rumdl_lib/rules/md013_line_length.rs
1/// Rule MD013: Line length
2///
3/// See [docs/md013.md](../../docs/md013.md) for full documentation, configuration, and examples.
4use crate::rule::{LintError, LintResult, LintWarning, Rule, RuleCategory, Severity};
5use crate::rule_config_serde::RuleConfig;
6use crate::utils::mkdocs_admonitions;
7use crate::utils::mkdocs_attr_list::is_standalone_attr_list;
8use crate::utils::mkdocs_snippets::is_snippet_block_delimiter;
9use crate::utils::mkdocs_tabs;
10use crate::utils::range_utils::LineIndex;
11use crate::utils::range_utils::calculate_excess_range;
12use crate::utils::regex_cache::{IMAGE_REF_PATTERN, LINK_REF_PATTERN, URL_PATTERN};
13use crate::utils::table_utils::TableUtils;
14use crate::utils::text_reflow::{
15 BlockquoteLineData, blockquote_continuation_style, dominant_blockquote_prefix, reflow_blockquote_content,
16 split_into_sentences,
17};
18use pulldown_cmark::LinkType;
19use toml;
20
21mod block_builder;
22mod helpers;
23pub mod md013_config;
24use crate::rules::md030_list_marker_space::MD030Config;
25use crate::utils::is_template_directive_only;
26use block_builder::{Block, BlockBuilder};
27use helpers::{
28 extract_list_marker_and_content, has_hard_break, is_github_alert_marker, is_horizontal_rule, is_html_only_line,
29 is_list_item, is_standalone_link_or_image_line, is_unwrappable_line, source_list_marker, split_into_segments,
30 trim_preserving_hard_break,
31};
32pub use md013_config::MD013Config;
33use md013_config::{LengthMode, ReflowMode};
34
35#[cfg(test)]
36mod tests;
37use unicode_width::UnicodeWidthStr;
38
39#[derive(Clone, Default)]
40pub struct MD013LineLength {
41 pub(crate) config: MD013Config,
42 /// MD030 list-marker spacing, applied when reflowing list items so the rewrite
43 /// uses the configured post-marker spacing rather than a hard-coded single
44 /// space. Defaults to MD030's defaults (a single space everywhere), which
45 /// reproduces the previous behaviour exactly. See [`MD030Config::expected_spaces`].
46 pub(crate) list_spacing: MD030Config,
47}
48
49/// Blockquote paragraph line collected for reflow, with original line index for range computation.
50struct CollectedBlockquoteLine {
51 line_idx: usize,
52 data: BlockquoteLineData,
53}
54
55impl MD013LineLength {
56 pub fn new(line_length: usize, code_blocks: bool, tables: bool, headings: bool, strict: bool) -> Self {
57 Self {
58 config: MD013Config {
59 line_length: crate::types::LineLength::new(line_length),
60 code_blocks,
61 code_spans: true,
62 tables,
63 headings,
64 paragraphs: true, // Default to true for backwards compatibility
65 blockquotes: true, // Default to true for backwards compatibility
66 strict,
67 stern: false,
68 heading_line_length: None,
69 code_block_line_length: None,
70 reflow: false,
71 reflow_mode: ReflowMode::default(),
72 length_mode: LengthMode::default(),
73 abbreviations: Vec::new(),
74 require_sentence_capital: true,
75 ignore_link_urls: true,
76 },
77 list_spacing: MD030Config::default(),
78 }
79 }
80
81 pub fn from_config_struct(config: MD013Config) -> Self {
82 Self {
83 config,
84 list_spacing: MD030Config::default(),
85 }
86 }
87
88 /// Return a clone with code block checking disabled.
89 /// Used for doc comment linting where code blocks are Rust code managed by rustfmt.
90 pub fn with_code_blocks_disabled(&self) -> Self {
91 let mut clone = self.clone();
92 clone.config.code_blocks = false;
93 clone
94 }
95
96 /// Convert MD013 LengthMode to text_reflow ReflowLengthMode
97 /// Normalized set of reference labels defined in the document.
98 ///
99 /// Passed to reflow so a bare shortcut reference (`[text]`) is treated as an
100 /// atomic link only when its label is actually defined; an undefined
101 /// bracketed run reflows as literal prose.
102 fn defined_reference_labels(ctx: &crate::lint_context::LintContext) -> std::collections::HashSet<String> {
103 ctx.reference_defs
104 .iter()
105 .map(|d| crate::utils::text_reflow::normalize_reference_label(&d.id))
106 .collect()
107 }
108
109 /// Build the reflow options shared by every MD013 fix path.
110 ///
111 /// `line_length` varies per call site (some subtract a list-marker or
112 /// blockquote prefix); every other field is derived uniformly from the
113 /// effective config and the document flavor. Callers that need a different
114 /// `max_list_continuation_indent` override it via struct update.
115 fn reflow_options(
116 ctx: &crate::lint_context::LintContext,
117 config: &MD013Config,
118 line_length: usize,
119 ) -> crate::utils::text_reflow::ReflowOptions {
120 crate::utils::text_reflow::ReflowOptions {
121 line_length,
122 break_on_sentences: true,
123 preserve_breaks: false,
124 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
125 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
126 abbreviations: config.abbreviations_for_reflow(),
127 length_mode: config.reflow_length_mode(),
128 attr_lists: ctx.flavor.supports_attr_lists(),
129 myst_roles: ctx.flavor.supports_myst_roles(),
130 require_sentence_capital: config.require_sentence_capital,
131 max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
132 Some(4)
133 } else {
134 None
135 },
136 defined_references: Some(Self::defined_reference_labels(ctx)),
137 }
138 }
139
140 fn should_ignore_line(
141 &self,
142 line: &str,
143 _lines: &[&str],
144 current_line: usize,
145 ctx: &crate::lint_context::LintContext,
146 ) -> bool {
147 if self.config.strict {
148 return false;
149 }
150
151 // Quick check for common patterns before expensive regex
152 let trimmed = line.trim();
153
154 // Only skip if the entire line is a URL (quick check first)
155 if (trimmed.starts_with("http://") || trimmed.starts_with("https://")) && URL_PATTERN.is_match(trimmed) {
156 return true;
157 }
158
159 // Only skip if the entire line is an image reference (quick check first)
160 if trimmed.starts_with("![") && trimmed.ends_with(']') && IMAGE_REF_PATTERN.is_match(trimmed) {
161 return true;
162 }
163
164 // Note: link reference definitions are handled as always-exempt (even in strict mode)
165 // in the main check loop, so they don't need to be checked here.
166
167 // Code blocks with long strings (only check if in code block)
168 if ctx.line_info(current_line + 1).is_some_and(|info| info.in_code_block)
169 && !trimmed.is_empty()
170 && !line.contains(' ')
171 && !line.contains('\t')
172 {
173 return true;
174 }
175
176 false
177 }
178
179 /// Check if rule should skip based on provided config (used for inline config support)
180 fn should_skip_with_config(&self, ctx: &crate::lint_context::LintContext, config: &MD013Config) -> bool {
181 // Skip if content is empty
182 if ctx.content.is_empty() {
183 return true;
184 }
185
186 // For sentence-per-line, semantic-line-breaks, or normalize mode, never skip based on line length
187 if config.reflow
188 && (config.reflow_mode == ReflowMode::SentencePerLine
189 || config.reflow_mode == ReflowMode::SemanticLineBreaks
190 || config.reflow_mode == ReflowMode::Normalize)
191 {
192 return false;
193 }
194
195 // Use the smallest applicable budget across line/heading/code-block
196 // contexts so a stricter context-specific limit doesn't get masked by
197 // the document-wide budget.
198 let min_limit = config.min_effective_line_length();
199 if min_limit.is_unlimited() {
200 return true;
201 }
202 let min_limit_bytes = min_limit.get();
203
204 // Quick check: if total content is shorter than the smallest line limit,
205 // definitely skip.
206 if ctx.content.len() <= min_limit_bytes {
207 return true;
208 }
209
210 // Skip if no line exceeds the smallest applicable limit.
211 !ctx.lines.iter().any(|line| line.byte_len > min_limit_bytes)
212 }
213
214 fn normalize_mode_needs_reflow<'a, I>(&self, lines: I, config: &MD013Config) -> bool
215 where
216 I: IntoIterator<Item = &'a str>,
217 {
218 let mut line_count = 0;
219 let check_length = !config.line_length.is_unlimited();
220
221 for line in lines {
222 line_count += 1;
223 if check_length && self.calculate_effective_length(line) > config.line_length.get() {
224 return true;
225 }
226 }
227
228 line_count > 1
229 }
230}
231
232impl Rule for MD013LineLength {
233 fn name(&self) -> &'static str {
234 "MD013"
235 }
236
237 fn description(&self) -> &'static str {
238 "Line length should not be excessive"
239 }
240
241 fn check(&self, ctx: &crate::lint_context::LintContext) -> LintResult {
242 // Use pre-parsed inline config from LintContext
243 let config_override = ctx.inline_config().get_rule_config("MD013");
244
245 // Apply configuration override if present
246 let effective_config = if let Some(json_config) = config_override {
247 if let Some(obj) = json_config.as_object() {
248 let mut config = self.config.clone();
249 if let Some(line_length) = obj.get("line_length").and_then(serde_json::Value::as_u64) {
250 config.line_length = crate::types::LineLength::new(line_length as usize);
251 }
252 if let Some(code_blocks) = obj.get("code_blocks").and_then(serde_json::Value::as_bool) {
253 config.code_blocks = code_blocks;
254 }
255 if let Some(code_spans) = obj.get("code_spans").and_then(serde_json::Value::as_bool) {
256 config.code_spans = code_spans;
257 }
258 if let Some(tables) = obj.get("tables").and_then(serde_json::Value::as_bool) {
259 config.tables = tables;
260 }
261 if let Some(headings) = obj.get("headings").and_then(serde_json::Value::as_bool) {
262 config.headings = headings;
263 }
264 if let Some(blockquotes) = obj.get("blockquotes").and_then(serde_json::Value::as_bool) {
265 config.blockquotes = blockquotes;
266 }
267 if let Some(strict) = obj.get("strict").and_then(serde_json::Value::as_bool) {
268 config.strict = strict;
269 }
270 if let Some(stern) = obj.get("stern").and_then(serde_json::Value::as_bool) {
271 config.stern = stern;
272 }
273 if let Some(v) = obj
274 .get("ignore_link_urls")
275 .or_else(|| obj.get("ignore-link-urls"))
276 .or_else(|| obj.get("semantic_link_understanding"))
277 .or_else(|| obj.get("semantic-link-understanding"))
278 .and_then(serde_json::Value::as_bool)
279 {
280 config.ignore_link_urls = v;
281 }
282 if let Some(reflow) = obj.get("reflow").and_then(serde_json::Value::as_bool) {
283 config.reflow = reflow;
284 }
285 if let Some(reflow_mode) = obj.get("reflow_mode").and_then(|v| v.as_str()) {
286 config.reflow_mode = match reflow_mode {
287 "default" => ReflowMode::Default,
288 "normalize" => ReflowMode::Normalize,
289 "sentence-per-line" => ReflowMode::SentencePerLine,
290 "semantic-line-breaks" => ReflowMode::SemanticLineBreaks,
291 _ => ReflowMode::default(),
292 };
293 }
294 config
295 } else {
296 self.config.clone()
297 }
298 } else {
299 self.config.clone()
300 };
301
302 // Fast early return using should_skip with EFFECTIVE config (after inline overrides)
303 // But don't skip if we're in reflow mode with Normalize or SentencePerLine
304 if self.should_skip_with_config(ctx, &effective_config)
305 && !(effective_config.reflow
306 && (effective_config.reflow_mode == ReflowMode::Normalize
307 || effective_config.reflow_mode == ReflowMode::SentencePerLine
308 || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
309 {
310 return Ok(Vec::new());
311 }
312
313 // Direct implementation without DocumentStructure
314 let mut warnings = Vec::new();
315
316 // Special handling: line_length = 0 means "no line length limit"
317 // Skip all line length checks, but still allow reflow if enabled
318 let skip_length_checks = effective_config.line_length.is_unlimited();
319
320 // Pre-filter lines that could be problematic to avoid processing all lines.
321 // Use the smallest applicable budget across line/heading/code-block contexts
322 // so candidates aren't dropped when a stricter context-specific budget applies.
323 let prefilter_limit = effective_config.min_effective_line_length();
324 let prefilter_skip = prefilter_limit.is_unlimited();
325 let mut candidate_lines = Vec::new();
326 if !skip_length_checks && !prefilter_skip {
327 for (line_idx, line_info) in ctx.lines.iter().enumerate() {
328 // Skip front matter - it should never be linted
329 if line_info.in_front_matter {
330 continue;
331 }
332
333 // Quick length check first
334 if line_info.byte_len > prefilter_limit.get() {
335 candidate_lines.push(line_idx);
336 }
337 }
338 }
339
340 // If no candidate lines and not in normalize or sentence-per-line mode, early return
341 if candidate_lines.is_empty()
342 && !(effective_config.reflow
343 && (effective_config.reflow_mode == ReflowMode::Normalize
344 || effective_config.reflow_mode == ReflowMode::SentencePerLine
345 || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
346 {
347 return Ok(warnings);
348 }
349
350 let lines = ctx.raw_lines();
351
352 // Whether a 1-indexed line is a heading. `LineInfo::heading` is an O(1)
353 // per-line field, so check it directly at each use site instead of
354 // materializing a full-document HashSet (an extra O(n) pass and
355 // allocation on a rule that runs on virtually every file).
356 let is_heading_line_num = |line_number: usize| -> bool {
357 line_number
358 .checked_sub(1)
359 .and_then(|idx| ctx.lines.get(idx))
360 .is_some_and(|line| line.heading.is_some())
361 };
362
363 // Use pre-computed table blocks from context
364 // We need this for both the table skip check AND the paragraphs check
365 let table_blocks = &ctx.table_blocks;
366 let mut table_lines_set = std::collections::HashSet::new();
367 for table in table_blocks {
368 table_lines_set.insert(table.header_line + 1);
369 table_lines_set.insert(table.delimiter_line + 1);
370 for &line in &table.content_lines {
371 table_lines_set.insert(line + 1);
372 }
373 }
374
375 // Process candidate lines for line length checks
376 'line_loop: for &line_idx in &candidate_lines {
377 let line_number = line_idx + 1;
378 let line = lines[line_idx];
379
380 // Calculate actual line length (used in warning messages)
381 let effective_length = self.calculate_effective_length(line);
382
383 // Pick the context-specific limit: heading > code-block > paragraph.
384 // Headings dominate over code-block context if a setext underline ever
385 // overlaps a fenced range (defensive — these are mutually exclusive in
386 // practice, but the explicit ordering documents intent).
387 let is_heading_line = is_heading_line_num(line_number);
388 let in_code_block = ctx.line_info(line_number).is_some_and(|info| info.in_code_block);
389 let line_limit = if is_heading_line {
390 effective_config.effective_heading_line_length().get()
391 } else if in_code_block {
392 effective_config.effective_code_block_line_length().get()
393 } else {
394 effective_config.line_length.get()
395 };
396
397 // A context-specific limit of 0 means "unlimited for this context".
398 if line_limit == 0 {
399 continue;
400 }
401
402 // Stern mode: like default, but the trailing-token forgiveness is
403 // disabled — a line with whitespace that exceeds the limit is a
404 // violation even if the excess is the final token. The "unwrappable"
405 // line exemption (single token, optionally prefixed by # or >) is
406 // still honored. Strict overrides stern entirely.
407 if effective_config.stern && !effective_config.strict && is_unwrappable_line(line) {
408 continue;
409 }
410
411 // Trailing-token forgiveness: only in default mode (not strict, not stern).
412 // If the line only exceeds the limit because of a long token at the end
413 // (URL, link chain, identifier), it passes. This matches markdownlint's
414 // behavior: line.replace(/\S*$/u, "#")
415 let check_length = if effective_config.strict || effective_config.stern {
416 effective_length
417 } else {
418 match line.rfind(char::is_whitespace) {
419 Some(pos) => {
420 let ws_char = line[pos..].chars().next().unwrap();
421 let prefix_end = pos + ws_char.len_utf8();
422 self.calculate_string_length(&line[..prefix_end]) + 1
423 }
424 None => 1, // No whitespace — entire line is a single token
425 }
426 };
427
428 // Skip lines where the check length is within the limit
429 if check_length <= line_limit {
430 continue;
431 }
432
433 // Ignore inline link/image URLs: suppress when excess comes entirely from inline URLs.
434 // Disabled by `strict` (all forgiveness off) and by `ignore_link_urls = false`
435 // (count link/image URLs toward the line length so such lines are flagged).
