLs
1526 // that cannot be shortened.
1527 let is_link_ref_def =
1528 lines[i].trim().starts_with('[') && !lines[i].trim().starts_with("[^") && lines[i].contains("]:");
1529
1530 if should_skip_due_to_line_info
1531 || lines[i].trim().starts_with('#')
1532 || TableUtils::is_potential_table_row(lines[i])
1533 || lines[i].trim().is_empty()
1534 || is_horizontal_rule(lines[i])
1535 || is_template_directive_only(lines[i])
1536 || is_link_ref_def
1537 || ctx.line_info(line_num).is_some_and(|info| info.is_div_marker)
1538 || is_html_only_line(lines[i])
1539 || (!config.strict && is_standalone_link_or_image_line(lines[i]))
1540 {
1541 i += 1;
1542 continue;
1543 }
1544
1545 // Handle footnote definitions: `[^id]: prose text that can be reflowed`
1546 // Supports multi-paragraph footnotes with code blocks, blockquotes,
1547 // tables, and lists preserved verbatim.
1548 // Validate structure: must start with `[^`, contain `]:`, and the ID
1549 // must not contain `[` or `]` (prevents false matches on nested brackets)
1550 if lines[i].trim().starts_with("[^") && lines[i].contains("]:") && {
1551 let after_caret = &lines[i].trim()[2..];
1552 after_caret
1553 .find("]:")
1554 .is_some_and(|pos| pos > 0 && !after_caret[..pos].contains(['[', ']']))
1555 } {
1556 let footnote_start = i;
1557 let line = lines[i];
1558
1559 // Extract the prefix `[^id]:`
1560 let Some(colon_pos) = line.find("]:") else {
1561 i += 1;
1562 continue;
1563 };
1564 let prefix_end = colon_pos + 2;
1565 let prefix = &line[..prefix_end];
1566
1567 // Content starts after `]: ` (with optional space)
1568 let content_start = if line[prefix_end..].starts_with(' ') {
1569 prefix_end + 1
1570 } else {
1571 prefix_end
1572 };
1573 let first_content = &line[content_start..];
1574
1575 // CommonMark footnotes use 4-space continuation indent
1576 const FN_INDENT: usize = 4;
1577
1578 // --- Line classification for footnote content ---
1579 #[derive(Debug, Clone)]
1580 enum FnLineType {
1581 Content(String),
1582 Verbatim(String, usize), // preserved text, original indent
1583 Empty,
1584 }
1585
1586 // Helper: compute visual indent (tabs = 4 spaces)
1587 let visual_indent = |s: &str| -> usize {
1588 s.chars()
1589 .take_while(|c| c.is_whitespace())
1590 .map(|c| if c == '\t' { 4 } else { 1 })
1591 .sum::<usize>()
1592 };
1593
1594 // Helper: check if a trimmed line is a fence marker (homogeneous chars)
1595 let is_fence = |s: &str| -> bool {
1596 let t = s.trim();
1597 let fence_char = t.chars().next();
1598 matches!(fence_char, Some('`') | Some('~'))
1599 && t.chars().take_while(|&c| c == fence_char.unwrap()).count() >= 3
1600 };
1601
1602 // Helper: check if a trimmed line is a setext underline
1603 let is_setext_underline = |s: &str| -> bool {
1604 let t = s.trim();
1605 !t.is_empty()
1606 && (t.chars().all(|c| c == '=' || c == ' ') || t.chars().all(|c| c == '-' || c == ' '))
1607 && t.contains(['=', '-'])
1608 };
1609
1610 // Deferred body: `[^id]:\n content` — first line has no content,
1611 // actual content starts on the next indented line
1612 let deferred_body = first_content.trim().is_empty();
1613
1614 // Collect all lines belonging to this footnote definition
1615 let mut fn_lines: Vec<FnLineType> = Vec::new();
1616 if !deferred_body {
1617 fn_lines.push(FnLineType::Content(first_content.to_string()));
1618 }
1619 let mut last_consumed = i;
1620 i += 1;
1621
1622 // Strip only the footnote continuation indent, preserving
1623 // internal indentation (e.g., code block body indent)
1624 let strip_fn_indent = |s: &str| -> String {
1625 let mut chars = s.chars();
1626 let mut stripped = 0;
1627 while stripped < FN_INDENT {
1628 match chars.next() {
1629 Some('\t') => stripped += 4,
1630 Some(c) if c.is_whitespace() => stripped += 1,
1631 _ => break,
1632 }
1633 }
1634 chars.as_str().to_string()
1635 };
1636
1637 let mut in_fenced_code = false;
1638 let mut consecutive_blanks = 0u32;
1639
1640 while i < lines.len() {
1641 let next = lines[i];
1642 let next_trimmed = next.trim();
1643
1644 // Blank line handling
1645 if next_trimmed.is_empty() {
1646 consecutive_blanks += 1;
1647 // 2+ consecutive blanks terminate the footnote
1648 if consecutive_blanks >= 2 {
1649 break;
1650 }
1651
1652 // Inside a fenced code block, blank lines are part of the code
1653 if in_fenced_code {
1654 consecutive_blanks = 0; // Don't count blanks inside code blocks
1655 fn_lines.push(FnLineType::Verbatim(String::new(), 0));
1656 last_consumed = i;
1657 i += 1;
1658 continue;
1659 }
1660
1661 // Peek ahead: if next non-blank line is indented >= FN_INDENT,
1662 // this blank is an internal paragraph separator
1663 if i + 1 < lines.len() {
1664 let peek = lines[i + 1];
1665 let peek_indent = visual_indent(peek);
1666 if !peek.trim().is_empty() && peek_indent >= FN_INDENT {
1667 fn_lines.push(FnLineType::Empty);
1668 last_consumed = i;
1669 i += 1;
1670 continue;
1671 }
1672 }
1673 // No valid continuation after blank — end of footnote
1674 break;
1675 }
1676
1677 consecutive_blanks = 0;
1678 let indent = visual_indent(next);
1679
1680 // Not indented enough — end of footnote
1681 if indent < FN_INDENT {
1682 break;
1683 }
1684
1685 // Inside a fenced code block: everything is verbatim until closing fence
1686 if in_fenced_code {
1687 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1688 if is_fence(next_trimmed) {
1689 in_fenced_code = false;
1690 }
1691 last_consumed = i;
1692 i += 1;
1693 continue;
1694 }
1695
1696 // Fence opener — start verbatim code block
1697 if is_fence(next_trimmed) {
1698 in_fenced_code = true;
1699 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1700 last_consumed = i;
1701 i += 1;
1702 continue;
1703 }
1704
1705 // A multi-line display-math block is verbatim: its line breaks
1706 // carry meaning (see `line_in_multiline_math_block`).
1707 if self.line_in_multiline_math_block(i + 1, ctx) {
1708 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1709 last_consumed = i;
1710 i += 1;
1711 continue;
1712 }
1713
1714 // Indented code block: indent >= FN_INDENT + 4 (= 8 spaces)
1715 if indent >= FN_INDENT + 4 {
1716 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1717 last_consumed = i;
1718 i += 1;
1719 continue;
1720 }
1721
1722 // Structural content that must be preserved verbatim
1723 if next_trimmed.starts_with('#')
1724 || is_list_item(next_trimmed)
1725 || next_trimmed.starts_with('>')
1726 || TableUtils::is_potential_table_row(next_trimmed)
1727 || is_setext_underline(next_trimmed)
1728 || is_horizontal_rule(next_trimmed)
1729 || crate::utils::mkdocs_footnotes::is_footnote_definition(next_trimmed)
1730 {
1731 // Preserve verbatim: blockquotes, tables, lists, setext
1732 // underlines, and horizontal rules inside the footnote
1733 if next_trimmed.starts_with('>')
1734 || TableUtils::is_potential_table_row(next_trimmed)
1735 || is_list_item(next_trimmed)
1736 || is_setext_underline(next_trimmed)
1737 || is_horizontal_rule(next_trimmed)
1738 {
1739 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1740 last_consumed = i;
1741 i += 1;
1742 continue;
1743 }
1744 // Headings, new footnote defs, link refs — end the footnote
1745 break;
1746 }
1747
1748 // Link reference definitions inside footnotes are not reflowable
1749 if next_trimmed.starts_with('[')
1750 && !next_trimmed.starts_with("[^")
1751 && next_trimmed.contains("]:")
1752 && LINK_REF_PATTERN.is_match(next_trimmed)
1753 {
1754 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1755 last_consumed = i;
1756 i += 1;
1757 continue;
1758 }
1759
1760 // HTML-only lines inside footnotes are not reflowable
1761 if is_html_only_line(next_trimmed) {
1762 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1763 last_consumed = i;
1764 i += 1;
1765 continue;
1766 }
1767
1768 // Regular prose content
1769 fn_lines.push(FnLineType::Content(next_trimmed.to_string()));
1770 last_consumed = i;
1771 i += 1;
1772 }
1773
1774 // Nothing collected or only empty lines
1775 if fn_lines.iter().all(|l| matches!(l, FnLineType::Empty)) || fn_lines.is_empty() {
1776 continue;
1777 }
1778
1779 // --- Group into blocks ---
1780 #[derive(Debug)]
1781 enum FnBlock {
1782 Paragraph(Vec<String>),
1783 Verbatim(Vec<(String, usize)>), // (content, indent) preserved as-is
1784 }
1785
1786 let mut blocks: Vec<FnBlock> = Vec::new();
1787 let mut current_para: Vec<String> = Vec::new();
1788 let mut current_verbatim: Vec<(String, usize)> = Vec::new();
1789
1790 for fl in &fn_lines {
1791 match fl {
1792 FnLineType::Content(s) => {
1793 if !current_verbatim.is_empty() {
1794 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
1795 }
1796 current_para.push(s.clone());
1797 }
1798 FnLineType::Verbatim(s, indent) => {
1799 if !current_para.is_empty() {
1800 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
1801 }
1802 current_verbatim.push((s.clone(), *indent));
1803 }
1804 FnLineType::Empty => {
1805 if !current_para.is_empty() {
1806 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
1807 }
1808 if !current_verbatim.is_empty() {
1809 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
1810 }
1811 }
1812 }
1813 }
1814 if !current_para.is_empty() {
1815 blocks.push(FnBlock::Paragraph(current_para));
1816 }
1817 if !current_verbatim.is_empty() {
1818 blocks.push(FnBlock::Verbatim(current_verbatim));
1819 }
1820
1821 // --- Reflow paragraphs and reconstruct ---
1822 let prefix_display_width = prefix.chars().count() + 1; // +1 for space
1823 let reflow_line_length = if config.line_length.is_unlimited() {
1824 usize::MAX
1825 } else {
1826 config
1827 .line_length
1828 .get()
1829 .saturating_sub(FN_INDENT.max(prefix_display_width))
1830 .max(20)
1831 };
1832 // Footnote continuation uses a fixed 4-space indent, so list
1833 // continuation capping does not apply here.
