rumdl_lib/rules/md013_line_length.rs
1/// Rule MD013: Line length
2///
3/// See [docs/md013.md](../../docs/md013.md) for full documentation, configuration, and examples.
4use crate::rule::{LintError, LintResult, LintWarning, Rule, RuleCategory, Severity};
5use crate::utils::mkdocs_admonitions;
6use crate::utils::mkdocs_attr_list::is_standalone_attr_list;
7use crate::utils::mkdocs_snippets::is_snippet_block_delimiter;
8use crate::utils::mkdocs_tabs;
9use crate::utils::range_utils::LineIndex;
10use crate::utils::range_utils::calculate_excess_range;
11use crate::utils::regex_cache::{IMAGE_REF_PATTERN, LINK_REF_PATTERN, URL_PATTERN};
12use crate::utils::table_utils::TableUtils;
13use crate::utils::text_reflow::{
14 BlockquoteLineData, blockquote_continuation_style, dominant_blockquote_prefix, reflow_blockquote_content,
15 split_into_sentences,
16};
17use pulldown_cmark::LinkType;
18use toml;
19
20mod block_builder;
21mod helpers;
22pub mod md013_config;
23use crate::rules::md030_list_marker_space::MD030Config;
24use crate::utils::is_template_directive_only;
25use block_builder::{Block, BlockBuilder};
26use helpers::{
27 extract_list_marker_and_content, has_hard_break, is_github_alert_marker, is_horizontal_rule, is_html_only_line,
28 is_list_item, is_standalone_link_or_image_line, is_unwrappable_line, source_list_marker, split_into_segments,
29 trim_preserving_hard_break,
30};
31pub use md013_config::MD013Config;
32use md013_config::{LengthMode, ReflowMode};
33
34#[cfg(test)]
35mod tests;
36use unicode_width::UnicodeWidthStr;
37
38#[derive(Clone, Default)]
39pub struct MD013LineLength {
40 pub(crate) config: MD013Config,
41 /// MD030 list-marker spacing, applied when reflowing list items so the rewrite
42 /// uses the configured post-marker spacing rather than a hard-coded single
43 /// space. Defaults to MD030's defaults (a single space everywhere), which
44 /// reproduces the previous behaviour exactly. See [`MD030Config::expected_spaces`].
45 pub(crate) list_spacing: MD030Config,
46}
47
48/// Blockquote paragraph line collected for reflow, with original line index for range computation.
49struct CollectedBlockquoteLine {
50 line_idx: usize,
51 data: BlockquoteLineData,
52}
53
54impl MD013LineLength {
55 pub fn new(line_length: usize, code_blocks: bool, tables: bool, headings: bool, strict: bool) -> Self {
56 Self {
57 config: MD013Config {
58 line_length: crate::types::LineLength::new(line_length),
59 code_blocks,
60 code_spans: true,
61 tables,
62 headings,
63 math_blocks: true,
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 atomic_spans: true,
77 reflow_length_exemptions: false,
78 },
79 list_spacing: MD030Config::default(),
80 }
81 }
82
83 pub fn from_config_struct(config: MD013Config) -> Self {
84 Self {
85 config,
86 list_spacing: MD030Config::default(),
87 }
88 }
89
90 /// Return a clone with code block checking disabled.
91 /// Used for doc comment linting where code blocks are Rust code managed by rustfmt.
92 pub fn with_code_blocks_disabled(&self) -> Self {
93 let mut clone = self.clone();
94 clone.config.code_blocks = false;
95 clone
96 }
97
98 /// Convert MD013 LengthMode to text_reflow ReflowLengthMode
99 /// Normalized set of reference labels defined in the document.
100 ///
101 /// Passed to reflow so a bare shortcut reference (`[text]`) is treated as an
102 /// atomic link only when its label is actually defined; an undefined
103 /// bracketed run reflows as literal prose.
104 fn defined_reference_labels(ctx: &crate::lint_context::LintContext) -> std::collections::HashSet<String> {
105 ctx.reference_defs
106 .iter()
107 .map(|d| crate::utils::text_reflow::normalize_reference_label(&d.id))
108 .collect()
109 }
110
111 /// Build the reflow options shared by every MD013 fix path.
112 ///
113 /// `line_length` varies per call site (some subtract a list-marker or
114 /// blockquote prefix); every other field is derived uniformly from the
115 /// effective config and the document flavor. Callers that need a different
116 /// `max_list_continuation_indent` override it via struct update.
117 fn reflow_options(
118 ctx: &crate::lint_context::LintContext,
119 config: &MD013Config,
120 line_length: usize,
121 ) -> crate::utils::text_reflow::ReflowOptions {
122 crate::utils::text_reflow::ReflowOptions {
123 line_length,
124 break_on_sentences: true,
125 preserve_breaks: false,
126 sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
127 semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
128 abbreviations: config.abbreviations_for_reflow(),
129 length_mode: config.reflow_length_mode(),
130 attr_lists: ctx.flavor.supports_attr_lists(),
131 myst_roles: ctx.flavor.supports_myst_roles(),
132 require_sentence_capital: config.require_sentence_capital,
133 max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
134 Some(4)
135 } else {
136 None
137 },
138 defined_references: Some(Self::defined_reference_labels(ctx)),
139 atomic_spans: config.atomic_spans,
140 length_exemptions: config.length_exemptions_for_reflow(),
141 }
142 }
143
144 fn should_ignore_line(
145 &self,
146 line: &str,
147 _lines: &[&str],
148 current_line: usize,
149 ctx: &crate::lint_context::LintContext,
150 ) -> bool {
151 if self.config.strict {
152 return false;
153 }
154
155 // Quick check for common patterns before expensive regex
156 let trimmed = line.trim();
157
158 // Only skip if the entire line is a URL (quick check first)
159 if (trimmed.starts_with("http://") || trimmed.starts_with("https://")) && URL_PATTERN.is_match(trimmed) {
160 return true;
161 }
162
163 // Only skip if the entire line is an image reference (quick check first)
164 if trimmed.starts_with("![") && trimmed.ends_with(']') && IMAGE_REF_PATTERN.is_match(trimmed) {
165 return true;
166 }
167
168 // Note: link reference definitions are handled as always-exempt (even in strict mode)
169 // in the main check loop, so they don't need to be checked here.
170
171 // Code blocks with long strings (only check if in code block)
172 if ctx.line_info(current_line + 1).is_some_and(|info| info.in_code_block)
173 && !trimmed.is_empty()
174 && !line.contains(' ')
175 && !line.contains('\t')
176 {
177 return true;
178 }
179
180 false
181 }
182
183 /// Check if rule should skip based on provided config (used for inline config support)
184 fn should_skip_with_config(&self, ctx: &crate::lint_context::LintContext, config: &MD013Config) -> bool {
185 // Skip if content is empty
186 if ctx.content.is_empty() {
187 return true;
188 }
189
190 // For sentence-per-line, semantic-line-breaks, or normalize mode, never skip based on line length
191 if config.reflow
192 && (config.reflow_mode == ReflowMode::SentencePerLine
193 || config.reflow_mode == ReflowMode::SemanticLineBreaks
194 || config.reflow_mode == ReflowMode::Normalize)
195 {
196 return false;
197 }
198
199 // Use the smallest applicable budget across line/heading/code-block
200 // contexts so a stricter context-specific limit doesn't get masked by
201 // the document-wide budget.
202 let min_limit = config.min_effective_line_length();
203 if min_limit.is_unlimited() {
204 return true;
205 }
206 let min_limit_bytes = min_limit.get();
207
208 // Quick check: if total content is shorter than the smallest line limit,
209 // definitely skip.
210 if ctx.content.len() <= min_limit_bytes {
211 return true;
212 }
213
214 // Skip if no line exceeds the smallest applicable limit.
215 !ctx.lines.iter().any(|line| line.byte_len > min_limit_bytes)
216 }
217
218 fn normalize_mode_needs_reflow<'a, I>(&self, lines: I, config: &MD013Config) -> bool
219 where
220 I: IntoIterator<Item = &'a str>,
221 {
222 let mut line_count = 0;
223 let check_length = !config.line_length.is_unlimited();
224
225 for line in lines {
226 line_count += 1;
227 if check_length && self.calculate_effective_length(line) > config.line_length.get() {
228 return true;
229 }
230 }
231
232 line_count > 1
233 }
234}
235
236impl Rule for MD013LineLength {
237 fn name(&self) -> &'static str {
238 "MD013"
239 }
240
241 fn description(&self) -> &'static str {
242 "Line length should not be excessive"
243 }
244
245 fn check(&self, ctx: &crate::lint_context::LintContext) -> LintResult {
246 // Use pre-parsed inline config from LintContext
247 let config_override = ctx.inline_config().get_rule_config("MD013");
248
249 // Apply configuration override if present
250 let effective_config = if let Some(json_config) = config_override {
251 if let Some(obj) = json_config.as_object() {
252 let mut config = self.config.clone();
253 if let Some(line_length) = obj.get("line_length").and_then(serde_json::Value::as_u64) {
254 config.line_length = crate::types::LineLength::new(line_length as usize);
255 }
256 if let Some(code_blocks) = obj.get("code_blocks").and_then(serde_json::Value::as_bool) {
257 config.code_blocks = code_blocks;
258 }
259 if let Some(code_spans) = obj.get("code_spans").and_then(serde_json::Value::as_bool) {
260 config.code_spans = code_spans;
261 }
262 if let Some(tables) = obj.get("tables").and_then(serde_json::Value::as_bool) {
263 config.tables = tables;
264 }
265 if let Some(headings) = obj.get("headings").and_then(serde_json::Value::as_bool) {
266 config.headings = headings;
267 }
268 if let Some(math_blocks) = obj
269 .get("math_blocks")
270 .or_else(|| obj.get("math-blocks"))
271 .and_then(serde_json::Value::as_bool)
272 {
273 config.math_blocks = math_blocks;
274 }
275 if let Some(blockquotes) = obj.get("blockquotes").and_then(serde_json::Value::as_bool) {
276 config.blockquotes = blockquotes;
277 }
278 if let Some(strict) = obj.get("strict").and_then(serde_json::Value::as_bool) {
279 config.strict = strict;
280 }
281 if let Some(stern) = obj.get("stern").and_then(serde_json::Value::as_bool) {
282 config.stern = stern;
283 }
284 if let Some(v) = obj
285 .get("ignore_link_urls")
286 .or_else(|| obj.get("ignore-link-urls"))
287 .or_else(|| obj.get("semantic_link_understanding"))
288 .or_else(|| obj.get("semantic-link-understanding"))
289 .and_then(serde_json::Value::as_bool)
290 {
291 config.ignore_link_urls = v;
292 }
293 if let Some(reflow) = obj.get("reflow").and_then(serde_json::Value::as_bool) {
294 config.reflow = reflow;
295 }
296 if let Some(reflow_mode) = obj.get("reflow_mode").and_then(|v| v.as_str()) {
297 config.reflow_mode = match reflow_mode {
298 "default" => ReflowMode::Default,
299 "normalize" => ReflowMode::Normalize,
300 "sentence-per-line" => ReflowMode::SentencePerLine,
301 "semantic-line-breaks" => ReflowMode::SemanticLineBreaks,
302 _ => ReflowMode::default(),
303 };
304 }
305 config
306 } else {
307 self.config.clone()
308 }
309 } else {
310 self.config.clone()
311 };
312
313 // Fast early return using should_skip with EFFECTIVE config (after inline overrides)
314 // But don't skip if we're in reflow mode with Normalize or SentencePerLine
315 if self.should_skip_with_config(ctx, &effective_config)
316 && !(effective_config.reflow
317 && (effective_config.reflow_mode == ReflowMode::Normalize
318 || effective_config.reflow_mode == ReflowMode::SentencePerLine
319 || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
320 {
321 return Ok(Vec::new());
322 }
323
324 // Direct implementation without DocumentStructure
325 let mut warnings = Vec::new();
326
327 // Special handling: line_length = 0 means "no line length limit"
328 // Skip all line length checks, but still allow reflow if enabled
329 let skip_length_checks = effective_config.line_length.is_unlimited();
330
331 // Pre-filter lines that could be problematic to avoid processing all lines.
332 // Use the smallest applicable budget across line/heading/code-block contexts
333 // so candidates aren't dropped when a stricter context-specific budget applies.
334 let prefilter_limit = effective_config.min_effective_line_length();
335 let prefilter_skip = prefilter_limit.is_unlimited();
336 let mut candidate_lines = Vec::new();
337 if !skip_length_checks && !prefilter_skip {
338 for (line_idx, line_info) in ctx.lines.iter().enumerate() {
339 // Skip front matter - it should never be linted
340 if line_info.in_front_matter {
341 continue;
342 }
343
344 // Quick length check first
345 if line_info.byte_len > prefilter_limit.get() {
346 candidate_lines.push(line_idx);
347 }
348 }
349 }
350
351 // If no candidate lines and not in normalize or sentence-per-line mode, early return
352 if candidate_lines.is_empty()
353 && !(effective_config.reflow
354 && (effective_config.reflow_mode == ReflowMode::Normalize
355 || effective_config.reflow_mode == ReflowMode::SentencePerLine
356 || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
357 {
358 return Ok(warnings);
359 }
360
361 let lines = ctx.raw_lines();
362
363 // Whether a 1-indexed line is a heading. `LineInfo::heading` is an O(1)
364 // per-line field, so check it directly at each use site instead of
365 // materializing a full-document HashSet (an extra O(n) pass and
366 // allocation on a rule that runs on virtually every file).
367 let is_heading_line_num = |line_number: usize| -> bool {
368 line_number
369 .checked_sub(1)
370 .and_then(|idx| ctx.lines.get(idx))
371 .is_some_and(|line| line.heading.is_some())
372 };
373
374 // Use pre-computed table blocks from context
375 // We need this for both the table skip check AND the paragraphs check
376 let table_blocks = &ctx.table_blocks;
377 let mut table_lines_set = std::collections::HashSet::new();
378 for table in table_blocks {
379 table_lines_set.insert(table.header_line + 1);
380 table_lines_set.insert(table.delimiter_line + 1);
381 for &line in &table.content_lines {
382 table_lines_set.insert(line + 1);
383 }
384 }
385
386 // Process candidate lines for line length checks
387 'line_loop: for &line_idx in &candidate_lines {
388 let line_number = line_idx + 1;
389 let line = lines[line_idx];
390
391 // Calculate actual line length (used in warning messages)
392 let effective_length = self.calculate_effective_length(line);
393
394 // Pick the context-specific limit: heading > code-block > paragraph.
395 // Headings dominate over code-block context if a setext underline ever
396 // overlaps a fenced range (defensive — these are mutually exclusive in
397 // practice, but the explicit ordering documents intent).
398 let is_heading_line = is_heading_line_num(line_number);
399 let in_code_block = ctx.line_info(line_number).is_some_and(|info| info.in_code_block);
400 let line_limit = if is_heading_line {
401 effective_config.effective_heading_line_length().get()
402 } else if in_code_block {
403 effective_config.effective_code_block_line_length().get()
404 } else {
405 effective_config.line_length.get()
406 };
407
408 // A context-specific limit of 0 means "unlimited for this context".
409 if line_limit == 0 {
410 continue;
411 }
412
413 // Stern mode: like default, but the trailing-token forgiveness is
414 // disabled — a line with whitespace that exceeds the limit is a
415 // violation even if the excess is the final token. The "unwrappable"
416 // line exemption (single token, optionally prefixed by # or >) is
417 // still honored. Strict overrides stern entirely.
418 if effective_config.stern && !effective_config.strict && is_unwrappable_line(line) {
419 continue;
420 }
421
422 // Trailing-token forgiveness: only in default mode (not strict, not stern).
423 // If the line only exceeds the limit because of a long token at the end
424 // (URL, link chain, identifier), it passes. This matches markdownlint's
425 // behavior: line.replace(/\S*$/u, "#")
426 let check_length = if effective_config.strict || effective_config.stern {
427 effective_length
428 } else {
429 match line.rfind(char::is_whitespace) {
430 Some(pos) => {
431 let ws_char = line[pos..].chars().next().unwrap();
432 let prefix_end = pos + ws_char.len_utf8();
433 self.calculate_string_length(&line[..prefix_end]) + 1
434 }
435 None => 1, // No whitespace — entire line is a single token
436 }
437 };
438
439 // Skip lines where the check length is within the limit
440 if check_length <= line_limit {
441 continue;
442 }
443
444 // Ignore inline link/image URLs: suppress when excess comes entirely from inline URLs.
445 // Disabled by `strict` (all forgiveness off) and by `ignore_link_urls = false`
446 // (count link/image URLs toward the line length so such lines are flagged).
447 if !effective_config.strict && effective_config.ignore_link_urls {
448 let length_without_urls = self.length_without_inline_link_urls(effective_length, line_number, ctx);
449 if length_without_urls <= line_limit {
450 continue;
451 }
452 }
453
454 // Inline code spans cannot be wrapped, so reflow cannot shorten a line
455 // whose excess length is one. When code-span checking is disabled,
456 // suppress a violation that would fit once inline code spans are excluded.
457 if !effective_config.code_spans {
458 let code_span_width: usize = ctx
459 .code_spans()
460 .iter()
461 .filter(|span| span.line == line_number && span.end_line == line_number)
462 .map(|span| self.calculate_string_length(&ctx.content[span.byte_offset..span.byte_end]))
463 .sum();
464 if effective_length.saturating_sub(code_span_width) <= line_limit {
465 continue;
466 }
467 }
468
469 // Skip mkdocstrings and pymdown blocks (already handled by LintContext)
470 if ctx.lines[line_idx].in_mkdocstrings || ctx.lines[line_idx].in_pymdown_block {
471 continue;
472 }
473
474 // Skip MyST comments (% comment) — structural lines, not prose
475 if ctx.lines[line_idx].is_myst_comment {
476 continue;
477 }
478
479 // Link reference definitions are always exempt, even in strict mode.
