rumdl_lib/rules/

md013_line_length.rs

/// Rule MD013: Line length
///
/// See [docs/md013.md](../../docs/md013.md) for full documentation, configuration, and examples.
use crate::rule::{LintError, LintResult, LintWarning, Rule, RuleCategory, Severity};
use crate::utils::mkdocs_admonitions;
use crate::utils::mkdocs_attr_list::is_standalone_attr_list;
use crate::utils::mkdocs_snippets::is_snippet_block_delimiter;
use crate::utils::mkdocs_tabs;
use crate::utils::range_utils::calculate_excess_range;
use crate::utils::regex_cache::{IMAGE_REF_PATTERN, LINK_REF_PATTERN, URL_PATTERN};
use crate::utils::table_utils::TableUtils;
use crate::utils::text_reflow::{
    BlockquoteLineData, blockquote_continuation_style, dominant_blockquote_prefix, reflow_blockquote_content,
    split_into_sentences,
};
use pulldown_cmark::LinkType;
use toml;

mod block_builder;
mod helpers;
pub mod md013_config;
use crate::rules::md030_list_marker_space::MD030Config;
use crate::utils::is_template_directive_only;
use block_builder::{Block, BlockBuilder};
use helpers::{
    extract_list_marker_and_content, has_hard_break, is_github_alert_marker, is_horizontal_rule, is_html_only_line,
    is_list_item, is_setext_heading_text_line, is_setext_underline_content, is_standalone_link_or_image_line,
    is_unwrappable_line, source_list_marker, split_into_segments, trim_preserving_hard_break,
};
pub use md013_config::MD013Config;
use md013_config::{LengthMode, ReflowMode};

#[cfg(test)]
mod tests;
use unicode_width::UnicodeWidthStr;

#[derive(Clone, Default)]
pub struct MD013LineLength {
    pub(crate) config: MD013Config,
    /// MD030 list-marker spacing, applied when reflowing list items so the rewrite
    /// uses the configured post-marker spacing rather than a hard-coded single
    /// space. Defaults to MD030's defaults (a single space everywhere), which
    /// reproduces the previous behaviour exactly. See [`MD030Config::expected_spaces`].
    pub(crate) list_spacing: MD030Config,
}

/// Blockquote paragraph line collected for reflow, with original line index for range computation.
struct CollectedBlockquoteLine {
    line_idx: usize,
    data: BlockquoteLineData,
}

/// A deliberately conservative MDG step candidate.
///
/// Gherkin keywords are localized, so recognizing the word after the marker is
/// not reliable without the complete dialect table. MDG represents steps as
/// unordered Markdown list items; withholding reflow from all such items in the
/// flavor is safer than splitting a step and changing its Gherkin text.
fn is_potential_mdg_step(ctx: &crate::lint_context::LintContext, line_num: usize) -> bool {
    let Some(item) = ctx.list_item_on_line(line_num) else {
        return false;
    };
    !item.is_ordered() && matches!(item.marker_char(), Some('*' | '-' | '+'))
}

impl MD013LineLength {
    pub fn new(line_length: usize, code_blocks: bool, tables: bool, headings: bool, strict: bool) -> Self {
        Self {
            config: MD013Config {
                line_length: crate::types::LineLength::new(line_length),
                code_blocks,
                code_spans: true,
                tables,
                headings,
                math_blocks: true,
                paragraphs: true,  // Default to true for backwards compatibility
                blockquotes: true, // Default to true for backwards compatibility
                strict,
                stern: false,
                heading_line_length: None,
                code_block_line_length: None,
                reflow: false,
                reflow_mode: ReflowMode::default(),
                length_mode: LengthMode::default(),
                abbreviations: Vec::new(),
                require_sentence_capital: true,
                ignore_link_urls: true,
                atomic_spans: true,
                reflow_length_exemptions: false,
            },
            list_spacing: MD030Config::default(),
        }
    }

    pub fn from_config_struct(config: MD013Config) -> Self {
        Self {
            config,
            list_spacing: MD030Config::default(),
        }
    }

    /// Return a clone with code block checking disabled.
    /// Used for doc comment linting where code blocks are Rust code managed by rustfmt.
    pub fn with_code_blocks_disabled(&self) -> Self {
        let mut clone = self.clone();
        clone.config.code_blocks = false;
        clone
    }

    /// Convert MD013 LengthMode to text_reflow ReflowLengthMode
    /// Normalized set of reference labels defined in the document.
    ///
    /// Passed to reflow so a bare shortcut reference (`[text]`) is treated as an
    /// atomic link only when its label is actually defined; an undefined
    /// bracketed run reflows as literal prose.
    fn defined_reference_labels(ctx: &crate::lint_context::LintContext) -> std::collections::HashSet<String> {
        ctx.reference_definitions()
            .iter()
            .map(|d| crate::utils::text_reflow::normalize_reference_label(&d.id))
            .collect()
    }

    /// Build the reflow options shared by every MD013 fix path.
    ///
    /// `line_length` varies per call site (some subtract a list-marker or
    /// blockquote prefix); every other field is derived uniformly from the
    /// effective config and the document flavor. Callers that need a different
    /// `max_list_continuation_indent` override it via struct update.
    fn reflow_options(
        ctx: &crate::lint_context::LintContext,
        config: &MD013Config,
        line_length: usize,
    ) -> crate::utils::text_reflow::ReflowOptions {
        crate::utils::text_reflow::ReflowOptions {
            line_length,
            break_on_sentences: true,
            preserve_breaks: false,
            sentence_per_line: config.reflow_mode == ReflowMode::SentencePerLine,
            semantic_line_breaks: config.reflow_mode == ReflowMode::SemanticLineBreaks,
            abbreviations: config.abbreviations_for_reflow(),
            length_mode: config.reflow_length_mode(),
            attr_lists: ctx.flavor.supports_attr_lists(),
            myst_roles: ctx.flavor.supports_myst_roles(),
            require_sentence_capital: config.require_sentence_capital,
            max_list_continuation_indent: if ctx.flavor.requires_strict_list_indent() {
                Some(4)
            } else {
                None
            },
            defined_references: Some(Self::defined_reference_labels(ctx)),
            atomic_spans: config.atomic_spans,
            length_exemptions: config.length_exemptions_for_reflow(),
        }
    }

    fn should_ignore_line(
        &self,
        line: &str,
        _lines: &[&str],
        current_line: usize,
        ctx: &crate::lint_context::LintContext,
    ) -> bool {
        if self.config.strict {
            return false;
        }

        // Quick check for common patterns before expensive regex
        let trimmed = line.trim();

        // Only skip if the entire line is a URL (quick check first)
        if (trimmed.starts_with("http://") || trimmed.starts_with("https://")) && URL_PATTERN.is_match(trimmed) {
            return true;
        }

        // Only skip if the entire line is an image reference (quick check first)
        if trimmed.starts_with("![") && trimmed.ends_with(']') && IMAGE_REF_PATTERN.is_match(trimmed) {
            return true;
        }

        // Note: link reference definitions are handled as always-exempt (even in strict mode)
        // in the main check loop, so they don't need to be checked here.

        // Code blocks with long strings (only check if in code block)
        if ctx.line_info(current_line + 1).is_some_and(|info| info.in_code_block)
            && !trimmed.is_empty()
            && !line.contains(' ')
            && !line.contains('\t')
        {
            return true;
        }

        false
    }

    /// Check if rule should skip based on provided config (used for inline config support)
    fn should_skip_with_config(&self, ctx: &crate::lint_context::LintContext, config: &MD013Config) -> bool {
        // Skip if content is empty
        if ctx.content.is_empty() {
            return true;
        }

        // For sentence-per-line, semantic-line-breaks, or normalize mode, never skip based on line length
        if config.reflow
            && (config.reflow_mode == ReflowMode::SentencePerLine
                || config.reflow_mode == ReflowMode::SemanticLineBreaks
                || config.reflow_mode == ReflowMode::Normalize)
        {
            return false;
        }

        // Use the smallest applicable budget across line/heading/code-block
        // contexts so a stricter context-specific limit doesn't get masked by
        // the document-wide budget.
        let min_limit = config.min_effective_line_length();
        if min_limit.is_unlimited() {
            return true;
        }
        let min_limit_bytes = min_limit.get();

        // Quick check: if total content is shorter than the smallest line limit,
        // definitely skip.
        if ctx.content.len() <= min_limit_bytes {
            return true;
        }

        // Skip if no line exceeds the smallest applicable limit.
        !ctx.lines.iter().any(|line| line.byte_len > min_limit_bytes)
    }

    fn normalize_mode_needs_reflow<'a, I>(&self, lines: I, config: &MD013Config) -> bool
    where
        I: IntoIterator<Item = &'a str>,
    {
        let mut line_count = 0;
        let check_length = !config.line_length.is_unlimited();

        for line in lines {
            line_count += 1;
            if check_length && self.calculate_effective_length(line) > config.line_length.get() {
                return true;
            }
        }

        line_count > 1
    }
}

impl Rule for MD013LineLength {
    fn name(&self) -> &'static str {
        "MD013"
    }

    fn description(&self) -> &'static str {
        "Line length should not be excessive"
    }

    fn check(&self, ctx: &crate::lint_context::LintContext) -> LintResult {
        // Use pre-parsed inline config from LintContext
        let config_override = ctx.inline_config().get_rule_config("MD013");