436 if !effective_config.strict && effective_config.ignore_link_urls {
437 let length_without_urls = self.length_without_inline_link_urls(effective_length, line_number, ctx);
438 if length_without_urls <= line_limit {
439 continue;
440 }
441 }
442
443 // Inline code spans cannot be wrapped, so reflow cannot shorten a line
444 // whose excess length is one. When code-span checking is disabled,
445 // suppress a violation that would fit once inline code spans are excluded.
446 if !effective_config.code_spans {
447 let code_span_width: usize = ctx
448 .code_spans()
449 .iter()
450 .filter(|span| span.line == line_number && span.end_line == line_number)
451 .map(|span| self.calculate_string_length(&ctx.content[span.byte_offset..span.byte_end]))
452 .sum();
453 if effective_length.saturating_sub(code_span_width) <= line_limit {
454 continue;
455 }
456 }
457
458 // Skip mkdocstrings and pymdown blocks (already handled by LintContext)
459 if ctx.lines[line_idx].in_mkdocstrings || ctx.lines[line_idx].in_pymdown_block {
460 continue;
461 }
462
463 // Skip MyST comments (% comment) — structural lines, not prose
464 if ctx.lines[line_idx].is_myst_comment {
465 continue;
466 }
467
468 // Link reference definitions are always exempt, even in strict mode.
469 // There's no way to shorten them without breaking the URL.
470 // Also check after stripping list markers, since list items may
471 // contain link ref defs as their content.
472 {
473 let trimmed = line.trim();
474 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
475 continue;
476 }
477 if is_list_item(trimmed) {
478 let (_, content) = extract_list_marker_and_content(trimmed);
479 let content_trimmed = content.trim();
480 if content_trimmed.starts_with('[')
481 && content_trimmed.contains("]:")
482 && LINK_REF_PATTERN.is_match(content_trimmed)
483 {
484 continue;
485 }
486 }
487 }
488
489 // Skip various block types efficiently
490 if !effective_config.strict {
491 // Lines whose only content is a link/image are exempt.
492 // After stripping list markers, blockquote markers, and emphasis,
493 // if only a link or image remains, there is no way to shorten it.
494 if is_standalone_link_or_image_line(line) {
495 continue;
496 }
497
498 // Lines consisting entirely of HTML tags are exempt.
499 // Badge lines, images with attributes, and similar inline HTML
500 // are long due to URLs in attributes and can't be meaningfully shortened.
501 if is_html_only_line(line) {
502 continue;
503 }
504
505 // Skip setext heading underlines
506 if !line.trim().is_empty() && line.trim().chars().all(|c| c == '=' || c == '-') {
507 continue;
508 }
509
510 // Skip block elements according to config flags
511 // The flags mean: true = check these elements, false = skip these elements
512 // So we skip when the flag is FALSE and the line is in that element type
513 if (!effective_config.headings && is_heading_line_num(line_number))
514 || (!effective_config.code_blocks
515 && ctx.line_info(line_number).is_some_and(|info| info.in_code_block))
516 || (!effective_config.tables && table_lines_set.contains(&line_number))
517 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
518 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
519 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
520 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
521 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
522 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
523 || ctx.line_info(line_number).is_some_and(|info| info.in_pymdown_block)
524 {
525 continue;
526 }
527
528 // Check if this is a paragraph/regular text line
529 // If paragraphs = false, skip lines that are NOT in special blocks
530 // Blockquote content is treated as paragraph text, so it's not
531 // included in the special blocks list here.
532 if !effective_config.paragraphs {
533 let is_special_block = is_heading_line_num(line_number)
534 || ctx.line_info(line_number).is_some_and(|info| info.in_code_block)
535 || table_lines_set.contains(&line_number)
536 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
537 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
538 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
539 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
540 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
541 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
542 || ctx
543 .line_info(line_number)
544 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container);
545
546 // Skip regular paragraph text when paragraphs = false
547 if !is_special_block {
548 continue;
549 }
550 }
551
552 // Skip blockquote lines when blockquotes = false.
553 // Also skip lazy continuation lines that belong to a blockquote
554 // (lines without `>` prefix that follow a blockquote line).
555 if !effective_config.blockquotes {
556 if ctx.lines[line_number - 1].blockquote.is_some() {
557 continue;
558 }
559 // Check for lazy continuation: scan backwards through
560 // non-blank lines to find if this paragraph started with
561 // a blockquote marker
562 if !line.trim().is_empty() {
563 let mut scan = line_number.saturating_sub(2);
564 loop {
565 if ctx.lines[scan].blockquote.is_some() {
566 // Found a blockquote ancestor — this is a lazy continuation
567 continue 'line_loop;
568 }
569 if lines[scan].trim().is_empty() || scan == 0 {
570 break;
571 }
572 scan -= 1;
573 }
574 }
575 }
576
577 // Skip lines that are only a URL, image ref, or link ref
578 if self.should_ignore_line(line, lines, line_idx, ctx) {
579 continue;
580 }
581 }
582
583 // In sentence-per-line mode, check if this is a single long sentence
584 // If so, emit a warning without a fix (user must manually rephrase)
585 if effective_config.reflow_mode == ReflowMode::SentencePerLine {
586 let sentences = split_into_sentences(line.trim());
587 if sentences.len() == 1 {
588 // Single sentence that's too long - warn but don't auto-fix
589 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
590
591 let (start_line, start_col, end_line, end_col) =
592 calculate_excess_range(line_number, line, line_limit);
593
594 warnings.push(LintWarning {
595 rule_name: Some(self.name().to_string()),
596 message,
597 line: start_line,
598 column: start_col,
599 end_line,
600 end_column: end_col,
601 severity: Severity::Warning,
602 fix: None, // No auto-fix for long single sentences
603 });
604 continue;
605 }
606 // Multiple sentences will be handled by paragraph-based reflow
607 continue;
608 }
609
610 // In semantic-line-breaks mode, skip per-line checks —
611 // all reflow is handled at the paragraph level with cascading splits
612 if effective_config.reflow_mode == ReflowMode::SemanticLineBreaks {
613 continue;
614 }
615
616 // Don't provide fix for individual lines when reflow is enabled
617 // Paragraph-based fixes will be handled separately
618 let fix = None;
619
620 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
621
622 // Calculate precise character range for the excess portion
623 let (start_line, start_col, end_line, end_col) = calculate_excess_range(line_number, line, line_limit);
624
625 warnings.push(LintWarning {
626 rule_name: Some(self.name().to_string()),
627 message,
628 line: start_line,
629 column: start_col,
630 end_line,
631 end_column: end_col,
632 severity: Severity::Warning,
633 fix,
634 });
635 }
636
637 // If reflow is enabled, generate paragraph-based fixes
638 if effective_config.reflow {
639 let paragraph_warnings = self.generate_paragraph_fixes(ctx, &effective_config, lines);
640 // Merge paragraph warnings with line warnings, removing duplicates
641 for pw in paragraph_warnings {
642 // Remove any line warnings that overlap with this paragraph
643 warnings.retain(|w| w.line < pw.line || w.line > pw.end_line);
644 warnings.push(pw);
645 }
646 }
647
648 Ok(warnings)
649 }
650
651 fn fix(&self, ctx: &crate::lint_context::LintContext) -> Result<String, LintError> {
652 // For CLI usage, apply fixes from warnings
653 // LSP will use the warning-based fixes directly
654 let warnings = self.check(ctx)?;
655 let warnings =
656 crate::utils::fix_utils::filter_warnings_by_inline_config(warnings, ctx.inline_config(), self.name());
657
658 // If there are no fixes, return content unchanged
659 if !warnings.iter().any(|w| w.fix.is_some()) {
660 return Ok(ctx.content.to_string());
661 }
662
663 // Apply warning-based fixes
664 crate::utils::fix_utils::apply_warning_fixes(ctx.content, &warnings)
665 .map_err(|e| LintError::FixFailed(format!("Failed to apply fixes: {e}")))
666 }
667
668 fn as_any(&self) -> &dyn std::any::Any {
669 self
670 }
671
672 fn category(&self) -> RuleCategory {
673 RuleCategory::Whitespace
674 }
675
676 fn should_skip(&self, ctx: &crate::lint_context::LintContext) -> bool {
677 self.should_skip_with_config(ctx, &self.config)
678 }
679
680 fn default_config_section(&self) -> Option<(String, toml::Value)> {
681 let table = crate::rule_config_serde::config_schema_table(&MD013Config::default())?;
682 if table.is_empty() {
683 None
684 } else {
685 Some((MD013Config::RULE_NAME.to_string(), toml::Value::Table(table)))
686 }
687 }
688
689 fn config_aliases(&self) -> Option<std::collections::HashMap<String, String>> {
690 let mut aliases = std::collections::HashMap::new();
691 aliases.insert("enable_reflow".to_string(), "reflow".to_string());
692 aliases.insert("strict_sentences".to_string(), "require-sentence-capital".to_string());
693 aliases.insert("strict-sentences".to_string(), "require-sentence-capital".to_string());
694 Some(aliases)
695 }
696
697 fn from_config(config: &crate::config::Config) -> Box<dyn Rule>
698 where
699 Self: Sized,
700 {
701 let mut rule_config = crate::rule_config_serde::load_rule_config::<MD013Config>(config);
702 // Use global line_length if rule-specific config still has default value
703 if rule_config.line_length.get() == 80 {
704 rule_config.line_length = config.global.line_length;
705 }
706 let mut rule = Self::from_config_struct(rule_config);
707 // Pull list-marker spacing from MD030 (via the shared serde config loader)
708 // so reflow rewrites list items with the configured spacing rather than a
709 // hard-coded single space.
710 rule.list_spacing = crate::rule_config_serde::load_rule_config::<MD030Config>(config);
711 Box::new(rule)
712 }
713}
714
715impl MD013LineLength {
716 /// True when `line_num` (1-based) sits inside a structure whose lines must be
717 /// preserved verbatim (code block, front matter, HTML/JSX/MDX block, MkDocs
718 /// container, div marker, ...). Used to keep blockquote reflow from touching
719 /// quoted-looking text embedded in such structures.
720 fn line_in_verbatim_context(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
721 ctx.line_info(line_num).is_some_and(|info| {
722 info.in_code_block
723 || info.in_front_matter
724 || info.in_html_block
725 || info.in_html_comment
726 || info.in_esm_block
727 || info.in_jsx_expression
728 || info.in_jsx_block
729 || info.in_mdx_comment
730 || info.in_mkdocstrings
731 || info.in_pymdown_block
732 || info.in_mkdocs_container()
733 || info.is_div_marker
734 })
735 }
736
737 fn is_blockquote_content_boundary(
738 &self,
739 content: &str,
740 line_num: usize,
741 ctx: &crate::lint_context::LintContext,
742 strict: bool,
743 ) -> bool {
744 let trimmed = content.trim();
745
746 trimmed.is_empty()
747 || self.line_in_verbatim_context(line_num, ctx)
748 || trimmed.starts_with('#')
749 || trimmed.starts_with("```")
750 || trimmed.starts_with("~~~")
751 || trimmed.starts_with('>')
752 || TableUtils::is_potential_table_row(content)
753 || is_list_item(trimmed)
754 || is_horizontal_rule(content)
755 || (trimmed.starts_with('[') && content.contains("]:"))
756 || is_template_directive_only(content)
757 || is_standalone_attr_list(content)
758 || is_snippet_block_delimiter(content)
759 || is_github_alert_marker(trimmed)
760 || is_html_only_line(content)
761 // A standalone link/image line is exempt from MD013 (non-strict mode),
762 // so it must end the blockquote paragraph rather than be absorbed into
763 // it, mirroring the top-level paragraph reflow boundary.
764 || (!strict && is_standalone_link_or_image_line(content))
765 }
766
767 #[allow(clippy::too_many_arguments)]
768 fn generate_blockquote_paragraph_fix(
769 &self,
770 ctx: &crate::lint_context::LintContext,
771 config: &MD013Config,
772 lines: &[&str],
773 line_index: &LineIndex,
774 start_idx: usize,
775 line_ending: &str,
776 // Extra indent (spaces) to prepend to the emitted `>` prefix so a blockquote
777 // nested in a list item moves with its parent's widened marker. Zero unless a
778 // non-default MD030 widened an ancestor list item.
779 ancestor_shift: isize,
780 ) -> (Option<LintWarning>, usize) {
781 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
782 return (None, start_idx + 1);
783 };
784 let target_level = start_bq.nesting_level;
785
786 let mut collected: Vec<CollectedBlockquoteLine> = Vec::new();
787 let mut i = start_idx;
788
789 while i < lines.len() {
790 if !collected.is_empty() && has_hard_break(&collected[collected.len() - 1].data.content) {
791 break;
792 }
793
794 let line_num = i + 1;
795 if line_num > ctx.lines.len() {
796 break;
797 }
798
799 if lines[i].trim().is_empty() {
800 break;
801 }
802
803 let line_bq = ctx.lines[i].blockquote.as_deref();
804 if let Some(bq) = line_bq {
805 if bq.nesting_level != target_level {
806 break;
807 }
808
809 if self.is_blockquote_content_boundary(&bq.content, line_num, ctx, config.strict) {
810 break;
811 }
812
813 collected.push(CollectedBlockquoteLine {
814 line_idx: i,
815 data: BlockquoteLineData::explicit(trim_preserving_hard_break(&bq.content), bq.prefix.clone()),
816 });
817 i += 1;
818 continue;
819 }
820
821 let lazy_content = lines[i].trim_start();
822 if self.is_blockquote_content_boundary(lazy_content, line_num, ctx, config.strict) {
823 break;
824 }
825
826 collected.push(CollectedBlockquoteLine {
827 line_idx: i,
828 data: BlockquoteLineData::lazy(trim_preserving_hard_break(lazy_content)),
829 });
830 i += 1;
831 }
832
833 if collected.is_empty() {
834 return (None, start_idx + 1);
835 }
836
837 let next_idx = i;
838 let paragraph_start = collected[0].line_idx;
839 let end_line = collected[collected.len() - 1].line_idx;
840 let line_data: Vec<BlockquoteLineData> = collected.iter().map(|l| l.data.clone()).collect();
841 let paragraph_text = line_data
842 .iter()
843 .map(|d| d.content.as_str())
844 .collect::<Vec<_>>()
845 .join(" ");
846
847 let contains_definition_list = line_data
848 .iter()
849 .any(|d| crate::utils::is_definition_list_item(&d.content));
850 if contains_definition_list {
851 return (None, next_idx);
852 }
853
854 let contains_snippets = line_data.iter().any(|d| is_snippet_block_delimiter(&d.content));
855 if contains_snippets {
856 return (None, next_idx);
857 }
858
859 let needs_reflow = match config.reflow_mode {
860 ReflowMode::Normalize => {
861 self.normalize_mode_needs_reflow(line_data.iter().map(|d| d.content.as_str()), config)
862 }
863 ReflowMode::SentencePerLine => {
864 let sentences = split_into_sentences(¶graph_text);
865 sentences.len() > 1 || line_data.len() > 1
866 }
867 ReflowMode::SemanticLineBreaks => {
868 let sentences = split_into_sentences(¶graph_text);
869 sentences.len() > 1
870 || line_data.len() > 1
871 || collected
872 .iter()
873 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get())
874 }
875 ReflowMode::Default => collected
876 .iter()
877 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get()),
878 };
879
880 if !needs_reflow {
881 return (None, next_idx);
882 }
883
884 let fallback_prefix = start_bq.prefix.clone();
885 let explicit_prefix = dominant_blockquote_prefix(&line_data, &fallback_prefix);
886 // Shift the whole quote right to track a widened parent list item's content
887 // column (only widening matters for nesting; a narrowed parent leaves the quote
888 // harmlessly over-indented, which MD027/MD030 tidy).