480 // There's no way to shorten them without breaking the URL.
481 // Also check after stripping list markers, since list items may
482 // contain link ref defs as their content.
483 {
484 let trimmed = line.trim();
485 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
486 continue;
487 }
488 if is_list_item(trimmed) {
489 let (_, content) = extract_list_marker_and_content(trimmed);
490 let content_trimmed = content.trim();
491 if content_trimmed.starts_with('[')
492 && content_trimmed.contains("]:")
493 && LINK_REF_PATTERN.is_match(content_trimmed)
494 {
495 continue;
496 }
497 }
498 }
499
500 // Skip various block types efficiently
501 if !effective_config.strict {
502 // Lines whose only content is a link/image are exempt.
503 // After stripping list markers, blockquote markers, and emphasis,
504 // if only a link or image remains, there is no way to shorten it.
505 if is_standalone_link_or_image_line(ctx, line_number) {
506 continue;
507 }
508
509 // Lines consisting entirely of HTML tags are exempt.
510 // Badge lines, images with attributes, and similar inline HTML
511 // are long due to URLs in attributes and can't be meaningfully shortened.
512 if is_html_only_line(line) {
513 continue;
514 }
515
516 // Skip setext heading underlines
517 if !line.trim().is_empty() && line.trim().chars().all(|c| c == '=' || c == '-') {
518 continue;
519 }
520
521 // Skip block elements according to config flags
522 // The flags mean: true = check these elements, false = skip these elements
523 // So we skip when the flag is FALSE and the line is in that element type
524 if (!effective_config.headings && is_heading_line_num(line_number))
525 || (!effective_config.code_blocks
526 && ctx.line_info(line_number).is_some_and(|info| info.in_code_block))
527 || (!effective_config.tables && table_lines_set.contains(&line_number))
528 || (!effective_config.math_blocks && self.line_is_display_math(line_number, ctx))
529 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
530 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
531 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
532 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
533 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
534 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
535 || ctx.line_info(line_number).is_some_and(|info| info.in_pymdown_block)
536 {
537 continue;
538 }
539
540 // Check if this is a paragraph/regular text line
541 // If paragraphs = false, skip lines that are NOT in special blocks
542 // Blockquote content is treated as paragraph text, so it's not
543 // included in the special blocks list here.
544 if !effective_config.paragraphs {
545 let is_special_block = is_heading_line_num(line_number)
546 || ctx.line_info(line_number).is_some_and(|info| info.in_code_block)
547 || table_lines_set.contains(&line_number)
548 || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
549 || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
550 || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
551 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
552 || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
553 || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
554 || ctx
555 .line_info(line_number)
556 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container);
557
558 // Skip regular paragraph text when paragraphs = false
559 if !is_special_block {
560 continue;
561 }
562 }
563
564 // Skip blockquote lines when blockquotes = false.
565 // Also skip lazy continuation lines that belong to a blockquote
566 // (lines without `>` prefix that follow a blockquote line).
567 if !effective_config.blockquotes {
568 if ctx.lines[line_number - 1].blockquote.is_some() {
569 continue;
570 }
571 // Check for lazy continuation: scan backwards through
572 // non-blank lines to find if this paragraph started with
573 // a blockquote marker
574 if !line.trim().is_empty() {
575 let mut scan = line_number.saturating_sub(2);
576 loop {
577 if ctx.lines[scan].blockquote.is_some() {
578 // Found a blockquote ancestor — this is a lazy continuation
579 continue 'line_loop;
580 }
581 if lines[scan].trim().is_empty() || scan == 0 {
582 break;
583 }
584 scan -= 1;
585 }
586 }
587 }
588
589 // Skip lines that are only a URL, image ref, or link ref
590 if self.should_ignore_line(line, lines, line_idx, ctx) {
591 continue;
592 }
593 }
594
595 // In sentence-per-line mode, check if this is a single long sentence
596 // If so, emit a warning without a fix (user must manually rephrase)
597 if effective_config.reflow_mode == ReflowMode::SentencePerLine {
598 let sentences = split_into_sentences(line.trim());
599 if sentences.len() == 1 {
600 // Single sentence that's too long - warn but don't auto-fix
601 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
602
603 let (start_line, start_col, end_line, end_col) =
604 calculate_excess_range(line_number, line, line_limit);
605
606 warnings.push(LintWarning {
607 rule_name: Some(self.name().to_string()),
608 message,
609 line: start_line,
610 column: start_col,
611 end_line,
612 end_column: end_col,
613 severity: Severity::Warning,
614 fix: None, // No auto-fix for long single sentences
615 });
616 continue;
617 }
618 // Multiple sentences will be handled by paragraph-based reflow
619 continue;
620 }
621
622 // In semantic-line-breaks mode, skip per-line checks —
623 // all reflow is handled at the paragraph level with cascading splits
624 if effective_config.reflow_mode == ReflowMode::SemanticLineBreaks {
625 continue;
626 }
627
628 // Don't provide fix for individual lines when reflow is enabled
629 // Paragraph-based fixes will be handled separately
630 let fix = None;
631
632 let message = format!("Line length {effective_length} exceeds {line_limit} characters");
633
634 // Calculate precise character range for the excess portion
635 let (start_line, start_col, end_line, end_col) = calculate_excess_range(line_number, line, line_limit);
636
637 warnings.push(LintWarning {
638 rule_name: Some(self.name().to_string()),
639 message,
640 line: start_line,
641 column: start_col,
642 end_line,
643 end_column: end_col,
644 severity: Severity::Warning,
645 fix,
646 });
647 }
648
649 // If reflow is enabled, generate paragraph-based fixes
650 if effective_config.reflow {
651 let paragraph_warnings = self.generate_paragraph_fixes(ctx, &effective_config, lines);
652 // Merge paragraph warnings with line warnings, removing duplicates
653 for pw in paragraph_warnings {
654 // Remove any line warnings that overlap with this paragraph
655 warnings.retain(|w| w.line < pw.line || w.line > pw.end_line);
656 warnings.push(pw);
657 }
658 }
659
660 Ok(warnings)
661 }
662
663 fn fix(&self, ctx: &crate::lint_context::LintContext) -> Result<String, LintError> {
664 // For CLI usage, apply fixes from warnings
665 // LSP will use the warning-based fixes directly
666 let warnings = self.check(ctx)?;
667 let warnings =
668 crate::utils::fix_utils::filter_warnings_by_inline_config(warnings, ctx.inline_config(), self.name());
669
670 // If there are no fixes, return content unchanged
671 if !warnings.iter().any(|w| w.fix.is_some()) {
672 return Ok(ctx.content.to_string());
673 }
674
675 // Apply warning-based fixes
676 crate::utils::fix_utils::apply_warning_fixes(ctx.content, &warnings)
677 .map_err(|e| LintError::FixFailed(format!("Failed to apply fixes: {e}")))
678 }
679
680 fn as_any(&self) -> &dyn std::any::Any {
681 self
682 }
683
684 fn category(&self) -> RuleCategory {
685 RuleCategory::Whitespace
686 }
687
688 fn should_skip(&self, ctx: &crate::lint_context::LintContext) -> bool {
689 self.should_skip_with_config(ctx, &self.config)
690 }
691
692 crate::impl_rule_config_sections!(MD013Config);
693
694 fn config_aliases(&self) -> Option<std::collections::HashMap<String, String>> {
695 let mut aliases = std::collections::HashMap::new();
696 aliases.insert("enable_reflow".to_string(), "reflow".to_string());
697 aliases.insert("strict_sentences".to_string(), "require-sentence-capital".to_string());
698 aliases.insert("strict-sentences".to_string(), "require-sentence-capital".to_string());
699 // Kept in step with the `alias` attributes on `MD013Config::ignore_link_urls`.
700 // Serde accepts these spellings, so a config using one is honored; without
701 // them here the key validator reports a documented, working key as unknown.
702 aliases.insert(
703 "semantic-link-understanding".to_string(),
704 "ignore-link-urls".to_string(),
705 );
706 aliases.insert(
707 "semantic_link_understanding".to_string(),
708 "ignore-link-urls".to_string(),
709 );
710 Some(aliases)
711 }
712
713 fn from_config(config: &crate::config::Config) -> Box<dyn Rule>
714 where
715 Self: Sized,
716 {
717 let mut rule_config = crate::rule_config_serde::load_rule_config::<MD013Config>(config);
718 // Use global line_length if rule-specific config still has default value
719 if rule_config.line_length.get() == 80 {
720 rule_config.line_length = config.global.line_length;
721 }
722 let mut rule = Self::from_config_struct(rule_config);
723 // Pull list-marker spacing from MD030 (via the shared serde config loader)
724 // so reflow rewrites list items with the configured spacing rather than a
725 // hard-coded single space.
726 rule.list_spacing = crate::rule_config_serde::load_rule_config::<MD030Config>(config);
727 Box::new(rule)
728 }
729}
730
731impl MD013LineLength {
732 /// True when `line_num` (1-indexed) sits inside a `$$` span covering more
733 /// than one line, as seen by the byte-level math parser.
734 fn line_in_multiline_math_span(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
735 ctx.math_spans()
736 .iter()
737 .any(|span| span.is_display && span.end_line > span.line && (span.line..=span.end_line).contains(&line_num))
738 }
739
740 /// True when `line_num` (1-indexed) holds nothing but display math, whether
741 /// that is one line of a multi-line block, a delimiter line, or a whole line
742 /// that is a single complete `$$...$$` span. This is what `math-blocks =
743 /// false` exempts from the length check.
744 ///
745 /// rumdl models math twice and the two models miss different containers, so
746 /// this consults both. `math_spans()` is byte-level and sees a block opened on
747 /// a list marker line (`- $$`), which the line-level map cannot because that
748 /// line does not begin with `$$`. `LineInfo::in_math_block` is line-level and
749 /// sees a four-space-indented block inside a footnote, which the byte-level
750 /// parser reads as an indented code block. Taking the union only ever adds
751 /// coverage: neither signal fires on ordinary prose, and an unmatched `$$`
752 /// opener is flagged by neither, so a stray delimiter cannot exempt the rest
753 /// of the document.
754 fn line_is_display_math(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
755 self.line_holds_only_multiline_math(line_num, ctx)
756 || ctx.line_info(line_num).is_some_and(|info| info.in_math_block)
757 }
758
759 /// True when a multi-line `$$` span covers `line_num` (1-indexed) and the line
760 /// holds nothing besides that math.
761 ///
762 /// A delimiter line can carry Markdown outside the delimiter: `$$ trailing
763 /// prose` closes a block and then continues in prose, and `leading prose $$`
764 /// opens one at the end of a sentence. That prose is ordinary text and counts
765 /// toward the line's length like any other, so exempting the whole line would
766 /// hide arbitrarily long prose behind a delimiter. The line-level map already
767 /// leaves such mixed lines unflagged; this is the byte-level half of the same
768 /// judgement.
769 fn line_holds_only_multiline_math(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
770 let Some(info) = ctx.line_info(line_num) else {
771 return false;
772 };
773 let line = info.content(ctx.content);
774
775 ctx.math_spans().iter().any(|span| {
776 if !span.is_display || span.end_line <= span.line || !(span.line..=span.end_line).contains(&line_num) {
777 return false;
778 }
779 if line_num == span.line {
780 let before = line.get(..span.byte_offset.saturating_sub(info.byte_offset));
781 if !before.is_none_or(|before| Self::only_structure_precedes_math(before, info)) {
782 return false;
783 }
784 }
785 if line_num == span.end_line {
786 let after = line.get(span.byte_end.saturating_sub(info.byte_offset)..);
787 if !after.is_none_or(|after| after.trim().is_empty()) {
788 return false;
789 }
790 }
791 true
792 })
793 }
794
795 /// True when the text before a block's opening delimiter is only the structure
796 /// the block sits in: indentation, a blockquote marker, or the list marker
797 /// introducing it. Such a block still owns its whole line.
798 fn only_structure_precedes_math(before: &str, info: &crate::lint_context::LineInfo) -> bool {
799 let after_marker =
800 crate::utils::blockquote::parse_blockquote_prefix(before).map_or(before, |prefix| prefix.content);
801 if after_marker.trim().is_empty() {
802 return true;
803 }
804 info.list_item.as_ref().is_some_and(|item| {
805 let mut chars = before.chars();
806 chars.by_ref().take(item.content_column).count() == item.content_column && chars.as_str().trim().is_empty()
807 })
808 }
809
810 /// True when `line_num` (1-indexed) falls inside a display-math block that
811 /// spans more than one line.
812 ///
813 /// Line breaks carry meaning inside such a block: a TeX `%` comment runs to
814 /// the end of its line, so joining the lines pulls whatever followed on later
815 /// lines into the comment and drops it from the rendered equation, which can
816 /// also leave an environment unclosed. Reflow therefore leaves these blocks
817 /// alone regardless of the `math_blocks` setting, which governs only whether
818 /// their length is reported.
819 ///
820 /// This is `line_is_display_math` minus the case of a whole line that is one
821 /// complete `$$...$$` span: such a line is a single atomic element reflow can
822 /// move around freely, and only a multi-line block has meaningful internal
823 /// line breaks.
824 fn line_in_multiline_math_block(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
825 self.line_in_multiline_math_span(line_num, ctx)
826 || ctx.line_info(line_num).is_some_and(|info| {
827 info.in_math_block && !Self::is_self_contained_display_math_line(info.content(ctx.content))
828 })
829 }
830
831 /// True when `line` is a whole line holding exactly one closed `$$...$$` span.
832 fn is_self_contained_display_math_line(line: &str) -> bool {
833 let trimmed = line.trim();
834 let inner = crate::utils::blockquote::parse_blockquote_prefix(trimmed).map_or(trimmed, |p| p.content.trim());
835 inner.strip_prefix("$$").is_some_and(|rest| rest.contains("$$"))
836 }
837
838 /// True when `line_num` (1-based) sits inside a structure whose lines must be
839 /// preserved verbatim (code block, front matter, HTML/JSX/MDX block, MkDocs
840 /// container, div marker, multi-line math block, ...). Used to keep blockquote
841 /// reflow from touching quoted-looking text embedded in such structures.
842 fn line_in_verbatim_context(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
843 if self.line_in_multiline_math_block(line_num, ctx) {
844 return true;
845 }
846 ctx.line_info(line_num).is_some_and(|info| {
847 info.in_code_block
848 || info.in_front_matter
849 || info.in_html_block
850 || info.in_html_comment
851 || info.in_esm_block
852 || info.in_jsx_expression
853 || info.in_jsx_block
854 || info.in_mdx_comment
855 || info.in_mkdocstrings
856 || info.in_pymdown_block
857 || info.in_mkdocs_container()
858 || info.is_div_marker
859 })
860 }
861
862 fn is_blockquote_content_boundary(
863 &self,
864 content: &str,
865 line_num: usize,
866 ctx: &crate::lint_context::LintContext,
867 strict: bool,
868 ) -> bool {
869 let trimmed = content.trim();
870
871 trimmed.is_empty()
872 || self.line_in_verbatim_context(line_num, ctx)
873 || trimmed.starts_with('#')
874 || trimmed.starts_with("```")
875 || trimmed.starts_with("~~~")
876 || trimmed.starts_with('>')
877 || TableUtils::is_potential_table_row(content)
878 || is_list_item(trimmed)
879 || is_horizontal_rule(content)
880 || (trimmed.starts_with('[') && content.contains("]:"))
881 || is_template_directive_only(content)
882 || is_standalone_attr_list(content)
883 || is_snippet_block_delimiter(content)
884 || is_github_alert_marker(trimmed)
885 || is_html_only_line(content)
886 // A standalone link/image line is exempt from MD013 (non-strict mode),
887 // so it must end the blockquote paragraph rather than be absorbed into
888 // it, mirroring the top-level paragraph reflow boundary.
889 || (!strict && is_standalone_link_or_image_line(ctx, line_num))
890 }
891
892 #[allow(clippy::too_many_arguments)]
893 fn generate_blockquote_paragraph_fix(
894 &self,
895 ctx: &crate::lint_context::LintContext,
896 config: &MD013Config,
897 lines: &[&str],
898 line_index: &LineIndex,
899 start_idx: usize,
900 line_ending: &str,
901 // Extra indent (spaces) to prepend to the emitted `>` prefix so a blockquote
902 // nested in a list item moves with its parent's widened marker. Zero unless a
903 // non-default MD030 widened an ancestor list item.