        // Apply configuration override if present
        let effective_config = if let Some(json_config) = config_override {
            if let Some(obj) = json_config.as_object() {
                let mut config = self.config.clone();
                if let Some(line_length) = obj.get("line_length").and_then(serde_json::Value::as_u64) {
                    config.line_length = crate::types::LineLength::new(line_length as usize);
                }
                if let Some(code_blocks) = obj.get("code_blocks").and_then(serde_json::Value::as_bool) {
                    config.code_blocks = code_blocks;
                }
                if let Some(code_spans) = obj.get("code_spans").and_then(serde_json::Value::as_bool) {
                    config.code_spans = code_spans;
                }
                if let Some(tables) = obj.get("tables").and_then(serde_json::Value::as_bool) {
                    config.tables = tables;
                }
                if let Some(headings) = obj.get("headings").and_then(serde_json::Value::as_bool) {
                    config.headings = headings;
                }
                if let Some(math_blocks) = obj
                    .get("math_blocks")
                    .or_else(|| obj.get("math-blocks"))
                    .and_then(serde_json::Value::as_bool)
                {
                    config.math_blocks = math_blocks;
                }
                if let Some(blockquotes) = obj.get("blockquotes").and_then(serde_json::Value::as_bool) {
                    config.blockquotes = blockquotes;
                }
                if let Some(strict) = obj.get("strict").and_then(serde_json::Value::as_bool) {
                    config.strict = strict;
                }
                if let Some(stern) = obj.get("stern").and_then(serde_json::Value::as_bool) {
                    config.stern = stern;
                }
                if let Some(v) = obj
                    .get("ignore_link_urls")
                    .or_else(|| obj.get("ignore-link-urls"))
                    .or_else(|| obj.get("semantic_link_understanding"))
                    .or_else(|| obj.get("semantic-link-understanding"))
                    .and_then(serde_json::Value::as_bool)
                {
                    config.ignore_link_urls = v;
                }
                if let Some(reflow) = obj.get("reflow").and_then(serde_json::Value::as_bool) {
                    config.reflow = reflow;
                }
                if let Some(reflow_mode) = obj.get("reflow_mode").and_then(|v| v.as_str()) {
                    config.reflow_mode = match reflow_mode {
                        "default" => ReflowMode::Default,
                        "normalize" => ReflowMode::Normalize,
                        "sentence-per-line" => ReflowMode::SentencePerLine,
                        "semantic-line-breaks" => ReflowMode::SemanticLineBreaks,
                        _ => ReflowMode::default(),
                    };
                }
                config
            } else {
                self.config.clone()
            }
        } else {
            self.config.clone()
        };

        // Fast early return using should_skip with EFFECTIVE config (after inline overrides)
        // But don't skip if we're in reflow mode with Normalize or SentencePerLine
        if self.should_skip_with_config(ctx, &effective_config)
            && !(effective_config.reflow
                && (effective_config.reflow_mode == ReflowMode::Normalize
                    || effective_config.reflow_mode == ReflowMode::SentencePerLine
                    || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
        {
            return Ok(Vec::new());
        }

        // Direct implementation without DocumentStructure
        let mut warnings = Vec::new();

        // Special handling: line_length = 0 means "no line length limit"
        // Skip all line length checks, but still allow reflow if enabled
        let skip_length_checks = effective_config.line_length.is_unlimited();

        // Pre-filter lines that could be problematic to avoid processing all lines.
        // Use the smallest applicable budget across line/heading/code-block contexts
        // so candidates aren't dropped when a stricter context-specific budget applies.
        let prefilter_limit = effective_config.min_effective_line_length();
        let prefilter_skip = prefilter_limit.is_unlimited();
        let mut candidate_lines = Vec::new();
        if !skip_length_checks && !prefilter_skip {
            for (line_idx, line_info) in ctx.lines.iter().enumerate() {
                // Skip front matter - it should never be linted
                if line_info.in_front_matter {
                    continue;
                }

                // Quick length check first
                if line_info.byte_len > prefilter_limit.get() {
                    candidate_lines.push(line_idx);
                }
            }
        }

        // If no candidate lines and not in normalize or sentence-per-line mode, early return
        if candidate_lines.is_empty()
            && !(effective_config.reflow
                && (effective_config.reflow_mode == ReflowMode::Normalize
                    || effective_config.reflow_mode == ReflowMode::SentencePerLine
                    || effective_config.reflow_mode == ReflowMode::SemanticLineBreaks))
        {
            return Ok(warnings);
        }

        let lines = ctx.raw_lines();

        // Whether a 1-indexed line is a heading. `LineInfo::heading` is an O(1)
        // per-line field, so check it directly at each use site instead of
        // materializing a full-document HashSet (an extra O(n) pass and
        // allocation on a rule that runs on virtually every file).
        let is_heading_line_num = |line_number: usize| -> bool {
            line_number
                .checked_sub(1)
                .and_then(|idx| ctx.lines.get(idx))
                .is_some_and(|line| line.heading.is_some())
        };

        // Use pre-computed table blocks from context
        // We need this for both the table skip check AND the paragraphs check
        let table_blocks = &ctx.table_blocks;
        let mut table_lines_set = std::collections::HashSet::new();
        for table in table_blocks {
            table_lines_set.insert(table.header_line + 1);
            table_lines_set.insert(table.delimiter_line + 1);
            for &line in &table.content_lines {
                table_lines_set.insert(line + 1);
            }
        }

        let defined_references = Self::defined_reference_labels(ctx);

        // Process candidate lines for line length checks
        'line_loop: for &line_idx in &candidate_lines {
            let line_number = line_idx + 1;
            let line = lines[line_idx];

            // Calculate actual line length (used in warning messages)
            let effective_length = self.calculate_effective_length(line);

            // Pick the context-specific limit: heading > code-block > paragraph.
            // Headings dominate over code-block context if a setext underline ever
            // overlaps a fenced range (defensive — these are mutually exclusive in
            // practice, but the explicit ordering documents intent).
            let is_heading_line = is_heading_line_num(line_number);
            let in_code_block = ctx.line_info(line_number).is_some_and(|info| info.in_code_block);
            let line_limit = if is_heading_line {
                effective_config.effective_heading_line_length().get()
            } else if in_code_block {
                effective_config.effective_code_block_line_length().get()
            } else {
                effective_config.line_length.get()
            };

            // A context-specific limit of 0 means "unlimited for this context".
            if line_limit == 0 {
                continue;
            }

            // Stern mode: like default, but the trailing-token forgiveness is
            // disabled — a line with whitespace that exceeds the limit is a
            // violation even if the excess is the final token. The "unwrappable"
            // line exemption (single token, optionally prefixed by # or >) is
            // still honored. Strict overrides stern entirely.
            if effective_config.stern && !effective_config.strict && is_unwrappable_line(line) {
                continue;
            }

            // Trailing-token forgiveness: only in default mode (not strict, not stern).
            // If the line only exceeds the limit because of a long token at the end
            // (URL, link chain, identifier), it passes. This matches markdownlint's
            // behavior: line.replace(/\S*$/u, "#")
            let check_length = if effective_config.strict || effective_config.stern {
                effective_length
            } else {
                match line.rfind(char::is_whitespace) {
                    Some(pos) => {
                        let ws_char = line[pos..].chars().next().unwrap();
                        let prefix_end = pos + ws_char.len_utf8();
                        self.calculate_string_length(&line[..prefix_end]) + 1
                    }
                    None => 1, // No whitespace — entire line is a single token
                }
            };

            // Skip lines where the check length is within the limit
            if check_length <= line_limit {
                continue;
            }

            // Ignore inline link/image URLs: suppress when excess comes entirely from inline URLs.
            // Disabled by `strict` (all forgiveness off) and by `ignore_link_urls = false`
            // (count link/image URLs toward the line length so such lines are flagged).
            if !effective_config.strict && effective_config.ignore_link_urls {
                let length_without_urls = self.length_without_inline_link_urls(effective_length, line_number, ctx);
                if length_without_urls <= line_limit {
                    continue;
                }
            }

            // Inline code spans cannot be wrapped, so reflow cannot shorten a line
            // whose excess length is one. When code-span checking is disabled,
            // suppress a violation that would fit once inline code spans are excluded.
            if !effective_config.code_spans {
                let code_span_width: usize = ctx
                    .code_spans()
                    .iter()
                    .filter(|span| span.line == line_number && span.end_line == line_number)
                    .map(|span| self.calculate_string_length(&ctx.content[span.byte_offset..span.byte_end]))
                    .sum();
                if effective_length.saturating_sub(code_span_width) <= line_limit {
                    continue;
                }
            }

            // Skip mkdocstrings and pymdown blocks (already handled by LintContext)
            if ctx.lines[line_idx].in_mkdocstrings || ctx.lines[line_idx].in_pymdown_block {
                continue;
            }

            // Skip MyST comments (% comment) — structural lines, not prose
            if ctx.lines[line_idx].is_myst_comment {
                continue;
            }

            // Link reference definitions are always exempt, even in strict mode.
            // There's no way to shorten them without breaking the URL.
            // Also check after stripping list markers, since list items may
            // contain link ref defs as their content.
            {
                let trimmed = line.trim();
                if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
                    continue;
                }
                if is_list_item(trimmed) {
                    let (_, content) = extract_list_marker_and_content(trimmed);
                    let content_trimmed = content.trim();
                    if content_trimmed.starts_with('[')
                        && content_trimmed.contains("]:")
                        && LINK_REF_PATTERN.is_match(content_trimmed)
                    {
                        continue;
                    }
                }
            }

            // Skip various block types efficiently
            if !effective_config.strict {
                // Lines whose only content is a link/image are exempt.
                // After stripping list markers, blockquote markers, and emphasis,
                // if only a link or image remains, there is no way to shorten it.
                if is_standalone_link_or_image_line(ctx, line_number) {
                    continue;
                }

                // Lines consisting entirely of HTML tags are exempt.
                // Badge lines, images with attributes, and similar inline HTML
                // are long due to URLs in attributes and can't be meaningfully shortened.
                if is_html_only_line(line) {
                    continue;
                }

                // Skip setext heading underlines
                if !line.trim().is_empty() && line.trim().chars().all(|c| c == '=' || c == '-') {
                    continue;
                }

                // Skip block elements according to config flags
                // The flags mean: true = check these elements, false = skip these elements
                // So we skip when the flag is FALSE and the line is in that element type
                if (!effective_config.headings && is_heading_line_num(line_number))
                    || (!effective_config.code_blocks
                        && ctx.line_info(line_number).is_some_and(|info| info.in_code_block))
                    || (!effective_config.tables && table_lines_set.contains(&line_number))
                    || (!effective_config.math_blocks && self.line_is_display_math(line_number, ctx))
                    || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
                    || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
                    || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
                    || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
                    || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
                    || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
                    || ctx.line_info(line_number).is_some_and(|info| info.in_pymdown_block)
                {
                    continue;
                }