889 let explicit_prefix = if ancestor_shift > 0 {
890 format!("{}{explicit_prefix}", " ".repeat(ancestor_shift as usize))
891 } else {
892 explicit_prefix
893 };
894 let continuation_style = blockquote_continuation_style(&line_data);
895
896 let reflow_line_length = if config.line_length.is_unlimited() {
897 usize::MAX
898 } else {
899 config
900 .line_length
901 .get()
902 .saturating_sub(self.calculate_string_length(&explicit_prefix))
903 .max(1)
904 };
905
906 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
907
908 let reflowed_with_style =
909 reflow_blockquote_content(&line_data, &explicit_prefix, continuation_style, &reflow_options);
910
911 if reflowed_with_style.is_empty() {
912 return (None, next_idx);
913 }
914
915 let reflowed_text = reflowed_with_style.join(line_ending);
916
917 let start_range = line_index.whole_line_range(paragraph_start + 1);
918 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
919 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
920 } else {
921 line_index.whole_line_range(end_line + 1)
922 };
923 let byte_range = start_range.start..end_range.end;
924
925 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
926 format!("{reflowed_text}{line_ending}")
927 } else {
928 reflowed_text
929 };
930
931 let original_text = &ctx.content[byte_range.clone()];
932 if original_text == replacement {
933 return (None, next_idx);
934 }
935
936 let (warning_line, warning_end_line) = match config.reflow_mode {
937 ReflowMode::Normalize => (paragraph_start + 1, end_line + 1),
938 ReflowMode::SentencePerLine | ReflowMode::SemanticLineBreaks => (paragraph_start + 1, end_line + 1),
939 ReflowMode::Default => {
940 let violating_line = collected
941 .iter()
942 .find(|line| self.calculate_effective_length(lines[line.line_idx]) > config.line_length.get())
943 .map_or(paragraph_start + 1, |line| line.line_idx + 1);
944 (violating_line, violating_line)
945 }
946 };
947
948 let warning = LintWarning {
949 rule_name: Some(self.name().to_string()),
950 message: match config.reflow_mode {
951 ReflowMode::Normalize => format!(
952 "Paragraph could be normalized to use line length of {} characters",
953 config.line_length.get()
954 ),
955 ReflowMode::SentencePerLine => {
956 let num_sentences = split_into_sentences(¶graph_text).len();
957 if line_data.len() == 1 {
958 format!("Line contains {num_sentences} sentences (one sentence per line required)")
959 } else {
960 let num_lines = line_data.len();
961 format!(
962 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
963 )
964 }
965 }
966 ReflowMode::SemanticLineBreaks => {
967 let num_sentences = split_into_sentences(¶graph_text).len();
968 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
969 }
970 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
971 },
972 line: warning_line,
973 column: 1,
974 end_line: warning_end_line,
975 end_column: lines[warning_end_line.saturating_sub(1)].chars().count() + 1,
976 severity: Severity::Warning,
977 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
978 };
979
980 (Some(warning), next_idx)
981 }
982
983 /// Reflow a single list item that lives inside a blockquote.
984 ///
985 /// The blockquote paragraph reflow treats a list marker as a content boundary,
986 /// so `> - long item ...` is never wrapped even though the identical top-level
987 /// item is. This handles that case: it collects one tight, prose-only list item
988 /// at the starting blockquote level, reflows the item body to the configured
989 /// width, and re-emits it with the blockquote prefix preserved and continuation
990 /// lines aligned under the list content.
991 ///
992 /// Sibling and nested list items end collection and are reflowed independently
993 /// by the caller's outer loop (a nested item carries its indent folded into the
994 /// blockquote prefix, so it reflows correctly on its own). Items that are not
995 /// simple tight prose - those embedding a code block, table, fence, or hard
996 /// break - are left untouched (`None`) and the cursor still advances past the
997 /// whole item so its inner lines are never reprocessed as loose prose.
998 #[allow(clippy::too_many_arguments)]
999 fn generate_blockquote_list_item_fix(
1000 &self,
1001 ctx: &crate::lint_context::LintContext,
1002 config: &MD013Config,
1003 lines: &[&str],
1004 line_index: &LineIndex,
1005 start_idx: usize,
1006 line_ending: &str,
1007 // Extra indent (spaces) to prepend to the emitted `>` prefix when this quoted
1008 // list lives inside a list item whose marker widened. Zero unless a non-default
1009 // MD030 widened an ancestor list item.
1010 ancestor_shift: isize,
1011 ) -> (Option<LintWarning>, usize) {
1012 use crate::utils::blockquote::effective_indent_in_blockquote;
1013
1014 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
1015 return (None, start_idx + 1);
1016 };
1017
1018 // A `>`-prefixed line can be marked as a blockquote even inside a fenced code
1019 // block (or other verbatim structure); such content must never be reflowed.
1020 if self.line_in_verbatim_context(start_idx + 1, ctx) {
1021 return (None, start_idx + 1);
1022 }
1023
1024 let target_level = start_bq.nesting_level;
1025
1026 // The marker line carries the canonical blockquote prefix: its content begins
1027 // with the list marker, so no list indent has been folded into the prefix.
1028 // Track a widened parent list item's content column so the quote stays nested
1029 // (only widening can detach it; narrowing just over-indents).
1030 let bq_prefix = if ancestor_shift > 0 {
1031 format!("{}{}", " ".repeat(ancestor_shift as usize), start_bq.prefix)
1032 } else {
1033 start_bq.prefix.clone()
1034 };
1035
1036 // A thematic break opens with what looks like a bullet marker (`- - -`).
1037 // It is not a list item, and reflowing it as prose destroys the break.
1038 // The top-level reflow path applies the same exemption.
1039 if is_horizontal_rule(&start_bq.content) {
1040 return (None, start_idx + 1);
1041 }
1042
1043 let (marker, first_body) = extract_list_marker_and_content(&start_bq.content);
1044 if marker.is_empty() {
1045 return (None, start_idx + 1);
1046 }
1047 let marker_width = marker.chars().count();
1048
1049 // Continuation lines of a checkbox item align under the bullet+checkbox, but
1050 // are recognized from the bullet width, matching the top-level list reflow.
1051 let base_marker_width = ["[ ] ", "[x] ", "[X] "]
1052 .iter()
1053 .find_map(|cb| marker.find(*cb))
1054 .unwrap_or(marker_width);
1055
1056 // Collect the item: the marker line plus its tight prose continuation lines.
1057 // `end_idx` always tracks the last consumed line so the cursor advances past
1058 // the entire item, even when it turns out to be too complex to reflow safely.
1059 let first_piece = trim_preserving_hard_break(&first_body);
1060 let mut simple = !has_hard_break(&first_piece);
1061 let mut body_pieces: Vec<String> = vec![first_piece];
1062 let mut end_idx = start_idx;
1063 let mut i = start_idx + 1;
1064
1065 while i < lines.len() {
1066 let Some(bq) = ctx.lines[i].blockquote.as_deref() else {
1067 // A blank line ends the item.
1068 if lines[i].trim().is_empty() {
1069 break;
1070 }
1071 // A lazy continuation (no `>` marker) is too ambiguous to reflow
1072 // safely. Consume the whole lazy run into this item's span and leave
1073 // the item untouched, so the caller does not reflow the continuation
1074 // in isolation and leave the marker line partially fixed.
1075 simple = false;
1076 while i < lines.len() && ctx.lines[i].blockquote.is_none() && !lines[i].trim().is_empty() {
1077 end_idx = i;
1078 i += 1;
1079 }
1080 break;
1081 };
1082 if bq.nesting_level != target_level {
1083 break;
1084 }
1085
1086 let content = bq.content.as_str();
1087 if content.trim().is_empty() {
1088 // Blank quoted line ends the tight paragraph. A following indented
1089 // paragraph (loose item) is reflowed on its own by the prose path.
1090 break;
1091 }
1092
1093 let eff_indent = effective_indent_in_blockquote(lines[i], target_level, 0);
1094 if eff_indent < base_marker_width {
1095 // Dedented: a sibling list item or text outside this item. Stop here
1096 // and let the outer loop classify it.
1097 break;
1098 }
1099 if is_list_item(content) {
1100 // A nested list item: its own item, handled independently.
1101 break;
1102 }
1103
1104 // An embedded structure (code block, table, fence, nested quote, ...)
1105 // means the item is not simple prose: keep consuming so the cursor clears
1106 // the whole structure, but do not produce a fix.
1107 if self.is_blockquote_content_boundary(content, i + 1, ctx, config.strict) {
1108 simple = false;
1109 }
1110
1111 let piece = trim_preserving_hard_break(content);
1112 if has_hard_break(&piece) {
1113 simple = false;
1114 }
1115 body_pieces.push(piece);
1116 end_idx = i;
1117 i += 1;
1118 }
1119
1120 let next_idx = end_idx + 1;
1121
1122 if !simple {
1123 return (None, next_idx);
1124 }
1125
1126 let exceeds_limit =
1127 || (start_idx..=end_idx).any(|idx| self.calculate_effective_length(lines[idx]) > config.line_length.get());
1128 let body_text = body_pieces.join(" ");
1129 let body_text = body_text.trim();
1130
1131 // Some bodies cannot be shortened and must stay verbatim, matching the
1132 // exemptions the top-level list reflow applies: link reference definitions
1133 // always, and (in non-strict mode) standalone links/images and HTML-only
1134 // lines. Reflowing a link reference definition would split it after the
1135 // colon/URL and break the definition.
1136 let is_link_ref_def =
1137 body_text.starts_with('[') && body_text.contains("]:") && LINK_REF_PATTERN.is_match(body_text);
1138 let raw_marker_line = lines[start_idx];
1139 let body_is_unwrappable = is_link_ref_def
1140 || (!config.strict && is_standalone_link_or_image_line(raw_marker_line))
1141 || (!config.strict && is_html_only_line(raw_marker_line));
1142 if body_is_unwrappable {
1143 return (None, next_idx);
1144 }
1145
1146 let needs_reflow = match config.reflow_mode {
1147 ReflowMode::Normalize => body_pieces.len() > 1 || exceeds_limit(),
1148 ReflowMode::Default => exceeds_limit(),
1149 ReflowMode::SentencePerLine => split_into_sentences(body_text).len() > 1 || body_pieces.len() > 1,
1150 ReflowMode::SemanticLineBreaks => split_into_sentences(body_text).len() > 1 || exceeds_limit(),
1151 };
1152 if !needs_reflow {
1153 return (None, next_idx);
1154 }
1155
1156 // Apply MD030 list-marker spacing in the spacing-normalizing modes, mirroring
1157 // the top-level list reflow: derive the post-marker spacing from MD030 and let
1158 // the continuation indent follow the resulting content width. Default MD030 (a
1159 // single space) leaves the marker unchanged. MkDocs keeps its rigid indent.
1160 let (marker, marker_width) = if matches!(config.reflow_mode, ReflowMode::Default | ReflowMode::Normalize)
1161 && !ctx.flavor.requires_strict_list_indent()
1162 {
1163 let is_ordered = marker.starts_with(|c: char| c.is_ascii_digit());
1164 // Bullet/number portion only (e.g. `-`, `1.`); the checkbox is content.
1165 let bullet = marker.split(' ').next().unwrap_or("");
1166 let bullet_len = bullet.chars().count();
1167 let checkbox_tail = &marker[base_marker_width..];
1168 // Decide single- vs multi-line spacing from the rewritten shape. This path
1169 // only handles a single tight prose paragraph (structural or multi-paragraph
1170 // items set `simple = false` and bail out above), so the emitted item stays
1171 // multi-line solely when the joined body wraps past one line. A multi-line
1172 // *source* that collapses onto the marker line must use MD030's single-line
1173 // spacing, matching the top-level reflow. The wrap test measures at the
1174 // single-line content column so it does not depend on the spacing chosen here.
1175 let single_col = self.calculate_string_length(&bq_prefix)
1176 + bullet_len
1177 + self.list_spacing.expected_spaces(is_ordered, false, bullet_len)
1178 + checkbox_tail.chars().count();
1179 let is_multi = !body_text.is_empty()
1180 && self.calculate_effective_length(&format!("{}{body_text}", " ".repeat(single_col)))
1181 > config.line_length.effective_limit();
1182 let spaces = self.list_spacing.expected_spaces(is_ordered, is_multi, bullet_len);
1183 let new_marker = format!("{bullet}{}{checkbox_tail}", " ".repeat(spaces));
1184 let width = new_marker.chars().count();
1185 (new_marker, width)
1186 } else {
1187 (marker, marker_width)
1188 };
1189
1190 let prefix_width = self.calculate_string_length(&bq_prefix) + self.calculate_string_length(&marker);
1191 let reflow_line_length = if config.line_length.is_unlimited() {
1192 usize::MAX
1193 } else {
1194 config.line_length.get().saturating_sub(prefix_width).max(1)
1195 };
1196
1197 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
1198
1199 let reflowed = crate::utils::text_reflow::reflow_line(body_text, &reflow_options);
1200 if reflowed.is_empty() {
1201 return (None, next_idx);
1202 }
1203
1204 let continuation_indent = " ".repeat(marker_width);
1205 let reflowed_text = reflowed
1206 .iter()
1207 .enumerate()
1208 .map(|(idx, line)| {
1209 if idx == 0 {
1210 format!("{bq_prefix}{marker}{line}")
1211 } else {
1212 format!("{bq_prefix}{continuation_indent}{line}")
1213 }
1214 })
1215 .collect::<Vec<_>>()
1216 .join(line_ending);
1217
1218 let start_range = line_index.whole_line_range(start_idx + 1);
1219 let end_range = if end_idx == lines.len() - 1 && !ctx.content.ends_with('\n') {
1220 line_index.line_text_range(end_idx + 1, 1, lines[end_idx].len() + 1)
1221 } else {
1222 line_index.whole_line_range(end_idx + 1)
1223 };
1224 let byte_range = start_range.start..end_range.end;
1225
1226 let replacement = if end_idx < lines.len() - 1 || ctx.content.ends_with('\n') {
1227 format!("{reflowed_text}{line_ending}")
1228 } else {
1229 reflowed_text
1230 };
1231
1232 let original_text = &ctx.content[byte_range.clone()];
1233 if original_text == replacement {
1234 return (None, next_idx);
1235 }
1236
1237 let message = match config.reflow_mode {
1238 ReflowMode::Normalize => format!(
1239 "Paragraph could be normalized to use line length of {} characters",
1240 config.line_length.get()
1241 ),
1242 ReflowMode::SentencePerLine => {
1243 let num_sentences = split_into_sentences(body_text).len();
1244 format!("List item should have one sentence per line (found {num_sentences} sentences)")
1245 }
1246 ReflowMode::SemanticLineBreaks => {
1247 let num_sentences = split_into_sentences(body_text).len();
1248 format!("List item should use semantic line breaks ({num_sentences} sentences)")
1249 }
1250 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
1251 };
1252
1253 let warning = LintWarning {
1254 rule_name: Some(self.name().to_string()),
1255 message,
1256 line: start_idx + 1,
1257 column: 1,
1258 end_line: end_idx + 1,
1259 end_column: lines[end_idx].chars().count() + 1,
1260 severity: Severity::Warning,
1261 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1262 };
1263
1264 (Some(warning), next_idx)
1265 }
1266
1267 /// Generate paragraph-based fixes
1268 fn generate_paragraph_fixes(
1269 &self,
1270 ctx: &crate::lint_context::LintContext,
1271 config: &MD013Config,
1272 lines: &[&str],
1273 ) -> Vec<LintWarning> {
1274 let mut warnings = Vec::new();
1275 let line_index = LineIndex::new(ctx.content);
1276
1277 // Detect the content's line ending style to preserve it in replacements.
1278 // The LSP receives content from editors which may use CRLF (Windows).
1279 // Replacements must match the original line endings to avoid false positives.
1280 let line_ending = crate::utils::line_ending::detect_line_ending(ctx.content);
1281
1282 // Ancestor list-item indent shifts, innermost last. When a reflowed parent's
1283 // marker widens under a non-default MD030 (e.g. ul-multi = 3 moves the parent's
1284 // content from column 2 to 4), its nested list/blockquote children are reflowed
1285 // independently and would otherwise keep their original indent — leaving them
1286 // under the parent's new content column, where a CommonMark parser reparses them
1287 // as siblings rather than children. Each frame is (normalized marker width,
1288 // cumulative shift applied to that item's content column); a descendant adds its
1289 // innermost open ancestor's shift to its own indent so the whole subtree moves
1290 // together. With a default MD030 and no marker padding every shift is 0, so this
1291 // is inert and the output is byte-identical.
1292 let mut list_shift_stack: Vec<(usize, isize)> = Vec::new();
1293
1294 let mut i = 0;
1295 while i < lines.len() {
1296 let line_num = i + 1;
1297
1298 // Close ancestor frames whose list item has ended at this line: a non-blank
1299 // line indented less than the frame's source content column is no longer
1300 // inside that item. Blank lines alone don't close a (loose) list item.
1301 if !list_shift_stack.is_empty()
1302 && let Some(info) = ctx.lines.get(i)
1303 && !info.is_blank
1304 {
1305 while let Some(&(content_column, _)) = list_shift_stack.last() {
1306 if info.indent < content_column {
1307 list_shift_stack.pop();
1308 } else {
1309 break;
1310 }
1311 }
1312 }
1313
1314 // Handle blockquote paragraphs with style-preserving reflow.
1315 // Skip blockquotes when blockquotes=false or paragraphs=false
1316 if line_num > 0 && line_num <= ctx.lines.len() && ctx.lines[line_num - 1].blockquote.is_some() {
1317 if !config.blockquotes || !config.paragraphs {
1318 // Skip past all blockquote lines (explicit and lazy continuations).
1319 // A lazy continuation is a non-blank line without `>` that follows
1320 // a blockquote line and isn't a structural element.