1834 let reflow_options = crate::utils::text_reflow::ReflowOptions {
1835 max_list_continuation_indent: None,
1836 ..Self::reflow_options(ctx, config, reflow_line_length)
1837 };
1838
1839 let indent_str = " ".repeat(FN_INDENT);
1840 let mut result_lines: Vec<String> = Vec::new();
1841 let mut is_first_block = true;
1842
1843 for block in &blocks {
1844 match block {
1845 FnBlock::Paragraph(para_lines) => {
1846 let paragraph_text = para_lines.join(" ");
1847 let paragraph_text = paragraph_text.trim();
1848 if paragraph_text.is_empty() {
1849 continue;
1850 }
1851
1852 let reflowed = crate::utils::text_reflow::reflow_line(paragraph_text, &reflow_options);
1853 if reflowed.is_empty() {
1854 continue;
1855 }
1856
1857 // Blank line separator between blocks
1858 if !result_lines.is_empty() {
1859 result_lines.push(String::new());
1860 }
1861
1862 for (idx, rline) in reflowed.iter().enumerate() {
1863 if is_first_block && idx == 0 {
1864 result_lines.push(format!("{prefix} {rline}"));
1865 } else {
1866 result_lines.push(format!("{indent_str}{rline}"));
1867 }
1868 }
1869 is_first_block = false;
1870 }
1871 FnBlock::Verbatim(verb_lines) => {
1872 // Blank line separator between blocks
1873 if !result_lines.is_empty() {
1874 result_lines.push(String::new());
1875 }
1876
1877 if is_first_block {
1878 // Verbatim as first block in a deferred-body footnote
1879 if deferred_body {
1880 result_lines.push(prefix.to_string());
1881 }
1882 is_first_block = false;
1883 }
1884 for (content, _orig_indent) in verb_lines {
1885 result_lines.push(format!("{indent_str}{content}"));
1886 }
1887 }
1888 }
1889 }
1890
1891 // If nothing was produced, skip
1892 if result_lines.is_empty() {
1893 continue;
1894 }
1895
1896 let reflowed_text = result_lines.join(line_ending);
1897
1898 // Calculate byte range using last_consumed
1899 let start_range = line_index.whole_line_range(footnote_start + 1);
1900 let end_range = if last_consumed == lines.len() - 1 && !ctx.content.ends_with('\n') {
1901 line_index.line_text_range(last_consumed + 1, 1, lines[last_consumed].len() + 1)
1902 } else {
1903 line_index.whole_line_range(last_consumed + 1)
1904 };
1905 let byte_range = start_range.start..end_range.end;
1906
1907 let replacement = if last_consumed < lines.len() - 1 || ctx.content.ends_with('\n') {
1908 format!("{reflowed_text}{line_ending}")
1909 } else {
1910 reflowed_text
1911 };
1912
1913 let original_text = &ctx.content[byte_range.clone()];
1914 let max_length = (footnote_start..=last_consumed)
1915 .map(|idx| self.calculate_effective_length(lines[idx]))
1916 .max()
1917 .unwrap_or(0);
1918 let line_limit = if config.line_length.is_unlimited() {
1919 usize::MAX
1920 } else {
1921 config.line_length.get()
1922 };
1923 if original_text != replacement && max_length > line_limit {
1924 warnings.push(LintWarning {
1925 rule_name: Some(self.name().to_string()),
1926 message: format!(
1927 "Line length {} exceeds {} characters",
1928 max_length,
1929 config.line_length.get()
1930 ),
1931 line: footnote_start + 1,
1932 column: 1,
1933 end_line: last_consumed + 1,
1934 end_column: lines[last_consumed].chars().count() + 1,
1935 severity: Severity::Warning,
1936 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1937 });
1938 }
1939 continue;
1940 }
1941
1942 // Handle MkDocs container content (admonitions and tabs) with indent-preserving reflow
1943 if ctx
1944 .line_info(line_num)
1945 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
1946 {
1947 // Skip admonition/tab marker lines — only reflow their indented content
1948 let current_line = lines[i];
1949 if mkdocs_admonitions::is_admonition_start(current_line) || mkdocs_tabs::is_tab_marker(current_line) {
1950 i += 1;
1951 continue;
1952 }
1953
1954 let container_start = i;
1955
1956 // Detect the actual indent level from the first content line
1957 // (supports nested admonitions with 8+ spaces)
1958 let first_line = lines[i];
1959 let base_indent_len = first_line.len() - first_line.trim_start().len();
1960 let base_indent: String = " ".repeat(base_indent_len);
1961
1962 // Collect consecutive MkDocs container paragraph lines
1963 let mut container_lines: Vec<&str> = Vec::new();
1964 while i < lines.len() {
1965 let current_line_num = i + 1;
1966 let line_info = ctx.line_info(current_line_num);
1967
1968 // Stop if we leave the MkDocs container
1969 if !line_info.is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container) {
1970 break;
1971 }
1972
1973 let line = lines[i];
1974
1975 // Stop at paragraph boundaries within the container
1976 if line.trim().is_empty() {
1977 break;
1978 }
1979
1980 // Skip list items, code blocks, headings, HTML-only lines within containers
1981 if is_list_item(line.trim())
1982 || line.trim().starts_with("```")
1983 || line.trim().starts_with("~~~")
1984 || line.trim().starts_with('#')
1985 || is_html_only_line(line)
1986 {
1987 break;
1988 }
1989
1990 container_lines.push(line);
1991 i += 1;
1992 }
1993
1994 if container_lines.is_empty() {
1995 // Must advance i to avoid infinite loop when we encounter
1996 // non-paragraph content (code block, list, heading, empty line)
1997 // at the start of an MkDocs container
1998 i += 1;
1999 continue;
2000 }
2001
2002 // Strip the base indent from each line and join for reflow
2003 let stripped_lines: Vec<&str> = container_lines
2004 .iter()
2005 .map(|line| {
2006 if line.starts_with(&base_indent) {
2007 &line[base_indent_len..]
2008 } else {
2009 line.trim_start()
2010 }
2011 })
2012 .collect();
2013 let paragraph_text = stripped_lines.join(" ");
2014
2015 // Check if reflow is needed
2016 let needs_reflow = match config.reflow_mode {
2017 ReflowMode::Normalize => self.normalize_mode_needs_reflow(container_lines.iter().copied(), config),
2018 ReflowMode::SentencePerLine => {
2019 let sentences = split_into_sentences(¶graph_text);
2020 sentences.len() > 1 || container_lines.len() > 1
2021 }
2022 ReflowMode::SemanticLineBreaks => {
2023 let sentences = split_into_sentences(¶graph_text);
2024 sentences.len() > 1
2025 || container_lines.len() > 1
2026 || container_lines
2027 .iter()
2028 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
2029 }
2030 ReflowMode::Default => container_lines
2031 .iter()
2032 .any(|line| self.calculate_effective_length(line) > config.line_length.get()),
2033 };
2034
2035 if !needs_reflow {
2036 continue;
2037 }
2038
2039 // Calculate byte range for this container paragraph
2040 let start_range = line_index.whole_line_range(container_start + 1);
2041 let end_line = container_start + container_lines.len() - 1;
2042 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
2043 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
2044 } else {
2045 line_index.whole_line_range(end_line + 1)
2046 };
2047 let byte_range = start_range.start..end_range.end;
2048
2049 // Reflow with adjusted line length (accounting for the 4-space indent)
2050 let reflow_line_length = if config.line_length.is_unlimited() {
2051 usize::MAX
2052 } else {
2053 config.line_length.get().saturating_sub(base_indent_len).max(1)
2054 };
2055 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
2056 let reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
2057
2058 // Re-add the 4-space indent to each reflowed line
2059 let reflowed_with_indent: Vec<String> =
2060 reflowed.iter().map(|line| format!("{base_indent}{line}")).collect();
2061 let reflowed_text = reflowed_with_indent.join(line_ending);
2062
2063 // Preserve trailing newline
2064 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
2065 format!("{reflowed_text}{line_ending}")
2066 } else {
2067 reflowed_text
2068 };
2069
2070 // Only generate a warning if the replacement is different
2071 let original_text = &ctx.content[byte_range.clone()];
2072 if original_text != replacement {
2073 warnings.push(LintWarning {
2074 rule_name: Some(self.name().to_string()),
2075 message: format!(
2076 "Line length {} exceeds {} characters (in MkDocs container)",
2077 container_lines.iter().map(|l| l.len()).max().unwrap_or(0),
2078 config.line_length.get()
2079 ),
2080 line: container_start + 1,
2081 column: 1,
2082 end_line: end_line + 1,
2083 end_column: lines[end_line].chars().count() + 1,
2084 severity: Severity::Warning,
2085 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
2086 });
2087 }
2088 continue;
2089 }
2090
2091 // Helper function to detect semantic line markers
2092 let is_semantic_line = |content: &str| -> bool {
2093 let trimmed = content.trim_start();
2094 let semantic_markers = [
2095 "NOTE:",
2096 "WARNING:",
2097 "IMPORTANT:",
2098 "CAUTION:",
2099 "TIP:",
2100 "DANGER:",
2101 "HINT:",
2102 "INFO:",
2103 ];
2104 semantic_markers.iter().any(|marker| trimmed.starts_with(marker))
2105 };
2106
2107 // Helper function to detect fence markers (opening or closing)
2108 let is_fence_marker = |content: &str| -> bool {
2109 let trimmed = content.trim_start();
2110 trimmed.starts_with("```") || trimmed.starts_with("~~~")
2111 };
2112
2113 // Check if this is a list item - handle it specially
2114 let trimmed = lines[i].trim();
2115 if is_list_item(trimmed) {
2116 // Collect the entire list item including continuation lines
2117 let list_start = i;
2118 let (marker, first_content) = extract_list_marker_and_content(lines[i]);
2119 let marker_len = marker.len();
2120 // The normalized marker above is what gets re-emitted; the source marker
2121 // is where the item's content actually starts. Nested blocks move by the
2122 // difference between the two content columns, so the shift must be
2123 // measured against the source, not against the normalized width.