                // Check if this is a paragraph/regular text line
                // If paragraphs = false, skip lines that are NOT in special blocks
                // Blockquote content is treated as paragraph text, so it's not
                // included in the special blocks list here.
                if !effective_config.paragraphs {
                    let is_special_block = is_heading_line_num(line_number)
                        || ctx.line_info(line_number).is_some_and(|info| info.in_code_block)
                        || table_lines_set.contains(&line_number)
                        || ctx.line_info(line_number).is_some_and(|info| info.in_html_block)
                        || ctx.line_info(line_number).is_some_and(|info| info.in_html_comment)
                        || ctx.line_info(line_number).is_some_and(|info| info.in_esm_block)
                        || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_expression)
                        || ctx.line_info(line_number).is_some_and(|info| info.in_jsx_block)
                        || ctx.line_info(line_number).is_some_and(|info| info.in_mdx_comment)
                        || ctx
                            .line_info(line_number)
                            .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container);

                    // Skip regular paragraph text when paragraphs = false
                    if !is_special_block {
                        continue;
                    }
                }

                // Skip blockquote lines when blockquotes = false.
                // Also skip lazy continuation lines that belong to a blockquote
                // (lines without `>` prefix that follow a blockquote line).
                if !effective_config.blockquotes {
                    if ctx.lines[line_number - 1].blockquote.is_some() {
                        continue;
                    }
                    // Check for lazy continuation: scan backwards through
                    // non-blank lines to find if this paragraph started with
                    // a blockquote marker
                    if !line.trim().is_empty() {
                        let mut scan = line_number.saturating_sub(2);
                        loop {
                            if ctx.lines[scan].blockquote.is_some() {
                                // Found a blockquote ancestor — this is a lazy continuation
                                continue 'line_loop;
                            }
                            if lines[scan].trim().is_empty() || scan == 0 {
                                break;
                            }
                            scan -= 1;
                        }
                    }
                }

                // Skip lines that are only a URL, image ref, or link ref
                if self.should_ignore_line(line, lines, line_idx, ctx) {
                    continue;
                }
            }

            // In sentence-per-line mode, check if this is a single long sentence
            // If so, emit a warning without a fix (user must manually rephrase)
            if effective_config.reflow_mode == ReflowMode::SentencePerLine {
                let sentences = split_into_sentences(line.trim(), Some(&defined_references));
                if sentences.len() == 1 {
                    // Single sentence that's too long - warn but don't auto-fix
                    let message = format!("Line length {effective_length} exceeds {line_limit} characters");

                    let (start_line, start_col, end_line, end_col) =
                        calculate_excess_range(line_number, line, line_limit);

                    warnings.push(LintWarning {
                        rule_name: Some(self.name().to_string()),
                        message,
                        line: start_line,
                        column: start_col,
                        end_line,
                        end_column: end_col,
                        severity: Severity::Warning,
                        fix: None, // No auto-fix for long single sentences
                    });
                    continue;
                }
                // Multiple sentences will be handled by paragraph-based reflow
                continue;
            }

            // In semantic-line-breaks mode, skip per-line checks —
            // all reflow is handled at the paragraph level with cascading splits
            if effective_config.reflow_mode == ReflowMode::SemanticLineBreaks {
                continue;
            }

            // Don't provide fix for individual lines when reflow is enabled
            // Paragraph-based fixes will be handled separately
            let fix = None;

            let message = format!("Line length {effective_length} exceeds {line_limit} characters");

            // Calculate precise character range for the excess portion
            let (start_line, start_col, end_line, end_col) = calculate_excess_range(line_number, line, line_limit);

            warnings.push(LintWarning {
                rule_name: Some(self.name().to_string()),
                message,
                line: start_line,
                column: start_col,
                end_line,
                end_column: end_col,
                severity: Severity::Warning,
                fix,
            });
        }

        // If reflow is enabled, generate paragraph-based fixes
        if effective_config.reflow {
            let paragraph_warnings = self.generate_paragraph_fixes(ctx, &effective_config, lines);
            // Merge paragraph warnings with line warnings, removing duplicates
            for mut pw in paragraph_warnings {
                if ctx.flavor == crate::config::MarkdownFlavor::MDG
                    && (pw.line..=pw.end_line).any(|line_number| is_potential_mdg_step(ctx, line_number))
                {
                    // Keep reporting the excessive line, but do not offer a fix
                    // whose replacement would split a Gherkin step across lines.
                    pw.fix = None;
                }
                // Remove any line warnings that overlap with this paragraph
                warnings.retain(|w| w.line < pw.line || w.line > pw.end_line);
                warnings.push(pw);
            }
        }

        Ok(warnings)
    }

    fn fix(&self, ctx: &crate::lint_context::LintContext) -> Result<String, LintError> {
        // For CLI usage, apply fixes from warnings
        // LSP will use the warning-based fixes directly
        let warnings = self.check(ctx)?;
        let warnings =
            crate::utils::fix_utils::filter_warnings_by_inline_config(warnings, ctx.inline_config(), self.name());

        // If there are no fixes, return content unchanged
        if !warnings.iter().any(|w| w.fix.is_some()) {
            return Ok(ctx.content.to_string());
        }

        // Apply warning-based fixes
        crate::utils::fix_utils::apply_warning_fixes(ctx.content, &warnings)
            .map_err(|e| LintError::FixFailed(format!("Failed to apply fixes: {e}")))
    }

    fn as_any(&self) -> &dyn std::any::Any {
        self
    }

    fn category(&self) -> RuleCategory {
        RuleCategory::Whitespace
    }

    fn should_skip(&self, ctx: &crate::lint_context::LintContext) -> bool {
        self.should_skip_with_config(ctx, &self.config)
    }

    crate::impl_rule_config_sections!(MD013Config);

    fn config_aliases(&self) -> Option<std::collections::HashMap<String, String>> {
        let mut aliases = std::collections::HashMap::new();
        aliases.insert("enable_reflow".to_string(), "reflow".to_string());
        aliases.insert("strict_sentences".to_string(), "require-sentence-capital".to_string());
        aliases.insert("strict-sentences".to_string(), "require-sentence-capital".to_string());
        // Kept in step with the `alias` attributes on `MD013Config::ignore_link_urls`.
        // Serde accepts these spellings, so a config using one is honored; without
        // them here the key validator reports a documented, working key as unknown.
        aliases.insert(
            "semantic-link-understanding".to_string(),
            "ignore-link-urls".to_string(),
        );
        aliases.insert(
            "semantic_link_understanding".to_string(),
            "ignore-link-urls".to_string(),
        );
        Some(aliases)
    }

    fn from_config(config: &crate::config::Config) -> Box<dyn Rule>
    where
        Self: Sized,
    {
        let mut rule_config = crate::rule_config_serde::load_rule_config::<MD013Config>(config);
        // Use global line_length if rule-specific config still has default value
        if rule_config.line_length.get() == 80 {
            rule_config.line_length = config.global.line_length;
        }
        let mut rule = Self::from_config_struct(rule_config);
        // Pull list-marker spacing from MD030 (via the shared serde config loader)
        // so reflow rewrites list items with the configured spacing rather than a
        // hard-coded single space.
        rule.list_spacing = crate::rule_config_serde::load_rule_config::<MD030Config>(config);
        Box::new(rule)
    }
}

impl MD013LineLength {
    /// True when `line_num` (1-indexed) sits inside a `$$` span covering more
    /// than one line, as seen by the byte-level math parser.
    fn line_in_multiline_math_span(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
        ctx.math_spans()
            .iter()
            .any(|span| span.is_display && span.end_line > span.line && (span.line..=span.end_line).contains(&line_num))
    }

    /// True when `line_num` (1-indexed) holds nothing but display math, whether
    /// that is one line of a multi-line block, a delimiter line, or a whole line
    /// that is a single complete `$$...$$` span. This is what `math-blocks =
    /// false` exempts from the length check.
    ///
    /// rumdl models math twice and the two models miss different containers, so
    /// this consults both. `math_spans()` is byte-level and sees a block opened on
    /// a list marker line (`- $$`), which the line-level map cannot because that
    /// line does not begin with `$$`. `LineInfo::in_math_block` is line-level and
    /// sees a four-space-indented block inside a footnote, which the byte-level
    /// parser reads as an indented code block. Taking the union only ever adds
    /// coverage: neither signal fires on ordinary prose, and an unmatched `$$`
    /// opener is flagged by neither, so a stray delimiter cannot exempt the rest
    /// of the document.
    fn line_is_display_math(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
        self.line_holds_only_multiline_math(line_num, ctx)
            || ctx.line_info(line_num).is_some_and(|info| info.in_math_block)
    }

    /// True when a multi-line `$$` span covers `line_num` (1-indexed) and the line
    /// holds nothing besides that math.
    ///
    /// A delimiter line can carry Markdown outside the delimiter: `$$ trailing
    /// prose` closes a block and then continues in prose, and `leading prose $$`
    /// opens one at the end of a sentence. That prose is ordinary text and counts
    /// toward the line's length like any other, so exempting the whole line would
    /// hide arbitrarily long prose behind a delimiter. The line-level map already
    /// leaves such mixed lines unflagged; this is the byte-level half of the same
    /// judgement.
    fn line_holds_only_multiline_math(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
        let Some(info) = ctx.line_info(line_num) else {
            return false;
        };
        let line = info.content(ctx.content);

        ctx.math_spans().iter().any(|span| {
            if !span.is_display || span.end_line <= span.line || !(span.line..=span.end_line).contains(&line_num) {
                return false;
            }
            if line_num == span.line {
                let before = line.get(..span.byte_offset.saturating_sub(info.byte_offset));
                if !before.is_none_or(|before| Self::only_structure_precedes_math(before, info)) {
                    return false;
                }
            }
            if line_num == span.end_line {
                let after = line.get(span.byte_end.saturating_sub(info.byte_offset)..);
                if !after.is_none_or(|after| after.trim().is_empty()) {
                    return false;
                }
            }
            true
        })
    }