1321 let mut saw_explicit_bq = false;
1322 while i < lines.len() && i < ctx.lines.len() {
1323 if ctx.lines[i].blockquote.is_some() {
1324 saw_explicit_bq = true;
1325 i += 1;
1326 } else if saw_explicit_bq
1327 && !lines[i].trim().is_empty()
1328 && !lines[i].trim_start().starts_with('#')
1329 && !lines[i].trim_start().starts_with('>')
1330 {
1331 // Lazy continuation of preceding blockquote
1332 i += 1;
1333 } else {
1334 break;
1335 }
1336 }
1337 continue;
1338 }
1339 // A blockquote nested in a list item moves with its parent when the
1340 // parent's marker widens (see `list_shift_stack`); pass that shift so the
1341 // emitted `>` prefix lands under the parent's new content column instead
1342 // of detaching into a sibling.
1343 let ancestor_shift = list_shift_stack.last().map_or(0isize, |&(_, shift)| shift);
1344 // A list item inside the blockquote needs list-aware reflow (marker +
1345 // continuation indent); plain prose goes through the paragraph path.
1346 let is_bq_list_item = ctx.lines[i]
1347 .blockquote
1348 .as_deref()
1349 .is_some_and(|bq| is_list_item(&bq.content));
1350 let (warning, next_idx) = if is_bq_list_item {
1351 self.generate_blockquote_list_item_fix(
1352 ctx,
1353 config,
1354 lines,
1355 &line_index,
1356 i,
1357 line_ending,
1358 ancestor_shift,
1359 )
1360 } else {
1361 self.generate_blockquote_paragraph_fix(
1362 ctx,
1363 config,
1364 lines,
1365 &line_index,
1366 i,
1367 line_ending,
1368 ancestor_shift,
1369 )
1370 };
1371 if let Some(warning) = warning {
1372 warnings.push(warning);
1373 }
1374 i = next_idx;
1375 continue;
1376 }
1377
1378 // Skip special structures (but NOT MkDocs containers - those get special handling)
1379 let should_skip_due_to_line_info = ctx.line_info(line_num).is_some_and(|info| {
1380 info.in_code_block
1381 || info.in_front_matter
1382 || info.in_html_block
1383 || info.in_html_comment
1384 || info.in_esm_block
1385 || info.in_jsx_expression
1386 || info.in_jsx_block
1387 || info.in_mdx_comment
1388 || info.in_mkdocstrings
1389 || info.in_pymdown_block
1390 });
1391
1392 // Skip link reference definitions but NOT footnote definitions.
1393 // Footnote definitions (`[^id]: prose`) contain reflowable text,
1394 // while link reference definitions (`[ref]: URL`) contain URLs
1395 // that cannot be shortened.
1396 let is_link_ref_def =
1397 lines[i].trim().starts_with('[') && !lines[i].trim().starts_with("[^") && lines[i].contains("]:");
1398
1399 if should_skip_due_to_line_info
1400 || lines[i].trim().starts_with('#')
1401 || TableUtils::is_potential_table_row(lines[i])
1402 || lines[i].trim().is_empty()
1403 || is_horizontal_rule(lines[i])
1404 || is_template_directive_only(lines[i])
1405 || is_link_ref_def
1406 || ctx.line_info(line_num).is_some_and(|info| info.is_div_marker)
1407 || is_html_only_line(lines[i])
1408 || (!config.strict && is_standalone_link_or_image_line(lines[i]))
1409 {
1410 i += 1;
1411 continue;
1412 }
1413
1414 // Handle footnote definitions: `[^id]: prose text that can be reflowed`
1415 // Supports multi-paragraph footnotes with code blocks, blockquotes,
1416 // tables, and lists preserved verbatim.
1417 // Validate structure: must start with `[^`, contain `]:`, and the ID
1418 // must not contain `[` or `]` (prevents false matches on nested brackets)
1419 if lines[i].trim().starts_with("[^") && lines[i].contains("]:") && {
1420 let after_caret = &lines[i].trim()[2..];
1421 after_caret
1422 .find("]:")
1423 .is_some_and(|pos| pos > 0 && !after_caret[..pos].contains(['[', ']']))
1424 } {
1425 let footnote_start = i;
1426 let line = lines[i];
1427
1428 // Extract the prefix `[^id]:`
1429 let Some(colon_pos) = line.find("]:") else {
1430 i += 1;
1431 continue;
1432 };
1433 let prefix_end = colon_pos + 2;
1434 let prefix = &line[..prefix_end];
1435
1436 // Content starts after `]: ` (with optional space)
1437 let content_start = if line[prefix_end..].starts_with(' ') {
1438 prefix_end + 1
1439 } else {
1440 prefix_end
1441 };
1442 let first_content = &line[content_start..];
1443
1444 // CommonMark footnotes use 4-space continuation indent
1445 const FN_INDENT: usize = 4;
1446
1447 // --- Line classification for footnote content ---
1448 #[derive(Debug, Clone)]
1449 enum FnLineType {
1450 Content(String),
1451 Verbatim(String, usize), // preserved text, original indent
1452 Empty,
1453 }
1454
1455 // Helper: compute visual indent (tabs = 4 spaces)
1456 let visual_indent = |s: &str| -> usize {
1457 s.chars()
1458 .take_while(|c| c.is_whitespace())
1459 .map(|c| if c == '\t' { 4 } else { 1 })
1460 .sum::<usize>()
1461 };
1462
1463 // Helper: check if a trimmed line is a fence marker (homogeneous chars)
1464 let is_fence = |s: &str| -> bool {
1465 let t = s.trim();
1466 let fence_char = t.chars().next();
1467 matches!(fence_char, Some('`') | Some('~'))
1468 && t.chars().take_while(|&c| c == fence_char.unwrap()).count() >= 3
1469 };
1470
1471 // Helper: check if a trimmed line is a setext underline
1472 let is_setext_underline = |s: &str| -> bool {
1473 let t = s.trim();
1474 !t.is_empty()
1475 && (t.chars().all(|c| c == '=' || c == ' ') || t.chars().all(|c| c == '-' || c == ' '))
1476 && t.contains(['=', '-'])
1477 };
1478
1479 // Deferred body: `[^id]:\n content` — first line has no content,
1480 // actual content starts on the next indented line
1481 let deferred_body = first_content.trim().is_empty();
1482
1483 // Collect all lines belonging to this footnote definition
1484 let mut fn_lines: Vec<FnLineType> = Vec::new();
1485 if !deferred_body {
1486 fn_lines.push(FnLineType::Content(first_content.to_string()));
1487 }
1488 let mut last_consumed = i;
1489 i += 1;
1490
1491 // Strip only the footnote continuation indent, preserving
1492 // internal indentation (e.g., code block body indent)
1493 let strip_fn_indent = |s: &str| -> String {
1494 let mut chars = s.chars();
1495 let mut stripped = 0;
1496 while stripped < FN_INDENT {
1497 match chars.next() {
1498 Some('\t') => stripped += 4,
1499 Some(c) if c.is_whitespace() => stripped += 1,
1500 _ => break,
1501 }
1502 }
1503 chars.as_str().to_string()
1504 };
1505
1506 let mut in_fenced_code = false;
1507 let mut consecutive_blanks = 0u32;
1508
1509 while i < lines.len() {
1510 let next = lines[i];
1511 let next_trimmed = next.trim();
1512
1513 // Blank line handling
1514 if next_trimmed.is_empty() {
1515 consecutive_blanks += 1;
1516 // 2+ consecutive blanks terminate the footnote
1517 if consecutive_blanks >= 2 {
1518 break;
1519 }
1520
1521 // Inside a fenced code block, blank lines are part of the code
1522 if in_fenced_code {
1523 consecutive_blanks = 0; // Don't count blanks inside code blocks
1524 fn_lines.push(FnLineType::Verbatim(String::new(), 0));
1525 last_consumed = i;
1526 i += 1;
1527 continue;
1528 }
1529
1530 // Peek ahead: if next non-blank line is indented >= FN_INDENT,
1531 // this blank is an internal paragraph separator
1532 if i + 1 < lines.len() {
1533 let peek = lines[i + 1];
1534 let peek_indent = visual_indent(peek);
1535 if !peek.trim().is_empty() && peek_indent >= FN_INDENT {
1536 fn_lines.push(FnLineType::Empty);
1537 last_consumed = i;
1538 i += 1;
1539 continue;
1540 }
1541 }
1542 // No valid continuation after blank — end of footnote
1543 break;
1544 }
1545
1546 consecutive_blanks = 0;
1547 let indent = visual_indent(next);
1548
1549 // Not indented enough — end of footnote
1550 if indent < FN_INDENT {
1551 break;
1552 }
1553
1554 // Inside a fenced code block: everything is verbatim until closing fence
1555 if in_fenced_code {
1556 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1557 if is_fence(next_trimmed) {
1558 in_fenced_code = false;
1559 }
1560 last_consumed = i;
1561 i += 1;
1562 continue;
1563 }
1564
1565 // Fence opener — start verbatim code block
1566 if is_fence(next_trimmed) {
1567 in_fenced_code = true;
1568 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1569 last_consumed = i;
1570 i += 1;
1571 continue;
1572 }
1573
1574 // Indented code block: indent >= FN_INDENT + 4 (= 8 spaces)
1575 if indent >= FN_INDENT + 4 {
1576 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1577 last_consumed = i;
1578 i += 1;
1579 continue;
1580 }
1581
1582 // Structural content that must be preserved verbatim
1583 if next_trimmed.starts_with('#')
1584 || is_list_item(next_trimmed)
1585 || next_trimmed.starts_with('>')
1586 || TableUtils::is_potential_table_row(next_trimmed)
1587 || is_setext_underline(next_trimmed)
1588 || is_horizontal_rule(next_trimmed)
1589 || crate::utils::mkdocs_footnotes::is_footnote_definition(next_trimmed)
1590 {
1591 // Preserve verbatim: blockquotes, tables, lists, setext
1592 // underlines, and horizontal rules inside the footnote
1593 if next_trimmed.starts_with('>')
1594 || TableUtils::is_potential_table_row(next_trimmed)
1595 || is_list_item(next_trimmed)
1596 || is_setext_underline(next_trimmed)
1597 || is_horizontal_rule(next_trimmed)
1598 {
1599 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1600 last_consumed = i;
1601 i += 1;
1602 continue;
1603 }
1604 // Headings, new footnote defs, link refs — end the footnote
1605 break;
1606 }
1607
1608 // Link reference definitions inside footnotes are not reflowable
1609 if next_trimmed.starts_with('[')
1610 && !next_trimmed.starts_with("[^")
1611 && next_trimmed.contains("]:")
1612 && LINK_REF_PATTERN.is_match(next_trimmed)
1613 {
1614 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1615 last_consumed = i;
1616 i += 1;
1617 continue;
1618 }
1619
1620 // HTML-only lines inside footnotes are not reflowable
1621 if is_html_only_line(next_trimmed) {
1622 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1623 last_consumed = i;
1624 i += 1;
1625 continue;
1626 }
1627
1628 // Regular prose content
1629 fn_lines.push(FnLineType::Content(next_trimmed.to_string()));
1630 last_consumed = i;
1631 i += 1;
1632 }
1633
1634 // Nothing collected or only empty lines
1635 if fn_lines.iter().all(|l| matches!(l, FnLineType::Empty)) || fn_lines.is_empty() {
1636 continue;
1637 }
1638
1639 // --- Group into blocks ---
1640 #[derive(Debug)]
1641 enum FnBlock {
1642 Paragraph(Vec<String>),
1643 Verbatim(Vec<(String, usize)>), // (content, indent) preserved as-is
1644 }
1645
1646 let mut blocks: Vec<FnBlock> = Vec::new();
1647 let mut current_para: Vec<String> = Vec::new();
1648 let mut current_verbatim: Vec<(String, usize)> = Vec::new();
1649
1650 for fl in &fn_lines {
1651 match fl {
1652 FnLineType::Content(s) => {
1653 if !current_verbatim.is_empty() {
1654 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
1655 }
1656 current_para.push(s.clone());
1657 }
1658 FnLineType::Verbatim(s, indent) => {
1659 if !current_para.is_empty() {
1660 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
1661 }
1662 current_verbatim.push((s.clone(), *indent));
1663 }
1664 FnLineType::Empty => {
1665 if !current_para.is_empty() {
1666 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
1667 }
1668 if !current_verbatim.is_empty() {
1669 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
1670 }
1671 }
1672 }
1673 }
1674 if !current_para.is_empty() {
1675 blocks.push(FnBlock::Paragraph(current_para));
1676 }
1677 if !current_verbatim.is_empty() {
1678 blocks.push(FnBlock::Verbatim(current_verbatim));
1679 }
1680
1681 // --- Reflow paragraphs and reconstruct ---
1682 let prefix_display_width = prefix.chars().count() + 1; // +1 for space
1683 let reflow_line_length = if config.line_length.is_unlimited() {
1684 usize::MAX
1685 } else {
1686 config
1687 .line_length
1688 .get()
1689 .saturating_sub(FN_INDENT.max(prefix_display_width))
1690 .max(20)
1691 };
1692 // Footnote continuation uses a fixed 4-space indent, so list
1693 // continuation capping does not apply here.
1694 let reflow_options = crate::utils::text_reflow::ReflowOptions {
1695 max_list_continuation_indent: None,
1696 ..Self::reflow_options(ctx, config, reflow_line_length)
1697 };
1698
1699 let indent_str = " ".repeat(FN_INDENT);
1700 let mut result_lines: Vec<String> = Vec::new();
1701 let mut is_first_block = true;
1702
1703 for block in &blocks {
1704 match block {
1705 FnBlock::Paragraph(para_lines) => {
1706 let paragraph_text = para_lines.join(" ");
1707 let paragraph_text = paragraph_text.trim();
1708 if paragraph_text.is_empty() {
1709 continue;
1710 }
1711
1712 let reflowed = crate::utils::text_reflow::reflow_line(paragraph_text, &reflow_options);
1713 if reflowed.is_empty() {
1714 continue;
1715 }
1716
1717 // Blank line separator between blocks
1718 if !result_lines.is_empty() {
1719 result_lines.push(String::new());
1720 }
1721
1722 for (idx, rline) in reflowed.iter().enumerate() {
1723 if is_first_block && idx == 0 {
1724 result_lines.push(format!("{prefix} {rline}"));
1725 } else {
1726 result_lines.push(format!("{indent_str}{rline}"));
1727 }
1728 }
1729 is_first_block = false;
1730 }
1731 FnBlock::Verbatim(verb_lines) => {
1732 // Blank line separator between blocks
1733 if !result_lines.is_empty() {
1734 result_lines.push(String::new());
1735 }
1736
1737 if is_first_block {
1738 // Verbatim as first block in a deferred-body footnote
1739 if deferred_body {
1740 result_lines.push(prefix.to_string());
1741 }
1742 is_first_block = false;
1743 }
1744 for (content, _orig_indent) in verb_lines {
1745 result_lines.push(format!("{indent_str}{content}"));
1746 }
1747 }
1748 }
1749 }
1750
1751 // If nothing was produced, skip
1752 if result_lines.is_empty() {
1753 continue;
1754 }
1755
1756 let reflowed_text = result_lines.join(line_ending);
1757
1758 // Calculate byte range using last_consumed
1759 let start_range = line_index.whole_line_range(footnote_start + 1);
1760 let end_range = if last_consumed == lines.len() - 1 && !ctx.content.ends_with('\n') {
1761 line_index.line_text_range(last_consumed + 1, 1, lines[last_consumed].len() + 1)
1762 } else {
1763 line_index.whole_line_range(last_consumed + 1)
1764 };
1765 let byte_range = start_range.start..end_range.end;
1766
1767 let replacement = if last_consumed < lines.len() - 1 || ctx.content.ends_with('\n') {
1768 format!("{reflowed_text}{line_ending}")
1769 } else {
1770 reflowed_text
1771 };
1772
1773 let original_text = &ctx.content[byte_range.clone()];
1774 let max_length = (footnote_start..=last_consumed)
1775 .map(|idx| self.calculate_effective_length(lines[idx]))
1776 .max()
1777 .unwrap_or(0);
1778 let line_limit = if config.line_length.is_unlimited() {
1779 usize::MAX
1780 } else {
1781 config.line_length.get()
1782 };
1783 if original_text != replacement && max_length > line_limit {
1784 warnings.push(LintWarning {
1785 rule_name: Some(self.name().to_string()),
1786 message: format!(
1787 "Line length {} exceeds {} characters",
1788 max_length,
1789 config.line_length.get()
1790 ),
1791 line: footnote_start + 1,
1792 column: 1,
1793 end_line: last_consumed + 1,
1794 end_column: lines[last_consumed].chars().count() + 1,
1795 severity: Severity::Warning,
1796 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1797 });
1798 }
1799 continue;
1800 }
1801
1802 // Handle MkDocs container content (admonitions and tabs) with indent-preserving reflow
1803 if ctx
1804 .line_info(line_num)
1805 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
1806 {
1807 // Skip admonition/tab marker lines — only reflow their indented content
1808 let current_line = lines[i];
1809 if mkdocs_admonitions::is_admonition_start(current_line) || mkdocs_tabs::is_tab_marker(current_line) {
1810 i += 1;
1811 continue;
1812 }
1813
1814 let container_start = i;
1815
1816 // Detect the actual indent level from the first content line
1817 // (supports nested admonitions with 8+ spaces)
1818 let first_line = lines[i];
1819 let base_indent_len = first_line.len() - first_line.trim_start().len();
1820 let base_indent: String = " ".repeat(base_indent_len);
1821
1822 // Collect consecutive MkDocs container paragraph lines
1823 let mut container_lines: Vec<&str> = Vec::new();
1824 while i < lines.len() {
1825 let current_line_num = i + 1;
1826 let line_info = ctx.line_info(current_line_num);
1827
1828 // Stop if we leave the MkDocs container
1829 if !line_info.is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container) {
1830 break;
1831 }
1832
1833 let line = lines[i];
1834
1835 // Stop at paragraph boundaries within the container
1836 if line.trim().is_empty() {
1837 break;
1838 }
1839
1840 // Skip list items, code blocks, headings, HTML-only lines within containers
1841 if is_list_item(line.trim())
1842 || line.trim().starts_with("```")
1843 || line.trim().starts_with("~~~")
1844 || line.trim().starts_with('#')
1845 || is_html_only_line(line)
1846 {
1847 break;
1848 }
1849
1850 container_lines.push(line);
1851 i += 1;
1852 }
1853
1854 if container_lines.is_empty() {
1855 // Must advance i to avoid infinite loop when we encounter
1856 // non-paragraph content (code block, list, heading, empty line)
1857 // at the start of an MkDocs container
1858 i += 1;
1859 continue;
1860 }
1861
1862 // Strip the base indent from each line and join for reflow
1863 let stripped_lines: Vec<&str> = container_lines
1864 .iter()
1865 .map(|line| {
1866 if line.starts_with(&base_indent) {
1867 &line[base_indent_len..]