2124 let source_marker = source_list_marker(lines[i]);
2125 let source_content_col = source_marker.as_ref().map_or(marker_len, |m| m.content_col);
2126
2127 // Checkbox ([ ]/[x]/[X]) is inline content, not part of the list marker.
2128 // Use the base bullet/number marker width for continuation recognition
2129 // so that continuation lines at 2+ spaces are collected for "- [ ] " items.
2130 let base_marker_len = if marker.contains("[ ] ") || marker.contains("[x] ") || marker.contains("[X] ") {
2131 marker.find('[').unwrap_or(marker_len)
2132 } else {
2133 marker_len
2134 };
2135
2136 // MkDocs flavor requires at least 4 spaces for list continuation
2137 // after a blank line (multi-paragraph list items). For non-blank
2138 // continuation (lines directly following the marker line), use
2139 // the natural marker width so that 2-space indent is recognized.
2140 let item_indent = ctx.lines[i].indent;
2141 let min_continuation_indent = if ctx.flavor.requires_strict_list_indent() {
2142 // Use 4-space relative indent from the list item's nesting level
2143 item_indent + (base_marker_len - item_indent).max(4)
2144 } else {
2145 marker_len
2146 };
2147 let content_continuation_indent = base_marker_len;
2148
2149 // Track lines and their types (content, code block, fence, nested list)
2150 #[derive(Clone)]
2151 enum LineType {
2152 Content(String),
2153 CodeBlock(String, usize), // content and original indent
2154 SemanticLine(String), // Lines starting with NOTE:, WARNING:, etc that should stay separate
2155 SnippetLine(String), // MkDocs Snippets delimiters (-8<-) that must stay on their own line
2156 DivMarker(String), // Quarto/Pandoc div markers (::: opening or closing)
2157 AdmonitionHeader(String, usize), // header text (e.g. "!!! note") and original indent
2158 AdmonitionContent(String, usize), // body content text and original indent
2159 Table(String, usize), // GFM table row, preserved verbatim with original indent
2160 Empty,
2161 }
2162
2163 let mut list_item_lines: Vec<LineType> = vec![LineType::Content(first_content)];
2164 // Set when collection stops at a nested list item or a nested
2165 // blockquote that belongs to this item. Such structure is reflowed
2166 // independently and is therefore absent from `list_item_lines`/`blocks`,
2167 // but it still keeps the emitted item spanning multiple physical lines,
2168 // which the MD030 multi-line spacing decision below must account for.
2169 let mut has_trailing_nested_structure = false;
2170 i += 1;
2171
2172 // Collect continuation lines using ctx.lines for metadata
2173 while i < lines.len() {
2174 let line_info = &ctx.lines[i];
2175
2176 // Use pre-computed is_blank from ctx
2177 if line_info.is_blank {
2178 // Empty line - check if next line is indented (part of list item)
2179 if i + 1 < lines.len() {
2180 let next_info = &ctx.lines[i + 1];
2181
2182 // Check if next line is indented enough to be continuation
2183 if !next_info.is_blank && next_info.indent >= min_continuation_indent {
2184 // This blank line is between paragraphs/blocks in the list item
2185 list_item_lines.push(LineType::Empty);
2186 i += 1;
2187 continue;
2188 }
2189 }
2190 // No indented line after blank, end of list item
2191 break;
2192 }
2193
2194 // Use pre-computed indent from ctx
2195 let indent = line_info.indent;
2196
2197 // Valid continuation must be indented at least content_continuation_indent.
2198 // For non-blank continuation, use marker_len (e.g. 2 for "- ").
2199 // MkDocs strict 4-space requirement applies only after blank lines.
2200 if indent >= content_continuation_indent {
2201 let trimmed = line_info.content(ctx.content).trim();
2202
2203 // Check for MkDocs admonition lines inside list items BEFORE
2204 // checking in_code_block. Lines inside code blocks within
2205 // admonitions have both in_admonition and in_code_block set;
2206 // admonition membership takes priority so the entire admonition
2207 // structure (including embedded code blocks) is preserved.
2208 if line_info.in_admonition {
2209 let raw_content = line_info.content(ctx.content);
2210 if mkdocs_admonitions::is_admonition_start(raw_content) {
2211 let header_text = raw_content[indent..].trim_end().to_string();
2212 list_item_lines.push(LineType::AdmonitionHeader(header_text, indent));
2213 } else {
2214 let body_text = raw_content[indent..].trim_end().to_string();
2215 list_item_lines.push(LineType::AdmonitionContent(body_text, indent));
2216 }
2217 i += 1;
2218 continue;
2219 }
2220
2221 // Use pre-computed in_code_block from ctx
2222 if line_info.in_code_block {
2223 list_item_lines.push(LineType::CodeBlock(
2224 line_info.content(ctx.content)[indent..].to_string(),
2225 indent,
2226 ));
2227 i += 1;
2228 continue;
2229 }
2230
2231 // A multi-line display-math block inside the item is verbatim:
2232 // its line breaks carry meaning (see
2233 // `line_in_multiline_math_block`), so reuse the code-block
2234 // carrier to re-emit it unchanged.
2235 if self.line_in_multiline_math_block(i + 1, ctx) {
2236 list_item_lines.push(LineType::CodeBlock(
2237 line_info.content(ctx.content)[indent..].to_string(),
2238 indent,
2239 ));
2240 i += 1;
2241 continue;
2242 }
2243
2244 // A blockquote nested inside the list item is reflowed by the
2245 // blockquote-aware path (it preserves the `>` prefix, including the
2246 // list indent), not as list-item prose. Collecting it as Content
2247 // would strip the markers and reflow `>` as words, collapsing the
2248 // blank `>` line and dropping `>` from wrapped continuations. End
2249 // the item here so the outer loop routes the blockquote line to
2250 // generate_blockquote_paragraph_fix. Uncollect a pending blank so
2251 // the separator between the list prose and the blockquote survives.
2252 if line_info.blockquote.is_some() {
2253 has_trailing_nested_structure = true;
2254 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
2255 list_item_lines.pop();
2256 i -= 1;
2257 }
2258 break;
2259 }
2260
2261 // Check if this is a SIBLING list item (breaks parent)
2262 // Nested lists are indented >= marker_len and are PART of the parent item
2263 // Siblings are at indent < marker_len (at or before parent marker)
2264 if is_list_item(trimmed) && indent < marker_len {
2265 // This is a sibling item at same or higher level - end parent item
2266 break;
2267 }
2268
2269 // Nested list items are always processed independently
2270 // by the outer loop, so break when we encounter one.
2271 // If a blank line was collected before this, uncollect it
2272 // so the outer loop preserves the blank between parent and nested.
2273 if is_list_item(trimmed) && indent >= marker_len {
2274 has_trailing_nested_structure = true;
2275 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
2276 list_item_lines.pop();
2277 i -= 1;
2278 }
2279 break;
2280 }
2281
2282 // Normal continuation vs indented code block.
2283 // Use min_continuation_indent for the threshold since
2284 // code blocks start 4 spaces beyond the expected content
2285 // level (which is min_continuation_indent for MkDocs).
2286 if indent <= min_continuation_indent + 3 {
2287 // Extract content (remove indentation and trailing whitespace)
2288 // Preserve hard breaks (2 trailing spaces) while removing excessive whitespace
2289 // See: https://github.com/rvben/rumdl/issues/76
2290 let content = trim_preserving_hard_break(&line_info.content(ctx.content)[indent..]);
2291
2292 // Check if this is a div marker (::: opening or closing)
2293 // These must be preserved on their own line, not merged into paragraphs
2294 if line_info.is_div_marker {
2295 list_item_lines.push(LineType::DivMarker(content));
2296 }
2297 // Check if this is a fence marker (opening or closing)
2298 // These should be treated as code block lines, not paragraph content
2299 else if is_fence_marker(&content) {
2300 list_item_lines.push(LineType::CodeBlock(content, indent));
2301 }
2302 // Check if this is a semantic line (NOTE:, WARNING:, etc.)
2303 else if is_semantic_line(&content) {
2304 list_item_lines.push(LineType::SemanticLine(content));
2305 }
2306 // Check if this is a snippet block delimiter (-8<- or --8<--)
2307 // These must be preserved on their own lines for MkDocs Snippets extension
2308 else if is_snippet_block_delimiter(&content) {
2309 list_item_lines.push(LineType::SnippetLine(content));
2310 }
2311 // Check if this is a GFM table row. Tables nested inside list
2312 // items must be preserved verbatim — joining them with prose
2313 // breaks the column structure.