    /// True when the text before a block's opening delimiter is only the structure
    /// the block sits in: indentation, a blockquote marker, or the list marker
    /// introducing it. Such a block still owns its whole line.
    fn only_structure_precedes_math(before: &str, info: &crate::lint_context::LineInfo) -> bool {
        let after_marker =
            crate::utils::blockquote::parse_blockquote_prefix(before).map_or(before, |prefix| prefix.content);
        if after_marker.trim().is_empty() {
            return true;
        }
        info.list_item.as_ref().is_some_and(|item| {
            let mut chars = before.chars();
            chars.by_ref().take(item.content_column).count() == item.content_column && chars.as_str().trim().is_empty()
        })
    }

    /// True when `line_num` (1-indexed) falls inside a display-math block that
    /// spans more than one line.
    ///
    /// Line breaks carry meaning inside such a block: a TeX `%` comment runs to
    /// the end of its line, so joining the lines pulls whatever followed on later
    /// lines into the comment and drops it from the rendered equation, which can
    /// also leave an environment unclosed. Reflow therefore leaves these blocks
    /// alone regardless of the `math_blocks` setting, which governs only whether
    /// their length is reported.
    ///
    /// This is `line_is_display_math` minus the case of a whole line that is one
    /// complete `$$...$$` span: such a line is a single atomic element reflow can
    /// move around freely, and only a multi-line block has meaningful internal
    /// line breaks.
    fn line_in_multiline_math_block(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
        self.line_in_multiline_math_span(line_num, ctx)
            || ctx.line_info(line_num).is_some_and(|info| {
                info.in_math_block && !Self::is_self_contained_display_math_line(info.content(ctx.content))
            })
    }

    /// True when `line` is a whole line holding exactly one closed `$$...$$` span.
    fn is_self_contained_display_math_line(line: &str) -> bool {
        let trimmed = line.trim();
        let inner = crate::utils::blockquote::parse_blockquote_prefix(trimmed).map_or(trimmed, |p| p.content.trim());
        inner.strip_prefix("$$").is_some_and(|rest| rest.contains("$$"))
    }

    /// True when `line_num` (1-based) sits inside a structure whose lines must be
    /// preserved verbatim (code block, front matter, HTML/JSX/MDX block, MkDocs
    /// container, div marker, multi-line math block, ...). Used to keep blockquote
    /// reflow from touching quoted-looking text embedded in such structures.
    fn line_in_verbatim_context(&self, line_num: usize, ctx: &crate::lint_context::LintContext) -> bool {
        if self.line_in_multiline_math_block(line_num, ctx) {
            return true;
        }
        ctx.line_info(line_num).is_some_and(|info| {
            info.in_code_block
                || info.in_front_matter
                || info.in_html_block
                || info.in_html_comment
                || info.in_esm_block
                || info.in_jsx_expression
                || info.in_jsx_block
                || info.in_mdx_comment
                || info.in_mkdocstrings
                || info.in_pymdown_block
                || info.in_mkdocs_container()
                || info.is_div_marker
        })
    }

    fn is_blockquote_content_boundary(
        &self,
        content: &str,
        line_num: usize,
        ctx: &crate::lint_context::LintContext,
        strict: bool,
    ) -> bool {
        let trimmed = content.trim();

        trimmed.is_empty()
            || self.line_in_verbatim_context(line_num, ctx)
            || trimmed.starts_with('#')
            || trimmed.starts_with("```")
            || trimmed.starts_with("~~~")
            || trimmed.starts_with('>')
            || TableUtils::is_potential_table_row_with_flavor(content, ctx.flavor)
            || is_list_item(trimmed)
            || is_horizontal_rule(content)
            // A setext underline inside a blockquote, judged from the content
            // text: the parser skips any line starting with `>`, so a
            // blockquoted setext heading carries no HeadingInfo to consult.
            || is_setext_underline_content(content)
            || (trimmed.starts_with('[') && content.contains("]:"))
            || is_template_directive_only(content)
            || is_standalone_attr_list(content)
            || is_snippet_block_delimiter(content)
            || is_github_alert_marker(trimmed)
            || is_html_only_line(content)
            // A standalone link/image line is exempt from MD013 (non-strict mode),
            // so it must end the blockquote paragraph rather than be absorbed into
            // it, mirroring the top-level paragraph reflow boundary.
            || (!strict && is_standalone_link_or_image_line(ctx, line_num))
    }

    #[allow(clippy::too_many_arguments)]
    fn generate_blockquote_paragraph_fix(
        &self,
        ctx: &crate::lint_context::LintContext,
        config: &MD013Config,
        lines: &[&str],
        start_idx: usize,
        line_ending: &str,
        // Extra indent (spaces) to prepend to the emitted `>` prefix so a blockquote
        // nested in a list item moves with its parent's widened marker. Zero unless a
        // non-default MD030 widened an ancestor list item.
        ancestor_shift: isize,
    ) -> (Option<LintWarning>, usize) {
        let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
            return (None, start_idx + 1);
        };
        let target_level = start_bq.nesting_level;
        let defined_references = Self::defined_reference_labels(ctx);

        let mut collected: Vec<CollectedBlockquoteLine> = Vec::new();
        let mut i = start_idx;

        while i < lines.len() {
            if !collected.is_empty() && has_hard_break(&collected[collected.len() - 1].data.content) {
                break;
            }

            let line_num = i + 1;
            if line_num > ctx.lines.len() {
                break;
            }

            if lines[i].trim().is_empty() {
                break;
            }

            let line_bq = ctx.lines[i].blockquote.as_deref();
            if let Some(bq) = line_bq {
                if bq.nesting_level != target_level {
                    break;
                }

                if self.is_blockquote_content_boundary(&bq.content, line_num, ctx, config.strict) {
                    break;
                }

                collected.push(CollectedBlockquoteLine {
                    line_idx: i,
                    data: BlockquoteLineData::explicit(trim_preserving_hard_break(&bq.content), bq.prefix.clone()),
                });
                i += 1;
                continue;
            }

            let lazy_content = lines[i].trim_start();
            if self.is_blockquote_content_boundary(lazy_content, line_num, ctx, config.strict) {
                break;
            }

            collected.push(CollectedBlockquoteLine {
                line_idx: i,
                data: BlockquoteLineData::lazy(trim_preserving_hard_break(lazy_content)),
            });
            i += 1;
        }

        if collected.is_empty() {
            return (None, start_idx + 1);
        }

        let next_idx = i;
        let paragraph_start = collected[0].line_idx;
        let end_line = collected[collected.len() - 1].line_idx;
        let line_data: Vec<BlockquoteLineData> = collected.iter().map(|l| l.data.clone()).collect();
        let paragraph_text = line_data
            .iter()
            .map(|d| d.content.as_str())
            .collect::<Vec<_>>()
            .join(" ");

        let contains_definition_list = line_data
            .iter()
            .any(|d| crate::utils::is_definition_list_item(&d.content));
        if contains_definition_list {
            return (None, next_idx);
        }

        let contains_snippets = line_data.iter().any(|d| is_snippet_block_delimiter(&d.content));
        if contains_snippets {
            return (None, next_idx);
        }

        let needs_reflow = match config.reflow_mode {
            ReflowMode::Normalize => {
                self.normalize_mode_needs_reflow(line_data.iter().map(|d| d.content.as_str()), config)
            }
            ReflowMode::SentencePerLine => {
                let sentences = split_into_sentences(&paragraph_text, Some(&defined_references));
                sentences.len() > 1 || line_data.len() > 1
            }
            ReflowMode::SemanticLineBreaks => {
                let sentences = split_into_sentences(&paragraph_text, Some(&defined_references));
                sentences.len() > 1
                    || line_data.len() > 1
                    || collected
                        .iter()
                        .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get())
            }
            ReflowMode::Default => collected
                .iter()
                .any(|l| self.calculate_effective_length(lines[l.line_idx]) > config.line_length.get()),
        };

        if !needs_reflow {
            return (None, next_idx);
        }

        let fallback_prefix = start_bq.prefix.clone();
        let explicit_prefix = dominant_blockquote_prefix(&line_data, &fallback_prefix);
        // Shift the whole quote right to track a widened parent list item's content
        // column (only widening matters for nesting; a narrowed parent leaves the quote
        // harmlessly over-indented, which MD027/MD030 tidy).
        let explicit_prefix = if ancestor_shift > 0 {
            format!("{}{explicit_prefix}", " ".repeat(ancestor_shift as usize))
        } else {
            explicit_prefix
        };
        let continuation_style = blockquote_continuation_style(&line_data);

        let reflow_line_length = if config.line_length.is_unlimited() {
            usize::MAX
        } else {
            config
                .line_length
                .get()
                .saturating_sub(self.calculate_string_length(&explicit_prefix))
                .max(1)
        };

        let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);

        let reflowed_with_style =
            reflow_blockquote_content(&line_data, &explicit_prefix, continuation_style, &reflow_options);

        if reflowed_with_style.is_empty() {
            return (None, next_idx);
        }

        let reflowed_text = reflowed_with_style.join(line_ending);

        let start_range = ctx.whole_line_byte_range(paragraph_start + 1);
        let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
            ctx.line_text_byte_range(end_line + 1, 1, lines[end_line].len() + 1)
        } else {
            ctx.whole_line_byte_range(end_line + 1)
        };
        let byte_range = start_range.start..end_range.end;

        let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
            format!("{reflowed_text}{line_ending}")
        } else {
            reflowed_text
        };

        let original_text = &ctx.content[byte_range.clone()];
        if original_text == replacement {
            return (None, next_idx);
        }

        let (warning_line, warning_end_line) = match config.reflow_mode {
            ReflowMode::Normalize => (paragraph_start + 1, end_line + 1),
            ReflowMode::SentencePerLine | ReflowMode::SemanticLineBreaks => (paragraph_start + 1, end_line + 1),
            ReflowMode::Default => {
                let violating_line = collected
                    .iter()
                    .find(|line| self.calculate_effective_length(lines[line.line_idx]) > config.line_length.get())
                    .map_or(paragraph_start + 1, |line| line.line_idx + 1);
                (violating_line, violating_line)
            }
        };

        let warning = LintWarning {
            rule_name: Some(self.name().to_string()),
            message: match config.reflow_mode {
                ReflowMode::Normalize => format!(
                    "Paragraph could be normalized to use line length of {} characters",
                    config.line_length.get()
                ),
                ReflowMode::SentencePerLine => {
                    let num_sentences = split_into_sentences(&paragraph_text, Some(&defined_references)).len();
                    if line_data.len() == 1 {
                        format!("Line contains {num_sentences} sentences (one sentence per line required)")
                    } else {
                        let num_lines = line_data.len();
                        format!(
                            "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
                        )
                    }
                }
                ReflowMode::SemanticLineBreaks => {
                    let num_sentences = split_into_sentences(&paragraph_text, Some(&defined_references)).len();
                    format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
                }
                ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
            },
            line: warning_line,
            column: 1,
            end_line: warning_end_line,
            end_column: lines[warning_end_line.saturating_sub(1)].chars().count() + 1,
            severity: Severity::Warning,
            fix: Some(crate::rule::Fix::new(byte_range, replacement)),
        };