1868 } else {
1869 line.trim_start()
1870 }
1871 })
1872 .collect();
1873 let paragraph_text = stripped_lines.join(" ");
1874
1875 // Check if reflow is needed
1876 let needs_reflow = match config.reflow_mode {
1877 ReflowMode::Normalize => self.normalize_mode_needs_reflow(container_lines.iter().copied(), config),
1878 ReflowMode::SentencePerLine => {
1879 let sentences = split_into_sentences(¶graph_text);
1880 sentences.len() > 1 || container_lines.len() > 1
1881 }
1882 ReflowMode::SemanticLineBreaks => {
1883 let sentences = split_into_sentences(¶graph_text);
1884 sentences.len() > 1
1885 || container_lines.len() > 1
1886 || container_lines
1887 .iter()
1888 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
1889 }
1890 ReflowMode::Default => container_lines
1891 .iter()
1892 .any(|line| self.calculate_effective_length(line) > config.line_length.get()),
1893 };
1894
1895 if !needs_reflow {
1896 continue;
1897 }
1898
1899 // Calculate byte range for this container paragraph
1900 let start_range = line_index.whole_line_range(container_start + 1);
1901 let end_line = container_start + container_lines.len() - 1;
1902 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
1903 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
1904 } else {
1905 line_index.whole_line_range(end_line + 1)
1906 };
1907 let byte_range = start_range.start..end_range.end;
1908
1909 // Reflow with adjusted line length (accounting for the 4-space indent)
1910 let reflow_line_length = if config.line_length.is_unlimited() {
1911 usize::MAX
1912 } else {
1913 config.line_length.get().saturating_sub(base_indent_len).max(1)
1914 };
1915 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
1916 let reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
1917
1918 // Re-add the 4-space indent to each reflowed line
1919 let reflowed_with_indent: Vec<String> =
1920 reflowed.iter().map(|line| format!("{base_indent}{line}")).collect();
1921 let reflowed_text = reflowed_with_indent.join(line_ending);
1922
1923 // Preserve trailing newline
1924 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
1925 format!("{reflowed_text}{line_ending}")
1926 } else {
1927 reflowed_text
1928 };
1929
1930 // Only generate a warning if the replacement is different
1931 let original_text = &ctx.content[byte_range.clone()];
1932 if original_text != replacement {
1933 warnings.push(LintWarning {
1934 rule_name: Some(self.name().to_string()),
1935 message: format!(
1936 "Line length {} exceeds {} characters (in MkDocs container)",
1937 container_lines.iter().map(|l| l.len()).max().unwrap_or(0),
1938 config.line_length.get()
1939 ),
1940 line: container_start + 1,
1941 column: 1,
1942 end_line: end_line + 1,
1943 end_column: lines[end_line].chars().count() + 1,
1944 severity: Severity::Warning,
1945 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1946 });
1947 }
1948 continue;
1949 }
1950
1951 // Helper function to detect semantic line markers
1952 let is_semantic_line = |content: &str| -> bool {
1953 let trimmed = content.trim_start();
1954 let semantic_markers = [
1955 "NOTE:",
1956 "WARNING:",
1957 "IMPORTANT:",
1958 "CAUTION:",
1959 "TIP:",
1960 "DANGER:",
1961 "HINT:",
1962 "INFO:",
1963 ];
1964 semantic_markers.iter().any(|marker| trimmed.starts_with(marker))
1965 };
1966
1967 // Helper function to detect fence markers (opening or closing)
1968 let is_fence_marker = |content: &str| -> bool {
1969 let trimmed = content.trim_start();
1970 trimmed.starts_with("```") || trimmed.starts_with("~~~")
1971 };
1972
1973 // Check if this is a list item - handle it specially
1974 let trimmed = lines[i].trim();
1975 if is_list_item(trimmed) {
1976 // Collect the entire list item including continuation lines
1977 let list_start = i;
1978 let (marker, first_content) = extract_list_marker_and_content(lines[i]);
1979 let marker_len = marker.len();
1980 // The normalized marker above is what gets re-emitted; the source marker
1981 // is where the item's content actually starts. Nested blocks move by the
1982 // difference between the two content columns, so the shift must be
1983 // measured against the source, not against the normalized width.
1984 let source_marker = source_list_marker(lines[i]);
1985 let source_content_col = source_marker.as_ref().map_or(marker_len, |m| m.content_col);
1986
1987 // Checkbox ([ ]/[x]/[X]) is inline content, not part of the list marker.
1988 // Use the base bullet/number marker width for continuation recognition
1989 // so that continuation lines at 2+ spaces are collected for "- [ ] " items.
1990 let base_marker_len = if marker.contains("[ ] ") || marker.contains("[x] ") || marker.contains("[X] ") {
1991 marker.find('[').unwrap_or(marker_len)
1992 } else {
1993 marker_len
1994 };
1995
1996 // MkDocs flavor requires at least 4 spaces for list continuation
1997 // after a blank line (multi-paragraph list items). For non-blank
1998 // continuation (lines directly following the marker line), use
1999 // the natural marker width so that 2-space indent is recognized.
2000 let item_indent = ctx.lines[i].indent;
2001 let min_continuation_indent = if ctx.flavor.requires_strict_list_indent() {
2002 // Use 4-space relative indent from the list item's nesting level
2003 item_indent + (base_marker_len - item_indent).max(4)
2004 } else {
2005 marker_len
2006 };
2007 let content_continuation_indent = base_marker_len;
2008
2009 // Track lines and their types (content, code block, fence, nested list)
2010 #[derive(Clone)]
2011 enum LineType {
2012 Content(String),
2013 CodeBlock(String, usize), // content and original indent
2014 SemanticLine(String), // Lines starting with NOTE:, WARNING:, etc that should stay separate
2015 SnippetLine(String), // MkDocs Snippets delimiters (-8<-) that must stay on their own line
2016 DivMarker(String), // Quarto/Pandoc div markers (::: opening or closing)
2017 AdmonitionHeader(String, usize), // header text (e.g. "!!! note") and original indent
2018 AdmonitionContent(String, usize), // body content text and original indent
2019 Table(String, usize), // GFM table row, preserved verbatim with original indent
2020 Empty,
2021 }
2022
2023 let mut list_item_lines: Vec<LineType> = vec![LineType::Content(first_content)];
2024 // Set when collection stops at a nested list item or a nested
2025 // blockquote that belongs to this item. Such structure is reflowed
2026 // independently and is therefore absent from `list_item_lines`/`blocks`,
2027 // but it still keeps the emitted item spanning multiple physical lines,
2028 // which the MD030 multi-line spacing decision below must account for.
2029 let mut has_trailing_nested_structure = false;
2030 i += 1;
2031
2032 // Collect continuation lines using ctx.lines for metadata
2033 while i < lines.len() {
2034 let line_info = &ctx.lines[i];
2035
2036 // Use pre-computed is_blank from ctx
2037 if line_info.is_blank {
2038 // Empty line - check if next line is indented (part of list item)
2039 if i + 1 < lines.len() {
2040 let next_info = &ctx.lines[i + 1];
2041
2042 // Check if next line is indented enough to be continuation
2043 if !next_info.is_blank && next_info.indent >= min_continuation_indent {
2044 // This blank line is between paragraphs/blocks in the list item
2045 list_item_lines.push(LineType::Empty);
2046 i += 1;
2047 continue;
2048 }
2049 }
2050 // No indented line after blank, end of list item
2051 break;
2052 }
2053
2054 // Use pre-computed indent from ctx
2055 let indent = line_info.indent;
2056
2057 // Valid continuation must be indented at least content_continuation_indent.
2058 // For non-blank continuation, use marker_len (e.g. 2 for "- ").
2059 // MkDocs strict 4-space requirement applies only after blank lines.
2060 if indent >= content_continuation_indent {
2061 let trimmed = line_info.content(ctx.content).trim();
2062
2063 // Check for MkDocs admonition lines inside list items BEFORE
2064 // checking in_code_block. Lines inside code blocks within
2065 // admonitions have both in_admonition and in_code_block set;
2066 // admonition membership takes priority so the entire admonition
2067 // structure (including embedded code blocks) is preserved.
2068 if line_info.in_admonition {
2069 let raw_content = line_info.content(ctx.content);
2070 if mkdocs_admonitions::is_admonition_start(raw_content) {
2071 let header_text = raw_content[indent..].trim_end().to_string();
2072 list_item_lines.push(LineType::AdmonitionHeader(header_text, indent));
2073 } else {
2074 let body_text = raw_content[indent..].trim_end().to_string();
2075 list_item_lines.push(LineType::AdmonitionContent(body_text, indent));
2076 }
2077 i += 1;
2078 continue;
2079 }
2080
2081 // Use pre-computed in_code_block from ctx
2082 if line_info.in_code_block {
2083 list_item_lines.push(LineType::CodeBlock(
2084 line_info.content(ctx.content)[indent..].to_string(),
2085 indent,
2086 ));
2087 i += 1;
2088 continue;
2089 }
2090
2091 // A blockquote nested inside the list item is reflowed by the
2092 // blockquote-aware path (it preserves the `>` prefix, including the
2093 // list indent), not as list-item prose. Collecting it as Content
2094 // would strip the markers and reflow `>` as words, collapsing the
2095 // blank `>` line and dropping `>` from wrapped continuations. End
2096 // the item here so the outer loop routes the blockquote line to
2097 // generate_blockquote_paragraph_fix. Uncollect a pending blank so
2098 // the separator between the list prose and the blockquote survives.
2099 if line_info.blockquote.is_some() {
2100 has_trailing_nested_structure = true;
2101 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
2102 list_item_lines.pop();
2103 i -= 1;
2104 }
2105 break;
2106 }
2107
2108 // Check if this is a SIBLING list item (breaks parent)
2109 // Nested lists are indented >= marker_len and are PART of the parent item
2110 // Siblings are at indent < marker_len (at or before parent marker)
2111 if is_list_item(trimmed) && indent < marker_len {
2112 // This is a sibling item at same or higher level - end parent item
2113 break;
2114 }
2115
2116 // Nested list items are always processed independently
2117 // by the outer loop, so break when we encounter one.
2118 // If a blank line was collected before this, uncollect it
2119 // so the outer loop preserves the blank between parent and nested.
2120 if is_list_item(trimmed) && indent >= marker_len {
2121 has_trailing_nested_structure = true;
2122 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
2123 list_item_lines.pop();
2124 i -= 1;
2125 }
2126 break;
2127 }
2128
2129 // Normal continuation vs indented code block.
2130 // Use min_continuation_indent for the threshold since
2131 // code blocks start 4 spaces beyond the expected content
2132 // level (which is min_continuation_indent for MkDocs).
2133 if indent <= min_continuation_indent + 3 {
2134 // Extract content (remove indentation and trailing whitespace)
2135 // Preserve hard breaks (2 trailing spaces) while removing excessive whitespace
2136 // See: https://github.com/rvben/rumdl/issues/76
2137 let content = trim_preserving_hard_break(&line_info.content(ctx.content)[indent..]);
2138
2139 // Check if this is a div marker (::: opening or closing)
2140 // These must be preserved on their own line, not merged into paragraphs
2141 if line_info.is_div_marker {
2142 list_item_lines.push(LineType::DivMarker(content));
2143 }
2144 // Check if this is a fence marker (opening or closing)
2145 // These should be treated as code block lines, not paragraph content
2146 else if is_fence_marker(&content) {
2147 list_item_lines.push(LineType::CodeBlock(content, indent));
2148 }
2149 // Check if this is a semantic line (NOTE:, WARNING:, etc.)
2150 else if is_semantic_line(&content) {
2151 list_item_lines.push(LineType::SemanticLine(content));
2152 }
2153 // Check if this is a snippet block delimiter (-8<- or --8<--)
2154 // These must be preserved on their own lines for MkDocs Snippets extension
2155 else if is_snippet_block_delimiter(&content) {
2156 list_item_lines.push(LineType::SnippetLine(content));
2157 }
2158 // Check if this is a GFM table row. Tables nested inside list
2159 // items must be preserved verbatim — joining them with prose
2160 // breaks the column structure.
2161 //
2162 // `is_potential_table_row` is intentionally permissive at the
2163 // row level: any line with `|` and 2+ cells qualifies. To avoid
2164 // misclassifying prose continuation lines that contain a literal
2165 // pipe (e.g. "use grep | sort to ..."), require one of:
2166 // - the row is pipe-bordered (`| ... |`), the canonical form
2167 // for tables nested in lists; or
2168 // - the next line is a delimiter row (this is a header); or
2169 // - the previous classified line was already a Table (this is
2170 // a continuation row).
2171 else if TableUtils::is_potential_table_row(&content) && {
2172 let pipe_bordered = content.trim().starts_with('|') && content.trim().ends_with('|');
2173 let next_is_delim = ctx
2174 .lines
2175 .get(i + 1)
2176 .is_some_and(|next| TableUtils::is_delimiter_row(next.content(ctx.content)));
2177 let prev_was_table = matches!(list_item_lines.last(), Some(LineType::Table(..)));
2178 pipe_bordered || next_is_delim || prev_was_table
2179 } {
2180 list_item_lines.push(LineType::Table(content, indent));
2181 } else {
2182 list_item_lines.push(LineType::Content(content));
2183 }
2184 i += 1;
2185 } else {
2186 // indent >= min_continuation_indent + 4: indented code block
2187 list_item_lines.push(LineType::CodeBlock(
2188 line_info.content(ctx.content)[indent..].to_string(),
2189 indent,
2190 ));
2191 i += 1;
2192 }
2193 } else {
2194 // Not indented enough, end of list item
2195 break;
2196 }
2197 }
2198
2199 // Determine the output continuation indent.
2200 // Normalize/Default modes canonicalize to min_continuation_indent
2201 // (fixing over-indented continuation). Semantic/SentencePerLine
2202 // modes preserve the user's actual indent since they only fix
2203 // line breaking, not indentation.
2204 let indent_size = match config.reflow_mode {
2205 ReflowMode::SemanticLineBreaks | ReflowMode::SentencePerLine => {
2206 // Find indent of the first plain text continuation line,
2207 // skipping the marker line (index 0), nested list items,
2208 // code blocks, and blank lines.