2314 //
2315 // `is_potential_table_row` is intentionally permissive at the
2316 // row level: any line with `|` and 2+ cells qualifies. To avoid
2317 // misclassifying prose continuation lines that contain a literal
2318 // pipe (e.g. "use grep | sort to ..."), require one of:
2319 // - the row is pipe-bordered (`| ... |`), the canonical form
2320 // for tables nested in lists; or
2321 // - the next line is a delimiter row (this is a header); or
2322 // - the previous classified line was already a Table (this is
2323 // a continuation row).
2324 else if TableUtils::is_potential_table_row(&content) && {
2325 let pipe_bordered = content.trim().starts_with('|') && content.trim().ends_with('|');
2326 let next_is_delim = ctx
2327 .lines
2328 .get(i + 1)
2329 .is_some_and(|next| TableUtils::is_delimiter_row(next.content(ctx.content)));
2330 let prev_was_table = matches!(list_item_lines.last(), Some(LineType::Table(..)));
2331 pipe_bordered || next_is_delim || prev_was_table
2332 } {
2333 list_item_lines.push(LineType::Table(content, indent));
2334 } else {
2335 list_item_lines.push(LineType::Content(content));
2336 }
2337 i += 1;
2338 } else {
2339 // indent >= min_continuation_indent + 4: indented code block
2340 list_item_lines.push(LineType::CodeBlock(
2341 line_info.content(ctx.content)[indent..].to_string(),
2342 indent,
2343 ));
2344 i += 1;
2345 }
2346 } else {
2347 // Not indented enough, end of list item
2348 break;
2349 }
2350 }
2351
2352 // Determine the output continuation indent.
2353 // Normalize/Default modes canonicalize to min_continuation_indent
2354 // (fixing over-indented continuation). Semantic/SentencePerLine
2355 // modes preserve the user's actual indent since they only fix
2356 // line breaking, not indentation.
2357 let indent_size = match config.reflow_mode {
2358 ReflowMode::SemanticLineBreaks | ReflowMode::SentencePerLine => {
2359 // Find indent of the first plain text continuation line,
2360 // skipping the marker line (index 0), nested list items,
2361 // code blocks, and blank lines.
2362 list_item_lines
2363 .iter()
2364 .enumerate()
2365 .skip(1)
2366 .find_map(|(k, lt)| {
2367 if matches!(lt, LineType::Content(_)) {
2368 Some(ctx.lines[list_start + k].indent)
2369 } else {
2370 None
2371 }
2372 })
2373 .unwrap_or(min_continuation_indent)
2374 }
2375 _ => min_continuation_indent,
2376 };
2377 // For checkbox items in mkdocs flavor, enforce minimum indent so
2378 // continuation lines use the structural list indent (4), not the
2379 // content-aligned indent (6) which Python-Markdown doesn't support
2380 let has_checkbox = base_marker_len < marker_len;
2381 let indent_size = if has_checkbox && ctx.flavor.requires_strict_list_indent() {
2382 indent_size.max(min_continuation_indent)
2383 } else {
2384 indent_size
2385 };
2386
2387 // Split list_item_lines into blocks (paragraphs, code blocks, nested lists, semantic lines, and HTML blocks)
2388 let mut builder = BlockBuilder::new();
2389 for line in &list_item_lines {
2390 match line {
2391 LineType::Empty => builder.feed_blank_line(),
2392 LineType::Content(content) => builder.feed_content(content),
2393 LineType::CodeBlock(content, indent) => builder.feed_code_line(content, *indent),
2394 LineType::SemanticLine(content) => builder.feed_semantic_line(content),
2395 LineType::SnippetLine(content) => builder.feed_snippet_line(content),
2396 LineType::DivMarker(content) => builder.feed_div_marker(content),
2397 LineType::AdmonitionHeader(header_text, indent) => {
2398 builder.feed_admonition_header(header_text, *indent)
2399 }
2400 LineType::AdmonitionContent(content, indent) => {
2401 builder.feed_admonition_content(content, *indent)
2402 }
2403 LineType::Table(content, indent) => builder.feed_table_line(content, *indent),
2404 }
2405 }
2406 let blocks = builder.finalize();
2407
2408 // Helper: check if a line (raw source or stripped content) is exempt
2409 // from line-length checks. Link reference definitions are always exempt;
2410 // standalone link/image lines are exempt when strict mode is off.
2411 // Also checks content after stripping list markers, since list item
2412 // continuation lines may contain link ref defs.
2413 let is_exempt_line = |raw_line: &str| -> bool {
2414 let trimmed = raw_line.trim();
2415 // Link reference definitions: always exempt
2416 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
2417 return true;
2418 }
2419 // Also check after stripping list markers (for list item content)
2420 if is_list_item(trimmed) {
2421 let (_, content) = extract_list_marker_and_content(trimmed);
2422 let content_trimmed = content.trim();
2423 if content_trimmed.starts_with('[')
2424 && content_trimmed.contains("]:")
2425 && LINK_REF_PATTERN.is_match(content_trimmed)
2426 {
2427 return true;
2428 }
2429 }
2430 // Standalone link/image lines: exempt when not strict
2431 if !config.strict && is_standalone_link_or_image_line(raw_line) {
2432 return true;
2433 }
2434 // HTML-only lines: exempt when not strict
2435 if !config.strict && is_html_only_line(raw_line) {
2436 return true;
2437 }
2438 false
2439 };
2440
2441 // Check if reflowing is needed (only for content paragraphs, not code blocks or nested lists)
2442 // Exclude link reference definitions and standalone link lines from content
2443 // so they don't pollute combined_content or trigger false reflow.
2444 let content_lines: Vec<String> = list_item_lines
2445 .iter()
2446 .filter_map(|line| {
2447 if let LineType::Content(s) = line {
2448 if is_exempt_line(s) {
2449 return None;
2450 }
2451 Some(s.clone())
2452 } else {
2453 None
2454 }
2455 })
2456 .collect();
2457
2458 // Check if we need to reflow this list item
2459 // We check the combined content to see if it exceeds length limits
2460 let combined_content = content_lines.join(" ").trim().to_string();
2461
2462 // Helper to check if we should reflow in normalize mode
2463 let should_normalize = || {
2464 // Don't normalize if the list item only contains nested lists, code blocks, or semantic lines
2465 // DO normalize if it has plain text content that spans multiple lines
2466 let has_code_blocks = blocks.iter().any(|b| matches!(b, Block::Code { .. }));
2467 let has_semantic_lines = blocks.iter().any(|b| matches!(b, Block::SemanticLine(_)));
2468 let has_snippet_lines = blocks.iter().any(|b| matches!(b, Block::SnippetLine(_)));
2469 let has_div_markers = blocks.iter().any(|b| matches!(b, Block::DivMarker(_)));
2470 let has_admonitions = blocks.iter().any(|b| matches!(b, Block::Admonition { .. }));
2471 let has_tables = blocks.iter().any(|b| matches!(b, Block::Table { .. }));
2472 let has_paragraphs = blocks.iter().any(|b| matches!(b, Block::Paragraph(_)));
2473
2474 // If we have structural blocks but no paragraphs, don't normalize
2475 if (has_code_blocks
2476 || has_semantic_lines
2477 || has_snippet_lines
2478 || has_div_markers
2479 || has_admonitions
2480 || has_tables)
2481 && !has_paragraphs
2482 {
2483 return false;
2484 }
2485
2486 // If we have paragraphs, check if they span multiple lines or there are multiple blocks
2487 if has_paragraphs {
2488 // Count only paragraphs that contain at least one non-exempt line.
2489 // Paragraphs consisting entirely of link ref defs or standalone links
2490 // should not trigger normalization.
2491 let paragraph_count = blocks
2492 .iter()
2493 .filter(|b| {
2494 if let Block::Paragraph(para_lines) = b {
2495 !para_lines.iter().all(|line| is_exempt_line(line))
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.iter().all(|line| is_exempt_line(line)) {
2611 return false;
2612 }
2613 let joined = para_lines.join(" ");
2614 let with_marker = format!("{}{}", " ".repeat(indent_size), joined.trim());
2615 self.calculate_effective_length(&with_marker) > config.line_length.get()
2616 }
2617 Block::Admonition {
2618 content_lines,
2619 header_indent,
2620 ..
2621 } => content_lines.iter().any(|(content, indent)| {
2622 if content.is_empty() {
2623 return false;
2624 }
2625 let with_indent = format!("{}{}", " ".repeat(*indent.max(header_indent)), content);
2626 self.calculate_effective_length(&with_indent) > config.line_length.get()
2627 }),
2628 _ => false,
2629 });
2630 if any_paragraph_exceeds {
2631 true
2632 } else {
2633 should_normalize()
2634 }
2635 }
2636 ReflowMode::SentencePerLine => {
2637 // Check if list item has multiple sentences
2638 let sentences = split_into_sentences(&combined_content);
2639 sentences.len() > 1
2640 }
2641 ReflowMode::SemanticLineBreaks => {
2642 let sentences = split_into_sentences(&combined_content);
2643 sentences.len() > 1
2644 || (list_start..i).any(|line_idx| {
2645 let line = lines[line_idx];
2646 let trimmed = line.trim();
2647 if trimmed.is_empty() || is_exempt_line(line) {
2648 return false;
2649 }
2650 self.calculate_effective_length(line) > config.line_length.get()
2651 })
2652 }
2653 ReflowMode::Default => {
2654 // In default mode, only reflow if any individual non-exempt line exceeds limit
2655 (list_start..i).any(|line_idx| {
2656 let line = lines[line_idx];
2657 let trimmed = line.trim();
2658 // Skip blank lines and exempt lines
2659 if trimmed.is_empty() || is_exempt_line(line) {
2660 return false;
2661 }
2662 self.calculate_effective_length(line) > config.line_length.get()
2663 })
2664 }
2665 };
2666
2667 // Record this item's frame so its nested children inherit the shift.