        (Some(warning), next_idx)
    }

    /// Reflow a single list item that lives inside a blockquote.
    ///
    /// The blockquote paragraph reflow treats a list marker as a content boundary,
    /// so `> - long item ...` is never wrapped even though the identical top-level
    /// item is. This handles that case: it collects one tight, prose-only list item
    /// at the starting blockquote level, reflows the item body to the configured
    /// width, and re-emits it with the blockquote prefix preserved and continuation
    /// lines aligned under the list content.
    ///
    /// Sibling and nested list items end collection and are reflowed independently
    /// by the caller's outer loop (a nested item carries its indent folded into the
    /// blockquote prefix, so it reflows correctly on its own). Items that are not
    /// simple tight prose - those embedding a code block, table, fence, or hard
    /// break - are left untouched (`None`) and the cursor still advances past the
    /// whole item so its inner lines are never reprocessed as loose prose.
    #[allow(clippy::too_many_arguments)]
    fn generate_blockquote_list_item_fix(
        &self,
        ctx: &crate::lint_context::LintContext,
        config: &MD013Config,
        lines: &[&str],
        start_idx: usize,
        line_ending: &str,
        // Extra indent (spaces) to prepend to the emitted `>` prefix when this quoted
        // list lives inside a list item whose marker widened. Zero unless a non-default
        // MD030 widened an ancestor list item.
        ancestor_shift: isize,
    ) -> (Option<LintWarning>, usize) {
        use crate::utils::blockquote::effective_indent_in_blockquote;

        let Some(start_bq) = ctx.lines.get(start_idx).and_then(|line| line.blockquote.as_deref()) else {
            return (None, start_idx + 1);
        };

        let defined_references = Self::defined_reference_labels(ctx);

        // A `>`-prefixed line can be marked as a blockquote even inside a fenced code
        // block (or other verbatim structure); such content must never be reflowed.
        if self.line_in_verbatim_context(start_idx + 1, ctx) {
            return (None, start_idx + 1);
        }

        let target_level = start_bq.nesting_level;

        // The marker line carries the canonical blockquote prefix: its content begins
        // with the list marker, so no list indent has been folded into the prefix.
        // Track a widened parent list item's content column so the quote stays nested
        // (only widening can detach it; narrowing just over-indents).
        let bq_prefix = if ancestor_shift > 0 {
            format!("{}{}", " ".repeat(ancestor_shift as usize), start_bq.prefix)
        } else {
            start_bq.prefix.clone()
        };

        // A thematic break opens with what looks like a bullet marker (`- - -`).
        // It is not a list item, and reflowing it as prose destroys the break.
        // The top-level reflow path applies the same exemption.
        if is_horizontal_rule(&start_bq.content) {
            return (None, start_idx + 1);
        }

        let (marker, first_body) = extract_list_marker_and_content(&start_bq.content);
        if marker.is_empty() {
            return (None, start_idx + 1);
        }
        let marker_width = marker.chars().count();

        // Continuation lines of a checkbox item align under the bullet+checkbox, but
        // are recognized from the bullet width, matching the top-level list reflow.
        let base_marker_width = ["[ ] ", "[x] ", "[X] "]
            .iter()
            .find_map(|cb| marker.find(*cb))
            .unwrap_or(marker_width);

        // Collect the item: the marker line plus its tight prose continuation lines.
        // `end_idx` always tracks the last consumed line so the cursor advances past
        // the entire item, even when it turns out to be too complex to reflow safely.
        let first_piece = trim_preserving_hard_break(&first_body);
        let mut simple = !has_hard_break(&first_piece);
        let mut body_pieces: Vec<String> = vec![first_piece];
        let mut end_idx = start_idx;
        let mut i = start_idx + 1;

        while i < lines.len() {
            let Some(bq) = ctx.lines[i].blockquote.as_deref() else {
                // A blank line ends the item.
                if lines[i].trim().is_empty() {
                    break;
                }
                // A lazy continuation (no `>` marker) is too ambiguous to reflow
                // safely. Consume the whole lazy run into this item's span and leave
                // the item untouched, so the caller does not reflow the continuation
                // in isolation and leave the marker line partially fixed.
                simple = false;
                while i < lines.len() && ctx.lines[i].blockquote.is_none() && !lines[i].trim().is_empty() {
                    end_idx = i;
                    i += 1;
                }
                break;
            };
            if bq.nesting_level != target_level {
                break;
            }

            let content = bq.content.as_str();
            if content.trim().is_empty() {
                // Blank quoted line ends the tight paragraph. A following indented
                // paragraph (loose item) is reflowed on its own by the prose path.
                break;
            }

            let eff_indent = effective_indent_in_blockquote(lines[i], target_level, 0);
            if eff_indent < base_marker_width {
                // Dedented: a sibling list item or text outside this item. Stop here
                // and let the outer loop classify it.
                break;
            }
            if is_list_item(content) {
                // A nested list item: its own item, handled independently.
                break;
            }

            // An embedded structure (code block, table, fence, nested quote, ...)
            // means the item is not simple prose: keep consuming so the cursor clears
            // the whole structure, but do not produce a fix.
            if self.is_blockquote_content_boundary(content, i + 1, ctx, config.strict) {
                simple = false;
            }

            let piece = trim_preserving_hard_break(content);
            if has_hard_break(&piece) {
                simple = false;
            }
            body_pieces.push(piece);
            end_idx = i;
            i += 1;
        }

        let next_idx = end_idx + 1;

        if !simple {
            return (None, next_idx);
        }

        let exceeds_limit =
            || (start_idx..=end_idx).any(|idx| self.calculate_effective_length(lines[idx]) > config.line_length.get());
        let body_text = body_pieces.join(" ");
        let body_text = body_text.trim();

        // Some bodies cannot be shortened and must stay verbatim, matching the
        // exemptions the top-level list reflow applies: link reference definitions
        // always, and (in non-strict mode) standalone links/images and HTML-only
        // lines. Reflowing a link reference definition would split it after the
        // colon/URL and break the definition.
        let is_link_ref_def =
            body_text.starts_with('[') && body_text.contains("]:") && LINK_REF_PATTERN.is_match(body_text);
        let raw_marker_line = lines[start_idx];
        let body_is_unwrappable = is_link_ref_def
            || (!config.strict && is_standalone_link_or_image_line(ctx, start_idx + 1))
            || (!config.strict && is_html_only_line(raw_marker_line));
        if body_is_unwrappable {
            return (None, next_idx);
        }

        let needs_reflow = match config.reflow_mode {
            ReflowMode::Normalize => body_pieces.len() > 1 || exceeds_limit(),
            ReflowMode::Default => exceeds_limit(),
            ReflowMode::SentencePerLine => {
                split_into_sentences(body_text, Some(&defined_references)).len() > 1 || body_pieces.len() > 1
            }
            ReflowMode::SemanticLineBreaks => {
                split_into_sentences(body_text, Some(&defined_references)).len() > 1 || exceeds_limit()
            }
        };
        if !needs_reflow {
            return (None, next_idx);
        }

        // Apply MD030 list-marker spacing in the spacing-normalizing modes, mirroring
        // the top-level list reflow: derive the post-marker spacing from MD030 and let
        // the continuation indent follow the resulting content width. Default MD030 (a
        // single space) leaves the marker unchanged. MkDocs keeps its rigid indent.
        let (marker, marker_width) = if matches!(config.reflow_mode, ReflowMode::Default | ReflowMode::Normalize)
            && !ctx.flavor.requires_strict_list_indent()
        {
            let is_ordered = marker.starts_with(|c: char| c.is_ascii_digit());
            // Bullet/number portion only (e.g. `-`, `1.`); the checkbox is content.
            let bullet = marker.split(' ').next().unwrap_or("");
            let bullet_len = bullet.chars().count();
            let checkbox_tail = &marker[base_marker_width..];
            // Decide single- vs multi-line spacing from the rewritten shape. This path
            // only handles a single tight prose paragraph (structural or multi-paragraph
            // items set `simple = false` and bail out above), so the emitted item stays
            // multi-line solely when the joined body wraps past one line. A multi-line
            // *source* that collapses onto the marker line must use MD030's single-line
            // spacing, matching the top-level reflow. The wrap test measures at the
            // single-line content column so it does not depend on the spacing chosen here.
            let single_col = self.calculate_string_length(&bq_prefix)
                + bullet_len
                + self.list_spacing.expected_spaces(is_ordered, false, bullet_len)
                + checkbox_tail.chars().count();
            let is_multi = !body_text.is_empty()
                && self.calculate_effective_length(&format!("{}{body_text}", " ".repeat(single_col)))
                    > config.line_length.effective_limit();
            let spaces = self.list_spacing.expected_spaces(is_ordered, is_multi, bullet_len);
            let new_marker = format!("{bullet}{}{checkbox_tail}", " ".repeat(spaces));
            let width = new_marker.chars().count();
            (new_marker, width)
        } else {
            (marker, marker_width)
        };