2209 list_item_lines
2210 .iter()
2211 .enumerate()
2212 .skip(1)
2213 .find_map(|(k, lt)| {
2214 if matches!(lt, LineType::Content(_)) {
2215 Some(ctx.lines[list_start + k].indent)
2216 } else {
2217 None
2218 }
2219 })
2220 .unwrap_or(min_continuation_indent)
2221 }
2222 _ => min_continuation_indent,
2223 };
2224 // For checkbox items in mkdocs flavor, enforce minimum indent so
2225 // continuation lines use the structural list indent (4), not the
2226 // content-aligned indent (6) which Python-Markdown doesn't support
2227 let has_checkbox = base_marker_len < marker_len;
2228 let indent_size = if has_checkbox && ctx.flavor.requires_strict_list_indent() {
2229 indent_size.max(min_continuation_indent)
2230 } else {
2231 indent_size
2232 };
2233
2234 // Split list_item_lines into blocks (paragraphs, code blocks, nested lists, semantic lines, and HTML blocks)
2235 let mut builder = BlockBuilder::new();
2236 for line in &list_item_lines {
2237 match line {
2238 LineType::Empty => builder.feed_blank_line(),
2239 LineType::Content(content) => builder.feed_content(content),
2240 LineType::CodeBlock(content, indent) => builder.feed_code_line(content, *indent),
2241 LineType::SemanticLine(content) => builder.feed_semantic_line(content),
2242 LineType::SnippetLine(content) => builder.feed_snippet_line(content),
2243 LineType::DivMarker(content) => builder.feed_div_marker(content),
2244 LineType::AdmonitionHeader(header_text, indent) => {
2245 builder.feed_admonition_header(header_text, *indent)
2246 }
2247 LineType::AdmonitionContent(content, indent) => {
2248 builder.feed_admonition_content(content, *indent)
2249 }
2250 LineType::Table(content, indent) => builder.feed_table_line(content, *indent),
2251 }
2252 }
2253 let blocks = builder.finalize();
2254
2255 // Helper: check if a line (raw source or stripped content) is exempt
2256 // from line-length checks. Link reference definitions are always exempt;
2257 // standalone link/image lines are exempt when strict mode is off.
2258 // Also checks content after stripping list markers, since list item
2259 // continuation lines may contain link ref defs.
2260 let is_exempt_line = |raw_line: &str| -> bool {
2261 let trimmed = raw_line.trim();
2262 // Link reference definitions: always exempt
2263 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
2264 return true;
2265 }
2266 // Also check after stripping list markers (for list item content)
2267 if is_list_item(trimmed) {
2268 let (_, content) = extract_list_marker_and_content(trimmed);
2269 let content_trimmed = content.trim();
2270 if content_trimmed.starts_with('[')
2271 && content_trimmed.contains("]:")
2272 && LINK_REF_PATTERN.is_match(content_trimmed)
2273 {
2274 return true;
2275 }
2276 }
2277 // Standalone link/image lines: exempt when not strict
2278 if !config.strict && is_standalone_link_or_image_line(raw_line) {
2279 return true;
2280 }
2281 // HTML-only lines: exempt when not strict
2282 if !config.strict && is_html_only_line(raw_line) {
2283 return true;
2284 }
2285 false
2286 };
2287
2288 // Check if reflowing is needed (only for content paragraphs, not code blocks or nested lists)
2289 // Exclude link reference definitions and standalone link lines from content
2290 // so they don't pollute combined_content or trigger false reflow.
2291 let content_lines: Vec<String> = list_item_lines
2292 .iter()
2293 .filter_map(|line| {
2294 if let LineType::Content(s) = line {
2295 if is_exempt_line(s) {
2296 return None;
2297 }
2298 Some(s.clone())
2299 } else {
2300 None
2301 }
2302 })
2303 .collect();
2304
2305 // Check if we need to reflow this list item
2306 // We check the combined content to see if it exceeds length limits
2307 let combined_content = content_lines.join(" ").trim().to_string();
2308
2309 // Helper to check if we should reflow in normalize mode
2310 let should_normalize = || {
2311 // Don't normalize if the list item only contains nested lists, code blocks, or semantic lines
2312 // DO normalize if it has plain text content that spans multiple lines
2313 let has_code_blocks = blocks.iter().any(|b| matches!(b, Block::Code { .. }));
2314 let has_semantic_lines = blocks.iter().any(|b| matches!(b, Block::SemanticLine(_)));
2315 let has_snippet_lines = blocks.iter().any(|b| matches!(b, Block::SnippetLine(_)));
2316 let has_div_markers = blocks.iter().any(|b| matches!(b, Block::DivMarker(_)));
2317 let has_admonitions = blocks.iter().any(|b| matches!(b, Block::Admonition { .. }));
2318 let has_tables = blocks.iter().any(|b| matches!(b, Block::Table { .. }));
2319 let has_paragraphs = blocks.iter().any(|b| matches!(b, Block::Paragraph(_)));
2320
2321 // If we have structural blocks but no paragraphs, don't normalize
2322 if (has_code_blocks
2323 || has_semantic_lines
2324 || has_snippet_lines
2325 || has_div_markers
2326 || has_admonitions
2327 || has_tables)
2328 && !has_paragraphs
2329 {
2330 return false;
2331 }
2332
2333 // If we have paragraphs, check if they span multiple lines or there are multiple blocks
2334 if has_paragraphs {
2335 // Count only paragraphs that contain at least one non-exempt line.
2336 // Paragraphs consisting entirely of link ref defs or standalone links
2337 // should not trigger normalization.
2338 let paragraph_count = blocks
2339 .iter()
2340 .filter(|b| {
2341 if let Block::Paragraph(para_lines) = b {
2342 !para_lines.iter().all(|line| is_exempt_line(line))
2343 } else {
2344 false
2345 }
2346 })
2347 .count();
2348 if paragraph_count > 1 {
2349 // Multiple non-exempt paragraph blocks should be normalized
2350 return true;
2351 }
2352
2353 // Single paragraph block: normalize if it has multiple content lines
2354 if content_lines.len() > 1 {
2355 return true;
2356 }
2357 }
2358
2359 false
2360 };
2361
2362 // Integrate MD030 list-marker spacing (and MD007's text-aligned
2363 // continuation). In Default/Normalize modes — the modes that already
2364 // canonicalize spacing/indent — derive the number of spaces after the
2365 // marker from the configured MD030 values instead of forcing a single
2366 // space, then align continuation lines to the resulting content column.
2367 // Sentence/Semantic modes only adjust line breaks, so they keep the
2368 // marker spacing and indentation already present in the source.
2369 //
2370 // With default MD030 (a single space everywhere) the rebuilt marker is
2371 // byte-identical to the source marker, so this is a no-op and existing
2372 // behaviour is preserved; only a non-default MD030 changes the output.
2373 // The MkDocs flavor enforces a rigid structural indent (4 spaces,
2374 // capped via max_list_continuation_indent) that Python-Markdown
2375 // requires; leave its specialized handling untouched.
2376 let (marker, indent_size, code_indent_shift) =
2377 if matches!(config.reflow_mode, ReflowMode::Default | ReflowMode::Normalize)
2378 && !ctx.flavor.requires_strict_list_indent()
2379 && let Some(li) = ctx.lines[list_start].list_item.as_deref()
2380 {
2381 let bullet_len = li.marker.len();
2382 // The checkbox (e.g. `[ ] `) is content, not part of the list
2383 // marker MD030 governs; carry it over verbatim after the spacing.
2384 let checkbox_tail = marker[base_marker_len..].to_string();
2385 // Shift this item right by its ancestors' cumulative marker
2386 // widening so a nested item stays under its parent's (widened)
2387 // content column. Zero for top-level items and for the whole
2388 // tree under default MD030, where the source indent is preserved
2389 // verbatim (byte-identical output).
2390 let ancestor_shift = list_shift_stack.last().map_or(0isize, |&(_, shift)| shift);
2391 let shifted_indent = (item_indent as isize + ancestor_shift).max(0) as usize;
2392 let indent_prefix = if ancestor_shift == 0 {
2393 marker[..item_indent].to_string()
2394 } else {
2395 " ".repeat(shifted_indent)
2396 };
2397
2398 // Decide single- vs multi-line spacing from the *rewritten* shape,
2399 // not the source. A multi-line source is not enough: plain prose
2400 // continuation collapses onto the marker line during reflow, so a
2401 // two-line bullet that fits becomes a single physical line and must
2402 // use MD030's single-line spacing (otherwise MD013 emits a result
2403 // that MD030 immediately rewrites). The emitted item stays
2404 // multi-line only when reflow cannot collapse it:
2405 // - the prose wraps past the line length, or
2406 // - a structural block remains (code, table, admonition, semantic
2407 // line, snippet, div marker, HTML) that is not joinable prose, or
2408 // - more than one paragraph remains (blank-separated), or
2409 // - a nested list/blockquote follows (reflowed independently, so it
2410 // is absent from `blocks` but still keeps the item multi-line).
2411 // The wrap test uses the single-line content column so it is
2412 // independent of the spacing we are about to choose (avoiding a
2413 // circular result). `ol-align-column` ignores this flag entirely in
2414 // expected_spaces(), so ordered lists are unaffected.
2415 //
2416 // This is the rewritten-shape counterpart of MD030's
2417 // `is_multi_line_list_item` (which keys off the *source*). The two
2418 // are related but technically distinct and intentionally separate;
2419 // if the notion of "multi-line" changes in one, revisit the other.
2420 let single_col = shifted_indent
2421 + bullet_len
2422 + self.list_spacing.expected_spaces(li.is_ordered, false, bullet_len)
2423 + checkbox_tail.len();
2424 let prose_wraps = !combined_content.is_empty()
2425 && self
2426 .calculate_effective_length(&format!("{}{combined_content}", " ".repeat(single_col)))
2427 > config.line_length.effective_limit();
2428 let has_structural_block = blocks.iter().any(|b| !matches!(b, Block::Paragraph(_)));
2429 let multiple_paragraphs =
2430 blocks.iter().filter(|b| matches!(b, Block::Paragraph(_))).count() > 1;
2431 let is_multi =
2432 prose_wraps || has_structural_block || multiple_paragraphs || has_trailing_nested_structure;
2433
2434 let spaces = self.list_spacing.expected_spaces(li.is_ordered, is_multi, bullet_len);
2435 let new_marker = format!("{indent_prefix}{}{}{checkbox_tail}", li.marker, " ".repeat(spaces));
2436 let new_col = new_marker.chars().count();
2437 let shift = new_col as isize - source_content_col as isize;
2438 (new_marker, new_col, shift)
2439 } else {
2440 // MkDocs enforces a rigid structural indent, so the item is emitted
2441 // exactly as written and its nested blocks never move. Re-emitting
2442 // the normalized marker here would narrow the content column while
2443 // leaving those blocks behind.
2444 let marker = source_marker.map_or(marker, |m| m.text);
2445 (marker, indent_size, 0isize)
2446 };
2447 let expected_indent = " ".repeat(indent_size);
2448
2449 let needs_reflow = match config.reflow_mode {
2450 ReflowMode::Normalize => {
2451 // Only reflow if:
2452 // 1. Any non-exempt paragraph, when joined, exceeds the limit, OR
2453 // 2. Any admonition content line exceeds the limit, OR
2454 // 3. The list item should be normalized (has multi-line plain text)
2455 let any_paragraph_exceeds = blocks.iter().any(|block| match block {
2456 Block::Paragraph(para_lines) => {
2457 if para_lines.iter().all(|line| is_exempt_line(line)) {
2458 return false;
2459 }
2460 let joined = para_lines.join(" ");
2461 let with_marker = format!("{}{}", " ".repeat(indent_size), joined.trim());
2462 self.calculate_effective_length(&with_marker) > config.line_length.get()
2463 }
2464 Block::Admonition {
2465 content_lines,
2466 header_indent,
2467 ..
2468 } => content_lines.iter().any(|(content, indent)| {
2469 if content.is_empty() {
2470 return false;
2471 }
2472 let with_indent = format!("{}{}", " ".repeat(*indent.max(header_indent)), content);
2473 self.calculate_effective_length(&with_indent) > config.line_length.get()
2474 }),
2475 _ => false,
2476 });
2477 if any_paragraph_exceeds {
2478 true
2479 } else {
2480 should_normalize()
2481 }
2482 }
2483 ReflowMode::SentencePerLine => {
2484 // Check if list item has multiple sentences
2485 let sentences = split_into_sentences(&combined_content);
2486 sentences.len() > 1
2487 }
2488 ReflowMode::SemanticLineBreaks => {
2489 let sentences = split_into_sentences(&combined_content);
2490 sentences.len() > 1
2491 || (list_start..i).any(|line_idx| {
2492 let line = lines[line_idx];
2493 let trimmed = line.trim();
2494 if trimmed.is_empty() || is_exempt_line(line) {
2495 return false;
2496 }
2497 self.calculate_effective_length(line) > config.line_length.get()
2498 })
2499 }
2500 ReflowMode::Default => {
2501 // In default mode, only reflow if any individual non-exempt line exceeds limit
2502 (list_start..i).any(|line_idx| {
2503 let line = lines[line_idx];
2504 let trimmed = line.trim();
2505 // Skip blank lines and exempt lines
2506 if trimmed.is_empty() || is_exempt_line(line) {
2507 return false;
2508 }
2509 self.calculate_effective_length(line) > config.line_length.get()
2510 })
2511 }
2512 };
2513
2514 // Record this item's frame so its nested children inherit the shift.
2515 // Only a reflowed item's marker actually moves; an unreflowed one keeps
2516 // its source position and so contributes no shift to its children. The
2517 // threshold that decides which following lines are inside this item is
2518 // the normalized marker width, NOT `source_content_col`: the collection
2519 // loop above gathers continuations by that same width, so the frame
2520 // boundary must match it or the two would disagree about ownership of
2521 // lines indented between the normalized and the source content column.
2522 list_shift_stack.push((marker_len, if needs_reflow { code_indent_shift } else { 0 }));
2523
2524 if needs_reflow {
2525 let start_range = line_index.whole_line_range(list_start + 1);
2526 let end_line = i - 1;
2527 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
2528 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
2529 } else {
2530 line_index.whole_line_range(end_line + 1)
2531 };
2532 let byte_range = start_range.start..end_range.end;
2533
2534 // Reflow each block (paragraphs only, preserve code blocks)
2535 // When line_length = 0 (no limit), use a very large value for reflow
2536 let reflow_line_length = if config.line_length.is_unlimited() {
2537 usize::MAX
2538 } else {
2539 config.line_length.get().saturating_sub(indent_size).max(1)
2540 };
2541 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
2542
2543 let mut result: Vec<String> = Vec::new();
2544 let mut is_first_block = true;
2545
2546 for (block_idx, block) in blocks.iter().enumerate() {
2547 match block {
2548 Block::Paragraph(para_lines) => {
2549 // If every line in this paragraph is exempt (link ref defs,
2550 // standalone links), preserve the paragraph verbatim instead
2551 // of reflowing it. Reflowing would corrupt link ref defs.
2552 let all_exempt = para_lines.iter().all(|line| is_exempt_line(line));
2553
2554 if all_exempt {
2555 for (idx, line) in para_lines.iter().enumerate() {
2556 if is_first_block && idx == 0 {
2557 result.push(format!("{marker}{line}"));
2558 is_first_block = false;
2559 } else {
2560 result.push(format!("{expected_indent}{line}"));
2561 }
2562 }
2563 } else {
2564 // Split the paragraph into segments at hard break boundaries
2565 // Each segment can be reflowed independently
2566 let segments = split_into_segments(para_lines);
2567
2568 for (segment_idx, segment) in segments.iter().enumerate() {
2569 // Check if this segment ends with a hard break and what type
2570 let hard_break_type = segment.last().and_then(|line| {
2571 let line = line.strip_suffix('\r').unwrap_or(line);
2572 if line.ends_with('\\') {
2573 Some("\\")
2574 } else if line.ends_with(" ") {
2575 Some(" ")
2576 } else {
2577 None
2578 }
2579 });
2580
2581 // Join and reflow the segment (removing the hard break marker for processing)
2582 let segment_for_reflow: Vec<String> = segment
2583 .iter()
2584 .map(|line| {
2585 // Strip hard break marker (2 spaces or backslash) for reflow processing
2586 if line.ends_with('\\') {
2587 line[..line.len() - 1].trim_end().to_string()
2588 } else if line.ends_with(" ") {
2589 line[..line.len() - 2].trim_end().to_string()
2590 } else {
2591 line.clone()
2592 }
2593 })
2594 .collect();
2595
2596 let segment_text = segment_for_reflow.join(" ").trim().to_string();
2597 if !segment_text.is_empty() {
2598 let reflowed =
2599 crate::utils::text_reflow::reflow_line(&segment_text, &reflow_options);
2600
2601 if is_first_block && segment_idx == 0 {
2602 // First segment of first block starts with marker
2603 result.push(format!("{marker}{}", reflowed[0]));
2604 for line in reflowed.iter().skip(1) {
2605 result.push(format!("{expected_indent}{line}"));
2606 }
2607 is_first_block = false;
2608 } else {
2609 // Subsequent segments
2610 for line in reflowed {
2611 result.push(format!("{expected_indent}{line}"));
2612 }
2613 }
2614
2615 // If this segment had a hard break, add it back to the last line
2616 // Preserve the original hard break format (backslash or two spaces)
2617 if let Some(break_marker) = hard_break_type
2618 && let Some(last_line) = result.last_mut()
2619 {
2620 last_line.push_str(break_marker);
2621 }
2622 }
2623 }
2624 }
2625
2626 // Add blank line after paragraph block if there's a next block.