904 ancestor_shift: isize,
905 ) -> (Option<LintWarning>, usize) {
906 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
907 return (None, start_idx + 1);
908 };
909 let target_level = start_bq.nesting_level;
910
911 let mut collected: Vec<CollectedBlockquoteLine> = Vec::new();
912 let mut i = start_idx;
913
914 while i < lines.len() {
915 if !collected.is_empty() && has_hard_break(&collected[collected.len() - 1].data.content) {
916 break;
917 }
918
919 let line_num = i + 1;
920 if line_num > ctx.lines.len() {
921 break;
922 }
923
924 if lines[i].trim().is_empty() {
925 break;
926 }
927
928 let line_bq = ctx.lines[i].blockquote.as_deref();
929 if let Some(bq) = line_bq {
930 if bq.nesting_level != target_level {
931 break;
932 }
933
934 if self.is_blockquote_content_boundary(&bq.content, line_num, ctx, config.strict) {
935 break;
936 }
937
938 collected.push(CollectedBlockquoteLine {
939 line_idx: i,
940 data: BlockquoteLineData::explicit(trim_preserving_hard_break(&bq.content), bq.prefix.clone()),
941 });
942 i += 1;
943 continue;
944 }
945
946 let lazy_content = lines[i].trim_start();
947 if self.is_blockquote_content_boundary(lazy_content, line_num, ctx, config.strict) {
948 break;
949 }
950
951 collected.push(CollectedBlockquoteLine {
952 line_idx: i,
953 data: BlockquoteLineData::lazy(trim_preserving_hard_break(lazy_content)),
954 });
955 i += 1;
956 }
957
958 if collected.is_empty() {
959 return (None, start_idx + 1);
960 }
961
962 let next_idx = i;
963 let paragraph_start = collected[0].line_idx;
964 let end_line = collected[collected.len() - 1].line_idx;
965 let line_data: Vec<BlockquoteLineData> = collected.iter().map(|l| l.data.clone()).collect();
966 let paragraph_text = line_data
967 .iter()
968 .map(|d| d.content.as_str())
969 .collect::<Vec<_>>()
970 .join(" ");
971
972 let contains_definition_list = line_data
973 .iter()
974 .any(|d| crate::utils::is_definition_list_item(&d.content));
975 if contains_definition_list {
976 return (None, next_idx);
977 }
978
979 let contains_snippets = line_data.iter().any(|d| is_snippet_block_delimiter(&d.content));
980 if contains_snippets {
981 return (None, next_idx);
982 }
983
984 let needs_reflow = match config.reflow_mode {
985 ReflowMode::Normalize => {
986 self.normalize_mode_needs_reflow(line_data.iter().map(|d| d.content.as_str()), config)
987 }
988 ReflowMode::SentencePerLine => {
989 let sentences = split_into_sentences(¶graph_text);
990 sentences.len() > 1 || line_data.len() > 1
991 }
992 ReflowMode::SemanticLineBreaks => {
993 let sentences = split_into_sentences(¶graph_text);
994 sentences.len() > 1
995 || line_data.len() > 1
996 || collected
997 .iter()
998 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get())
999 }
1000 ReflowMode::Default => collected
1001 .iter()
1002 .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get()),
1003 };
1004
1005 if !needs_reflow {
1006 return (None, next_idx);
1007 }
1008
1009 let fallback_prefix = start_bq.prefix.clone();
1010 let explicit_prefix = dominant_blockquote_prefix(&line_data, &fallback_prefix);
1011 // Shift the whole quote right to track a widened parent list item's content
1012 // column (only widening matters for nesting; a narrowed parent leaves the quote
1013 // harmlessly over-indented, which MD027/MD030 tidy).
1014 let explicit_prefix = if ancestor_shift > 0 {
1015 format!("{}{explicit_prefix}", " ".repeat(ancestor_shift as usize))
1016 } else {
1017 explicit_prefix
1018 };
1019 let continuation_style = blockquote_continuation_style(&line_data);
1020
1021 let reflow_line_length = if config.line_length.is_unlimited() {
1022 usize::MAX
1023 } else {
1024 config
1025 .line_length
1026 .get()
1027 .saturating_sub(self.calculate_string_length(&explicit_prefix))
1028 .max(1)
1029 };
1030
1031 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
1032
1033 let reflowed_with_style =
1034 reflow_blockquote_content(&line_data, &explicit_prefix, continuation_style, &reflow_options);
1035
1036 if reflowed_with_style.is_empty() {
1037 return (None, next_idx);
1038 }
1039
1040 let reflowed_text = reflowed_with_style.join(line_ending);
1041
1042 let start_range = line_index.whole_line_range(paragraph_start + 1);
1043 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
1044 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
1045 } else {
1046 line_index.whole_line_range(end_line + 1)
1047 };
1048 let byte_range = start_range.start..end_range.end;
1049
1050 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
1051 format!("{reflowed_text}{line_ending}")
1052 } else {
1053 reflowed_text
1054 };
1055
1056 let original_text = &ctx.content[byte_range.clone()];
1057 if original_text == replacement {
1058 return (None, next_idx);
1059 }
1060
1061 let (warning_line, warning_end_line) = match config.reflow_mode {
1062 ReflowMode::Normalize => (paragraph_start + 1, end_line + 1),
1063 ReflowMode::SentencePerLine | ReflowMode::SemanticLineBreaks => (paragraph_start + 1, end_line + 1),
1064 ReflowMode::Default => {
1065 let violating_line = collected
1066 .iter()
1067 .find(|line| self.calculate_effective_length(lines[line.line_idx]) > config.line_length.get())
1068 .map_or(paragraph_start + 1, |line| line.line_idx + 1);
1069 (violating_line, violating_line)
1070 }
1071 };
1072
1073 let warning = LintWarning {
1074 rule_name: Some(self.name().to_string()),
1075 message: match config.reflow_mode {
1076 ReflowMode::Normalize => format!(
1077 "Paragraph could be normalized to use line length of {} characters",
1078 config.line_length.get()
1079 ),
1080 ReflowMode::SentencePerLine => {
1081 let num_sentences = split_into_sentences(¶graph_text).len();
1082 if line_data.len() == 1 {
1083 format!("Line contains {num_sentences} sentences (one sentence per line required)")
1084 } else {
1085 let num_lines = line_data.len();
1086 format!(
1087 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
1088 )
1089 }
1090 }
1091 ReflowMode::SemanticLineBreaks => {
1092 let num_sentences = split_into_sentences(¶graph_text).len();
1093 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
1094 }
1095 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
1096 },
1097 line: warning_line,
1098 column: 1,
1099 end_line: warning_end_line,
1100 end_column: lines[warning_end_line.saturating_sub(1)].chars().count() + 1,
1101 severity: Severity::Warning,
1102 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1103 };
1104
1105 (Some(warning), next_idx)
1106 }
1107
1108 /// Reflow a single list item that lives inside a blockquote.
1109 ///
1110 /// The blockquote paragraph reflow treats a list marker as a content boundary,
1111 /// so `> - long item ...` is never wrapped even though the identical top-level
1112 /// item is. This handles that case: it collects one tight, prose-only list item
1113 /// at the starting blockquote level, reflows the item body to the configured
1114 /// width, and re-emits it with the blockquote prefix preserved and continuation
1115 /// lines aligned under the list content.
1116 ///
1117 /// Sibling and nested list items end collection and are reflowed independently
1118 /// by the caller's outer loop (a nested item carries its indent folded into the
1119 /// blockquote prefix, so it reflows correctly on its own). Items that are not
1120 /// simple tight prose - those embedding a code block, table, fence, or hard
1121 /// break - are left untouched (`None`) and the cursor still advances past the
1122 /// whole item so its inner lines are never reprocessed as loose prose.
1123 #[allow(clippy::too_many_arguments)]
1124 fn generate_blockquote_list_item_fix(
1125 &self,
1126 ctx: &crate::lint_context::LintContext,
1127 config: &MD013Config,
1128 lines: &[&str],
1129 line_index: &LineIndex,
1130 start_idx: usize,
1131 line_ending: &str,
1132 // Extra indent (spaces) to prepend to the emitted `>` prefix when this quoted
1133 // list lives inside a list item whose marker widened. Zero unless a non-default
1134 // MD030 widened an ancestor list item.
1135 ancestor_shift: isize,
1136 ) -> (Option<LintWarning>, usize) {
1137 use crate::utils::blockquote::effective_indent_in_blockquote;
1138
1139 let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
1140 return (None, start_idx + 1);
1141 };
1142
1143 // A `>`-prefixed line can be marked as a blockquote even inside a fenced code
1144 // block (or other verbatim structure); such content must never be reflowed.
1145 if self.line_in_verbatim_context(start_idx + 1, ctx) {
1146 return (None, start_idx + 1);
1147 }
1148
1149 let target_level = start_bq.nesting_level;
1150
1151 // The marker line carries the canonical blockquote prefix: its content begins
1152 // with the list marker, so no list indent has been folded into the prefix.
1153 // Track a widened parent list item's content column so the quote stays nested
1154 // (only widening can detach it; narrowing just over-indents).
1155 let bq_prefix = if ancestor_shift > 0 {
1156 format!("{}{}", " ".repeat(ancestor_shift as usize), start_bq.prefix)
1157 } else {
1158 start_bq.prefix.clone()
1159 };
1160
1161 // A thematic break opens with what looks like a bullet marker (`- - -`).
1162 // It is not a list item, and reflowing it as prose destroys the break.
1163 // The top-level reflow path applies the same exemption.
1164 if is_horizontal_rule(&start_bq.content) {
1165 return (None, start_idx + 1);
1166 }
1167
1168 let (marker, first_body) = extract_list_marker_and_content(&start_bq.content);
1169 if marker.is_empty() {
1170 return (None, start_idx + 1);
1171 }
1172 let marker_width = marker.chars().count();
1173
1174 // Continuation lines of a checkbox item align under the bullet+checkbox, but
1175 // are recognized from the bullet width, matching the top-level list reflow.
1176 let base_marker_width = ["[ ] ", "[x] ", "[X] "]
1177 .iter()
1178 .find_map(|cb| marker.find(*cb))
1179 .unwrap_or(marker_width);
1180
1181 // Collect the item: the marker line plus its tight prose continuation lines.
1182 // `end_idx` always tracks the last consumed line so the cursor advances past
1183 // the entire item, even when it turns out to be too complex to reflow safely.
1184 let first_piece = trim_preserving_hard_break(&first_body);
1185 let mut simple = !has_hard_break(&first_piece);
1186 let mut body_pieces: Vec<String> = vec![first_piece];
1187 let mut end_idx = start_idx;
1188 let mut i = start_idx + 1;
1189
1190 while i < lines.len() {
1191 let Some(bq) = ctx.lines[i].blockquote.as_deref() else {
1192 // A blank line ends the item.
1193 if lines[i].trim().is_empty() {
1194 break;
1195 }
1196 // A lazy continuation (no `>` marker) is too ambiguous to reflow
1197 // safely. Consume the whole lazy run into this item's span and leave
1198 // the item untouched, so the caller does not reflow the continuation
1199 // in isolation and leave the marker line partially fixed.
1200 simple = false;
1201 while i < lines.len() && ctx.lines[i].blockquote.is_none() && !lines[i].trim().is_empty() {
1202 end_idx = i;
1203 i += 1;
1204 }
1205 break;
1206 };
1207 if bq.nesting_level != target_level {
1208 break;
1209 }
1210
1211 let content = bq.content.as_str();
1212 if content.trim().is_empty() {
1213 // Blank quoted line ends the tight paragraph. A following indented
1214 // paragraph (loose item) is reflowed on its own by the prose path.
1215 break;
1216 }
1217
1218 let eff_indent = effective_indent_in_blockquote(lines[i], target_level, 0);
1219 if eff_indent < base_marker_width {
1220 // Dedented: a sibling list item or text outside this item. Stop here
1221 // and let the outer loop classify it.
1222 break;
1223 }
1224 if is_list_item(content) {
1225 // A nested list item: its own item, handled independently.
1226 break;
1227 }
1228
1229 // An embedded structure (code block, table, fence, nested quote, ...)
1230 // means the item is not simple prose: keep consuming so the cursor clears
1231 // the whole structure, but do not produce a fix.
1232 if self.is_blockquote_content_boundary(content, i + 1, ctx, config.strict) {
1233 simple = false;
1234 }
1235
1236 let piece = trim_preserving_hard_break(content);
1237 if has_hard_break(&piece) {
1238 simple = false;
1239 }
1240 body_pieces.push(piece);
1241 end_idx = i;
1242 i += 1;
1243 }
1244
1245 let next_idx = end_idx + 1;
1246
1247 if !simple {
1248 return (None, next_idx);
1249 }
1250
1251 let exceeds_limit =
1252 || (start_idx..=end_idx).any(|idx| self.calculate_effective_length(lines[idx]) > config.line_length.get());
1253 let body_text = body_pieces.join(" ");
1254 let body_text = body_text.trim();
1255
1256 // Some bodies cannot be shortened and must stay verbatim, matching the
1257 // exemptions the top-level list reflow applies: link reference definitions
1258 // always, and (in non-strict mode) standalone links/images and HTML-only
1259 // lines. Reflowing a link reference definition would split it after the
1260 // colon/URL and break the definition.
1261 let is_link_ref_def =
1262 body_text.starts_with('[') && body_text.contains("]:") && LINK_REF_PATTERN.is_match(body_text);
1263 let raw_marker_line = lines[start_idx];
1264 let body_is_unwrappable = is_link_ref_def
1265 || (!config.strict && is_standalone_link_or_image_line(ctx, start_idx + 1))
1266 || (!config.strict && is_html_only_line(raw_marker_line));
1267 if body_is_unwrappable {
1268 return (None, next_idx);
1269 }
1270
1271 let needs_reflow = match config.reflow_mode {
1272 ReflowMode::Normalize => body_pieces.len() > 1 || exceeds_limit(),
1273 ReflowMode::Default => exceeds_limit(),
1274 ReflowMode::SentencePerLine => split_into_sentences(body_text).len() > 1 || body_pieces.len() > 1,
1275 ReflowMode::SemanticLineBreaks => split_into_sentences(body_text).len() > 1 || exceeds_limit(),
1276 };
1277 if !needs_reflow {
1278 return (None, next_idx);
1279 }
1280
1281 // Apply MD030 list-marker spacing in the spacing-normalizing modes, mirroring
1282 // the top-level list reflow: derive the post-marker spacing from MD030 and let
1283 // the continuation indent follow the resulting content width. Default MD030 (a
1284 // single space) leaves the marker unchanged. MkDocs keeps its rigid indent.
1285 let (marker, marker_width) = if matches!(config.reflow_mode, ReflowMode::Default | ReflowMode::Normalize)
1286 && !ctx.flavor.requires_strict_list_indent()
1287 {
1288 let is_ordered = marker.starts_with(|c: char| c.is_ascii_digit());
1289 // Bullet/number portion only (e.g. `-`, `1.`); the checkbox is content.
1290 let bullet = marker.split(' ').next().unwrap_or("");
1291 let bullet_len = bullet.chars().count();
1292 let checkbox_tail = &marker[base_marker_width..];
1293 // Decide single- vs multi-line spacing from the rewritten shape. This path
1294 // only handles a single tight prose paragraph (structural or multi-paragraph
1295 // items set `simple = false` and bail out above), so the emitted item stays
1296 // multi-line solely when the joined body wraps past one line. A multi-line
1297 // *source* that collapses onto the marker line must use MD030's single-line
1298 // spacing, matching the top-level reflow. The wrap test measures at the
1299 // single-line content column so it does not depend on the spacing chosen here.
1300 let single_col = self.calculate_string_length(&bq_prefix)
1301 + bullet_len
1302 + self.list_spacing.expected_spaces(is_ordered, false, bullet_len)
1303 + checkbox_tail.chars().count();
1304 let is_multi = !body_text.is_empty()
1305 && self.calculate_effective_length(&format!("{}{body_text}", " ".repeat(single_col)))
1306 > config.line_length.effective_limit();
1307 let spaces = self.list_spacing.expected_spaces(is_ordered, is_multi, bullet_len);
1308 let new_marker = format!("{bullet}{}{checkbox_tail}", " ".repeat(spaces));
1309 let width = new_marker.chars().count();
1310 (new_marker, width)
1311 } else {
1312 (marker, marker_width)
1313 };
1314
1315 let prefix_width = self.calculate_string_length(&bq_prefix) + self.calculate_string_length(&marker);
1316 let reflow_line_length = if config.line_length.is_unlimited() {
1317 usize::MAX
1318 } else {
1319 config.line_length.get().saturating_sub(prefix_width).max(1)
1320 };
1321
1322 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
1323
1324 let reflowed = crate::utils::text_reflow::reflow_line(body_text, &reflow_options);
1325 if reflowed.is_empty() {
1326 return (None, next_idx);
1327 }
1328
1329 let continuation_indent = " ".repeat(marker_width);
1330 let reflowed_text = reflowed
1331 .iter()
1332 .enumerate()
1333 .map(|(idx, line)| {
1334 if idx == 0 {
1335 format!("{bq_prefix}{marker}{line}")
1336 } else {
1337 format!("{bq_prefix}{continuation_indent}{line}")
1338 }
1339 })
1340 .collect::<Vec<_>>()
1341 .join(line_ending);
1342
1343 let start_range = line_index.whole_line_range(start_idx + 1);
1344 let end_range = if end_idx == lines.len() - 1 && !ctx.content.ends_with('\n') {
1345 line_index.line_text_range(end_idx + 1, 1, lines[end_idx].len() + 1)
1346 } else {
1347 line_index.whole_line_range(end_idx + 1)
1348 };
1349 let byte_range = start_range.start..end_range.end;
1350
1351 let replacement = if end_idx < lines.len() - 1 || ctx.content.ends_with('\n') {
1352 format!("{reflowed_text}{line_ending}")
1353 } else {
1354 reflowed_text
1355 };
1356
1357 let original_text = &ctx.content[byte_range.clone()];
1358 if original_text == replacement {
1359 return (None, next_idx);
1360 }
1361
1362 let message = match config.reflow_mode {
1363 ReflowMode::Normalize => format!(
1364 "Paragraph could be normalized to use line length of {} characters",
1365 config.line_length.get()
1366 ),
1367 ReflowMode::SentencePerLine => {
1368 let num_sentences = split_into_sentences(body_text).len();
1369 format!("List item should have one sentence per line (found {num_sentences} sentences)")
1370 }
1371 ReflowMode::SemanticLineBreaks => {
1372 let num_sentences = split_into_sentences(body_text).len();
1373 format!("List item should use semantic line breaks ({num_sentences} sentences)")
1374 }
1375 ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
1376 };
1377
1378 let warning = LintWarning {
1379 rule_name: Some(self.name().to_string()),
1380 message,
1381 line: start_idx + 1,
1382 column: 1,
1383 end_line: end_idx + 1,
1384 end_column: lines[end_idx].chars().count() + 1,
1385 severity: Severity::Warning,
1386 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1387 };
1388
1389 (Some(warning), next_idx)
1390 }
1391
1392 /// Generate paragraph-based fixes
1393 fn generate_paragraph_fixes(
1394 &self,
1395 ctx: &crate::lint_context::LintContext,
1396 config: &MD013Config,
1397 lines: &[&str],
1398 ) -> Vec<LintWarning> {
1399 let mut warnings = Vec::new();
1400 let line_index = LineIndex::new(ctx.content);
1401
1402 // Detect the content's line ending style to preserve it in replacements.