        let prefix_width = self.calculate_string_length(&bq_prefix) + self.calculate_string_length(&marker);
        let reflow_line_length = if config.line_length.is_unlimited() {
            usize::MAX
        } else {
            config.line_length.get().saturating_sub(prefix_width).max(1)
        };

        let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);

        let reflowed = crate::utils::text_reflow::reflow_line(body_text, &reflow_options);
        if reflowed.is_empty() {
            return (None, next_idx);
        }

        let continuation_indent = " ".repeat(marker_width);
        let reflowed_text = reflowed
            .iter()
            .enumerate()
            .map(|(idx, line)| {
                if idx == 0 {
                    format!("{bq_prefix}{marker}{line}")
                } else {
                    format!("{bq_prefix}{continuation_indent}{line}")
                }
            })
            .collect::<Vec<_>>()
            .join(line_ending);

        let start_range = ctx.whole_line_byte_range(start_idx + 1);
        let end_range = if end_idx == lines.len() - 1 && !ctx.content.ends_with('\n') {
            ctx.line_text_byte_range(end_idx + 1, 1, lines[end_idx].len() + 1)
        } else {
            ctx.whole_line_byte_range(end_idx + 1)
        };
        let byte_range = start_range.start..end_range.end;

        let replacement = if end_idx < lines.len() - 1 || ctx.content.ends_with('\n') {
            format!("{reflowed_text}{line_ending}")
        } else {
            reflowed_text
        };

        let original_text = &ctx.content[byte_range.clone()];
        if original_text == replacement {
            return (None, next_idx);
        }

        let message = match config.reflow_mode {
            ReflowMode::Normalize => format!(
                "Paragraph could be normalized to use line length of {} characters",
                config.line_length.get()
            ),
            ReflowMode::SentencePerLine => {
                let num_sentences = split_into_sentences(body_text, Some(&defined_references)).len();
                format!("List item should have one sentence per line (found {num_sentences} sentences)")
            }
            ReflowMode::SemanticLineBreaks => {
                let num_sentences = split_into_sentences(body_text, Some(&defined_references)).len();
                format!("List item should use semantic line breaks ({num_sentences} sentences)")
            }
            ReflowMode::Default => format!("Line length exceeds {} characters", config.line_length.get()),
        };

        let warning = LintWarning {
            rule_name: Some(self.name().to_string()),
            message,
            line: start_idx + 1,
            column: 1,
            end_line: end_idx + 1,
            end_column: lines[end_idx].chars().count() + 1,
            severity: Severity::Warning,
            fix: Some(crate::rule::Fix::new(byte_range, replacement)),
        };

        (Some(warning), next_idx)
    }

    /// Generate paragraph-based fixes
    fn generate_paragraph_fixes(
        &self,
        ctx: &crate::lint_context::LintContext,
        config: &MD013Config,
        lines: &[&str],
    ) -> Vec<LintWarning> {
        let mut warnings = Vec::new();
        let defined_references = Self::defined_reference_labels(ctx);

        // Detect the content's line ending style to preserve it in replacements.
        // The LSP receives content from editors which may use CRLF (Windows).
        // Replacements must match the original line endings to avoid false positives.
        let line_ending = crate::utils::line_ending::detect_line_ending(ctx.content);

        // Ancestor list-item indent shifts, innermost last. When a reflowed parent's
        // marker widens under a non-default MD030 (e.g. ul-multi = 3 moves the parent's
        // content from column 2 to 4), its nested list/blockquote children are reflowed
        // independently and would otherwise keep their original indent — leaving them
        // under the parent's new content column, where a CommonMark parser reparses them
        // as siblings rather than children. Each frame is (normalized marker width,
        // cumulative shift applied to that item's content column); a descendant adds its
        // innermost open ancestor's shift to its own indent so the whole subtree moves
        // together. With a default MD030 and no marker padding every shift is 0, so this
        // is inert and the output is byte-identical.
        let mut list_shift_stack: Vec<(usize, isize)> = Vec::new();

        let mut i = 0;
        while i < lines.len() {
            let line_num = i + 1;

            // Close ancestor frames whose list item has ended at this line: a non-blank
            // line indented less than the frame's source content column is no longer
            // inside that item. Blank lines alone don't close a (loose) list item.
            if !list_shift_stack.is_empty()
                && let Some(info) = ctx.lines.get(i)
                && !info.is_blank
            {
                while let Some(&(content_column, _)) = list_shift_stack.last() {
                    if info.indent < content_column {
                        list_shift_stack.pop();
                    } else {
                        break;
                    }
                }
            }

            // Handle blockquote paragraphs with style-preserving reflow.
            // Skip blockquotes when blockquotes=false or paragraphs=false
            if line_num > 0 && line_num <= ctx.lines.len() && ctx.lines[line_num - 1].blockquote.is_some() {
                if !config.blockquotes || !config.paragraphs {
                    // Skip past all blockquote lines (explicit and lazy continuations).
                    // A lazy continuation is a non-blank line without `>` that follows
                    // a blockquote line and isn't a structural element.
                    let mut saw_explicit_bq = false;
                    while i < lines.len() && i < ctx.lines.len() {
                        if ctx.lines[i].blockquote.is_some() {
                            saw_explicit_bq = true;
                            i += 1;
                        } else if saw_explicit_bq
                            && !lines[i].trim().is_empty()
                            && !lines[i].trim_start().starts_with('#')
                            && !lines[i].trim_start().starts_with('>')
                        {
                            // Lazy continuation of preceding blockquote
                            i += 1;
                        } else {
                            break;
                        }
                    }
                    continue;
                }
                // A blockquote nested in a list item moves with its parent when the
                // parent's marker widens (see `list_shift_stack`); pass that shift so the
                // emitted `>` prefix lands under the parent's new content column instead
                // of detaching into a sibling.
                let ancestor_shift = list_shift_stack.last().map_or(0isize, |&(_, shift)| shift);
                // A list item inside the blockquote needs list-aware reflow (marker +
                // continuation indent); plain prose goes through the paragraph path.
                let is_bq_list_item = ctx.lines[i]
                    .blockquote
                    .as_deref()
                    .is_some_and(|bq| is_list_item(&bq.content));
                let (warning, next_idx) = if is_bq_list_item {
                    self.generate_blockquote_list_item_fix(ctx, config, lines, i, line_ending, ancestor_shift)
                } else {
                    self.generate_blockquote_paragraph_fix(ctx, config, lines, i, line_ending, ancestor_shift)
                };
                if let Some(warning) = warning {
                    warnings.push(warning);
                }
                i = next_idx;
                continue;
            }

            // Skip special structures (but NOT MkDocs containers - those get special handling)
            let should_skip_due_to_line_info = ctx.line_info(line_num).is_some_and(|info| {
                info.in_code_block
                    || info.in_front_matter
                    || info.in_html_block
                    || info.in_html_comment
                    || info.in_esm_block
                    || info.in_jsx_expression
                    || info.in_jsx_block
                    || info.in_mdx_comment
                    || info.in_mkdocstrings
                    || info.in_pymdown_block
            });

            // Skip link reference definitions but NOT footnote definitions.
            // Footnote definitions (`[^id]: prose`) contain reflowable text,
            // while link reference definitions (`[ref]: URL`) contain URLs
            // that cannot be shortened.
            let is_link_ref_def =
                lines[i].trim().starts_with('[') && !lines[i].trim().starts_with("[^") && lines[i].contains("]:");

            // A setext heading is a heading, not a paragraph: skip its text line
            // and its underline together, the way an ATX heading is skipped just
            // below. Reflowing either half rewrites the document's structure -
            // joining the underline onto the text demotes the heading to prose.
            if is_setext_heading_text_line(ctx, line_num) {
                i += 2;
                continue;
            }

            if should_skip_due_to_line_info
                || lines[i].trim().starts_with('#')
                || TableUtils::is_potential_table_row_with_flavor(lines[i], ctx.flavor)
                || lines[i].trim().is_empty()
                || is_horizontal_rule(lines[i])
                || is_template_directive_only(lines[i])
                || is_link_ref_def
                || ctx.line_info(line_num).is_some_and(|info| info.is_div_marker)
                || is_html_only_line(lines[i])
                || (!config.strict && is_standalone_link_or_image_line(ctx, line_num))
            {
                i += 1;
                continue;
            }

            // Handle footnote definitions: `[^id]: prose text that can be reflowed`
            // Supports multi-paragraph footnotes with code blocks, blockquotes,
            // tables, and lists preserved verbatim.
            // Validate structure: must start with `[^`, contain `]:`, and the ID
            // must not contain `[` or `]` (prevents false matches on nested brackets)
            if lines[i].trim().starts_with("[^") && lines[i].contains("]:") && {
                let after_caret = &lines[i].trim()[2..];
                after_caret
                    .find("]:")
                    .is_some_and(|pos| pos > 0 && !after_caret[..pos].contains(['[', ']']))
            } {
                let footnote_start = i;
                let line = lines[i];

                // Extract the prefix `[^id]:`
                let Some(colon_pos) = line.find("]:") else {
                    i += 1;
                    continue;
                };
                let prefix_end = colon_pos + 2;
                let prefix = &line[..prefix_end];

                // Content starts after `]: ` (with optional space)
                let content_start = if line[prefix_end..].starts_with(' ') {
                    prefix_end + 1
                } else {
                    prefix_end
                };
                let first_content = &line[content_start..];

                // CommonMark footnotes use 4-space continuation indent
                const FN_INDENT: usize = 4;

                // --- Line classification for footnote content ---
                #[derive(Debug, Clone)]
                enum FnLineType {
                    Content(String),
                    Verbatim(String, usize), // preserved text, original indent
                    Empty,
                }