2627 // Check if next block is a code block that doesn't want a preceding blank.
2628 // Also don't add blank lines before snippet lines (they should stay tight).
2629 // Only add if not already ending with one (avoids double blanks).
2630 if block_idx < blocks.len() - 1 {
2631 let next_block = &blocks[block_idx + 1];
2632 let should_add_blank = match next_block {
2633 Block::Code {
2634 has_preceding_blank, ..
2635 } => *has_preceding_blank,
2636 Block::Table {
2637 has_preceding_blank, ..
2638 } => *has_preceding_blank,
2639 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2640 _ => true, // For all other blocks, add blank line
2641 };
2642 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2643 result.push(String::new());
2644 }
2645 }
2646 }
2647 Block::Code {
2648 lines: code_lines,
2649 has_preceding_blank: _,
2650 } => {
2651 // Preserve code blocks as-is with original indentation
2652 // NOTE: Blank line before code block is handled by the previous block
2653 // (see paragraph block's logic above)
2654
2655 for (idx, (content, orig_indent)) in code_lines.iter().enumerate() {
2656 if is_first_block && idx == 0 {
2657 // First line of first block gets marker
2658 result.push(format!(
2659 "{marker}{}",
2660 " ".repeat(orig_indent - marker_len) + content
2661 ));
2662 is_first_block = false;
2663 } else if content.is_empty() {
2664 result.push(String::new());
2665 } else {
2666 // Shift nested code with the marker so it stays
2667 // aligned under content when MD030 widens spacing.
2668 result.push(format!(
2669 "{}{}",
2670 " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
2671 content
2672 ));
2673 }
2674 }
2675 }
2676 Block::SemanticLine(content) => {
2677 // Preserve semantic lines (NOTE:, WARNING:, etc.) as-is on their own line.
2678 // Only add blank before if not already ending with one.
2679 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
2680 result.push(String::new());
2681 }
2682
2683 if is_first_block {
2684 // First block starts with marker
2685 result.push(format!("{marker}{content}"));
2686 is_first_block = false;
2687 } else {
2688 // Subsequent blocks use expected indent
2689 result.push(format!("{expected_indent}{content}"));
2690 }
2691
2692 // Add blank line after semantic line if there's a next block.
2693 // Only add if not already ending with one.
2694 if block_idx < blocks.len() - 1 {
2695 let next_block = &blocks[block_idx + 1];
2696 let should_add_blank = match next_block {
2697 Block::Code {
2698 has_preceding_blank, ..
2699 } => *has_preceding_blank,
2700 Block::Table {
2701 has_preceding_blank, ..
2702 } => *has_preceding_blank,
2703 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2704 _ => true, // For all other blocks, add blank line
2705 };
2706 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2707 result.push(String::new());
2708 }
2709 }
2710 }
2711 Block::SnippetLine(content) => {
2712 // Preserve snippet delimiters (-8<-) as-is on their own line
2713 // Unlike semantic lines, snippet lines don't add extra blank lines
2714 if is_first_block {
2715 // First block starts with marker
2716 result.push(format!("{marker}{content}"));
2717 is_first_block = false;
2718 } else {
2719 // Subsequent blocks use expected indent
2720 result.push(format!("{expected_indent}{content}"));
2721 }
2722 // No blank lines added before or after snippet delimiters
2723 }
2724 Block::DivMarker(content) => {
2725 // Preserve div markers (::: opening or closing) as-is on their own line
2726 if is_first_block {
2727 result.push(format!("{marker}{content}"));
2728 is_first_block = false;
2729 } else {
2730 result.push(format!("{expected_indent}{content}"));
2731 }
2732 }
2733 Block::Html {
2734 lines: html_lines,
2735 has_preceding_blank: _,
2736 } => {
2737 // Preserve HTML blocks exactly as-is with original indentation
2738 // NOTE: Blank line before HTML block is handled by the previous block
2739
2740 for (idx, line) in html_lines.iter().enumerate() {
2741 if is_first_block && idx == 0 {
2742 // First line of first block gets marker
2743 result.push(format!("{marker}{line}"));
2744 is_first_block = false;
2745 } else if line.is_empty() {
2746 // Preserve blank lines inside HTML blocks
2747 result.push(String::new());
2748 } else {
2749 // Preserve lines with their original content (already includes indentation)
2750 result.push(format!("{expected_indent}{line}"));
2751 }
2752 }
2753
2754 // Add blank line after HTML block if there's a next block.
2755 // Only add if not already ending with one (avoids double blanks
2756 // when the HTML block itself contained a trailing blank line).
2757 if block_idx < blocks.len() - 1 {
2758 let next_block = &blocks[block_idx + 1];
2759 let should_add_blank = match next_block {
2760 Block::Code {
2761 has_preceding_blank, ..
2762 } => *has_preceding_blank,
2763 Block::Html {
2764 has_preceding_blank, ..
2765 } => *has_preceding_blank,
2766 Block::Table {
2767 has_preceding_blank, ..
2768 } => *has_preceding_blank,
2769 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2770 _ => true, // For all other blocks, add blank line
2771 };
2772 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2773 result.push(String::new());
2774 }
2775 }
2776 }
2777 Block::Table {
2778 lines: table_lines,
2779 has_preceding_blank: _,
2780 } => {
2781 // Preserve table rows verbatim with their original indentation.
2782 // Reflowing rows would corrupt column alignment and inject `|`
2783 // characters mid-paragraph (issue #590).
2784 // The leading blank line is emitted by the previous block.
2785 for (idx, (content, orig_indent)) in table_lines.iter().enumerate() {
2786 if is_first_block && idx == 0 {
2787 // First line of first block gets the list marker
2788 result.push(format!(
2789 "{marker}{}",
2790 " ".repeat(orig_indent.saturating_sub(marker_len)) + content
2791 ));
2792 is_first_block = false;
2793 } else {
2794 // Shift nested table rows with the marker so they
2795 // stay aligned when MD030 widens marker spacing.
2796 result.push(format!(
2797 "{}{}",
2798 " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
2799 content
2800 ));
2801 }
2802 }
2803
2804 // Add blank line after table block if there's a next block.
2805 if block_idx < blocks.len() - 1 {
2806 let next_block = &blocks[block_idx + 1];
2807 let should_add_blank = match next_block {
2808 Block::Code {
2809 has_preceding_blank, ..
2810 } => *has_preceding_blank,
2811 Block::Html {
2812 has_preceding_blank, ..
2813 } => *has_preceding_blank,
2814 Block::Table {
2815 has_preceding_blank, ..
2816 } => *has_preceding_blank,
2817 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2818 _ => true,
2819 };
2820 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2821 result.push(String::new());
2822 }
2823 }
2824 }
2825 Block::Admonition {
2826 header,
2827 header_indent,
2828 content_lines: admon_lines,
2829 } => {
2830 // Reconstruct admonition block with header at original indent
2831 // and body content reflowed to fit within the line length limit
2832
2833 // Add blank line before admonition if not first block
2834 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
2835 result.push(String::new());
2836 }
2837
2838 // Output the header at its original indent
2839 let header_indent_str = " ".repeat(*header_indent);
2840 if is_first_block {
2841 result.push(format!(
2842 "{marker}{}",
2843 " ".repeat(header_indent.saturating_sub(marker_len)) + header
2844 ));
2845 is_first_block = false;
2846 } else {
2847 result.push(format!("{header_indent_str}{header}"));
2848 }
2849
2850 // Derive body indent from the first non-empty content line's
2851 // stored indent, falling back to header_indent + 4 for
2852 // empty-body admonitions
2853 let body_indent = admon_lines
2854 .iter()
2855 .find(|(content, _)| !content.is_empty())
2856 .map_or(header_indent + 4, |(_, indent)| *indent);
2857 let body_indent_str = " ".repeat(body_indent);
2858
2859 // Segment body content into code blocks (verbatim) and
2860 // text paragraphs (reflowable), separated by blank lines.
2861 // Code lines store (content, orig_indent) to reconstruct
2862 // internal indentation relative to body_indent.
2863 enum AdmonSegment {
2864 Text(Vec<String>),
2865 Code(Vec<(String, usize)>),
2866 }
2867
2868 let mut segments: Vec<AdmonSegment> = Vec::new();
2869 let mut current_text: Vec<String> = Vec::new();
2870 let mut current_code: Vec<(String, usize)> = Vec::new();
2871 let mut in_admon_code = false;
2872 // Track the opening fence character so closing fences
2873 // must match (backticks close backticks, tildes close tildes)
2874 let mut fence_char: char = '`';
2875
2876 // Opening fences: ``` or ~~~ followed by optional info string
2877 let get_opening_fence = |s: &str| -> Option<(char, usize)> {
2878 let t = s.trim_start();
2879 if t.starts_with("```") {
2880 Some(('`', t.bytes().take_while(|&b| b == b'`').count()))
2881 } else if t.starts_with("~~~") {
2882 Some(('~', t.bytes().take_while(|&b| b == b'~').count()))
2883 } else {
2884 None
2885 }
2886 };
2887 // Closing fences: ONLY fence chars + optional trailing spaces
2888 let get_closing_fence = |s: &str| -> Option<(char, usize)> {
2889 let t = s.trim();
2890 if t.starts_with("```") && t.bytes().all(|b| b == b'`') {
2891 Some(('`', t.len()))
2892 } else if t.starts_with("~~~") && t.bytes().all(|b| b == b'~') {
2893 Some(('~', t.len()))
2894 } else {
2895 None
2896 }
2897 };
2898 let mut fence_len: usize = 3;
2899
2900 for (content, orig_indent) in admon_lines {
2901 if in_admon_code {
2902 // Closing fence must use the same character, be
2903 // at least as long, and have no info string
2904 if let Some((ch, len)) = get_closing_fence(content)
2905 && ch == fence_char
2906 && len >= fence_len
2907 {
2908 current_code.push((content.clone(), *orig_indent));
2909 in_admon_code = false;
2910 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
2911 continue;
2912 }
2913 current_code.push((content.clone(), *orig_indent));
2914 } else if let Some((ch, len)) = get_opening_fence(content) {
2915 if !current_text.is_empty() {
2916 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
2917 }
2918 in_admon_code = true;
2919 fence_char = ch;
2920 fence_len = len;
2921 current_code.push((content.clone(), *orig_indent));
2922 } else if content.is_empty() {
2923 if !current_text.is_empty() {
2924 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
2925 }
2926 } else {
2927 current_text.push(content.clone());
2928 }
2929 }
2930 if in_admon_code && !current_code.is_empty() {
2931 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
2932 }
2933 if !current_text.is_empty() {
2934 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
2935 }
2936
2937 // Build reflow options once for all text segments
2938 let admon_reflow_length = if config.line_length.is_unlimited() {
2939 usize::MAX
2940 } else {
2941 config.line_length.get().saturating_sub(body_indent).max(1)
2942 };
2943
2944 let admon_reflow_options = Self::reflow_options(ctx, config, admon_reflow_length);
2945
2946 // Output each segment
2947 for segment in &segments {
2948 // Blank line before each segment (after the header or previous segment)
2949 result.push(String::new());
2950
2951 match segment {
2952 AdmonSegment::Code(lines) => {
2953 for (line, orig_indent) in lines {
2954 if line.is_empty() {
2955 // Preserve blank lines inside code blocks
2956 result.push(String::new());
2957 } else {
2958 // Reconstruct with body_indent + any extra
2959 // indentation the line had beyond body_indent
2960 let extra = orig_indent.saturating_sub(body_indent);
2961 let indent_str = " ".repeat(body_indent + extra);
2962 result.push(format!("{indent_str}{line}"));
2963 }
2964 }
2965 }
2966 AdmonSegment::Text(lines) => {
2967 let paragraph_text = lines.join(" ").trim().to_string();
2968 if paragraph_text.is_empty() {
2969 continue;
2970 }
2971 let reflowed = crate::utils::text_reflow::reflow_line(
2972 ¶graph_text,
2973 &admon_reflow_options,
2974 );
2975 for line in &reflowed {
2976 result.push(format!("{body_indent_str}{line}"));
2977 }
2978 }
2979 }
2980 }
2981
2982 // Add blank line after admonition if there's a next block
2983 if block_idx < blocks.len() - 1 {
2984 let next_block = &blocks[block_idx + 1];
2985 let should_add_blank = match next_block {
2986 Block::Code {
2987 has_preceding_blank, ..
2988 } => *has_preceding_blank,
2989 Block::Table {
2990 has_preceding_blank, ..
2991 } => *has_preceding_blank,
2992 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2993 _ => true,
2994 };
2995 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2996 result.push(String::new());
2997 }
2998 }
2999 }
3000 }
3001 }
3002
3003 let reflowed_text = result.join(line_ending);
3004
3005 // Preserve trailing newline
3006 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
3007 format!("{reflowed_text}{line_ending}")
3008 } else {
3009 reflowed_text
3010 };
3011
3012 // Get the original text to compare
3013 let original_text = &ctx.content[byte_range.clone()];
3014
3015 // Physical-line-length scan, shared by the Normalize-mode gate and its
3016 // message. The list-item reflow preserves code blocks, HTML blocks,
3017 // admonition headers, fence markers, semantic markers, and snippet/div
3018 // markers verbatim; only paragraph content and admonition bodies are
3019 // restructured. Only those lines drive the length warning, so that
3020 // preserved-but-overlong content does not keep the paragraph-level
3021 // warning alive when the reflow would not fix that line.
3022 let should_count_for_length = |line_idx: usize| -> bool {
3023 let line = lines[line_idx];
3024 let trimmed = line.trim();
3025 if trimmed.is_empty() || is_exempt_line(line) {
3026 return false;
3027 }
3028 let info = &ctx.lines[line_idx];
3029 if info.in_code_block || info.in_html_block {
3030 return false;
3031 }
3032 if info.in_admonition && mkdocs_admonitions::is_admonition_start(line) {
3033 return false;
3034 }
3035 if is_fence_marker(line) || is_semantic_line(line) {
3036 return false;
3037 }
3038 if is_snippet_block_delimiter(line) {
3039 return false;
3040 }
3041 if line.trim_start().starts_with(":::") {
3042 return false;
3043 }
3044 true
3045 };
3046 let max_physical_length = (list_start..i)
3047 .filter(|&idx| should_count_for_length(idx))
3048 .map(|idx| self.calculate_effective_length(lines[idx]))
3049 .max()
3050 .unwrap_or(0);
3051 // `line-length = 0` means "no limit", so no physical line can be
3052 // "over"; the message below then describes a structural join rather
3053 // than a length violation.
3054 let any_paragraph_line_over =
3055 !config.line_length.is_unlimited() && max_physical_length > config.line_length.get();
3056
3057 // Normalize mode reflows list-item prose just like paragraphs:
3058 // joining continuation lines and re-wrapping to `line-length`.
3059 // `prose_changed` is true only when the reflow alters the words or
3060 // line breaks, not when it would merely re-indent continuation
3061 // lines or trim trailing whitespace. Comparing the texts with each
3062 // line's leading and trailing whitespace removed isolates "did the
3063 // words/line breaks change" from "did the surrounding whitespace
3064 // change". Continuation indentation is MD077's responsibility and
3065 // trailing whitespace is MD009's; an MD013 warning for either would
3066 // both duplicate those rules and resurface a persistent advisory on
3067 // already-fitting items that users disable MD013 fixing to avoid.
3068 let prose_changed = {
3069 let stripped = |text: &str| text.lines().map(str::trim).collect::<Vec<_>>().join("\n");
3070 stripped(original_text) != stripped(&replacement)
3071 };
3072 // Warn when the reflow rewraps prose (the normalize feature for
3073 // list items), or when a physical line genuinely exceeds the limit
3074 // and the reflow can change something (a true length violation,
3075 // even if all that changes is the continuation indent). A line that
3076 // is already optimal in both respects produces no warning.
3077 let gate_ok = prose_changed || (any_paragraph_line_over && original_text != replacement);
3078 if gate_ok {
3079 // Generate an appropriate message based on why reflow is needed
3080 let message = match config.reflow_mode {
3081 ReflowMode::SentencePerLine => {
3082 let num_sentences = split_into_sentences(&combined_content).len();
3083 let num_lines = content_lines.len();
3084 if num_lines == 1 {
3085 // Single line with multiple sentences
3086 format!("Line contains {num_sentences} sentences (one sentence per line required)")
3087 } else {
3088 // Multiple lines - could be split sentences or mixed
3089 format!(
3090 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
3091 )
3092 }
3093 }
3094 ReflowMode::SemanticLineBreaks => {
3095 let num_sentences = split_into_sentences(&combined_content).len();
3096 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
3097 }
3098 ReflowMode::Normalize => {
3099 // When a physical line genuinely exceeds the limit, report
3100 // it as a length violation. Otherwise the reflow is a
3101 // structural normalization (joining/re-wrapping multi-line
3102 // content that already fits), mirroring the paragraph path.