1403 // The LSP receives content from editors which may use CRLF (Windows).
1404 // Replacements must match the original line endings to avoid false positives.
1405 let line_ending = crate::utils::line_ending::detect_line_ending(ctx.content);
1406
1407 // Ancestor list-item indent shifts, innermost last. When a reflowed parent's
1408 // marker widens under a non-default MD030 (e.g. ul-multi = 3 moves the parent's
1409 // content from column 2 to 4), its nested list/blockquote children are reflowed
1410 // independently and would otherwise keep their original indent — leaving them
1411 // under the parent's new content column, where a CommonMark parser reparses them
1412 // as siblings rather than children. Each frame is (normalized marker width,
1413 // cumulative shift applied to that item's content column); a descendant adds its
1414 // innermost open ancestor's shift to its own indent so the whole subtree moves
1415 // together. With a default MD030 and no marker padding every shift is 0, so this
1416 // is inert and the output is byte-identical.
1417 let mut list_shift_stack: Vec<(usize, isize)> = Vec::new();
1418
1419 let mut i = 0;
1420 while i < lines.len() {
1421 let line_num = i + 1;
1422
1423 // Close ancestor frames whose list item has ended at this line: a non-blank
1424 // line indented less than the frame's source content column is no longer
1425 // inside that item. Blank lines alone don't close a (loose) list item.
1426 if !list_shift_stack.is_empty()
1427 && let Some(info) = ctx.lines.get(i)
1428 && !info.is_blank
1429 {
1430 while let Some(&(content_column, _)) = list_shift_stack.last() {
1431 if info.indent < content_column {
1432 list_shift_stack.pop();
1433 } else {
1434 break;
1435 }
1436 }
1437 }
1438
1439 // Handle blockquote paragraphs with style-preserving reflow.
1440 // Skip blockquotes when blockquotes=false or paragraphs=false
1441 if line_num > 0 && line_num <= ctx.lines.len() && ctx.lines[line_num - 1].blockquote.is_some() {
1442 if !config.blockquotes || !config.paragraphs {
1443 // Skip past all blockquote lines (explicit and lazy continuations).
1444 // A lazy continuation is a non-blank line without `>` that follows
1445 // a blockquote line and isn't a structural element.
1446 let mut saw_explicit_bq = false;
1447 while i < lines.len() && i < ctx.lines.len() {
1448 if ctx.lines[i].blockquote.is_some() {
1449 saw_explicit_bq = true;
1450 i += 1;
1451 } else if saw_explicit_bq
1452 && !lines[i].trim().is_empty()
1453 && !lines[i].trim_start().starts_with('#')
1454 && !lines[i].trim_start().starts_with('>')
1455 {
1456 // Lazy continuation of preceding blockquote
1457 i += 1;
1458 } else {
1459 break;
1460 }
1461 }
1462 continue;
1463 }
1464 // A blockquote nested in a list item moves with its parent when the
1465 // parent's marker widens (see `list_shift_stack`); pass that shift so the
1466 // emitted `>` prefix lands under the parent's new content column instead
1467 // of detaching into a sibling.
1468 let ancestor_shift = list_shift_stack.last().map_or(0isize, |&(_, shift)| shift);
1469 // A list item inside the blockquote needs list-aware reflow (marker +
1470 // continuation indent); plain prose goes through the paragraph path.
1471 let is_bq_list_item = ctx.lines[i]
1472 .blockquote
1473 .as_deref()
1474 .is_some_and(|bq| is_list_item(&bq.content));
1475 let (warning, next_idx) = if is_bq_list_item {
1476 self.generate_blockquote_list_item_fix(
1477 ctx,
1478 config,
1479 lines,
1480 &line_index,
1481 i,
1482 line_ending,
1483 ancestor_shift,
1484 )
1485 } else {
1486 self.generate_blockquote_paragraph_fix(
1487 ctx,
1488 config,
1489 lines,
1490 &line_index,
1491 i,
1492 line_ending,
1493 ancestor_shift,
1494 )
1495 };
1496 if let Some(warning) = warning {
1497 warnings.push(warning);
1498 }
1499 i = next_idx;
1500 continue;
1501 }
1502
1503 // Skip special structures (but NOT MkDocs containers - those get special handling)
1504 let should_skip_due_to_line_info = ctx.line_info(line_num).is_some_and(|info| {
1505 info.in_code_block
1506 || info.in_front_matter
1507 || info.in_html_block
1508 || info.in_html_comment
1509 || info.in_esm_block
1510 || info.in_jsx_expression
1511 || info.in_jsx_block
1512 || info.in_mdx_comment
1513 || info.in_mkdocstrings
1514 || info.in_pymdown_block
1515 });
1516
1517 // Skip link reference definitions but NOT footnote definitions.
1518 // Footnote definitions (`[^id]: prose`) contain reflowable text,
1519 // while link reference definitions (`[ref]: URL`) contain URLs
1520 // that cannot be shortened.
1521 let is_link_ref_def =
1522 lines[i].trim().starts_with('[') && !lines[i].trim().starts_with("[^") && lines[i].contains("]:");
1523
1524 if should_skip_due_to_line_info
1525 || lines[i].trim().starts_with('#')
1526 || TableUtils::is_potential_table_row(lines[i])
1527 || lines[i].trim().is_empty()
1528 || is_horizontal_rule(lines[i])
1529 || is_template_directive_only(lines[i])
1530 || is_link_ref_def
1531 || ctx.line_info(line_num).is_some_and(|info| info.is_div_marker)
1532 || is_html_only_line(lines[i])
1533 || (!config.strict && is_standalone_link_or_image_line(ctx, line_num))
1534 {
1535 i += 1;
1536 continue;
1537 }
1538
1539 // Handle footnote definitions: `[^id]: prose text that can be reflowed`
1540 // Supports multi-paragraph footnotes with code blocks, blockquotes,
1541 // tables, and lists preserved verbatim.
1542 // Validate structure: must start with `[^`, contain `]:`, and the ID
1543 // must not contain `[` or `]` (prevents false matches on nested brackets)
1544 if lines[i].trim().starts_with("[^") && lines[i].contains("]:") && {
1545 let after_caret = &lines[i].trim()[2..];
1546 after_caret
1547 .find("]:")
1548 .is_some_and(|pos| pos > 0 && !after_caret[..pos].contains(['[', ']']))
1549 } {
1550 let footnote_start = i;
1551 let line = lines[i];
1552
1553 // Extract the prefix `[^id]:`
1554 let Some(colon_pos) = line.find("]:") else {
1555 i += 1;
1556 continue;
1557 };
1558 let prefix_end = colon_pos + 2;
1559 let prefix = &line[..prefix_end];
1560
1561 // Content starts after `]: ` (with optional space)
1562 let content_start = if line[prefix_end..].starts_with(' ') {
1563 prefix_end + 1
1564 } else {
1565 prefix_end
1566 };
1567 let first_content = &line[content_start..];
1568
1569 // CommonMark footnotes use 4-space continuation indent
1570 const FN_INDENT: usize = 4;
1571
1572 // --- Line classification for footnote content ---
1573 #[derive(Debug, Clone)]
1574 enum FnLineType {
1575 Content(String),
1576 Verbatim(String, usize), // preserved text, original indent
1577 Empty,
1578 }
1579
1580 // Helper: compute visual indent (tabs = 4 spaces)
1581 let visual_indent = |s: &str| -> usize {
1582 s.chars()
1583 .take_while(|c| c.is_whitespace())
1584 .map(|c| if c == '\t' { 4 } else { 1 })
1585 .sum::<usize>()
1586 };
1587
1588 // Helper: check if a trimmed line is a fence marker (homogeneous chars)
1589 let is_fence = |s: &str| -> bool {
1590 let t = s.trim();
1591 let fence_char = t.chars().next();
1592 matches!(fence_char, Some('`') | Some('~'))
1593 && t.chars().take_while(|&c| c == fence_char.unwrap()).count() >= 3
1594 };
1595
1596 // Helper: check if a trimmed line is a setext underline
1597 let is_setext_underline = |s: &str| -> bool {
1598 let t = s.trim();
1599 !t.is_empty()
1600 && (t.chars().all(|c| c == '=' || c == ' ') || t.chars().all(|c| c == '-' || c == ' '))
1601 && t.contains(['=', '-'])
1602 };
1603
1604 // Deferred body: `[^id]:\n content` — first line has no content,
1605 // actual content starts on the next indented line
1606 let deferred_body = first_content.trim().is_empty();
1607
1608 // Collect all lines belonging to this footnote definition
1609 let mut fn_lines: Vec<FnLineType> = Vec::new();
1610 if !deferred_body {
1611 fn_lines.push(FnLineType::Content(first_content.to_string()));
1612 }
1613 let mut last_consumed = i;
1614 i += 1;
1615
1616 // Strip only the footnote continuation indent, preserving
1617 // internal indentation (e.g., code block body indent)
1618 let strip_fn_indent = |s: &str| -> String {
1619 let mut chars = s.chars();
1620 let mut stripped = 0;
1621 while stripped < FN_INDENT {
1622 match chars.next() {
1623 Some('\t') => stripped += 4,
1624 Some(c) if c.is_whitespace() => stripped += 1,
1625 _ => break,
1626 }
1627 }
1628 chars.as_str().to_string()
1629 };
1630
1631 let mut in_fenced_code = false;
1632 let mut consecutive_blanks = 0u32;
1633
1634 while i < lines.len() {
1635 let next = lines[i];
1636 let next_trimmed = next.trim();
1637
1638 // Blank line handling
1639 if next_trimmed.is_empty() {
1640 consecutive_blanks += 1;
1641 // 2+ consecutive blanks terminate the footnote
1642 if consecutive_blanks >= 2 {
1643 break;
1644 }
1645
1646 // Inside a fenced code block, blank lines are part of the code
1647 if in_fenced_code {
1648 consecutive_blanks = 0; // Don't count blanks inside code blocks
1649 fn_lines.push(FnLineType::Verbatim(String::new(), 0));
1650 last_consumed = i;
1651 i += 1;
1652 continue;
1653 }
1654
1655 // Peek ahead: if next non-blank line is indented >= FN_INDENT,
1656 // this blank is an internal paragraph separator
1657 if i + 1 < lines.len() {
1658 let peek = lines[i + 1];
1659 let peek_indent = visual_indent(peek);
1660 if !peek.trim().is_empty() && peek_indent >= FN_INDENT {
1661 fn_lines.push(FnLineType::Empty);
1662 last_consumed = i;
1663 i += 1;
1664 continue;
1665 }
1666 }
1667 // No valid continuation after blank — end of footnote
1668 break;
1669 }
1670
1671 consecutive_blanks = 0;
1672 let indent = visual_indent(next);
1673
1674 // Not indented enough — end of footnote
1675 if indent < FN_INDENT {
1676 break;
1677 }
1678
1679 // Inside a fenced code block: everything is verbatim until closing fence
1680 if in_fenced_code {
1681 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1682 if is_fence(next_trimmed) {
1683 in_fenced_code = false;
1684 }
1685 last_consumed = i;
1686 i += 1;
1687 continue;
1688 }
1689
1690 // Fence opener — start verbatim code block
1691 if is_fence(next_trimmed) {
1692 in_fenced_code = true;
1693 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1694 last_consumed = i;
1695 i += 1;
1696 continue;
1697 }
1698
1699 // A multi-line display-math block is verbatim: its line breaks
1700 // carry meaning (see `line_in_multiline_math_block`).
1701 if self.line_in_multiline_math_block(i + 1, ctx) {
1702 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1703 last_consumed = i;
1704 i += 1;
1705 continue;
1706 }
1707
1708 // Indented code block: indent >= FN_INDENT + 4 (= 8 spaces)
1709 if indent >= FN_INDENT + 4 {
1710 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1711 last_consumed = i;
1712 i += 1;
1713 continue;
1714 }
1715
1716 // Structural content that must be preserved verbatim
1717 if next_trimmed.starts_with('#')
1718 || is_list_item(next_trimmed)
1719 || next_trimmed.starts_with('>')
1720 || TableUtils::is_potential_table_row(next_trimmed)
1721 || is_setext_underline(next_trimmed)
1722 || is_horizontal_rule(next_trimmed)
1723 || crate::utils::mkdocs_footnotes::is_footnote_definition(next_trimmed)
1724 {
1725 // Preserve verbatim: blockquotes, tables, lists, setext
1726 // underlines, and horizontal rules inside the footnote
1727 if next_trimmed.starts_with('>')
1728 || TableUtils::is_potential_table_row(next_trimmed)
1729 || is_list_item(next_trimmed)
1730 || is_setext_underline(next_trimmed)
1731 || is_horizontal_rule(next_trimmed)
1732 {
1733 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1734 last_consumed = i;
1735 i += 1;
1736 continue;
1737 }
1738 // Headings, new footnote defs, link refs — end the footnote
1739 break;
1740 }
1741
1742 // Link reference definitions inside footnotes are not reflowable
1743 if next_trimmed.starts_with('[')
1744 && !next_trimmed.starts_with("[^")
1745 && next_trimmed.contains("]:")
1746 && LINK_REF_PATTERN.is_match(next_trimmed)
1747 {
1748 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1749 last_consumed = i;
1750 i += 1;
1751 continue;
1752 }
1753
1754 // HTML-only lines inside footnotes are not reflowable
1755 if is_html_only_line(next_trimmed) {
1756 fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
1757 last_consumed = i;
1758 i += 1;
1759 continue;
1760 }
1761
1762 // Regular prose content
1763 fn_lines.push(FnLineType::Content(next_trimmed.to_string()));
1764 last_consumed = i;
1765 i += 1;
1766 }
1767
1768 // Nothing collected or only empty lines
1769 if fn_lines.iter().all(|l| matches!(l, FnLineType::Empty)) || fn_lines.is_empty() {
1770 continue;
1771 }
1772
1773 // --- Group into blocks ---
1774 #[derive(Debug)]
1775 enum FnBlock {
1776 Paragraph(Vec<String>),
1777 Verbatim(Vec<(String, usize)>), // (content, indent) preserved as-is
1778 }
1779
1780 let mut blocks: Vec<FnBlock> = Vec::new();
1781 let mut current_para: Vec<String> = Vec::new();
1782 let mut current_verbatim: Vec<(String, usize)> = Vec::new();
1783
1784 for fl in &fn_lines {
1785 match fl {
1786 FnLineType::Content(s) => {
1787 if !current_verbatim.is_empty() {
1788 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
1789 }
1790 current_para.push(s.clone());
1791 }
1792 FnLineType::Verbatim(s, indent) => {
1793 if !current_para.is_empty() {
1794 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
1795 }
1796 current_verbatim.push((s.clone(), *indent));
1797 }
1798 FnLineType::Empty => {
1799 if !current_para.is_empty() {
1800 blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
1801 }
1802 if !current_verbatim.is_empty() {
1803 blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
1804 }
1805 }
1806 }
1807 }
1808 if !current_para.is_empty() {
1809 blocks.push(FnBlock::Paragraph(current_para));
1810 }
1811 if !current_verbatim.is_empty() {
1812 blocks.push(FnBlock::Verbatim(current_verbatim));
1813 }
1814
1815 // --- Reflow paragraphs and reconstruct ---
1816 let prefix_display_width = prefix.chars().count() + 1; // +1 for space
1817 let reflow_line_length = if config.line_length.is_unlimited() {
1818 usize::MAX
1819 } else {
1820 config
1821 .line_length
1822 .get()
1823 .saturating_sub(FN_INDENT.max(prefix_display_width))
1824 .max(20)
1825 };
1826 // Footnote continuation uses a fixed 4-space indent, so list
1827 // continuation capping does not apply here.