                // Helper: compute visual indent (tabs = 4 spaces)
                let visual_indent = |s: &str| -> usize {
                    s.chars()
                        .take_while(|c| c.is_whitespace())
                        .map(|c| if c == '\t' { 4 } else { 1 })
                        .sum::<usize>()
                };

                // Helper: check if a trimmed line is a fence marker (homogeneous chars)
                let is_fence = |s: &str| -> bool {
                    let t = s.trim();
                    let fence_char = t.chars().next();
                    matches!(fence_char, Some('`') | Some('~'))
                        && t.chars().take_while(|&c| c == fence_char.unwrap()).count() >= 3
                };

                // Helper: check if a trimmed line is a setext underline
                let is_setext_underline = |s: &str| -> bool {
                    let t = s.trim();
                    !t.is_empty()
                        && (t.chars().all(|c| c == '=' || c == ' ') || t.chars().all(|c| c == '-' || c == ' '))
                        && t.contains(['=', '-'])
                };

                // Deferred body: `[^id]:\n    content` — first line has no content,
                // actual content starts on the next indented line
                let deferred_body = first_content.trim().is_empty();

                // Collect all lines belonging to this footnote definition
                let mut fn_lines: Vec<FnLineType> = Vec::new();
                if !deferred_body {
                    fn_lines.push(FnLineType::Content(first_content.to_string()));
                }
                let mut last_consumed = i;
                i += 1;

                // Strip only the footnote continuation indent, preserving
                // internal indentation (e.g., code block body indent)
                let strip_fn_indent = |s: &str| -> String {
                    let mut chars = s.chars();
                    let mut stripped = 0;
                    while stripped < FN_INDENT {
                        match chars.next() {
                            Some('\t') => stripped += 4,
                            Some(c) if c.is_whitespace() => stripped += 1,
                            _ => break,
                        }
                    }
                    chars.as_str().to_string()
                };

                let mut in_fenced_code = false;
                let mut consecutive_blanks = 0u32;

                while i < lines.len() {
                    let next = lines[i];
                    let next_trimmed = next.trim();

                    // Blank line handling
                    if next_trimmed.is_empty() {
                        consecutive_blanks += 1;
                        // 2+ consecutive blanks terminate the footnote
                        if consecutive_blanks >= 2 {
                            break;
                        }

                        // Inside a fenced code block, blank lines are part of the code
                        if in_fenced_code {
                            consecutive_blanks = 0; // Don't count blanks inside code blocks
                            fn_lines.push(FnLineType::Verbatim(String::new(), 0));
                            last_consumed = i;
                            i += 1;
                            continue;
                        }

                        // Peek ahead: if next non-blank line is indented >= FN_INDENT,
                        // this blank is an internal paragraph separator
                        if i + 1 < lines.len() {
                            let peek = lines[i + 1];
                            let peek_indent = visual_indent(peek);
                            if !peek.trim().is_empty() && peek_indent >= FN_INDENT {
                                fn_lines.push(FnLineType::Empty);
                                last_consumed = i;
                                i += 1;
                                continue;
                            }
                        }
                        // No valid continuation after blank — end of footnote
                        break;
                    }

                    consecutive_blanks = 0;
                    let indent = visual_indent(next);

                    // Not indented enough — end of footnote
                    if indent < FN_INDENT {
                        break;
                    }

                    // Inside a fenced code block: everything is verbatim until closing fence
                    if in_fenced_code {
                        fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
                        if is_fence(next_trimmed) {
                            in_fenced_code = false;
                        }
                        last_consumed = i;
                        i += 1;
                        continue;
                    }

                    // Fence opener — start verbatim code block
                    if is_fence(next_trimmed) {
                        in_fenced_code = true;
                        fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
                        last_consumed = i;
                        i += 1;
                        continue;
                    }

                    // A multi-line display-math block is verbatim: its line breaks
                    // carry meaning (see `line_in_multiline_math_block`).
                    if self.line_in_multiline_math_block(i + 1, ctx) {
                        fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
                        last_consumed = i;
                        i += 1;
                        continue;
                    }

                    // Indented code block: indent >= FN_INDENT + 4 (= 8 spaces)
                    if indent >= FN_INDENT + 4 {
                        fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
                        last_consumed = i;
                        i += 1;
                        continue;
                    }

                    // Structural content that must be preserved verbatim
                    if next_trimmed.starts_with('#')
                        || is_list_item(next_trimmed)
                        || next_trimmed.starts_with('>')
                        || TableUtils::is_potential_table_row_with_flavor(next_trimmed, ctx.flavor)
                        || is_setext_underline(next_trimmed)
                        || is_horizontal_rule(next_trimmed)
                        || crate::utils::mkdocs_footnotes::is_footnote_definition(next_trimmed)
                    {
                        // Preserve verbatim: blockquotes, tables, lists, setext
                        // underlines, and horizontal rules inside the footnote
                        if next_trimmed.starts_with('>')
                            || TableUtils::is_potential_table_row_with_flavor(next_trimmed, ctx.flavor)
                            || is_list_item(next_trimmed)
                            || is_setext_underline(next_trimmed)
                            || is_horizontal_rule(next_trimmed)
                        {
                            fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
                            last_consumed = i;
                            i += 1;
                            continue;
                        }
                        // Headings, new footnote defs, link refs — end the footnote
                        break;
                    }

                    // Link reference definitions inside footnotes are not reflowable
                    if next_trimmed.starts_with('[')
                        && !next_trimmed.starts_with("[^")
                        && next_trimmed.contains("]:")
                        && LINK_REF_PATTERN.is_match(next_trimmed)
                    {
                        fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
                        last_consumed = i;
                        i += 1;
                        continue;
                    }

                    // HTML-only lines inside footnotes are not reflowable
                    if is_html_only_line(next_trimmed) {
                        fn_lines.push(FnLineType::Verbatim(strip_fn_indent(next), indent));
                        last_consumed = i;
                        i += 1;
                        continue;
                    }

                    // Regular prose content
                    fn_lines.push(FnLineType::Content(next_trimmed.to_string()));
                    last_consumed = i;
                    i += 1;
                }

                // Nothing collected or only empty lines
                if fn_lines.iter().all(|l| matches!(l, FnLineType::Empty)) || fn_lines.is_empty() {
                    continue;
                }

                // --- Group into blocks ---
                #[derive(Debug)]
                enum FnBlock {
                    Paragraph(Vec<String>),
                    Verbatim(Vec<(String, usize)>), // (content, indent) preserved as-is
                }

                let mut blocks: Vec<FnBlock> = Vec::new();
                let mut current_para: Vec<String> = Vec::new();
                let mut current_verbatim: Vec<(String, usize)> = Vec::new();

                for fl in &fn_lines {
                    match fl {
                        FnLineType::Content(s) => {
                            if !current_verbatim.is_empty() {
                                blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
                            }
                            current_para.push(s.clone());
                        }
                        FnLineType::Verbatim(s, indent) => {
                            if !current_para.is_empty() {
                                blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
                            }
                            current_verbatim.push((s.clone(), *indent));
                        }
                        FnLineType::Empty => {
                            if !current_para.is_empty() {
                                blocks.push(FnBlock::Paragraph(std::mem::take(&mut current_para)));
                            }
                            if !current_verbatim.is_empty() {
                                blocks.push(FnBlock::Verbatim(std::mem::take(&mut current_verbatim)));
                            }
                        }
                    }
                }
                if !current_para.is_empty() {
                    blocks.push(FnBlock::Paragraph(current_para));
                }
                if !current_verbatim.is_empty() {
                    blocks.push(FnBlock::Verbatim(current_verbatim));
                }

                // --- Reflow paragraphs and reconstruct ---
                let prefix_display_width = prefix.chars().count() + 1; // +1 for space
                let reflow_line_length = if config.line_length.is_unlimited() {
                    usize::MAX
                } else {
                    config
                        .line_length
                        .get()
                        .saturating_sub(FN_INDENT.max(prefix_display_width))
                        .max(20)
                };
                // Footnote continuation uses a fixed 4-space indent, so list
                // continuation capping does not apply here.
                let reflow_options = crate::utils::text_reflow::ReflowOptions {
                    max_list_continuation_indent: None,
                    ..Self::reflow_options(ctx, config, reflow_line_length)
                };

                let indent_str = " ".repeat(FN_INDENT);
                let mut result_lines: Vec<String> = Vec::new();
                let mut is_first_block = true;

                for block in &blocks {
                    match block {
                        FnBlock::Paragraph(para_lines) => {
                            let paragraph_text = para_lines.join(" ");
                            let paragraph_text = paragraph_text.trim();
                            if paragraph_text.is_empty() {
                                continue;
                            }

                            let reflowed = crate::utils::text_reflow::reflow_line(paragraph_text, &reflow_options);
                            if reflowed.is_empty() {
                                continue;
                            }

                            // Blank line separator between blocks
                            if !result_lines.is_empty() {
                                result_lines.push(String::new());
                            }

                            for (idx, rline) in reflowed.iter().enumerate() {
                                if is_first_block && idx == 0 {
                                    result_lines.push(format!("{prefix} {rline}"));
                                } else {
                                    result_lines.push(format!("{indent_str}{rline}"));
                                }
                            }
                            is_first_block = false;
                        }
                        FnBlock::Verbatim(verb_lines) => {
                            // Blank line separator between blocks
                            if !result_lines.is_empty() {
                                result_lines.push(String::new());
                            }

                            if is_first_block {
                                // Verbatim as first block in a deferred-body footnote
                                if deferred_body {
                                    result_lines.push(prefix.to_string());
                                }
                                is_first_block = false;
                            }
                            for (content, _orig_indent) in verb_lines {
                                result_lines.push(format!("{indent_str}{content}"));
                            }
                        }
                    }
                }

                // If nothing was produced, skip
                if result_lines.is_empty() {
                    continue;
                }

                let reflowed_text = result_lines.join(line_ending);