3103 if any_paragraph_line_over {
3104 format!(
3105 "Line length {} exceeds {} characters",
3106 max_physical_length,
3107 config.line_length.get()
3108 )
3109 } else {
3110 format!(
3111 "List item could be normalized to use line length of {} characters",
3112 config.line_length.get()
3113 )
3114 }
3115 }
3116 ReflowMode::Default => {
3117 // Report the actual longest non-exempt line, not the combined content
3118 let max_length = (list_start..i)
3119 .filter(|&line_idx| {
3120 let line = lines[line_idx];
3121 let trimmed = line.trim();
3122 !trimmed.is_empty() && !is_exempt_line(line)
3123 })
3124 .map(|line_idx| self.calculate_effective_length(lines[line_idx]))
3125 .max()
3126 .unwrap_or(0);
3127 format!(
3128 "Line length {} exceeds {} characters",
3129 max_length,
3130 config.line_length.get()
3131 )
3132 }
3133 };
3134
3135 warnings.push(LintWarning {
3136 rule_name: Some(self.name().to_string()),
3137 message,
3138 line: list_start + 1,
3139 column: 1,
3140 end_line: end_line + 1,
3141 end_column: lines[end_line].chars().count() + 1,
3142 severity: Severity::Warning,
3143 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
3144 });
3145 }
3146 }
3147 continue;
3148 }
3149
3150 // Found start of a paragraph - collect all lines in it
3151 let paragraph_start = i;
3152 let mut paragraph_lines = vec![lines[i]];
3153 i += 1;
3154
3155 while i < lines.len() {
3156 let next_line = lines[i];
3157 let next_line_num = i + 1;
3158 let next_trimmed = next_line.trim();
3159
3160 // Stop at paragraph boundaries
3161 if next_trimmed.is_empty()
3162 || ctx.line_info(next_line_num).is_some_and(|info| info.in_code_block)
3163 || ctx.line_info(next_line_num).is_some_and(|info| info.in_front_matter)
3164 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_block)
3165 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_comment)
3166 || ctx.line_info(next_line_num).is_some_and(|info| info.in_esm_block)
3167 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_expression)
3168 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_block)
3169 || ctx.line_info(next_line_num).is_some_and(|info| info.in_mdx_comment)
3170 || ctx
3171 .line_info(next_line_num)
3172 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
3173 || (next_line_num > 0
3174 && next_line_num <= ctx.lines.len()
3175 && ctx.lines[next_line_num - 1].blockquote.is_some())
3176 || next_trimmed.starts_with('#')
3177 || TableUtils::is_potential_table_row(next_line)
3178 || is_list_item(next_trimmed)
3179 || is_horizontal_rule(next_line)
3180 || (next_trimmed.starts_with('[') && next_line.contains("]:"))
3181 || is_template_directive_only(next_line)
3182 || is_standalone_attr_list(next_line)
3183 || is_snippet_block_delimiter(next_line)
3184 || ctx.line_info(next_line_num).is_some_and(|info| info.is_div_marker)
3185 || is_html_only_line(next_line)
3186 || (!config.strict && is_standalone_link_or_image_line(next_line))
3187 {
3188 break;
3189 }
3190
3191 // Check if the previous line ends with a hard break (2+ spaces or backslash)
3192 if i > 0 && has_hard_break(lines[i - 1]) {
3193 // Don't include lines after hard breaks in the same paragraph
3194 break;
3195 }
3196
3197 paragraph_lines.push(next_line);
3198 i += 1;
3199 }
3200
3201 // Compute the common leading indent of all non-empty paragraph lines,
3202 // but only when those lines are structurally inside a list block.
3203 // Indented continuation lines that follow a nested list arrive here
3204 // with their structural indentation intact (e.g. 2 spaces for a
3205 // top-level list item). Stripping the indent before reflow and
3206 // re-applying it afterward prevents the fixer from moving those
3207 // lines to column 0.
3208 //
3209 // The list-block guard is essential: top-level paragraphs that happen
3210 // to start with spaces (insignificant in Markdown) must NOT have those
3211 // spaces preserved or injected by the fixer.
3212 let common_indent: String = if ctx.is_in_list_block(paragraph_start + 1) {
3213 let min_len = paragraph_lines
3214 .iter()
3215 .filter(|l| !l.trim().is_empty())
3216 .map(|l| l.len() - l.trim_start().len())
3217 .min()
3218 .unwrap_or(0);
3219 paragraph_lines
3220 .iter()
3221 .find(|l| !l.trim().is_empty())
3222 .map(|l| l[..min_len].to_string())
3223 .unwrap_or_default()
3224 } else {
3225 String::new()
3226 };
3227
3228 // Combine paragraph lines into a single string for processing.
3229 // This must be done BEFORE the needs_reflow check for sentence-per-line mode.
3230 let paragraph_text = if common_indent.is_empty() {
3231 paragraph_lines.join(" ")
3232 } else {
3233 paragraph_lines
3234 .iter()
3235 .map(|l| {
3236 if l.starts_with(common_indent.as_str()) {
3237 &l[common_indent.len()..]
3238 } else {
3239 l.trim_start()
3240 }
3241 })
3242 .collect::<Vec<_>>()
3243 .join(" ")
3244 };
3245
3246 // Skip reflowing if this paragraph contains definition list items
3247 // Definition lists are multi-line structures that should not be joined
3248 let contains_definition_list = paragraph_lines
3249 .iter()
3250 .any(|line| crate::utils::is_definition_list_item(line));
3251
3252 if contains_definition_list {
3253 // Don't reflow definition lists - skip this paragraph
3254 i = paragraph_start + paragraph_lines.len();
3255 continue;
3256 }
3257
3258 // Skip reflowing if this paragraph contains MkDocs Snippets markers
3259 // Snippets blocks (-8<- ... -8<-) should be preserved exactly
3260 let contains_snippets = paragraph_lines.iter().any(|line| is_snippet_block_delimiter(line));
3261
3262 if contains_snippets {
3263 // Don't reflow Snippets blocks - skip this paragraph
3264 i = paragraph_start + paragraph_lines.len();
3265 continue;
3266 }
3267
3268 // Check if this paragraph needs reflowing
3269 let needs_reflow = match config.reflow_mode {
3270 ReflowMode::Normalize => self.normalize_mode_needs_reflow(paragraph_lines.iter().copied(), config),
3271 ReflowMode::SentencePerLine => {
3272 // In sentence-per-line mode, check if the JOINED paragraph has multiple sentences
3273 // Note: we check the joined text because sentences can span multiple lines
3274 let sentences = split_into_sentences(¶graph_text);
3275
3276 // Always reflow if multiple sentences on one line
3277 if sentences.len() > 1 {
3278 true
3279 } else if paragraph_lines.len() > 1 {
3280 // For single-sentence paragraphs spanning multiple lines:
3281 // Reflow if they COULD fit on one line (respecting line-length constraint)
3282 if config.line_length.is_unlimited() {
3283 // No line-length constraint - always join single sentences
3284 true
3285 } else {
3286 // Only join if it fits within line-length.
3287 // paragraph_text has the common indent stripped, so add it
3288 // back to get the true output length before comparing.
3289 let effective_length =
3290 self.calculate_effective_length(¶graph_text) + common_indent.len();
3291 effective_length <= config.line_length.get()
3292 }
3293 } else {
3294 false
3295 }
3296 }
3297 ReflowMode::SemanticLineBreaks => {
3298 let sentences = split_into_sentences(¶graph_text);
3299 // Reflow if multiple sentences, multiple lines, or any line exceeds limit
3300 sentences.len() > 1
3301 || paragraph_lines.len() > 1
3302 || paragraph_lines
3303 .iter()
3304 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3305 }
3306 ReflowMode::Default => {
3307 // In default mode, only reflow if lines exceed limit
3308 paragraph_lines
3309 .iter()
3310 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3311 }
3312 };
3313
3314 if needs_reflow {
3315 // Calculate byte range for this paragraph
3316 // Use whole_line_range for each line and combine
3317 let start_range = line_index.whole_line_range(paragraph_start + 1);
3318 let end_line = paragraph_start + paragraph_lines.len() - 1;
3319
3320 // For the last line, we want to preserve any trailing newline
3321 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
3322 // Last line without trailing newline - use line_text_range
3323 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
3324 } else {
3325 // Not the last line or has trailing newline - use whole_line_range
3326 line_index.whole_line_range(end_line + 1)
3327 };
3328
3329 let byte_range = start_range.start..end_range.end;
3330
3331 // Check if the paragraph ends with a hard break and what type
3332 let hard_break_type = paragraph_lines.last().and_then(|line| {
3333 let line = line.strip_suffix('\r').unwrap_or(line);
3334 if line.ends_with('\\') {
3335 Some("\\")
3336 } else if line.ends_with(" ") {
3337 Some(" ")
3338 } else {
3339 None
3340 }
3341 });
3342
3343 // Reflow the paragraph
3344 // When line_length = 0 (no limit), use a very large value for reflow
3345 let reflow_line_length = if config.line_length.is_unlimited() {
3346 usize::MAX
3347 } else {
3348 config.line_length.get()
3349 };
3350 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
3351 let mut reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
3352
3353 // Re-apply the common indent to each non-empty reflowed line so
3354 // that the replacement preserves the original structural indentation.
3355 if !common_indent.is_empty() {
3356 for line in &mut reflowed {
3357 if !line.is_empty() {
3358 *line = format!("{common_indent}{line}");
3359 }
3360 }
3361 }
3362
3363 // If the original paragraph ended with a hard break, preserve it
3364 // Preserve the original hard break format (backslash or two spaces)
3365 if let Some(break_marker) = hard_break_type
3366 && !reflowed.is_empty()
3367 {
3368 let last_idx = reflowed.len() - 1;
3369 if !has_hard_break(&reflowed[last_idx]) {
3370 reflowed[last_idx].push_str(break_marker);
3371 }
3372 }
3373
3374 let reflowed_text = reflowed.join(line_ending);
3375
3376 // Preserve trailing newline if the original paragraph had one
3377 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
3378 format!("{reflowed_text}{line_ending}")
3379 } else {
3380 reflowed_text
3381 };
3382
3383 // Get the original text to compare
3384 let original_text = &ctx.content[byte_range.clone()];
3385
3386 // Only generate a warning if the replacement is different from the original
3387 if original_text != replacement {
3388 // Determine which line ranges and messages to report based on the reflow mode.
3389 let warnings_to_report: Vec<(usize, usize, String)> = match config.reflow_mode {
3390 ReflowMode::Default => {
3391 // In default mode, report a warning for *every* line in the paragraph
3392 // that exceeds the limit. Each warning will carry the same paragraph-level
3393 // fix, making all of them auto-fixable.
3394 paragraph_lines
3395 .iter()
3396 .enumerate()
3397 .filter(|(_, line)| self.calculate_effective_length(line) > config.line_length.get())
3398 .map(|(idx, _)| {
3399 let violating_line = paragraph_start + idx + 1;
3400 (
3401 violating_line,
3402 violating_line,
3403 format!("Line length exceeds {} characters", config.line_length.get()),
3404 )
3405 })
3406 .collect()
3407 }
3408 ReflowMode::Normalize => {
3409 // In normalize mode, report the whole paragraph as needing normalization.
3410 vec![(
3411 paragraph_start + 1,
3412 end_line + 1,
3413 format!(
3414 "Paragraph could be normalized to use line length of {} characters",
3415 config.line_length.get()
3416 ),
3417 )]
3418 }
3419 ReflowMode::SentencePerLine => {
3420 // In sentence-per-line mode, highlight the entire paragraph that needs reformatting.
3421 let num_sentences = split_into_sentences(¶graph_text).len();
3422 let message = if paragraph_lines.len() == 1 {
3423 // Single line with multiple sentences
3424 format!("Line contains {num_sentences} sentences (one sentence per line required)")
3425 } else {
3426 // Multiple lines - could be split sentences or mixed
3427 let num_lines = paragraph_lines.len();
3428 format!(
3429 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
3430 )
3431 };
3432 vec![(paragraph_start + 1, paragraph_start + paragraph_lines.len(), message)]
3433 }
3434 ReflowMode::SemanticLineBreaks => {
3435 // In semantic-line-breaks mode, highlight the entire paragraph.
3436 let num_sentences = split_into_sentences(¶graph_text).len();
3437 vec![(
3438 paragraph_start + 1,
3439 paragraph_start + paragraph_lines.len(),
3440 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)"),
3441 )]
3442 }
3443 };
3444
3445 // Generate the actual lint warnings. All warnings for this paragraph
3446 // share the same paragraph-level fix.
3447 for (w_start, w_end, msg) in warnings_to_report {
3448 warnings.push(LintWarning {
3449 rule_name: Some(self.name().to_string()),
3450 message: msg,
3451 line: w_start,
3452 column: 1,
3453 end_line: w_end,
3454 end_column: lines[w_end.saturating_sub(1)].chars().count() + 1,
3455 severity: Severity::Warning,
3456 fix: Some(crate::rule::Fix::new(byte_range.clone(), replacement.clone())),
3457 });
3458 }
3459 }
3460 }
3461 }
3462
3463 warnings
3464 }
3465
3466 /// Calculate string length based on the configured length mode
3467 fn calculate_string_length(&self, s: &str) -> usize {
3468 match self.config.length_mode {
3469 LengthMode::Chars => s.chars().count(),
3470 LengthMode::Visual => s.width(),
3471 LengthMode::Bytes => s.len(),
3472 }
3473 }
3474
3475 /// Calculate effective line length
3476 ///
3477 /// Returns the actual display length of the line using the configured length mode.
3478 fn calculate_effective_length(&self, line: &str) -> usize {
3479 self.calculate_string_length(line)
3480 }
3481
3482 /// Calculate line length with inline link/image URLs removed.
3483 ///
3484 /// For each inline link `[text](url)` or image `` on the line,
3485 /// computes the "savings" from removing the URL portion (keeping only `[text]`
3486 /// or `![alt]`). Returns `effective_length - total_savings`.
3487 ///
3488 /// Handles nested constructs (e.g., `[](url)`) by only counting the
3489 /// outermost construct to avoid double-counting.
3490 fn length_without_inline_link_urls(
3491 &self,
3492 effective_length: usize,
3493 line_number: usize,
3494 ctx: &crate::lint_context::LintContext,
3495 ) -> usize {
3496 let line_range = ctx.line_index.line_content_range(line_number);
3497 let line_byte_end = line_range.end;
3498
3499 // Collect inline links/images on this line: (byte_offset, byte_end, text_only_display_len)
3500 let mut constructs: Vec<(usize, usize, usize)> = Vec::new();
3501
3502 // Binary search: links are sorted by byte_offset, so link.line is non-decreasing
3503 let link_start = ctx.links.partition_point(|l| l.line < line_number);
3504 for link in &ctx.links[link_start..] {
3505 if link.line != line_number {
3506 break;
3507 }
3508 if link.is_reference {
3509 continue;
3510 }
3511 if !matches!(link.link_type, LinkType::Inline) {
3512 continue;
3513 }
3514 if link.byte_end > line_byte_end {
3515 continue;
3516 }
3517 let text_only_len = 2 + self.calculate_string_length(&link.text);
3518 constructs.push((link.byte_offset, link.byte_end, text_only_len));
3519 }
3520
3521 let img_start = ctx.images.partition_point(|i| i.line < line_number);
3522 for image in &ctx.images[img_start..] {
3523 if image.line != line_number {
3524 break;
3525 }
3526 if image.is_reference {
3527 continue;
3528 }
3529 if !matches!(image.link_type, LinkType::Inline) {
3530 continue;
3531 }
3532 if image.byte_end > line_byte_end {
3533 continue;
3534 }
3535 let text_only_len = 3 + self.calculate_string_length(&image.alt_text);
3536 constructs.push((image.byte_offset, image.byte_end, text_only_len));
3537 }
3538
3539 if constructs.is_empty() {
3540 return effective_length;
3541 }
3542
3543 // Sort by byte offset to handle overlapping/nested constructs
3544 constructs.sort_by_key(|&(start, _, _)| start);
3545
3546 let mut total_savings: usize = 0;
3547 let mut last_end: usize = 0;
3548
3549 for (start, end, text_only_len) in &constructs {
3550 // Skip constructs nested inside a previously counted one
3551 if *start < last_end {
3552 continue;
3553 }
3554 // Full construct length in configured length mode
3555 let full_source = &ctx.content[*start..*end];
3556 let full_len = self.calculate_string_length(full_source);
3557 total_savings += full_len.saturating_sub(*text_only_len);
3558 last_end = *end;
3559 }
3560
3561 effective_length.saturating_sub(total_savings)
3562 }
3563}