1828 let reflow_options = crate::utils::text_reflow::ReflowOptions {
1829 max_list_continuation_indent: None,
1830 ..Self::reflow_options(ctx, config, reflow_line_length)
1831 };
1832
1833 let indent_str = " ".repeat(FN_INDENT);
1834 let mut result_lines: Vec<String> = Vec::new();
1835 let mut is_first_block = true;
1836
1837 for block in &blocks {
1838 match block {
1839 FnBlock::Paragraph(para_lines) => {
1840 let paragraph_text = para_lines.join(" ");
1841 let paragraph_text = paragraph_text.trim();
1842 if paragraph_text.is_empty() {
1843 continue;
1844 }
1845
1846 let reflowed = crate::utils::text_reflow::reflow_line(paragraph_text, &reflow_options);
1847 if reflowed.is_empty() {
1848 continue;
1849 }
1850
1851 // Blank line separator between blocks
1852 if !result_lines.is_empty() {
1853 result_lines.push(String::new());
1854 }
1855
1856 for (idx, rline) in reflowed.iter().enumerate() {
1857 if is_first_block && idx == 0 {
1858 result_lines.push(format!("{prefix} {rline}"));
1859 } else {
1860 result_lines.push(format!("{indent_str}{rline}"));
1861 }
1862 }
1863 is_first_block = false;
1864 }
1865 FnBlock::Verbatim(verb_lines) => {
1866 // Blank line separator between blocks
1867 if !result_lines.is_empty() {
1868 result_lines.push(String::new());
1869 }
1870
1871 if is_first_block {
1872 // Verbatim as first block in a deferred-body footnote
1873 if deferred_body {
1874 result_lines.push(prefix.to_string());
1875 }
1876 is_first_block = false;
1877 }
1878 for (content, _orig_indent) in verb_lines {
1879 result_lines.push(format!("{indent_str}{content}"));
1880 }
1881 }
1882 }
1883 }
1884
1885 // If nothing was produced, skip
1886 if result_lines.is_empty() {
1887 continue;
1888 }
1889
1890 let reflowed_text = result_lines.join(line_ending);
1891
1892 // Calculate byte range using last_consumed
1893 let start_range = line_index.whole_line_range(footnote_start + 1);
1894 let end_range = if last_consumed == lines.len() - 1 && !ctx.content.ends_with('\n') {
1895 line_index.line_text_range(last_consumed + 1, 1, lines[last_consumed].len() + 1)
1896 } else {
1897 line_index.whole_line_range(last_consumed + 1)
1898 };
1899 let byte_range = start_range.start..end_range.end;
1900
1901 let replacement = if last_consumed < lines.len() - 1 || ctx.content.ends_with('\n') {
1902 format!("{reflowed_text}{line_ending}")
1903 } else {
1904 reflowed_text
1905 };
1906
1907 let original_text = &ctx.content[byte_range.clone()];
1908 let max_length = (footnote_start..=last_consumed)
1909 .map(|idx| self.calculate_effective_length(lines[idx]))
1910 .max()
1911 .unwrap_or(0);
1912 let line_limit = if config.line_length.is_unlimited() {
1913 usize::MAX
1914 } else {
1915 config.line_length.get()
1916 };
1917 if original_text != replacement && max_length > line_limit {
1918 warnings.push(LintWarning {
1919 rule_name: Some(self.name().to_string()),
1920 message: format!(
1921 "Line length {} exceeds {} characters",
1922 max_length,
1923 config.line_length.get()
1924 ),
1925 line: footnote_start + 1,
1926 column: 1,
1927 end_line: last_consumed + 1,
1928 end_column: lines[last_consumed].chars().count() + 1,
1929 severity: Severity::Warning,
1930 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
1931 });
1932 }
1933 continue;
1934 }
1935
1936 // Handle MkDocs container content (admonitions and tabs) with indent-preserving reflow
1937 if ctx
1938 .line_info(line_num)
1939 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
1940 {
1941 // Skip admonition/tab marker lines — only reflow their indented content
1942 let current_line = lines[i];
1943 if mkdocs_admonitions::is_admonition_start(current_line) || mkdocs_tabs::is_tab_marker(current_line) {
1944 i += 1;
1945 continue;
1946 }
1947
1948 let container_start = i;
1949
1950 // Detect the actual indent level from the first content line
1951 // (supports nested admonitions with 8+ spaces)
1952 let first_line = lines[i];
1953 let base_indent_len = first_line.len() - first_line.trim_start().len();
1954 let base_indent: String = " ".repeat(base_indent_len);
1955
1956 // Collect consecutive MkDocs container paragraph lines
1957 let mut container_lines: Vec<&str> = Vec::new();
1958 while i < lines.len() {
1959 let current_line_num = i + 1;
1960 let line_info = ctx.line_info(current_line_num);
1961
1962 // Stop if we leave the MkDocs container
1963 if !line_info.is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container) {
1964 break;
1965 }
1966
1967 let line = lines[i];
1968
1969 // Stop at paragraph boundaries within the container
1970 if line.trim().is_empty() {
1971 break;
1972 }
1973
1974 // Skip list items, code blocks, headings, HTML-only lines within containers
1975 if is_list_item(line.trim())
1976 || line.trim().starts_with("```")
1977 || line.trim().starts_with("~~~")
1978 || line.trim().starts_with('#')
1979 || is_html_only_line(line)
1980 {
1981 break;
1982 }
1983
1984 container_lines.push(line);
1985 i += 1;
1986 }
1987
1988 if container_lines.is_empty() {
1989 // Must advance i to avoid infinite loop when we encounter
1990 // non-paragraph content (code block, list, heading, empty line)
1991 // at the start of an MkDocs container
1992 i += 1;
1993 continue;
1994 }
1995
1996 // Strip the base indent from each line and join for reflow
1997 let stripped_lines: Vec<&str> = container_lines
1998 .iter()
1999 .map(|line| {
2000 if line.starts_with(&base_indent) {
2001 &line[base_indent_len..]
2002 } else {
2003 line.trim_start()
2004 }
2005 })
2006 .collect();
2007 let paragraph_text = stripped_lines.join(" ");
2008
2009 // Check if reflow is needed
2010 let needs_reflow = match config.reflow_mode {
2011 ReflowMode::Normalize => self.normalize_mode_needs_reflow(container_lines.iter().copied(), config),
2012 ReflowMode::SentencePerLine => {
2013 let sentences = split_into_sentences(¶graph_text);
2014 sentences.len() > 1 || container_lines.len() > 1
2015 }
2016 ReflowMode::SemanticLineBreaks => {
2017 let sentences = split_into_sentences(¶graph_text);
2018 sentences.len() > 1
2019 || container_lines.len() > 1
2020 || container_lines
2021 .iter()
2022 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
2023 }
2024 ReflowMode::Default => container_lines
2025 .iter()
2026 .any(|line| self.calculate_effective_length(line) > config.line_length.get()),
2027 };
2028
2029 if !needs_reflow {
2030 continue;
2031 }
2032
2033 // Calculate byte range for this container paragraph
2034 let start_range = line_index.whole_line_range(container_start + 1);
2035 let end_line = container_start + container_lines.len() - 1;
2036 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
2037 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
2038 } else {
2039 line_index.whole_line_range(end_line + 1)
2040 };
2041 let byte_range = start_range.start..end_range.end;
2042
2043 // Reflow with adjusted line length (accounting for the 4-space indent)
2044 let reflow_line_length = if config.line_length.is_unlimited() {
2045 usize::MAX
2046 } else {
2047 config.line_length.get().saturating_sub(base_indent_len).max(1)
2048 };
2049 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
2050 let reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
2051
2052 // Re-add the 4-space indent to each reflowed line
2053 let reflowed_with_indent: Vec<String> =
2054 reflowed.iter().map(|line| format!("{base_indent}{line}")).collect();
2055 let reflowed_text = reflowed_with_indent.join(line_ending);
2056
2057 // Preserve trailing newline
2058 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
2059 format!("{reflowed_text}{line_ending}")
2060 } else {
2061 reflowed_text
2062 };
2063
2064 // Only generate a warning if the replacement is different
2065 let original_text = &ctx.content[byte_range.clone()];
2066 if original_text != replacement {
2067 warnings.push(LintWarning {
2068 rule_name: Some(self.name().to_string()),
2069 message: format!(
2070 "Line length {} exceeds {} characters (in MkDocs container)",
2071 container_lines.iter().map(|l| l.len()).max().unwrap_or(0),
2072 config.line_length.get()
2073 ),
2074 line: container_start + 1,
2075 column: 1,
2076 end_line: end_line + 1,
2077 end_column: lines[end_line].chars().count() + 1,
2078 severity: Severity::Warning,
2079 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
2080 });
2081 }
2082 continue;
2083 }
2084
2085 // Helper function to detect semantic line markers
2086 let is_semantic_line = |content: &str| -> bool {
2087 let trimmed = content.trim_start();
2088 let semantic_markers = [
2089 "NOTE:",
2090 "WARNING:",
2091 "IMPORTANT:",
2092 "CAUTION:",
2093 "TIP:",
2094 "DANGER:",
2095 "HINT:",
2096 "INFO:",
2097 ];
2098 semantic_markers.iter().any(|marker| trimmed.starts_with(marker))
2099 };
2100
2101 // Helper function to detect fence markers (opening or closing)
2102 let is_fence_marker = |content: &str| -> bool {
2103 let trimmed = content.trim_start();
2104 trimmed.starts_with("```") || trimmed.starts_with("~~~")
2105 };
2106
2107 // Check if this is a list item - handle it specially
2108 let trimmed = lines[i].trim();
2109 if is_list_item(trimmed) {
2110 // Collect the entire list item including continuation lines
2111 let list_start = i;
2112 let (marker, first_content) = extract_list_marker_and_content(lines[i]);
2113 let marker_len = marker.len();
2114 // The normalized marker above is what gets re-emitted; the source marker
2115 // is where the item's content actually starts. Nested blocks move by the
2116 // difference between the two content columns, so the shift must be
2117 // measured against the source, not against the normalized width.
2118 let source_marker = source_list_marker(lines[i]);
2119 let source_content_col = source_marker.as_ref().map_or(marker_len, |m| m.content_col);
2120
2121 // Checkbox ([ ]/[x]/[X]) is inline content, not part of the list marker.
2122 // Use the base bullet/number marker width for continuation recognition
2123 // so that continuation lines at 2+ spaces are collected for "- [ ] " items.
2124 let base_marker_len = if marker.contains("[ ] ") || marker.contains("[x] ") || marker.contains("[X] ") {
2125 marker.find('[').unwrap_or(marker_len)
2126 } else {
2127 marker_len
2128 };
2129
2130 // MkDocs flavor requires at least 4 spaces for list continuation
2131 // after a blank line (multi-paragraph list items). For non-blank
2132 // continuation (lines directly following the marker line), use
2133 // the natural marker width so that 2-space indent is recognized.
2134 let item_indent = ctx.lines[i].indent;
2135 let min_continuation_indent = if ctx.flavor.requires_strict_list_indent() {
2136 // Use 4-space relative indent from the list item's nesting level
2137 item_indent + (base_marker_len - item_indent).max(4)
2138 } else {
2139 marker_len
2140 };
2141 let content_continuation_indent = base_marker_len;
2142
2143 // Track lines and their types (content, code block, fence, nested list)
2144 #[derive(Clone)]
2145 enum LineType {
2146 Content(String, usize), // content and 1-indexed line number
2147 CodeBlock(String, usize), // content and original indent
2148 SemanticLine(String), // Lines starting with NOTE:, WARNING:, etc that should stay separate
2149 SnippetLine(String), // MkDocs Snippets delimiters (-8<-) that must stay on their own line
2150 DivMarker(String), // Quarto/Pandoc div markers (::: opening or closing)
2151 AdmonitionHeader(String, usize), // header text (e.g. "!!! note") and original indent
2152 AdmonitionContent(String, usize), // body content text and original indent
2153 Table(String, usize), // GFM table row, preserved verbatim with original indent
2154 Empty,
2155 }
2156
2157 let start_idx = i;
2158 let mut list_item_lines: Vec<LineType> = vec![LineType::Content(first_content, i + 1)];
2159 // Set when collection stops at a nested list item or a nested
2160 // blockquote that belongs to this item. Such structure is reflowed
2161 // independently and is therefore absent from `list_item_lines`/`blocks`,
2162 // but it still keeps the emitted item spanning multiple physical lines,
2163 // which the MD030 multi-line spacing decision below must account for.
2164 let mut has_trailing_nested_structure = false;
2165 i += 1;
2166
2167 // Collect continuation lines using ctx.lines for metadata
2168 while i < lines.len() {
2169 let line_info = &ctx.lines[i];
2170
2171 // Use pre-computed is_blank from ctx
2172 if line_info.is_blank {
2173 // Empty line - check if next line is indented (part of list item)
2174 if i + 1 < lines.len() {
2175 let next_info = &ctx.lines[i + 1];
2176
2177 // Check if next line is indented enough to be continuation
2178 if !next_info.is_blank && next_info.indent >= min_continuation_indent {
2179 // This blank line is between paragraphs/blocks in the list item
2180 list_item_lines.push(LineType::Empty);
2181 i += 1;
2182 continue;
2183 }
2184 }
2185 // No indented line after blank, end of list item
2186 break;
2187 }
2188
2189 // Use pre-computed indent from ctx
2190 let indent = line_info.indent;
2191
2192 // Valid continuation must be indented at least content_continuation_indent.
2193 // For non-blank continuation, use marker_len (e.g. 2 for "- ").
2194 // MkDocs strict 4-space requirement applies only after blank lines.
2195 if indent >= content_continuation_indent {
2196 let trimmed = line_info.content(ctx.content).trim();
2197
2198 // Check for MkDocs admonition lines inside list items BEFORE
2199 // checking in_code_block. Lines inside code blocks within
2200 // admonitions have both in_admonition and in_code_block set;
2201 // admonition membership takes priority so the entire admonition
2202 // structure (including embedded code blocks) is preserved.
2203 if line_info.in_admonition {
2204 let raw_content = line_info.content(ctx.content);
2205 if mkdocs_admonitions::is_admonition_start(raw_content) {
2206 let header_text = raw_content[indent..].trim_end().to_string();
2207 list_item_lines.push(LineType::AdmonitionHeader(header_text, indent));
2208 } else {
2209 let body_text = raw_content[indent..].trim_end().to_string();
2210 list_item_lines.push(LineType::AdmonitionContent(body_text, indent));
2211 }
2212 i += 1;
2213 continue;
2214 }
2215
2216 // Use pre-computed in_code_block from ctx
2217 if line_info.in_code_block {
2218 list_item_lines.push(LineType::CodeBlock(
2219 line_info.content(ctx.content)[indent..].to_string(),
2220 indent,
2221 ));
2222 i += 1;
2223 continue;
2224 }
2225
2226 // A multi-line display-math block inside the item is verbatim:
2227 // its line breaks carry meaning (see
2228 // `line_in_multiline_math_block`), so reuse the code-block
2229 // carrier to re-emit it unchanged.
2230 if self.line_in_multiline_math_block(i + 1, ctx) {
2231 list_item_lines.push(LineType::CodeBlock(
2232 line_info.content(ctx.content)[indent..].to_string(),
2233 indent,
2234 ));
2235 i += 1;
2236 continue;
2237 }
2238
2239 // A blockquote nested inside the list item is reflowed by the
2240 // blockquote-aware path (it preserves the `>` prefix, including the
2241 // list indent), not as list-item prose. Collecting it as Content
2242 // would strip the markers and reflow `>` as words, collapsing the
2243 // blank `>` line and dropping `>` from wrapped continuations. End
2244 // the item here so the outer loop routes the blockquote line to
2245 // generate_blockquote_paragraph_fix. Uncollect a pending blank so
2246 // the separator between the list prose and the blockquote survives.
2247 if line_info.blockquote.is_some() {
2248 has_trailing_nested_structure = true;
2249 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
2250 list_item_lines.pop();
2251 i -= 1;
2252 }
2253 break;
2254 }
2255
2256 // Check if this is a SIBLING list item (breaks parent)
2257 // Nested lists are indented >= marker_len and are PART of the parent item
2258 // Siblings are at indent < marker_len (at or before parent marker)
2259 if is_list_item(trimmed) && indent < marker_len {
2260 // This is a sibling item at same or higher level - end parent item
2261 break;
2262 }
2263
2264 // Nested list items are always processed independently
2265 // by the outer loop, so break when we encounter one.
2266 // If a blank line was collected before this, uncollect it
2267 // so the outer loop preserves the blank between parent and nested.
2268 if is_list_item(trimmed) && indent >= marker_len {
2269 has_trailing_nested_structure = true;
2270 if matches!(list_item_lines.last(), Some(LineType::Empty)) {
2271 list_item_lines.pop();
2272 i -= 1;
2273 }
2274 break;
2275 }
2276
2277 // Normal continuation vs indented code block.
2278 // Use min_continuation_indent for the threshold since
2279 // code blocks start 4 spaces beyond the expected content
2280 // level (which is min_continuation_indent for MkDocs).
2281 if indent <= min_continuation_indent + 3 {
2282 // Extract content (remove indentation and trailing whitespace)
2283 // Preserve hard breaks (2 trailing spaces) while removing excessive whitespace
2284 // See: https://github.com/rvben/rumdl/issues/76
2285 let content = trim_preserving_hard_break(&line_info.content(ctx.content)[indent..]);
2286
2287 // Check if this is a div marker (::: opening or closing)
2288 // These must be preserved on their own line, not merged into paragraphs
2289 if line_info.is_div_marker {
2290 list_item_lines.push(LineType::DivMarker(content));
2291 }
2292 // Check if this is a fence marker (opening or closing)
2293 // These should be treated as code block lines, not paragraph content
2294 else if is_fence_marker(&content) {
2295 list_item_lines.push(LineType::CodeBlock(content, indent));
2296 }
2297 // Check if this is a semantic line (NOTE:, WARNING:, etc.)
2298 else if is_semantic_line(&content) {
2299 list_item_lines.push(LineType::SemanticLine(content));
2300 }
2301 // Check if this is a snippet block delimiter (-8<- or --8<--)
2302 // These must be preserved on their own lines for MkDocs Snippets extension
2303 else if is_snippet_block_delimiter(&content) {
2304 list_item_lines.push(LineType::SnippetLine(content));
2305 }
2306 // Check if this is a GFM table row. Tables nested inside list
2307 // items must be preserved verbatim — joining them with prose
2308 // breaks the column structure.
2309 //
2310 // `is_potential_table_row` is intentionally permissive at the
2311 // row level: any line with `|` and 2+ cells qualifies. To avoid
2312 // misclassifying prose continuation lines that contain a literal
2313 // pipe (e.g. "use grep | sort to ..."), require one of:
2314 // - the row is pipe-bordered (`| ... |`), the canonical form
2315 // for tables nested in lists; or
2316 // - the next line is a delimiter row (this is a header); or
2317 // - the previous classified line was already a Table (this is
2318 // a continuation row).
2319 else if TableUtils::is_potential_table_row(&content) && {
2320 let pipe_bordered = content.trim().starts_with('|') && content.trim().ends_with('|');
2321 let next_is_delim = ctx
2322 .lines
2323 .get(i + 1)
2324 .is_some_and(|next| TableUtils::is_delimiter_row(next.content(ctx.content)));
2325 let prev_was_table = matches!(list_item_lines.last(), Some(LineType::Table(..)));
2326 pipe_bordered || next_is_delim || prev_was_table
2327 } {
2328 list_item_lines.push(LineType::Table(content, indent));
2329 } else {
2330 list_item_lines.push(LineType::Content(content, i + 1));
2331 }
2332 i += 1;
2333 } else {
2334 // indent >= min_continuation_indent + 4: indented code block
2335 list_item_lines.push(LineType::CodeBlock(
2336 line_info.content(ctx.content)[indent..].to_string(),
2337 indent,
2338 ));
2339 i += 1;
2340 }
2341 } else {
2342 // Not indented enough, end of list item
2343 break;
2344 }
2345 }
2346
2347 // Determine the output continuation indent.