2668 // Only a reflowed item's marker actually moves; an unreflowed one keeps
2669 // its source position and so contributes no shift to its children. The
2670 // threshold that decides which following lines are inside this item is
2671 // the normalized marker width, NOT `source_content_col`: the collection
2672 // loop above gathers continuations by that same width, so the frame
2673 // boundary must match it or the two would disagree about ownership of
2674 // lines indented between the normalized and the source content column.
2675 list_shift_stack.push((marker_len, if needs_reflow { code_indent_shift } else { 0 }));
2676
2677 if needs_reflow {
2678 let start_range = line_index.whole_line_range(list_start + 1);
2679 let end_line = i - 1;
2680 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
2681 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
2682 } else {
2683 line_index.whole_line_range(end_line + 1)
2684 };
2685 let byte_range = start_range.start..end_range.end;
2686
2687 // Reflow each block (paragraphs only, preserve code blocks)
2688 // When line_length = 0 (no limit), use a very large value for reflow
2689 let reflow_line_length = if config.line_length.is_unlimited() {
2690 usize::MAX
2691 } else {
2692 config.line_length.get().saturating_sub(indent_size).max(1)
2693 };
2694 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
2695
2696 let mut result: Vec<String> = Vec::new();
2697 let mut is_first_block = true;
2698
2699 for (block_idx, block) in blocks.iter().enumerate() {
2700 match block {
2701 Block::Paragraph(para_lines) => {
2702 // If every line in this paragraph is exempt (link ref defs,
2703 // standalone links), preserve the paragraph verbatim instead
2704 // of reflowing it. Reflowing would corrupt link ref defs.
2705 let all_exempt = para_lines.iter().all(|line| is_exempt_line(line));
2706
2707 if all_exempt {
2708 for (idx, line) in para_lines.iter().enumerate() {
2709 if is_first_block && idx == 0 {
2710 result.push(format!("{marker}{line}"));
2711 is_first_block = false;
2712 } else {
2713 result.push(format!("{expected_indent}{line}"));
2714 }
2715 }
2716 } else {
2717 // Split the paragraph into segments at hard break boundaries
2718 // Each segment can be reflowed independently
2719 let segments = split_into_segments(para_lines);
2720
2721 for (segment_idx, segment) in segments.iter().enumerate() {
2722 // Check if this segment ends with a hard break and what type
2723 let hard_break_type = segment.last().and_then(|line| {
2724 let line = line.strip_suffix('\r').unwrap_or(line);
2725 if line.ends_with('\\') {
2726 Some("\\")
2727 } else if line.ends_with(" ") {
2728 Some(" ")
2729 } else {
2730 None
2731 }
2732 });
2733
2734 // Join and reflow the segment (removing the hard break marker for processing)
2735 let segment_for_reflow: Vec<String> = segment
2736 .iter()
2737 .map(|line| {
2738 // Strip hard break marker (2 spaces or backslash) for reflow processing
2739 if line.ends_with('\\') {
2740 line[..line.len() - 1].trim_end().to_string()
2741 } else if line.ends_with(" ") {
2742 line[..line.len() - 2].trim_end().to_string()
2743 } else {
2744 line.clone()
2745 }
2746 })
2747 .collect();
2748
2749 let segment_text = segment_for_reflow.join(" ").trim().to_string();
2750 if !segment_text.is_empty() {
2751 let reflowed =
2752 crate::utils::text_reflow::reflow_line(&segment_text, &reflow_options);
2753
2754 if is_first_block && segment_idx == 0 {
2755 // First segment of first block starts with marker
2756 result.push(format!("{marker}{}", reflowed[0]));
2757 for line in reflowed.iter().skip(1) {
2758 result.push(format!("{expected_indent}{line}"));
2759 }
2760 is_first_block = false;
2761 } else {
2762 // Subsequent segments
2763 for line in reflowed {
2764 result.push(format!("{expected_indent}{line}"));
2765 }
2766 }
2767
2768 // If this segment had a hard break, add it back to the last line
2769 // Preserve the original hard break format (backslash or two spaces)
2770 if let Some(break_marker) = hard_break_type
2771 && let Some(last_line) = result.last_mut()
2772 {
2773 last_line.push_str(break_marker);
2774 }
2775 }
2776 }
2777 }
2778
2779 // Add blank line after paragraph block if there's a next block.
2780 // Check if next block is a code block that doesn't want a preceding blank.
2781 // Also don't add blank lines before snippet lines (they should stay tight).
2782 // Only add if not already ending with one (avoids double blanks).
2783 if block_idx < blocks.len() - 1 {
2784 let next_block = &blocks[block_idx + 1];
2785 let should_add_blank = match next_block {
2786 Block::Code {
2787 has_preceding_blank, ..
2788 } => *has_preceding_blank,
2789 Block::Table {
2790 has_preceding_blank, ..
2791 } => *has_preceding_blank,
2792 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2793 _ => true, // For all other blocks, add blank line
2794 };
2795 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2796 result.push(String::new());
2797 }
2798 }
2799 }
2800 Block::Code {
2801 lines: code_lines,
2802 has_preceding_blank: _,
2803 } => {
2804 // Preserve code blocks as-is with original indentation
2805 // NOTE: Blank line before code block is handled by the previous block
2806 // (see paragraph block's logic above)
2807
2808 for (idx, (content, orig_indent)) in code_lines.iter().enumerate() {
2809 if is_first_block && idx == 0 {
2810 // First line of first block gets marker
2811 result.push(format!(
2812 "{marker}{}",
2813 " ".repeat(orig_indent - marker_len) + content.as_str()
2814 ));
2815 is_first_block = false;
2816 } else if content.is_empty() {
2817 result.push(String::new());
2818 } else {
2819 // Shift nested code with the marker so it stays
2820 // aligned under content when MD030 widens spacing.
2821 result.push(format!(
2822 "{}{}",
2823 " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
2824 content
2825 ));
2826 }
2827 }
2828 }
2829 Block::SemanticLine(content) => {
2830 // Preserve semantic lines (NOTE:, WARNING:, etc.) as-is on their own line.
2831 // Only add blank before if not already ending with one.
2832 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
2833 result.push(String::new());
2834 }
2835
2836 if is_first_block {
2837 // First block starts with marker
2838 result.push(format!("{marker}{content}"));
2839 is_first_block = false;
2840 } else {
2841 // Subsequent blocks use expected indent
2842 result.push(format!("{expected_indent}{content}"));
2843 }
2844
2845 // Add blank line after semantic line if there's a next block.
2846 // Only add if not already ending with one.
2847 if block_idx < blocks.len() - 1 {
2848 let next_block = &blocks[block_idx + 1];
2849 let should_add_blank = match next_block {
2850 Block::Code {
2851 has_preceding_blank, ..
2852 } => *has_preceding_blank,
2853 Block::Table {
2854 has_preceding_blank, ..
2855 } => *has_preceding_blank,
2856 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2857 _ => true, // For all other blocks, add blank line
2858 };
2859 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2860 result.push(String::new());
2861 }
2862 }
2863 }
2864 Block::SnippetLine(content) => {
2865 // Preserve snippet delimiters (-8<-) as-is on their own line
2866 // Unlike semantic lines, snippet lines don't add extra blank lines
2867 if is_first_block {
2868 // First block starts with marker
2869 result.push(format!("{marker}{content}"));
2870 is_first_block = false;
2871 } else {
2872 // Subsequent blocks use expected indent
2873 result.push(format!("{expected_indent}{content}"));
2874 }
2875 // No blank lines added before or after snippet delimiters
2876 }
2877 Block::DivMarker(content) => {
2878 // Preserve div markers (::: opening or closing) as-is on their own line
2879 if is_first_block {
2880 result.push(format!("{marker}{content}"));
2881 is_first_block = false;
2882 } else {
2883 result.push(format!("{expected_indent}{content}"));
2884 }
2885 }
2886 Block::Html {
2887 lines: html_lines,
2888 has_preceding_blank: _,
2889 } => {
2890 // Preserve HTML blocks exactly as-is with original indentation
2891 // NOTE: Blank line before HTML block is handled by the previous block
2892
2893 for (idx, line) in html_lines.iter().enumerate() {
2894 if is_first_block && idx == 0 {
2895 // First line of first block gets marker
2896 result.push(format!("{marker}{line}"));
2897 is_first_block = false;
2898 } else if line.is_empty() {
2899 // Preserve blank lines inside HTML blocks
2900 result.push(String::new());
2901 } else {
2902 // Preserve lines with their original content (already includes indentation)
2903 result.push(format!("{expected_indent}{line}"));
2904 }
2905 }
2906
2907 // Add blank line after HTML block if there's a next block.
2908 // Only add if not already ending with one (avoids double blanks
2909 // when the HTML block itself contained a trailing blank line).
2910 if block_idx < blocks.len() - 1 {
2911 let next_block = &blocks[block_idx + 1];
2912 let should_add_blank = match next_block {
2913 Block::Code {
2914 has_preceding_blank, ..
2915 } => *has_preceding_blank,
2916 Block::Html {
2917 has_preceding_blank, ..
2918 } => *has_preceding_blank,
2919 Block::Table {
2920 has_preceding_blank, ..
2921 } => *has_preceding_blank,
2922 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2923 _ => true, // For all other blocks, add blank line
2924 };
2925 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2926 result.push(String::new());
2927 }
2928 }
2929 }
2930 Block::Table {
2931 lines: table_lines,
2932 has_preceding_blank: _,
2933 } => {
2934 // Preserve table rows verbatim with their original indentation.