2348 // Normalize/Default modes canonicalize to min_continuation_indent
2349 // (fixing over-indented continuation). Semantic/SentencePerLine
2350 // modes preserve the user's actual indent since they only fix
2351 // line breaking, not indentation.
2352 let indent_size = match config.reflow_mode {
2353 ReflowMode::SemanticLineBreaks | ReflowMode::SentencePerLine => {
2354 // Find indent of the first plain text continuation line,
2355 // skipping the marker line (index 0), nested list items,
2356 // code blocks, and blank lines.
2357 list_item_lines
2358 .iter()
2359 .enumerate()
2360 .skip(1)
2361 .find_map(|(k, lt)| {
2362 if matches!(lt, LineType::Content(..)) {
2363 Some(ctx.lines[list_start + k].indent)
2364 } else {
2365 None
2366 }
2367 })
2368 .unwrap_or(min_continuation_indent)
2369 }
2370 _ => min_continuation_indent,
2371 };
2372 // For checkbox items in mkdocs flavor, enforce minimum indent so
2373 // continuation lines use the structural list indent (4), not the
2374 // content-aligned indent (6) which Python-Markdown doesn't support
2375 let has_checkbox = base_marker_len < marker_len;
2376 let indent_size = if has_checkbox && ctx.flavor.requires_strict_list_indent() {
2377 indent_size.max(min_continuation_indent)
2378 } else {
2379 indent_size
2380 };
2381
2382 // Split list_item_lines into blocks (paragraphs, code blocks, nested lists, semantic lines, and HTML blocks)
2383 let mut builder = BlockBuilder::new_with_start_line(start_idx + 1);
2384 for line in &list_item_lines {
2385 match line {
2386 LineType::Empty => builder.feed_blank_line(),
2387 LineType::Content(content, _) => builder.feed_content(content),
2388 LineType::CodeBlock(content, indent) => builder.feed_code_line(content, *indent),
2389 LineType::SemanticLine(content) => builder.feed_semantic_line(content),
2390 LineType::SnippetLine(content) => builder.feed_snippet_line(content),
2391 LineType::DivMarker(content) => builder.feed_div_marker(content),
2392 LineType::AdmonitionHeader(header_text, indent) => {
2393 builder.feed_admonition_header(header_text, *indent)
2394 }
2395 LineType::AdmonitionContent(content, indent) => {
2396 builder.feed_admonition_content(content, *indent)
2397 }
2398 LineType::Table(content, indent) => builder.feed_table_line(content, *indent),
2399 }
2400 }
2401 let blocks = builder.finalize();
2402
2403 // Helper: check if a line (raw source or stripped content) is exempt
2404 // from line-length checks. Link reference definitions are always exempt;
2405 // standalone link/image lines are exempt when strict mode is off.
2406 // Also checks content after stripping list markers, since list item
2407 // continuation lines may contain link ref defs.
2408 let is_exempt_line = |raw_line: &str, line_num: usize| -> bool {
2409 let trimmed = raw_line.trim();
2410 // Link reference definitions: always exempt
2411 if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
2412 return true;
2413 }
2414 // Also check after stripping list markers (for list item content)
2415 if is_list_item(trimmed) {
2416 let (_, content) = extract_list_marker_and_content(trimmed);
2417 let content_trimmed = content.trim();
2418 if content_trimmed.starts_with('[')
2419 && content_trimmed.contains("]:")
2420 && LINK_REF_PATTERN.is_match(content_trimmed)
2421 {
2422 return true;
2423 }
2424 }
2425 // Standalone link/image lines: exempt when not strict
2426 if !config.strict && is_standalone_link_or_image_line(ctx, line_num) {
2427 return true;
2428 }
2429 // HTML-only lines: exempt when not strict
2430 if !config.strict && is_html_only_line(raw_line) {
2431 return true;
2432 }
2433 false
2434 };
2435
2436 // Check if reflowing is needed (only for content paragraphs, not code blocks or nested lists)
2437 // Exclude link reference definitions and standalone link lines from content
2438 // so they don't pollute combined_content or trigger false reflow.
2439 let content_lines: Vec<String> = list_item_lines
2440 .iter()
2441 .filter_map(|line| {
2442 if let LineType::Content(s, line_num) = line {
2443 if is_exempt_line(s, *line_num) {
2444 return None;
2445 }
2446 Some(s.clone())
2447 } else {
2448 None
2449 }
2450 })
2451 .collect();
2452
2453 // Check if we need to reflow this list item
2454 // We check the combined content to see if it exceeds length limits
2455 let combined_content = content_lines.join(" ").trim().to_string();
2456
2457 // Helper to check if we should reflow in normalize mode
2458 let should_normalize = || {
2459 // Don't normalize if the list item only contains nested lists, code blocks, or semantic lines
2460 // DO normalize if it has plain text content that spans multiple lines
2461 let has_code_blocks = blocks.iter().any(|b| matches!(b, Block::Code { .. }));
2462 let has_semantic_lines = blocks.iter().any(|b| matches!(b, Block::SemanticLine(_)));
2463 let has_snippet_lines = blocks.iter().any(|b| matches!(b, Block::SnippetLine(_)));
2464 let has_div_markers = blocks.iter().any(|b| matches!(b, Block::DivMarker(_)));
2465 let has_admonitions = blocks.iter().any(|b| matches!(b, Block::Admonition { .. }));
2466 let has_tables = blocks.iter().any(|b| matches!(b, Block::Table { .. }));
2467 let has_paragraphs = blocks.iter().any(|b| matches!(b, Block::Paragraph(_)));
2468
2469 // If we have structural blocks but no paragraphs, don't normalize
2470 if (has_code_blocks
2471 || has_semantic_lines
2472 || has_snippet_lines
2473 || has_div_markers
2474 || has_admonitions
2475 || has_tables)
2476 && !has_paragraphs
2477 {
2478 return false;
2479 }
2480
2481 // If we have paragraphs, check if they span multiple lines or there are multiple blocks
2482 if has_paragraphs {
2483 // Count only paragraphs that contain at least one non-exempt line.
2484 // Paragraphs consisting entirely of link ref defs or standalone links
2485 // should not trigger normalization.
2486 let paragraph_count = blocks
2487 .iter()
2488 .filter(|b| {
2489 if let Block::Paragraph(para_lines) = b {
2490 !para_lines
2491 .iter()
2492 .all(|(line, line_num)| is_exempt_line(line, *line_num))
2493 } else {
2494 false
2495 }
2496 })
2497 .count();
2498 if paragraph_count > 1 {
2499 // Multiple non-exempt paragraph blocks should be normalized
2500 return true;
2501 }
2502
2503 // Single paragraph block: normalize if it has multiple content lines
2504 if content_lines.len() > 1 {
2505 return true;
2506 }
2507 }
2508
2509 false
2510 };
2511
2512 // Integrate MD030 list-marker spacing (and MD007's text-aligned
2513 // continuation). In Default/Normalize modes — the modes that already
2514 // canonicalize spacing/indent — derive the number of spaces after the
2515 // marker from the configured MD030 values instead of forcing a single
2516 // space, then align continuation lines to the resulting content column.
2517 // Sentence/Semantic modes only adjust line breaks, so they keep the
2518 // marker spacing and indentation already present in the source.
2519 //
2520 // With default MD030 (a single space everywhere) the rebuilt marker is
2521 // byte-identical to the source marker, so this is a no-op and existing
2522 // behaviour is preserved; only a non-default MD030 changes the output.
2523 // The MkDocs flavor enforces a rigid structural indent (4 spaces,
2524 // capped via max_list_continuation_indent) that Python-Markdown
2525 // requires; leave its specialized handling untouched.
2526 let (marker, indent_size, code_indent_shift) =
2527 if matches!(config.reflow_mode, ReflowMode::Default | ReflowMode::Normalize)
2528 && !ctx.flavor.requires_strict_list_indent()
2529 && let Some(li) = ctx.lines[list_start].list_item.as_deref()
2530 {
2531 let bullet_len = li.marker.len();
2532 // The checkbox (e.g. `[ ] `) is content, not part of the list
2533 // marker MD030 governs; carry it over verbatim after the spacing.
2534 let checkbox_tail = marker[base_marker_len..].to_string();
2535 // Shift this item right by its ancestors' cumulative marker
2536 // widening so a nested item stays under its parent's (widened)
2537 // content column. Zero for top-level items and for the whole
2538 // tree under default MD030, where the source indent is preserved
2539 // verbatim (byte-identical output).
2540 let ancestor_shift = list_shift_stack.last().map_or(0isize, |&(_, shift)| shift);
2541 let shifted_indent = (item_indent as isize + ancestor_shift).max(0) as usize;
2542 let indent_prefix = if ancestor_shift == 0 {
2543 marker[..item_indent].to_string()
2544 } else {
2545 " ".repeat(shifted_indent)
2546 };
2547
2548 // Decide single- vs multi-line spacing from the *rewritten* shape,
2549 // not the source. A multi-line source is not enough: plain prose
2550 // continuation collapses onto the marker line during reflow, so a
2551 // two-line bullet that fits becomes a single physical line and must
2552 // use MD030's single-line spacing (otherwise MD013 emits a result
2553 // that MD030 immediately rewrites). The emitted item stays
2554 // multi-line only when reflow cannot collapse it:
2555 // - the prose wraps past the line length, or
2556 // - a structural block remains (code, table, admonition, semantic
2557 // line, snippet, div marker, HTML) that is not joinable prose, or
2558 // - more than one paragraph remains (blank-separated), or
2559 // - a nested list/blockquote follows (reflowed independently, so it
2560 // is absent from `blocks` but still keeps the item multi-line).
2561 // The wrap test uses the single-line content column so it is
2562 // independent of the spacing we are about to choose (avoiding a
2563 // circular result). `ol-align-column` ignores this flag entirely in
2564 // expected_spaces(), so ordered lists are unaffected.
2565 //
2566 // This is the rewritten-shape counterpart of MD030's
2567 // `is_multi_line_list_item` (which keys off the *source*). The two
2568 // are related but technically distinct and intentionally separate;
2569 // if the notion of "multi-line" changes in one, revisit the other.
2570 let single_col = shifted_indent
2571 + bullet_len
2572 + self.list_spacing.expected_spaces(li.is_ordered, false, bullet_len)
2573 + checkbox_tail.len();
2574 let prose_wraps = !combined_content.is_empty()
2575 && self
2576 .calculate_effective_length(&format!("{}{combined_content}", " ".repeat(single_col)))
2577 > config.line_length.effective_limit();
2578 let has_structural_block = blocks.iter().any(|b| !matches!(b, Block::Paragraph(_)));
2579 let multiple_paragraphs =
2580 blocks.iter().filter(|b| matches!(b, Block::Paragraph(_))).count() > 1;
2581 let is_multi =
2582 prose_wraps || has_structural_block || multiple_paragraphs || has_trailing_nested_structure;
2583
2584 let spaces = self.list_spacing.expected_spaces(li.is_ordered, is_multi, bullet_len);
2585 let new_marker = format!("{indent_prefix}{}{}{checkbox_tail}", li.marker, " ".repeat(spaces));
2586 let new_col = new_marker.chars().count();
2587 let shift = new_col as isize - source_content_col as isize;
2588 (new_marker, new_col, shift)
2589 } else {
2590 // MkDocs enforces a rigid structural indent, so the item is emitted
2591 // exactly as written and its nested blocks never move. Re-emitting
2592 // the normalized marker here would narrow the content column while
2593 // leaving those blocks behind.
2594 let marker = source_marker.map_or(marker, |m| m.text);
2595 (marker, indent_size, 0isize)
2596 };
2597 let expected_indent = " ".repeat(indent_size);
2598
2599 let needs_reflow = match config.reflow_mode {
2600 ReflowMode::Normalize => {
2601 // Only reflow if:
2602 // 1. Any non-exempt paragraph, when joined, exceeds the limit, OR
2603 // 2. Any admonition content line exceeds the limit, OR
2604 // 3. The list item should be normalized (has multi-line plain text)
2605 let any_paragraph_exceeds = blocks.iter().any(|block| match block {
2606 Block::Paragraph(para_lines) => {
2607 if para_lines
2608 .iter()
2609 .all(|(line, line_num)| is_exempt_line(line, *line_num))
2610 {
2611 return false;
2612 }
2613 let joined = para_lines.iter().map(|(l, _)| l.as_str()).collect::<Vec<_>>().join(" ");
2614 let with_marker = format!("{}{}", " ".repeat(indent_size), joined.trim());
2615 self.calculate_effective_length(&with_marker) > config.line_length.get()
2616 }
2617 Block::Admonition {
2618 content_lines,
2619 header_indent,
2620 ..
2621 } => content_lines.iter().any(|(content, indent)| {
2622 if content.is_empty() {
2623 return false;
2624 }
2625 let with_indent = format!("{}{}", " ".repeat(*indent.max(header_indent)), content);
2626 self.calculate_effective_length(&with_indent) > config.line_length.get()
2627 }),
2628 _ => false,
2629 });
2630 if any_paragraph_exceeds {
2631 true
2632 } else {
2633 should_normalize()
2634 }
2635 }
2636 ReflowMode::SentencePerLine => {
2637 // Check if list item has multiple sentences
2638 let sentences = split_into_sentences(&combined_content);
2639 sentences.len() > 1
2640 }
2641 ReflowMode::SemanticLineBreaks => {
2642 let sentences = split_into_sentences(&combined_content);
2643 sentences.len() > 1
2644 || (list_start..i).any(|line_idx| {
2645 let line = lines[line_idx];
2646 let trimmed = line.trim();
2647 if trimmed.is_empty() || is_exempt_line(line, line_idx + 1) {
2648 return false;
2649 }
2650 self.calculate_effective_length(line) > config.line_length.get()
2651 })
2652 }
2653 ReflowMode::Default => {
2654 // In default mode, only reflow if any individual non-exempt line exceeds limit
2655 (list_start..i).any(|line_idx| {
2656 let line = lines[line_idx];
2657 let trimmed = line.trim();
2658 // Skip blank lines and exempt lines
2659 if trimmed.is_empty() || is_exempt_line(line, line_idx + 1) {
2660 return false;
2661 }
2662 self.calculate_effective_length(line) > config.line_length.get()
2663 })
2664 }
2665 };
2666
2667 // Record this item's frame so its nested children inherit the shift.
2668 // Only a reflowed item's marker actually moves; an unreflowed one keeps
2669 // its source position and so contributes no shift to its children. The
2670 // threshold that decides which following lines are inside this item is
2671 // the normalized marker width, NOT `source_content_col`: the collection
2672 // loop above gathers continuations by that same width, so the frame
2673 // boundary must match it or the two would disagree about ownership of
2674 // lines indented between the normalized and the source content column.
2675 list_shift_stack.push((marker_len, if needs_reflow { code_indent_shift } else { 0 }));
2676
2677 if needs_reflow {
2678 let start_range = line_index.whole_line_range(list_start + 1);
2679 let end_line = i - 1;
2680 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
2681 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
2682 } else {
2683 line_index.whole_line_range(end_line + 1)
2684 };
2685 let byte_range = start_range.start..end_range.end;
2686
2687 // Reflow each block (paragraphs only, preserve code blocks)
2688 // When line_length = 0 (no limit), use a very large value for reflow
2689 let reflow_line_length = if config.line_length.is_unlimited() {
2690 usize::MAX
2691 } else {
2692 config.line_length.get().saturating_sub(indent_size).max(1)
2693 };
2694 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
2695
2696 let mut result: Vec<String> = Vec::new();
2697 let mut is_first_block = true;
2698
2699 for (block_idx, block) in blocks.iter().enumerate() {
2700 match block {
2701 Block::Paragraph(para_lines) => {
2702 // If every line in this paragraph is exempt (link ref defs,
2703 // standalone links), preserve the paragraph verbatim instead
2704 // of reflowing it. Reflowing would corrupt link ref defs.
2705 let all_exempt = para_lines
2706 .iter()
2707 .all(|(line, line_num)| is_exempt_line(line, *line_num));
2708
2709 if all_exempt {
2710 for (idx, (line, _)) in para_lines.iter().enumerate() {
2711 if is_first_block && idx == 0 {
2712 result.push(format!("{marker}{line}"));
2713 is_first_block = false;
2714 } else {
2715 result.push(format!("{expected_indent}{line}"));
2716 }
2717 }
2718 } else {
2719 // Split the paragraph into segments at hard break boundaries
2720 // Each segment can be reflowed independently
2721 let segments = split_into_segments(para_lines);
2722
2723 for (segment_idx, segment) in segments.iter().enumerate() {
2724 // Check if this segment ends with a hard break and what type
2725 let hard_break_type = segment.last().and_then(|(line, _)| {
2726 let line = line.strip_suffix('\r').unwrap_or(line);
2727 if line.ends_with('\\') {
2728 Some("\\")
2729 } else if line.ends_with(" ") {
2730 Some(" ")
2731 } else {
2732 None
2733 }
2734 });
2735
2736 // Join and reflow the segment (removing the hard break marker for processing)
2737 let segment_for_reflow: Vec<String> = segment
2738 .iter()
2739 .map(|(line, _)| {
2740 // Strip hard break marker (2 spaces or backslash) for reflow processing
2741 if line.ends_with('\\') {
2742 line[..line.len() - 1].trim_end().to_string()
2743 } else if line.ends_with(" ") {
2744 line[..line.len() - 2].trim_end().to_string()
2745 } else {
2746 line.clone()
2747 }
2748 })
2749 .collect();
2750
2751 let segment_text = segment_for_reflow.join(" ").trim().to_string();
2752 if !segment_text.is_empty() {
2753 let reflowed =
2754 crate::utils::text_reflow::reflow_line(&segment_text, &reflow_options);
2755
2756 if is_first_block && segment_idx == 0 {
2757 // First segment of first block starts with marker
2758 result.push(format!("{marker}{}", reflowed[0]));
2759 for line in reflowed.iter().skip(1) {
2760 result.push(format!("{expected_indent}{line}"));
2761 }
2762 is_first_block = false;
2763 } else {
2764 // Subsequent segments
2765 for line in reflowed {
2766 result.push(format!("{expected_indent}{line}"));
2767 }
2768 }
2769
2770 // If this segment had a hard break, add it back to the last line
2771 // Preserve the original hard break format (backslash or two spaces)
2772 if let Some(break_marker) = hard_break_type
2773 && let Some(last_line) = result.last_mut()
2774 {
2775 last_line.push_str(break_marker);
2776 }
2777 }
2778 }
2779 }
2780
2781 // Add blank line after paragraph block if there's a next block.