2935 // Reflowing rows would corrupt column alignment and inject `|`
2936 // characters mid-paragraph (issue #590).
2937 // The leading blank line is emitted by the previous block.
2938 for (idx, (content, orig_indent)) in table_lines.iter().enumerate() {
2939 if is_first_block && idx == 0 {
2940 // First line of first block gets the list marker
2941 result.push(format!(
2942 "{marker}{}",
2943 " ".repeat(orig_indent.saturating_sub(marker_len)) + content.as_str()
2944 ));
2945 is_first_block = false;
2946 } else {
2947 // Shift nested table rows with the marker so they
2948 // stay aligned when MD030 widens marker spacing.
2949 result.push(format!(
2950 "{}{}",
2951 " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
2952 content
2953 ));
2954 }
2955 }
2956
2957 // Add blank line after table block if there's a next block.
2958 if block_idx < blocks.len() - 1 {
2959 let next_block = &blocks[block_idx + 1];
2960 let should_add_blank = match next_block {
2961 Block::Code {
2962 has_preceding_blank, ..
2963 } => *has_preceding_blank,
2964 Block::Html {
2965 has_preceding_blank, ..
2966 } => *has_preceding_blank,
2967 Block::Table {
2968 has_preceding_blank, ..
2969 } => *has_preceding_blank,
2970 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2971 _ => true,
2972 };
2973 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2974 result.push(String::new());
2975 }
2976 }
2977 }
2978 Block::Admonition {
2979 header,
2980 header_indent,
2981 content_lines: admon_lines,
2982 } => {
2983 // Reconstruct admonition block with header at original indent
2984 // and body content reflowed to fit within the line length limit
2985
2986 // Add blank line before admonition if not first block
2987 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
2988 result.push(String::new());
2989 }
2990
2991 // Output the header at its original indent
2992 let header_indent_str = " ".repeat(*header_indent);
2993 if is_first_block {
2994 result.push(format!(
2995 "{marker}{}",
2996 " ".repeat(header_indent.saturating_sub(marker_len)) + header.as_str()
2997 ));
2998 is_first_block = false;
2999 } else {
3000 result.push(format!("{header_indent_str}{header}"));
3001 }
3002
3003 // Derive body indent from the first non-empty content line's
3004 // stored indent, falling back to header_indent + 4 for
3005 // empty-body admonitions
3006 let body_indent = admon_lines
3007 .iter()
3008 .find(|(content, _)| !content.is_empty())
3009 .map_or(header_indent + 4, |(_, indent)| *indent);
3010 let body_indent_str = " ".repeat(body_indent);
3011
3012 // Segment body content into code blocks (verbatim) and
3013 // text paragraphs (reflowable), separated by blank lines.
3014 // Code lines store (content, orig_indent) to reconstruct
3015 // internal indentation relative to body_indent.
3016 enum AdmonSegment {
3017 Text(Vec<String>),
3018 Code(Vec<(String, usize)>),
3019 }
3020
3021 let mut segments: Vec<AdmonSegment> = Vec::new();
3022 let mut current_text: Vec<String> = Vec::new();
3023 let mut current_code: Vec<(String, usize)> = Vec::new();
3024 let mut in_admon_code = false;
3025 // Track the opening fence character so closing fences
3026 // must match (backticks close backticks, tildes close tildes)
3027 let mut fence_char: char = '`';
3028
3029 // Opening fences: ``` or ~~~ followed by optional info string
3030 let get_opening_fence = |s: &str| -> Option<(char, usize)> {
3031 let t = s.trim_start();
3032 if t.starts_with("```") {
3033 Some(('`', t.bytes().take_while(|&b| b == b'`').count()))
3034 } else if t.starts_with("~~~") {
3035 Some(('~', t.bytes().take_while(|&b| b == b'~').count()))
3036 } else {
3037 None
3038 }
3039 };
3040 // Closing fences: ONLY fence chars + optional trailing spaces
3041 let get_closing_fence = |s: &str| -> Option<(char, usize)> {
3042 let t = s.trim();
3043 if t.starts_with("```") && t.bytes().all(|b| b == b'`') {
3044 Some(('`', t.len()))
3045 } else if t.starts_with("~~~") && t.bytes().all(|b| b == b'~') {
3046 Some(('~', t.len()))
3047 } else {
3048 None
3049 }
3050 };
3051 let mut fence_len: usize = 3;
3052
3053 for (content, orig_indent) in admon_lines {
3054 if in_admon_code {
3055 // Closing fence must use the same character, be
3056 // at least as long, and have no info string
3057 if let Some((ch, len)) = get_closing_fence(content)
3058 && ch == fence_char
3059 && len >= fence_len
3060 {
3061 current_code.push((content.clone(), *orig_indent));
3062 in_admon_code = false;
3063 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
3064 continue;
3065 }
3066 current_code.push((content.clone(), *orig_indent));
3067 } else if let Some((ch, len)) = get_opening_fence(content) {
3068 if !current_text.is_empty() {
3069 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
3070 }
3071 in_admon_code = true;
3072 fence_char = ch;
3073 fence_len = len;
3074 current_code.push((content.clone(), *orig_indent));
3075 } else if content.is_empty() {
3076 if !current_text.is_empty() {
3077 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
3078 }
3079 } else {
3080 current_text.push(content.clone());
3081 }
3082 }
3083 if in_admon_code && !current_code.is_empty() {
3084 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
3085 }
3086 if !current_text.is_empty() {
3087 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
3088 }
3089
3090 // Build reflow options once for all text segments
3091 let admon_reflow_length = if config.line_length.is_unlimited() {
3092 usize::MAX
3093 } else {
3094 config.line_length.get().saturating_sub(body_indent).max(1)
3095 };
3096
3097 let admon_reflow_options = Self::reflow_options(ctx, config, admon_reflow_length);
3098
3099 // Output each segment
3100 for segment in &segments {
3101 // Blank line before each segment (after the header or previous segment)
3102 result.push(String::new());
3103
3104 match segment {
3105 AdmonSegment::Code(lines) => {
3106 for (line, orig_indent) in lines {
3107 if line.is_empty() {
3108 // Preserve blank lines inside code blocks
3109 result.push(String::new());
3110 } else {
3111 // Reconstruct with body_indent + any extra
3112 // indentation the line had beyond body_indent
3113 let extra = orig_indent.saturating_sub(body_indent);
3114 let indent_str = " ".repeat(body_indent + extra);
3115 result.push(format!("{indent_str}{line}"));
3116 }
3117 }
3118 }
3119 AdmonSegment::Text(lines) => {
3120 let paragraph_text = lines.join(" ").trim().to_string();
3121 if paragraph_text.is_empty() {
3122 continue;
3123 }
3124 let reflowed = crate::utils::text_reflow::reflow_line(
3125 ¶graph_text,
3126 &admon_reflow_options,
3127 );
3128 for line in &reflowed {
3129 result.push(format!("{body_indent_str}{line}"));
3130 }
3131 }
3132 }
3133 }
3134
3135 // Add blank line after admonition if there's a next block
3136 if block_idx < blocks.len() - 1 {
3137 let next_block = &blocks[block_idx + 1];
3138 let should_add_blank = match next_block {
3139 Block::Code {
3140 has_preceding_blank, ..
3141 } => *has_preceding_blank,
3142 Block::Table {
3143 has_preceding_blank, ..
3144 } => *has_preceding_blank,
3145 Block::SnippetLine(_) | Block::DivMarker(_) => false,
3146 _ => true,
3147 };
3148 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
3149 result.push(String::new());
3150 }
3151 }
3152 }
3153 }
3154 }
3155
3156 let reflowed_text = result.join(line_ending);
3157
3158 // Preserve trailing newline
3159 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
3160 format!("{reflowed_text}{line_ending}")
3161 } else {
3162 reflowed_text
3163 };
3164
3165 // Get the original text to compare
3166 let original_text = &ctx.content[byte_range.clone()];
3167
3168 // Physical-line-length scan, shared by the Normalize-mode gate and its
3169 // message. The list-item reflow preserves code blocks, HTML blocks,
3170 // admonition headers, fence markers, semantic markers, and snippet/div
3171 // markers verbatim; only paragraph content and admonition bodies are
3172 // restructured. Only those lines drive the length warning, so that
3173 // preserved-but-overlong content does not keep the paragraph-level
3174 // warning alive when the reflow would not fix that line.
3175 let should_count_for_length = |line_idx: usize| -> bool {
3176 let line = lines[line_idx];
3177 let trimmed = line.trim();
3178 if trimmed.is_empty() || is_exempt_line(line) {
3179 return false;
3180 }
3181 let info = &ctx.lines[line_idx];
3182 if info.in_code_block || info.in_html_block {
3183 return false;
3184 }
3185 if info.in_admonition && mkdocs_admonitions::is_admonition_start(line) {
3186 return false;
3187 }
3188 if is_fence_marker(line) || is_semantic_line(line) {
3189 return false;
3190 }
3191 if is_snippet_block_delimiter(line) {
3192 return false;
3193 }
3194 if line.trim_start().starts_with(":::") {
3195 return false;
3196 }
3197 true
3198 };
3199 let max_physical_length = (list_start..i)
3200 .filter(|&idx| should_count_for_length(idx))
3201 .map(|idx| self.calculate_effective_length(lines[idx]))
3202 .max()
3203 .unwrap_or(0);
3204 // `line-length = 0` means "no limit", so no physical line can be
3205 // "over"; the message below then describes a structural join rather
3206 // than a length violation.
3207 let any_paragraph_line_over =
3208 !config.line_length.is_unlimited() && max_physical_length > config.line_length.get();
3209
3210 // Normalize mode reflows list-item prose just like paragraphs:
3211 // joining continuation lines and re-wrapping to `line-length`.
3212 // `prose_changed` is true only when the reflow alters the words or
3213 // line breaks, not when it would merely re-indent continuation
3214 // lines or trim trailing whitespace. Comparing the texts with each
3215 // line's leading and trailing whitespace removed isolates "did the
3216 // words/line breaks change" from "did the surrounding whitespace
3217 // change". Continuation indentation is MD077's responsibility and
3218 // trailing whitespace is MD009's; an MD013 warning for either would
3219 // both duplicate those rules and resurface a persistent advisory on
3220 // already-fitting items that users disable MD013 fixing to avoid.
3221 let prose_changed = {
3222 let stripped = |text: &str| text.lines().map(str::trim).collect::<Vec<_>>().join("\n");
3223 stripped(original_text) != stripped(&replacement)
3224 };
3225 // Warn when the reflow rewraps prose (the normalize feature for
3226 // list items), or when a physical line genuinely exceeds the limit
3227 // and the reflow can change something (a true length violation,
3228 // even if all that changes is the continuation indent). A line that
3229 // is already optimal in both respects produces no warning.
3230 let gate_ok = prose_changed || (any_paragraph_line_over && original_text != replacement);
3231 if gate_ok {
3232 // Generate an appropriate message based on why reflow is needed
3233 let message = match config.reflow_mode {
3234 ReflowMode::SentencePerLine => {
3235 let num_sentences = split_into_sentences(&combined_content).len();
3236 let num_lines = content_lines.len();
3237 if num_lines == 1 {
3238 // Single line with multiple sentences
3239 format!("Line contains {num_sentences} sentences (one sentence per line required)")
3240 } else {
3241 // Multiple lines - could be split sentences or mixed
3242 format!(
3243 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
3244 )
3245 }
3246 }
3247 ReflowMode::SemanticLineBreaks => {
3248 let num_sentences = split_into_sentences(&combined_content).len();
3249 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
3250 }
3251 ReflowMode::Normalize => {
3252 // When a physical line genuinely exceeds the limit, report
3253 // it as a length violation. Otherwise the reflow is a
3254 // structural normalization (joining/re-wrapping multi-line
3255 // content that already fits), mirroring the paragraph path.
3256 if any_paragraph_line_over {
3257 format!(
3258 "Line length {} exceeds {} characters",
3259 max_physical_length,
3260 config.line_length.get()
3261 )
3262 } else {
3263 format!(
3264 "List item could be normalized to use line length of {} characters",
3265 config.line_length.get()
3266 )
3267 }
3268 }
3269 ReflowMode::Default => {
3270 // Report the actual longest non-exempt line, not the combined content
3271 let max_length = (list_start..i)
3272 .filter(|&line_idx| {
3273 let line = lines[line_idx];
3274 let trimmed = line.trim();
3275 !trimmed.is_empty() && !is_exempt_line(line)
3276 })
3277 .map(|line_idx| self.calculate_effective_length(lines[line_idx]))
3278 .max()
3279 .unwrap_or(0);
3280 format!(
3281 "Line length {} exceeds {} characters",
3282 max_length,
3283 config.line_length.get()
3284 )
3285 }
3286 };
3287
3288 warnings.push(LintWarning {
3289 rule_name: Some(self.name().to_string()),
3290 message,
3291 line: list_start + 1,
3292 column: 1,
3293 end_line: end_line + 1,
3294 end_column: lines[end_line].chars().count() + 1,
3295 severity: Severity::Warning,
3296 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
3297 });
3298 }
3299 }
3300 continue;
3301 }
3302
3303 // Found start of a paragraph - collect all lines in it
3304 let paragraph_start = i;
3305 let mut paragraph_lines = vec![lines[i]];
3306 i += 1;
3307
3308 while i < lines.len() {
3309 let next_line = lines[i];
3310 let next_line_num = i + 1;
3311 let next_trimmed = next_line.trim();
3312
3313 // Stop at paragraph boundaries
3314 if next_trimmed.is_empty()
3315 || ctx.line_info(next_line_num).is_some_and(|info| info.in_code_block)
3316 || ctx.line_info(next_line_num).is_some_and(|info| info.in_front_matter)
3317 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_block)
3318 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_comment)
3319 || ctx.line_info(next_line_num).is_some_and(|info| info.in_esm_block)
3320 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_expression)
3321 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_block)
3322 || ctx.line_info(next_line_num).is_some_and(|info| info.in_mdx_comment)
3323 || ctx
3324 .line_info(next_line_num)
3325 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
3326 || (next_line_num > 0
3327 && next_line_num <= ctx.lines.len()
3328 && ctx.lines[next_line_num - 1].blockquote.is_some())
3329 || next_trimmed.starts_with('#')
3330 || TableUtils::is_potential_table_row(next_line)
3331 || is_list_item(next_trimmed)
3332 || is_horizontal_rule(next_line)
3333 || (next_trimmed.starts_with('[') && next_line.contains("]:"))
3334 || is_template_directive_only(next_line)
3335 || is_standalone_attr_list(next_line)
3336 || is_snippet_block_delimiter(next_line)
3337 || ctx.line_info(next_line_num).is_some_and(|info| info.is_div_marker)
3338 || is_html_only_line(next_line)
3339 || self.line_in_multiline_math_block(next_line_num, ctx)
3340 || (!config.strict && is_standalone_link_or_image_line(next_line))
3341 {
3342 break;
3343 }
3344
3345 // Check if the previous line ends with a hard break (2+ spaces or backslash)
3346 if i > 0 && has_hard_break(lines[i - 1]) {
3347 // Don't include lines after hard breaks in the same paragraph
3348 break;
3349 }
3350
3351 paragraph_lines.push(next_line);
3352 i += 1;
3353 }
3354
3355 // Compute the common leading indent of all non-empty paragraph lines,
3356 // but only when those lines are structurally inside a list block.
3357 // Indented continuation lines that follow a nested list arrive here
3358 // with their structural indentation intact (e.g. 2 spaces for a
3359 // top-level list item). Stripping the indent before reflow and
3360 // re-applying it afterward prevents the fixer from moving those
3361 // lines to column 0.
3362 //
3363 // The list-block guard is essential: top-level paragraphs that happen
3364 // to start with spaces (insignificant in Markdown) must NOT have those
3365 // spaces preserved or injected by the fixer.
3366 let common_indent: String = if ctx.is_in_list_block(paragraph_start + 1) {
3367 let min_len = paragraph_lines
3368 .iter()
3369 .filter(|l| !l.trim().is_empty())
3370 .map(|l| l.len() - l.trim_start().len())
3371 .min()
3372 .unwrap_or(0);
3373 paragraph_lines
3374 .iter()
3375 .find(|l| !l.trim().is_empty())
3376 .map(|l| l[..min_len].to_string())
3377 .unwrap_or_default()
3378 } else {
3379 String::new()
3380 };
3381
3382 // Combine paragraph lines into a single string for processing.
3383 // This must be done BEFORE the needs_reflow check for sentence-per-line mode.
3384 let paragraph_text = if common_indent.is_empty() {
3385 paragraph_lines.join(" ")
3386 } else {
3387 paragraph_lines
3388 .iter()
3389 .map(|l| {
3390 if l.starts_with(common_indent.as_str()) {
3391 &l[common_indent.len()..]
3392 } else {
3393 l.trim_start()
3394 }
3395 })
3396 .collect::<Vec<_>>()
3397 .join(" ")
3398 };
3399
3400 // Skip reflowing if this paragraph contains definition list items
3401 // Definition lists are multi-line structures that should not be joined
3402 let contains_definition_list = paragraph_lines
3403 .iter()
3404 .any(|line| crate::utils::is_definition_list_item(line));
3405
3406 if contains_definition_list {
3407 // Don't reflow definition lists - skip this paragraph
3408 i = paragraph_start + paragraph_lines.len();
3409 continue;
3410 }
3411
3412 // Skip reflowing if this paragraph contains MkDocs Snippets markers
3413 // Snippets blocks (-8<- ... -8<-) should be preserved exactly
3414 let contains_snippets = paragraph_lines.iter().any(|line| is_snippet_block_delimiter(line));
3415
3416 if contains_snippets {
3417 // Don't reflow Snippets blocks - skip this paragraph
3418 i = paragraph_start + paragraph_lines.len();
3419 continue;
3420 }
3421
3422 // Leave a line of a multi-line display-math block as it is, where
3423 // joining lines would corrupt the equation (see
3424 // `line_in_multiline_math_block`). Only the first line has to be
3425 // asked about: such a line ends the paragraph above it, so a
3426 // paragraph reaching here holds one only when it starts on one.
3427 //
3428 // Only that line is passed over, not the rest of what was collected
3429 // with it: prose written directly under the closing delimiter is an
3430 // ordinary paragraph and still reflows.
3431 if self.line_in_multiline_math_block(paragraph_start + 1, ctx) {
3432 i = paragraph_start + 1;
3433 continue;
3434 }
3435
3436 // Check if this paragraph needs reflowing
3437 let needs_reflow = match config.reflow_mode {
3438 ReflowMode::Normalize => self.normalize_mode_needs_reflow(paragraph_lines.iter().copied(), config),
3439 ReflowMode::SentencePerLine => {
3440 // In sentence-per-line mode, check if the JOINED paragraph has multiple sentences
3441 // Note: we check the joined text because sentences can span multiple lines
3442 let sentences = split_into_sentences(¶graph_text);
3443
3444 // Always reflow if multiple sentences on one line
3445 if sentences.len() > 1 {
3446 true
3447 } else if paragraph_lines.len() > 1 {
3448 // For single-sentence paragraphs spanning multiple lines:
3449 // Reflow if they COULD fit on one line (respecting line-length constraint)
3450 if config.line_length.is_unlimited() {
3451 // No line-length constraint - always join single sentences
3452 true
3453 } else {
3454 // Only join if it fits within line-length.