2782 // Check if next block is a code block that doesn't want a preceding blank.
2783 // Also don't add blank lines before snippet lines (they should stay tight).
2784 // Only add if not already ending with one (avoids double blanks).
2785 if block_idx < blocks.len() - 1 {
2786 let next_block = &blocks[block_idx + 1];
2787 let should_add_blank = match next_block {
2788 Block::Code {
2789 has_preceding_blank, ..
2790 } => *has_preceding_blank,
2791 Block::Table {
2792 has_preceding_blank, ..
2793 } => *has_preceding_blank,
2794 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2795 _ => true, // For all other blocks, add blank line
2796 };
2797 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2798 result.push(String::new());
2799 }
2800 }
2801 }
2802 Block::Code {
2803 lines: code_lines,
2804 has_preceding_blank: _,
2805 } => {
2806 // Preserve code blocks as-is with original indentation
2807 // NOTE: Blank line before code block is handled by the previous block
2808 // (see paragraph block's logic above)
2809
2810 for (idx, (content, orig_indent)) in code_lines.iter().enumerate() {
2811 if is_first_block && idx == 0 {
2812 // First line of first block gets marker
2813 result.push(format!(
2814 "{marker}{}",
2815 " ".repeat(orig_indent - marker_len) + content.as_str()
2816 ));
2817 is_first_block = false;
2818 } else if content.is_empty() {
2819 result.push(String::new());
2820 } else {
2821 // Shift nested code with the marker so it stays
2822 // aligned under content when MD030 widens spacing.
2823 result.push(format!(
2824 "{}{}",
2825 " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
2826 content
2827 ));
2828 }
2829 }
2830 }
2831 Block::SemanticLine(content) => {
2832 // Preserve semantic lines (NOTE:, WARNING:, etc.) as-is on their own line.
2833 // Only add blank before if not already ending with one.
2834 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
2835 result.push(String::new());
2836 }
2837
2838 if is_first_block {
2839 // First block starts with marker
2840 result.push(format!("{marker}{content}"));
2841 is_first_block = false;
2842 } else {
2843 // Subsequent blocks use expected indent
2844 result.push(format!("{expected_indent}{content}"));
2845 }
2846
2847 // Add blank line after semantic line if there's a next block.
2848 // Only add if not already ending with one.
2849 if block_idx < blocks.len() - 1 {
2850 let next_block = &blocks[block_idx + 1];
2851 let should_add_blank = match next_block {
2852 Block::Code {
2853 has_preceding_blank, ..
2854 } => *has_preceding_blank,
2855 Block::Table {
2856 has_preceding_blank, ..
2857 } => *has_preceding_blank,
2858 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2859 _ => true, // For all other blocks, add blank line
2860 };
2861 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2862 result.push(String::new());
2863 }
2864 }
2865 }
2866 Block::SnippetLine(content) => {
2867 // Preserve snippet delimiters (-8<-) as-is on their own line
2868 // Unlike semantic lines, snippet lines don't add extra blank lines
2869 if is_first_block {
2870 // First block starts with marker
2871 result.push(format!("{marker}{content}"));
2872 is_first_block = false;
2873 } else {
2874 // Subsequent blocks use expected indent
2875 result.push(format!("{expected_indent}{content}"));
2876 }
2877 // No blank lines added before or after snippet delimiters
2878 }
2879 Block::DivMarker(content) => {
2880 // Preserve div markers (::: opening or closing) as-is on their own line
2881 if is_first_block {
2882 result.push(format!("{marker}{content}"));
2883 is_first_block = false;
2884 } else {
2885 result.push(format!("{expected_indent}{content}"));
2886 }
2887 }
2888 Block::Html {
2889 lines: html_lines,
2890 has_preceding_blank: _,
2891 } => {
2892 // Preserve HTML blocks exactly as-is with original indentation
2893 // NOTE: Blank line before HTML block is handled by the previous block
2894
2895 for (idx, line) in html_lines.iter().enumerate() {
2896 if is_first_block && idx == 0 {
2897 // First line of first block gets marker
2898 result.push(format!("{marker}{line}"));
2899 is_first_block = false;
2900 } else if line.is_empty() {
2901 // Preserve blank lines inside HTML blocks
2902 result.push(String::new());
2903 } else {
2904 // Preserve lines with their original content (already includes indentation)
2905 result.push(format!("{expected_indent}{line}"));
2906 }
2907 }
2908
2909 // Add blank line after HTML block if there's a next block.
2910 // Only add if not already ending with one (avoids double blanks
2911 // when the HTML block itself contained a trailing blank line).
2912 if block_idx < blocks.len() - 1 {
2913 let next_block = &blocks[block_idx + 1];
2914 let should_add_blank = match next_block {
2915 Block::Code {
2916 has_preceding_blank, ..
2917 } => *has_preceding_blank,
2918 Block::Html {
2919 has_preceding_blank, ..
2920 } => *has_preceding_blank,
2921 Block::Table {
2922 has_preceding_blank, ..
2923 } => *has_preceding_blank,
2924 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2925 _ => true, // For all other blocks, add blank line
2926 };
2927 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2928 result.push(String::new());
2929 }
2930 }
2931 }
2932 Block::Table {
2933 lines: table_lines,
2934 has_preceding_blank: _,
2935 } => {
2936 // Preserve table rows verbatim with their original indentation.
2937 // Reflowing rows would corrupt column alignment and inject `|`
2938 // characters mid-paragraph (issue #590).
2939 // The leading blank line is emitted by the previous block.
2940 for (idx, (content, orig_indent)) in table_lines.iter().enumerate() {
2941 if is_first_block && idx == 0 {
2942 // First line of first block gets the list marker
2943 result.push(format!(
2944 "{marker}{}",
2945 " ".repeat(orig_indent.saturating_sub(marker_len)) + content.as_str()
2946 ));
2947 is_first_block = false;
2948 } else {
2949 // Shift nested table rows with the marker so they
2950 // stay aligned when MD030 widens marker spacing.
2951 result.push(format!(
2952 "{}{}",
2953 " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
2954 content
2955 ));
2956 }
2957 }
2958
2959 // Add blank line after table block if there's a next block.
2960 if block_idx < blocks.len() - 1 {
2961 let next_block = &blocks[block_idx + 1];
2962 let should_add_blank = match next_block {
2963 Block::Code {
2964 has_preceding_blank, ..
2965 } => *has_preceding_blank,
2966 Block::Html {
2967 has_preceding_blank, ..
2968 } => *has_preceding_blank,
2969 Block::Table {
2970 has_preceding_blank, ..
2971 } => *has_preceding_blank,
2972 Block::SnippetLine(_) | Block::DivMarker(_) => false,
2973 _ => true,
2974 };
2975 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
2976 result.push(String::new());
2977 }
2978 }
2979 }
2980 Block::Admonition {
2981 header,
2982 header_indent,
2983 content_lines: admon_lines,
2984 } => {
2985 // Reconstruct admonition block with header at original indent
2986 // and body content reflowed to fit within the line length limit
2987
2988 // Add blank line before admonition if not first block
2989 if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
2990 result.push(String::new());
2991 }
2992
2993 // Output the header at its original indent
2994 let header_indent_str = " ".repeat(*header_indent);
2995 if is_first_block {
2996 result.push(format!(
2997 "{marker}{}",
2998 " ".repeat(header_indent.saturating_sub(marker_len)) + header.as_str()
2999 ));
3000 is_first_block = false;
3001 } else {
3002 result.push(format!("{header_indent_str}{header}"));
3003 }
3004
3005 // Derive body indent from the first non-empty content line's
3006 // stored indent, falling back to header_indent + 4 for
3007 // empty-body admonitions
3008 let body_indent = admon_lines
3009 .iter()
3010 .find(|(content, _)| !content.is_empty())
3011 .map_or(header_indent + 4, |(_, indent)| *indent);
3012 let body_indent_str = " ".repeat(body_indent);
3013
3014 // Segment body content into code blocks (verbatim) and
3015 // text paragraphs (reflowable), separated by blank lines.
3016 // Code lines store (content, orig_indent) to reconstruct
3017 // internal indentation relative to body_indent.
3018 enum AdmonSegment {
3019 Text(Vec<String>),
3020 Code(Vec<(String, usize)>),
3021 }
3022
3023 let mut segments: Vec<AdmonSegment> = Vec::new();
3024 let mut current_text: Vec<String> = Vec::new();
3025 let mut current_code: Vec<(String, usize)> = Vec::new();
3026 let mut in_admon_code = false;
3027 // Track the opening fence character so closing fences
3028 // must match (backticks close backticks, tildes close tildes)
3029 let mut fence_char: char = '`';
3030
3031 // Opening fences: ``` or ~~~ followed by optional info string
3032 let get_opening_fence = |s: &str| -> Option<(char, usize)> {
3033 let t = s.trim_start();
3034 if t.starts_with("```") {
3035 Some(('`', t.bytes().take_while(|&b| b == b'`').count()))
3036 } else if t.starts_with("~~~") {
3037 Some(('~', t.bytes().take_while(|&b| b == b'~').count()))
3038 } else {
3039 None
3040 }
3041 };
3042 // Closing fences: ONLY fence chars + optional trailing spaces
3043 let get_closing_fence = |s: &str| -> Option<(char, usize)> {
3044 let t = s.trim();
3045 if t.starts_with("```") && t.bytes().all(|b| b == b'`') {
3046 Some(('`', t.len()))
3047 } else if t.starts_with("~~~") && t.bytes().all(|b| b == b'~') {
3048 Some(('~', t.len()))
3049 } else {
3050 None
3051 }
3052 };
3053 let mut fence_len: usize = 3;
3054
3055 for (content, orig_indent) in admon_lines {
3056 if in_admon_code {
3057 // Closing fence must use the same character, be
3058 // at least as long, and have no info string
3059 if let Some((ch, len)) = get_closing_fence(content)
3060 && ch == fence_char
3061 && len >= fence_len
3062 {
3063 current_code.push((content.clone(), *orig_indent));
3064 in_admon_code = false;
3065 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
3066 continue;
3067 }
3068 current_code.push((content.clone(), *orig_indent));
3069 } else if let Some((ch, len)) = get_opening_fence(content) {
3070 if !current_text.is_empty() {
3071 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
3072 }
3073 in_admon_code = true;
3074 fence_char = ch;
3075 fence_len = len;
3076 current_code.push((content.clone(), *orig_indent));
3077 } else if content.is_empty() {
3078 if !current_text.is_empty() {
3079 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
3080 }
3081 } else {
3082 current_text.push(content.clone());
3083 }
3084 }
3085 if in_admon_code && !current_code.is_empty() {
3086 segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
3087 }
3088 if !current_text.is_empty() {
3089 segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
3090 }
3091
3092 // Build reflow options once for all text segments
3093 let admon_reflow_length = if config.line_length.is_unlimited() {
3094 usize::MAX
3095 } else {
3096 config.line_length.get().saturating_sub(body_indent).max(1)
3097 };
3098
3099 let admon_reflow_options = Self::reflow_options(ctx, config, admon_reflow_length);
3100
3101 // Output each segment
3102 for segment in &segments {
3103 // Blank line before each segment (after the header or previous segment)
3104 result.push(String::new());
3105
3106 match segment {
3107 AdmonSegment::Code(lines) => {
3108 for (line, orig_indent) in lines {
3109 if line.is_empty() {
3110 // Preserve blank lines inside code blocks
3111 result.push(String::new());
3112 } else {
3113 // Reconstruct with body_indent + any extra
3114 // indentation the line had beyond body_indent
3115 let extra = orig_indent.saturating_sub(body_indent);
3116 let indent_str = " ".repeat(body_indent + extra);
3117 result.push(format!("{indent_str}{line}"));
3118 }
3119 }
3120 }
3121 AdmonSegment::Text(lines) => {
3122 let paragraph_text = lines.join(" ").trim().to_string();
3123 if paragraph_text.is_empty() {
3124 continue;
3125 }
3126 let reflowed = crate::utils::text_reflow::reflow_line(
3127 ¶graph_text,
3128 &admon_reflow_options,
3129 );
3130 for line in &reflowed {
3131 result.push(format!("{body_indent_str}{line}"));
3132 }
3133 }
3134 }
3135 }
3136
3137 // Add blank line after admonition if there's a next block
3138 if block_idx < blocks.len() - 1 {
3139 let next_block = &blocks[block_idx + 1];
3140 let should_add_blank = match next_block {
3141 Block::Code {
3142 has_preceding_blank, ..
3143 } => *has_preceding_blank,
3144 Block::Table {
3145 has_preceding_blank, ..
3146 } => *has_preceding_blank,
3147 Block::SnippetLine(_) | Block::DivMarker(_) => false,
3148 _ => true,
3149 };
3150 if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
3151 result.push(String::new());
3152 }
3153 }
3154 }
3155 }
3156 }
3157
3158 let reflowed_text = result.join(line_ending);
3159
3160 // Preserve trailing newline
3161 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
3162 format!("{reflowed_text}{line_ending}")
3163 } else {
3164 reflowed_text
3165 };
3166
3167 // Get the original text to compare
3168 let original_text = &ctx.content[byte_range.clone()];
3169
3170 // Physical-line-length scan, shared by the Normalize-mode gate and its
3171 // message. The list-item reflow preserves code blocks, HTML blocks,
3172 // admonition headers, fence markers, semantic markers, and snippet/div
3173 // markers verbatim; only paragraph content and admonition bodies are
3174 // restructured. Only those lines drive the length warning, so that
3175 // preserved-but-overlong content does not keep the paragraph-level
3176 // warning alive when the reflow would not fix that line.
3177 let should_count_for_length = |line_idx: usize| -> bool {
3178 let line = lines[line_idx];
3179 let trimmed = line.trim();
3180 if trimmed.is_empty() || is_exempt_line(line, line_idx + 1) {
3181 return false;
3182 }
3183 let info = &ctx.lines[line_idx];
3184 if info.in_code_block || info.in_html_block {
3185 return false;
3186 }
3187 if info.in_admonition && mkdocs_admonitions::is_admonition_start(line) {
3188 return false;
3189 }
3190 if is_fence_marker(line) || is_semantic_line(line) {
3191 return false;
3192 }
3193 if is_snippet_block_delimiter(line) {
3194 return false;
3195 }
3196 if line.trim_start().starts_with(":::") {
3197 return false;
3198 }
3199 true
3200 };
3201 let max_physical_length = (list_start..i)
3202 .filter(|&idx| should_count_for_length(idx))
3203 .map(|idx| self.calculate_effective_length(lines[idx]))
3204 .max()
3205 .unwrap_or(0);
3206 // `line-length = 0` means "no limit", so no physical line can be
3207 // "over"; the message below then describes a structural join rather
3208 // than a length violation.
3209 let any_paragraph_line_over =
3210 !config.line_length.is_unlimited() && max_physical_length > config.line_length.get();
3211
3212 // Normalize mode reflows list-item prose just like paragraphs:
3213 // joining continuation lines and re-wrapping to `line-length`.
3214 // `prose_changed` is true only when the reflow alters the words or
3215 // line breaks, not when it would merely re-indent continuation
3216 // lines or trim trailing whitespace. Comparing the texts with each
3217 // line's leading and trailing whitespace removed isolates "did the
3218 // words/line breaks change" from "did the surrounding whitespace
3219 // change". Continuation indentation is MD077's responsibility and
3220 // trailing whitespace is MD009's; an MD013 warning for either would
3221 // both duplicate those rules and resurface a persistent advisory on
3222 // already-fitting items that users disable MD013 fixing to avoid.
3223 let prose_changed = {
3224 let stripped = |text: &str| text.lines().map(str::trim).collect::<Vec<_>>().join("\n");
3225 stripped(original_text) != stripped(&replacement)
3226 };
3227 // Warn when the reflow rewraps prose (the normalize feature for
3228 // list items), or when a physical line genuinely exceeds the limit
3229 // and the reflow can change something (a true length violation,
3230 // even if all that changes is the continuation indent). A line that
3231 // is already optimal in both respects produces no warning.
3232 let gate_ok = prose_changed || (any_paragraph_line_over && original_text != replacement);
3233 if gate_ok {
3234 // Generate an appropriate message based on why reflow is needed
3235 let message = match config.reflow_mode {
3236 ReflowMode::SentencePerLine => {
3237 let num_sentences = split_into_sentences(&combined_content).len();
3238 let num_lines = content_lines.len();
3239 if num_lines == 1 {
3240 // Single line with multiple sentences
3241 format!("Line contains {num_sentences} sentences (one sentence per line required)")
3242 } else {
3243 // Multiple lines - could be split sentences or mixed
3244 format!(
3245 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
3246 )
3247 }
3248 }
3249 ReflowMode::SemanticLineBreaks => {
3250 let num_sentences = split_into_sentences(&combined_content).len();
3251 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
3252 }
3253 ReflowMode::Normalize => {
3254 // When a physical line genuinely exceeds the limit, report
3255 // it as a length violation. Otherwise the reflow is a
3256 // structural normalization (joining/re-wrapping multi-line
3257 // content that already fits), mirroring the paragraph path.