3455 // paragraph_text has the common indent stripped, so add it
3456 // back to get the true output length before comparing.
3457 let effective_length =
3458 self.calculate_effective_length(¶graph_text) + common_indent.len();
3459 effective_length <= config.line_length.get()
3460 }
3461 } else {
3462 false
3463 }
3464 }
3465 ReflowMode::SemanticLineBreaks => {
3466 let sentences = split_into_sentences(¶graph_text);
3467 // Reflow if multiple sentences, multiple lines, or any line exceeds limit
3468 sentences.len() > 1
3469 || paragraph_lines.len() > 1
3470 || paragraph_lines
3471 .iter()
3472 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3473 }
3474 ReflowMode::Default => {
3475 // In default mode, only reflow if lines exceed limit
3476 paragraph_lines
3477 .iter()
3478 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3479 }
3480 };
3481
3482 if needs_reflow {
3483 // Calculate byte range for this paragraph
3484 // Use whole_line_range for each line and combine
3485 let start_range = line_index.whole_line_range(paragraph_start + 1);
3486 let end_line = paragraph_start + paragraph_lines.len() - 1;
3487
3488 // For the last line, we want to preserve any trailing newline
3489 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
3490 // Last line without trailing newline - use line_text_range
3491 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
3492 } else {
3493 // Not the last line or has trailing newline - use whole_line_range
3494 line_index.whole_line_range(end_line + 1)
3495 };
3496
3497 let byte_range = start_range.start..end_range.end;
3498
3499 // Check if the paragraph ends with a hard break and what type
3500 let hard_break_type = paragraph_lines.last().and_then(|line| {
3501 let line = line.strip_suffix('\r').unwrap_or(line);
3502 if line.ends_with('\\') {
3503 Some("\\")
3504 } else if line.ends_with(" ") {
3505 Some(" ")
3506 } else {
3507 None
3508 }
3509 });
3510
3511 // Reflow the paragraph
3512 // When line_length = 0 (no limit), use a very large value for reflow
3513 let reflow_line_length = if config.line_length.is_unlimited() {
3514 usize::MAX
3515 } else {
3516 config.line_length.get()
3517 };
3518 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
3519 let mut reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
3520
3521 // Re-apply the common indent to each non-empty reflowed line so
3522 // that the replacement preserves the original structural indentation.
3523 if !common_indent.is_empty() {
3524 for line in &mut reflowed {
3525 if !line.is_empty() {
3526 *line = format!("{common_indent}{line}");
3527 }
3528 }
3529 }
3530
3531 // If the original paragraph ended with a hard break, preserve it
3532 // Preserve the original hard break format (backslash or two spaces)
3533 if let Some(break_marker) = hard_break_type
3534 && !reflowed.is_empty()
3535 {
3536 let last_idx = reflowed.len() - 1;
3537 if !has_hard_break(&reflowed[last_idx]) {
3538 reflowed[last_idx].push_str(break_marker);
3539 }
3540 }
3541
3542 let reflowed_text = reflowed.join(line_ending);
3543
3544 // Preserve trailing newline if the original paragraph had one
3545 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
3546 format!("{reflowed_text}{line_ending}")
3547 } else {
3548 reflowed_text
3549 };
3550
3551 // Get the original text to compare
3552 let original_text = &ctx.content[byte_range.clone()];
3553
3554 // Only generate a warning if the replacement is different from the original
3555 if original_text != replacement {
3556 // Determine which line ranges and messages to report based on the reflow mode.
3557 let warnings_to_report: Vec<(usize, usize, String)> = match config.reflow_mode {
3558 ReflowMode::Default => {
3559 // In default mode, report a warning for *every* line in the paragraph
3560 // that exceeds the limit. Each warning will carry the same paragraph-level
3561 // fix, making all of them auto-fixable.
3562 paragraph_lines
3563 .iter()
3564 .enumerate()
3565 .filter(|(_, line)| self.calculate_effective_length(line) > config.line_length.get())
3566 .map(|(idx, _)| {
3567 let violating_line = paragraph_start + idx + 1;
3568 (
3569 violating_line,
3570 violating_line,
3571 format!("Line length exceeds {} characters", config.line_length.get()),
3572 )
3573 })
3574 .collect()
3575 }
3576 ReflowMode::Normalize => {
3577 // In normalize mode, report the whole paragraph as needing normalization.
3578 vec![(
3579 paragraph_start + 1,
3580 end_line + 1,
3581 format!(
3582 "Paragraph could be normalized to use line length of {} characters",
3583 config.line_length.get()
3584 ),
3585 )]
3586 }
3587 ReflowMode::SentencePerLine => {
3588 // In sentence-per-line mode, highlight the entire paragraph that needs reformatting.
3589 let num_sentences = split_into_sentences(¶graph_text).len();
3590 let message = if paragraph_lines.len() == 1 {
3591 // Single line with multiple sentences
3592 format!("Line contains {num_sentences} sentences (one sentence per line required)")
3593 } else {
3594 // Multiple lines - could be split sentences or mixed
3595 let num_lines = paragraph_lines.len();
3596 format!(
3597 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
3598 )
3599 };
3600 vec![(paragraph_start + 1, paragraph_start + paragraph_lines.len(), message)]
3601 }
3602 ReflowMode::SemanticLineBreaks => {
3603 // In semantic-line-breaks mode, highlight the entire paragraph.
3604 let num_sentences = split_into_sentences(¶graph_text).len();
3605 vec![(
3606 paragraph_start + 1,
3607 paragraph_start + paragraph_lines.len(),
3608 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)"),
3609 )]
3610 }
3611 };
3612
3613 // Generate the actual lint warnings. All warnings for this paragraph
3614 // share the same paragraph-level fix.
3615 for (w_start, w_end, msg) in warnings_to_report {
3616 warnings.push(LintWarning {
3617 rule_name: Some(self.name().to_string()),
3618 message: msg,
3619 line: w_start,
3620 column: 1,
3621 end_line: w_end,
3622 end_column: lines[w_end.saturating_sub(1)].chars().count() + 1,
3623 severity: Severity::Warning,
3624 fix: Some(crate::rule::Fix::new(byte_range.clone(), replacement.clone())),
3625 });
3626 }
3627 }
3628 }
3629 }
3630
3631 warnings
3632 }
3633
3634 /// Calculate string length based on the configured length mode
3635 fn calculate_string_length(&self, s: &str) -> usize {
3636 match self.config.length_mode {
3637 LengthMode::Chars => s.chars().count(),
3638 LengthMode::Visual => s.width(),
3639 LengthMode::Bytes => s.len(),
3640 }
3641 }
3642
3643 /// Calculate effective line length
3644 ///
3645 /// Returns the actual display length of the line using the configured length mode.
3646 fn calculate_effective_length(&self, line: &str) -> usize {
3647 self.calculate_string_length(line)
3648 }
3649
3650 /// Calculate line length with inline link/image URLs removed.
3651 ///
3652 /// For each inline link `[text](url)` or image `` on the line,
3653 /// computes the "savings" from removing the URL portion (keeping only `[text]`
3654 /// or `![alt]`). Returns `effective_length - total_savings`.
3655 ///
3656 /// Handles nested constructs (e.g., `[](url)`) by only counting the
3657 /// outermost construct to avoid double-counting.
3658 fn length_without_inline_link_urls(
3659 &self,
3660 effective_length: usize,
3661 line_number: usize,
3662 ctx: &crate::lint_context::LintContext,
3663 ) -> usize {
3664 let line_range = ctx.line_index.line_content_range(line_number);
3665 let line_byte_end = line_range.end;
3666
3667 // Collect inline links/images on this line: (byte_offset, byte_end, text_only_display_len)
3668 let mut constructs: Vec<(usize, usize, usize)> = Vec::new();
3669
3670 // Binary search: links are sorted by byte_offset, so link.line is non-decreasing
3671 let link_start = ctx.links.partition_point(|l| l.line < line_number);
3672 for link in &ctx.links[link_start..] {
3673 if link.line != line_number {
3674 break;
3675 }
3676 if link.is_reference {
3677 continue;
3678 }
3679 if !matches!(link.link_type, LinkType::Inline) {
3680 continue;
3681 }
3682 if link.byte_end > line_byte_end {
3683 continue;
3684 }
3685 let text_only_len = 2 + self.calculate_string_length(&link.text);
3686 constructs.push((link.byte_offset, link.byte_end, text_only_len));
3687 }
3688
3689 let img_start = ctx.images.partition_point(|i| i.line < line_number);
3690 for image in &ctx.images[img_start..] {
3691 if image.line != line_number {
3692 break;
3693 }
3694 if image.is_reference {
3695 continue;
3696 }
3697 if !matches!(image.link_type, LinkType::Inline) {
3698 continue;
3699 }
3700 if image.byte_end > line_byte_end {
3701 continue;
3702 }
3703 let text_only_len = 3 + self.calculate_string_length(&image.alt_text);
3704 constructs.push((image.byte_offset, image.byte_end, text_only_len));
3705 }
3706
3707 if constructs.is_empty() {
3708 return effective_length;
3709 }
3710
3711 // Sort by byte offset to handle overlapping/nested constructs
3712 constructs.sort_by_key(|&(start, _, _)| start);
3713
3714 let mut total_savings: usize = 0;
3715 let mut last_end: usize = 0;
3716
3717 for (start, end, text_only_len) in &constructs {
3718 // Skip constructs nested inside a previously counted one
3719 if *start < last_end {
3720 continue;
3721 }
3722 // Full construct length in configured length mode
3723 let full_source = &ctx.content[*start..*end];
3724 let full_len = self.calculate_string_length(full_source);
3725 total_savings += full_len.saturating_sub(*text_only_len);
3726 last_end = *end;
3727 }
3728
3729 effective_length.saturating_sub(total_savings)
3730 }
3731}