3258 if any_paragraph_line_over {
3259 format!(
3260 "Line length {} exceeds {} characters",
3261 max_physical_length,
3262 config.line_length.get()
3263 )
3264 } else {
3265 format!(
3266 "List item could be normalized to use line length of {} characters",
3267 config.line_length.get()
3268 )
3269 }
3270 }
3271 ReflowMode::Default => {
3272 // Report the actual longest non-exempt line, not the combined content
3273 let max_length = (list_start..i)
3274 .filter(|&line_idx| {
3275 let line = lines[line_idx];
3276 let trimmed = line.trim();
3277 !trimmed.is_empty() && !is_exempt_line(line, line_idx + 1)
3278 })
3279 .map(|line_idx| self.calculate_effective_length(lines[line_idx]))
3280 .max()
3281 .unwrap_or(0);
3282 format!(
3283 "Line length {} exceeds {} characters",
3284 max_length,
3285 config.line_length.get()
3286 )
3287 }
3288 };
3289
3290 warnings.push(LintWarning {
3291 rule_name: Some(self.name().to_string()),
3292 message,
3293 line: list_start + 1,
3294 column: 1,
3295 end_line: end_line + 1,
3296 end_column: lines[end_line].chars().count() + 1,
3297 severity: Severity::Warning,
3298 fix: Some(crate::rule::Fix::new(byte_range, replacement)),
3299 });
3300 }
3301 }
3302 continue;
3303 }
3304
3305 // Found start of a paragraph - collect all lines in it
3306 let paragraph_start = i;
3307 let mut paragraph_lines = vec![lines[i]];
3308 i += 1;
3309
3310 while i < lines.len() {
3311 let next_line = lines[i];
3312 let next_line_num = i + 1;
3313 let next_trimmed = next_line.trim();
3314
3315 // Stop at paragraph boundaries
3316 if next_trimmed.is_empty()
3317 || ctx.line_info(next_line_num).is_some_and(|info| info.in_code_block)
3318 || ctx.line_info(next_line_num).is_some_and(|info| info.in_front_matter)
3319 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_block)
3320 || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_comment)
3321 || ctx.line_info(next_line_num).is_some_and(|info| info.in_esm_block)
3322 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_expression)
3323 || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_block)
3324 || ctx.line_info(next_line_num).is_some_and(|info| info.in_mdx_comment)
3325 || ctx
3326 .line_info(next_line_num)
3327 .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
3328 || (next_line_num > 0
3329 && next_line_num <= ctx.lines.len()
3330 && ctx.lines[next_line_num - 1].blockquote.is_some())
3331 || next_trimmed.starts_with('#')
3332 || TableUtils::is_potential_table_row(next_line)
3333 || is_list_item(next_trimmed)
3334 || is_horizontal_rule(next_line)
3335 || (next_trimmed.starts_with('[') && next_line.contains("]:"))
3336 || is_template_directive_only(next_line)
3337 || is_standalone_attr_list(next_line)
3338 || is_snippet_block_delimiter(next_line)
3339 || ctx.line_info(next_line_num).is_some_and(|info| info.is_div_marker)
3340 || is_html_only_line(next_line)
3341 || self.line_in_multiline_math_block(next_line_num, ctx)
3342 || (!config.strict && is_standalone_link_or_image_line(ctx, next_line_num))
3343 {
3344 break;
3345 }
3346
3347 // Check if the previous line ends with a hard break (2+ spaces or backslash)
3348 if i > 0 && has_hard_break(lines[i - 1]) {
3349 // Don't include lines after hard breaks in the same paragraph
3350 break;
3351 }
3352
3353 paragraph_lines.push(next_line);
3354 i += 1;
3355 }
3356
3357 // Compute the common leading indent of all non-empty paragraph lines,
3358 // but only when those lines are structurally inside a list block.
3359 // Indented continuation lines that follow a nested list arrive here
3360 // with their structural indentation intact (e.g. 2 spaces for a
3361 // top-level list item). Stripping the indent before reflow and
3362 // re-applying it afterward prevents the fixer from moving those
3363 // lines to column 0.
3364 //
3365 // The list-block guard is essential: top-level paragraphs that happen
3366 // to start with spaces (insignificant in Markdown) must NOT have those
3367 // spaces preserved or injected by the fixer.
3368 let common_indent: String = if ctx.is_in_list_block(paragraph_start + 1) {
3369 let min_len = paragraph_lines
3370 .iter()
3371 .filter(|l| !l.trim().is_empty())
3372 .map(|l| l.len() - l.trim_start().len())
3373 .min()
3374 .unwrap_or(0);
3375 paragraph_lines
3376 .iter()
3377 .find(|l| !l.trim().is_empty())
3378 .map(|l| l[..min_len].to_string())
3379 .unwrap_or_default()
3380 } else {
3381 String::new()
3382 };
3383
3384 // Combine paragraph lines into a single string for processing.
3385 // This must be done BEFORE the needs_reflow check for sentence-per-line mode.
3386 let paragraph_text = if common_indent.is_empty() {
3387 paragraph_lines.join(" ")
3388 } else {
3389 paragraph_lines
3390 .iter()
3391 .map(|l| {
3392 if l.starts_with(common_indent.as_str()) {
3393 &l[common_indent.len()..]
3394 } else {
3395 l.trim_start()
3396 }
3397 })
3398 .collect::<Vec<_>>()
3399 .join(" ")
3400 };
3401
3402 // Skip reflowing if this paragraph contains definition list items
3403 // Definition lists are multi-line structures that should not be joined
3404 let contains_definition_list = paragraph_lines
3405 .iter()
3406 .any(|line| crate::utils::is_definition_list_item(line));
3407
3408 if contains_definition_list {
3409 // Don't reflow definition lists - skip this paragraph
3410 i = paragraph_start + paragraph_lines.len();
3411 continue;
3412 }
3413
3414 // Skip reflowing if this paragraph contains MkDocs Snippets markers
3415 // Snippets blocks (-8<- ... -8<-) should be preserved exactly
3416 let contains_snippets = paragraph_lines.iter().any(|line| is_snippet_block_delimiter(line));
3417
3418 if contains_snippets {
3419 // Don't reflow Snippets blocks - skip this paragraph
3420 i = paragraph_start + paragraph_lines.len();
3421 continue;
3422 }
3423
3424 // Leave a line of a multi-line display-math block as it is, where
3425 // joining lines would corrupt the equation (see
3426 // `line_in_multiline_math_block`). Only the first line has to be
3427 // asked about: such a line ends the paragraph above it, so a
3428 // paragraph reaching here holds one only when it starts on one.
3429 //
3430 // Only that line is passed over, not the rest of what was collected
3431 // with it: prose written directly under the closing delimiter is an
3432 // ordinary paragraph and still reflows.
3433 if self.line_in_multiline_math_block(paragraph_start + 1, ctx) {
3434 i = paragraph_start + 1;
3435 continue;
3436 }
3437
3438 // Check if this paragraph needs reflowing
3439 let needs_reflow = match config.reflow_mode {
3440 ReflowMode::Normalize => self.normalize_mode_needs_reflow(paragraph_lines.iter().copied(), config),
3441 ReflowMode::SentencePerLine => {
3442 // In sentence-per-line mode, check if the JOINED paragraph has multiple sentences
3443 // Note: we check the joined text because sentences can span multiple lines
3444 let sentences = split_into_sentences(¶graph_text);
3445
3446 // Always reflow if multiple sentences on one line
3447 if sentences.len() > 1 {
3448 true
3449 } else if paragraph_lines.len() > 1 {
3450 // For single-sentence paragraphs spanning multiple lines:
3451 // Reflow if they COULD fit on one line (respecting line-length constraint)
3452 if config.line_length.is_unlimited() {
3453 // No line-length constraint - always join single sentences
3454 true
3455 } else {
3456 // Only join if it fits within line-length.
3457 // paragraph_text has the common indent stripped, so add it
3458 // back to get the true output length before comparing.
3459 let effective_length =
3460 self.calculate_effective_length(¶graph_text) + common_indent.len();
3461 effective_length <= config.line_length.get()
3462 }
3463 } else {
3464 false
3465 }
3466 }
3467 ReflowMode::SemanticLineBreaks => {
3468 let sentences = split_into_sentences(¶graph_text);
3469 // Reflow if multiple sentences, multiple lines, or any line exceeds limit
3470 sentences.len() > 1
3471 || paragraph_lines.len() > 1
3472 || paragraph_lines
3473 .iter()
3474 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3475 }
3476 ReflowMode::Default => {
3477 // In default mode, only reflow if lines exceed limit
3478 paragraph_lines
3479 .iter()
3480 .any(|line| self.calculate_effective_length(line) > config.line_length.get())
3481 }
3482 };
3483
3484 if needs_reflow {
3485 // Calculate byte range for this paragraph
3486 // Use whole_line_range for each line and combine
3487 let start_range = line_index.whole_line_range(paragraph_start + 1);
3488 let end_line = paragraph_start + paragraph_lines.len() - 1;
3489
3490 // For the last line, we want to preserve any trailing newline
3491 let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
3492 // Last line without trailing newline - use line_text_range
3493 line_index.line_text_range(end_line + 1, 1, lines[end_line].len() + 1)
3494 } else {
3495 // Not the last line or has trailing newline - use whole_line_range
3496 line_index.whole_line_range(end_line + 1)
3497 };
3498
3499 let byte_range = start_range.start..end_range.end;
3500
3501 // Check if the paragraph ends with a hard break and what type
3502 let hard_break_type = paragraph_lines.last().and_then(|line| {
3503 let line = line.strip_suffix('\r').unwrap_or(line);
3504 if line.ends_with('\\') {
3505 Some("\\")
3506 } else if line.ends_with(" ") {
3507 Some(" ")
3508 } else {
3509 None
3510 }
3511 });
3512
3513 // Reflow the paragraph
3514 // When line_length = 0 (no limit), use a very large value for reflow
3515 let reflow_line_length = if config.line_length.is_unlimited() {
3516 usize::MAX
3517 } else {
3518 config.line_length.get().saturating_sub(common_indent.len()).max(1)
3519 };
3520 let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
3521 let mut reflowed = crate::utils::text_reflow::reflow_line(¶graph_text, &reflow_options);
3522
3523 // Re-apply the common indent to each non-empty reflowed line so
3524 // that the replacement preserves the original structural indentation.
3525 if !common_indent.is_empty() {
3526 for line in &mut reflowed {
3527 if !line.is_empty() {
3528 *line = format!("{common_indent}{line}");
3529 }
3530 }
3531 }
3532
3533 // If the original paragraph ended with a hard break, preserve it
3534 // Preserve the original hard break format (backslash or two spaces)
3535 if let Some(break_marker) = hard_break_type
3536 && !reflowed.is_empty()
3537 {
3538 let last_idx = reflowed.len() - 1;
3539 if !has_hard_break(&reflowed[last_idx]) {
3540 reflowed[last_idx].push_str(break_marker);
3541 }
3542 }
3543
3544 let reflowed_text = reflowed.join(line_ending);
3545
3546 // Preserve trailing newline if the original paragraph had one
3547 let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
3548 format!("{reflowed_text}{line_ending}")
3549 } else {
3550 reflowed_text
3551 };
3552
3553 // Get the original text to compare
3554 let original_text = &ctx.content[byte_range.clone()];
3555
3556 // Only generate a warning if the replacement is different from the original
3557 if original_text != replacement {
3558 // Determine which line ranges and messages to report based on the reflow mode.
3559 let warnings_to_report: Vec<(usize, usize, String)> = match config.reflow_mode {
3560 ReflowMode::Default => {
3561 // In default mode, report a warning for *every* line in the paragraph
3562 // that exceeds the limit. Each warning will carry the same paragraph-level
3563 // fix, making all of them auto-fixable.
3564 paragraph_lines
3565 .iter()
3566 .enumerate()
3567 .filter(|(_, line)| self.calculate_effective_length(line) > config.line_length.get())
3568 .map(|(idx, _)| {
3569 let violating_line = paragraph_start + idx + 1;
3570 (
3571 violating_line,
3572 violating_line,
3573 format!("Line length exceeds {} characters", config.line_length.get()),
3574 )
3575 })
3576 .collect()
3577 }
3578 ReflowMode::Normalize => {
3579 // In normalize mode, report the whole paragraph as needing normalization.
3580 vec![(
3581 paragraph_start + 1,
3582 end_line + 1,
3583 format!(
3584 "Paragraph could be normalized to use line length of {} characters",
3585 config.line_length.get()
3586 ),
3587 )]
3588 }
3589 ReflowMode::SentencePerLine => {
3590 // In sentence-per-line mode, highlight the entire paragraph that needs reformatting.
3591 let num_sentences = split_into_sentences(¶graph_text).len();
3592 let message = if paragraph_lines.len() == 1 {
3593 // Single line with multiple sentences
3594 format!("Line contains {num_sentences} sentences (one sentence per line required)")
3595 } else {
3596 // Multiple lines - could be split sentences or mixed
3597 let num_lines = paragraph_lines.len();
3598 format!(
3599 "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
3600 )
3601 };
3602 vec![(paragraph_start + 1, paragraph_start + paragraph_lines.len(), message)]
3603 }
3604 ReflowMode::SemanticLineBreaks => {
3605 // In semantic-line-breaks mode, highlight the entire paragraph.
3606 let num_sentences = split_into_sentences(¶graph_text).len();
3607 vec![(
3608 paragraph_start + 1,
3609 paragraph_start + paragraph_lines.len(),
3610 format!("Paragraph should use semantic line breaks ({num_sentences} sentences)"),
3611 )]
3612 }
3613 };
3614
3615 // Generate the actual lint warnings. All warnings for this paragraph
3616 // share the same paragraph-level fix.
3617 for (w_start, w_end, msg) in warnings_to_report {
3618 warnings.push(LintWarning {
3619 rule_name: Some(self.name().to_string()),
3620 message: msg,
3621 line: w_start,
3622 column: 1,
3623 end_line: w_end,
3624 end_column: lines[w_end.saturating_sub(1)].chars().count() + 1,
3625 severity: Severity::Warning,
3626 fix: Some(crate::rule::Fix::new(byte_range.clone(), replacement.clone())),
3627 });
3628 }
3629 }
3630 }
3631 }
3632
3633 warnings
3634 }
3635
3636 /// Calculate string length based on the configured length mode
3637 fn calculate_string_length(&self, s: &str) -> usize {
3638 match self.config.length_mode {
3639 LengthMode::Chars => s.chars().count(),
3640 LengthMode::Visual => s.width(),
3641 LengthMode::Bytes => s.len(),
3642 }
3643 }
3644
3645 /// Calculate effective line length
3646 ///
3647 /// Returns the actual display length of the line using the configured length mode.
3648 fn calculate_effective_length(&self, line: &str) -> usize {
3649 self.calculate_string_length(line)
3650 }
3651
3652 /// Calculate line length with inline link/image URLs removed.
3653 ///
3654 /// For each inline link `[text](url)` or image `` on the line,
3655 /// computes the "savings" from removing the URL portion (keeping only `[text]`
3656 /// or `![alt]`). Returns `effective_length - total_savings`.
3657 ///
3658 /// Handles nested constructs (e.g., `[](url)`) by only counting the
3659 /// outermost construct to avoid double-counting.
3660 fn length_without_inline_link_urls(
3661 &self,
3662 effective_length: usize,
3663 line_number: usize,
3664 ctx: &crate::lint_context::LintContext,
3665 ) -> usize {
3666 let line_range = ctx.line_index.line_content_range(line_number);
3667 let line_byte_end = line_range.end;
3668
3669 // Collect inline links/images on this line: (byte_offset, byte_end, text_only_display_len)
3670 let mut constructs: Vec<(usize, usize, usize)> = Vec::new();
3671
3672 // Binary search: links are sorted by byte_offset, so link.line is non-decreasing
3673 let link_start = ctx.links.partition_point(|l| l.line < line_number);
3674 for link in &ctx.links[link_start..] {
3675 if link.line != line_number {
3676 break;
3677 }
3678 if link.is_reference {
3679 continue;
3680 }
3681 if !matches!(link.link_type, LinkType::Inline) {
3682 continue;
3683 }
3684 if link.byte_end > line_byte_end {
3685 continue;
3686 }
3687 let text_only_len = 2 + self.calculate_string_length(&link.text);
3688 constructs.push((link.byte_offset, link.byte_end, text_only_len));
3689 }
3690
3691 let img_start = ctx.images.partition_point(|i| i.line < line_number);
3692 for image in &ctx.images[img_start..] {
3693 if image.line != line_number {
3694 break;
3695 }
3696 if image.is_reference {
3697 continue;
3698 }
3699 if !matches!(image.link_type, LinkType::Inline) {
3700 continue;
3701 }
3702 if image.byte_end > line_byte_end {
3703 continue;
3704 }
3705 let text_only_len = 3 + self.calculate_string_length(&image.alt_text);
3706 constructs.push((image.byte_offset, image.byte_end, text_only_len));
3707 }
3708
3709 if constructs.is_empty() {
3710 return effective_length;
3711 }
3712
3713 // Sort by byte offset to handle overlapping/nested constructs
3714 constructs.sort_by_key(|&(start, _, _)| start);
3715
3716 let mut total_savings: usize = 0;
3717 let mut last_end: usize = 0;
3718
3719 for (start, end, text_only_len) in &constructs {
3720 // Skip constructs nested inside a previously counted one
3721 if *start < last_end {
3722 continue;
3723 }
3724 // Full construct length in configured length mode
3725 let full_source = &ctx.content[*start..*end];
3726 let full_len = self.calculate_string_length(full_source);
3727 total_savings += full_len.saturating_sub(*text_only_len);
3728 last_end = *end;
3729 }
3730
3731 effective_length.saturating_sub(total_savings)
3732 }
3733}