                // Calculate byte range using last_consumed
                let start_range = ctx.whole_line_byte_range(footnote_start + 1);
                let end_range = if last_consumed == lines.len() - 1 && !ctx.content.ends_with('\n') {
                    ctx.line_text_byte_range(last_consumed + 1, 1, lines[last_consumed].len() + 1)
                } else {
                    ctx.whole_line_byte_range(last_consumed + 1)
                };
                let byte_range = start_range.start..end_range.end;

                let replacement = if last_consumed < lines.len() - 1 || ctx.content.ends_with('\n') {
                    format!("{reflowed_text}{line_ending}")
                } else {
                    reflowed_text
                };

                let original_text = &ctx.content[byte_range.clone()];
                let max_length = (footnote_start..=last_consumed)
                    .map(|idx| self.calculate_effective_length(lines[idx]))
                    .max()
                    .unwrap_or(0);
                let line_limit = if config.line_length.is_unlimited() {
                    usize::MAX
                } else {
                    config.line_length.get()
                };
                if original_text != replacement && max_length > line_limit {
                    warnings.push(LintWarning {
                        rule_name: Some(self.name().to_string()),
                        message: format!(
                            "Line length {} exceeds {} characters",
                            max_length,
                            config.line_length.get()
                        ),
                        line: footnote_start + 1,
                        column: 1,
                        end_line: last_consumed + 1,
                        end_column: lines[last_consumed].chars().count() + 1,
                        severity: Severity::Warning,
                        fix: Some(crate::rule::Fix::new(byte_range, replacement)),
                    });
                }
                continue;
            }

            // Handle MkDocs container content (admonitions and tabs) with indent-preserving reflow
            if ctx
                .line_info(line_num)
                .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
            {
                // Skip admonition/tab marker lines — only reflow their indented content
                let current_line = lines[i];
                if mkdocs_admonitions::is_admonition_start(current_line) || mkdocs_tabs::is_tab_marker(current_line) {
                    i += 1;
                    continue;
                }

                let container_start = i;

                // Detect the actual indent level from the first content line
                // (supports nested admonitions with 8+ spaces)
                let first_line = lines[i];
                let base_indent_len = first_line.len() - first_line.trim_start().len();
                let base_indent: String = " ".repeat(base_indent_len);

                // Collect consecutive MkDocs container paragraph lines
                let mut container_lines: Vec<&str> = Vec::new();
                while i < lines.len() {
                    let current_line_num = i + 1;
                    let line_info = ctx.line_info(current_line_num);

                    // Stop if we leave the MkDocs container
                    if !line_info.is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container) {
                        break;
                    }

                    let line = lines[i];

                    // Stop at paragraph boundaries within the container
                    if line.trim().is_empty() {
                        break;
                    }

                    // Skip list items, code blocks, headings, HTML-only lines within containers
                    if is_list_item(line.trim())
                        || line.trim().starts_with("```")
                        || line.trim().starts_with("~~~")
                        || line.trim().starts_with('#')
                        || is_html_only_line(line)
                    {
                        break;
                    }

                    container_lines.push(line);
                    i += 1;
                }

                if container_lines.is_empty() {
                    // Must advance i to avoid infinite loop when we encounter
                    // non-paragraph content (code block, list, heading, empty line)
                    // at the start of an MkDocs container
                    i += 1;
                    continue;
                }

                // Strip the base indent from each line and join for reflow
                let stripped_lines: Vec<&str> = container_lines
                    .iter()
                    .map(|line| {
                        if line.starts_with(&base_indent) {
                            &line[base_indent_len..]
                        } else {
                            line.trim_start()
                        }
                    })
                    .collect();
                let paragraph_text = stripped_lines.join(" ");

                // Check if reflow is needed
                let needs_reflow = match config.reflow_mode {
                    ReflowMode::Normalize => self.normalize_mode_needs_reflow(container_lines.iter().copied(), config),
                    ReflowMode::SentencePerLine => {
                        let sentences = split_into_sentences(&paragraph_text, Some(&defined_references));
                        sentences.len() > 1 || container_lines.len() > 1
                    }
                    ReflowMode::SemanticLineBreaks => {
                        let sentences = split_into_sentences(&paragraph_text, Some(&defined_references));
                        sentences.len() > 1
                            || container_lines.len() > 1
                            || container_lines
                                .iter()
                                .any(|line| self.calculate_effective_length(line) > config.line_length.get())
                    }
                    ReflowMode::Default => container_lines
                        .iter()
                        .any(|line| self.calculate_effective_length(line) > config.line_length.get()),
                };

                if !needs_reflow {
                    continue;
                }

                // Calculate byte range for this container paragraph
                let start_range = ctx.whole_line_byte_range(container_start + 1);
                let end_line = container_start + container_lines.len() - 1;
                let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
                    ctx.line_text_byte_range(end_line + 1, 1, lines[end_line].len() + 1)
                } else {
                    ctx.whole_line_byte_range(end_line + 1)
                };
                let byte_range = start_range.start..end_range.end;

                // Reflow with adjusted line length (accounting for the 4-space indent)
                let reflow_line_length = if config.line_length.is_unlimited() {
                    usize::MAX
                } else {
                    config.line_length.get().saturating_sub(base_indent_len).max(1)
                };
                let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
                let reflowed = crate::utils::text_reflow::reflow_line(&paragraph_text, &reflow_options);

                // Re-add the 4-space indent to each reflowed line
                let reflowed_with_indent: Vec<String> =
                    reflowed.iter().map(|line| format!("{base_indent}{line}")).collect();
                let reflowed_text = reflowed_with_indent.join(line_ending);

                // Preserve trailing newline
                let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
                    format!("{reflowed_text}{line_ending}")
                } else {
                    reflowed_text
                };

                // Only generate a warning if the replacement is different
                let original_text = &ctx.content[byte_range.clone()];
                if original_text != replacement {
                    warnings.push(LintWarning {
                        rule_name: Some(self.name().to_string()),
                        message: format!(
                            "Line length {} exceeds {} characters (in MkDocs container)",
                            container_lines.iter().map(|l| l.len()).max().unwrap_or(0),
                            config.line_length.get()
                        ),
                        line: container_start + 1,
                        column: 1,
                        end_line: end_line + 1,
                        end_column: lines[end_line].chars().count() + 1,
                        severity: Severity::Warning,
                        fix: Some(crate::rule::Fix::new(byte_range, replacement)),
                    });
                }
                continue;
            }

            // Helper function to detect semantic line markers
            let is_semantic_line = |content: &str| -> bool {
                let trimmed = content.trim_start();
                let semantic_markers = [
                    "NOTE:",
                    "WARNING:",
                    "IMPORTANT:",
                    "CAUTION:",
                    "TIP:",
                    "DANGER:",
                    "HINT:",
                    "INFO:",
                ];
                semantic_markers.iter().any(|marker| trimmed.starts_with(marker))
            };

            // Helper function to detect fence markers (opening or closing)
            let is_fence_marker = |content: &str| -> bool {
                let trimmed = content.trim_start();
                trimmed.starts_with("```") || trimmed.starts_with("~~~")
            };

            // Check if this is a list item - handle it specially
            let trimmed = lines[i].trim();
            if is_list_item(trimmed) {
                // Collect the entire list item including continuation lines
                let list_start = i;
                let (marker, first_content) = extract_list_marker_and_content(lines[i]);
                let marker_len = marker.len();
                // The normalized marker above is what gets re-emitted; the source marker
                // is where the item's content actually starts. Nested blocks move by the
                // difference between the two content columns, so the shift must be
                // measured against the source, not against the normalized width.
                let source_marker = source_list_marker(lines[i]);
                let source_content_col = source_marker.as_ref().map_or(marker_len, |m| m.content_col);

                // Checkbox ([ ]/[x]/[X]) is inline content, not part of the list marker.
                // Use the base bullet/number marker width for continuation recognition
                // so that continuation lines at 2+ spaces are collected for "- [ ] " items.
                let base_marker_len = if marker.contains("[ ] ") || marker.contains("[x] ") || marker.contains("[X] ") {
                    marker.find('[').unwrap_or(marker_len)
                } else {
                    marker_len
                };

                // MkDocs flavor requires at least 4 spaces for list continuation
                // after a blank line (multi-paragraph list items). For non-blank
                // continuation (lines directly following the marker line), use
                // the natural marker width so that 2-space indent is recognized.
                let item_indent = ctx.lines[i].indent;
                let min_continuation_indent = if ctx.flavor.requires_strict_list_indent() {
                    // Use 4-space relative indent from the list item's nesting level
                    item_indent + (base_marker_len - item_indent).max(4)
                } else {
                    marker_len
                };
                let content_continuation_indent = base_marker_len;

                // Track lines and their types (content, code block, fence, nested list)
                #[derive(Clone)]
                enum LineType {
                    Content(String, usize),           // content and 1-indexed line number
                    CodeBlock(String, usize),         // content and original indent
                    SemanticLine(String), // Lines starting with NOTE:, WARNING:, etc that should stay separate
                    SnippetLine(String),  // MkDocs Snippets delimiters (-8<-) that must stay on their own line
                    DivMarker(String),    // Quarto/Pandoc div markers (::: opening or closing)
                    AdmonitionHeader(String, usize), // header text (e.g. "!!! note") and original indent
                    AdmonitionContent(String, usize), // body content text and original indent
                    Table(String, usize), // GFM table row, preserved verbatim with original indent
                    Empty,
                }

                let start_idx = i;
                let mut list_item_lines: Vec<LineType> = vec![LineType::Content(first_content, i + 1)];
                // Set when collection stops at a nested list item or a nested
                // blockquote that belongs to this item. Such structure is reflowed
                // independently and is therefore absent from `list_item_lines`/`blocks`,
                // but it still keeps the emitted item spanning multiple physical lines,
                // which the MD030 multi-line spacing decision below must account for.
                let mut has_trailing_nested_structure = false;
                i += 1;

                // Collect continuation lines using ctx.lines for metadata
                while i < lines.len() {
                    let line_info = &ctx.lines[i];

                    // Use pre-computed is_blank from ctx
                    if line_info.is_blank {
                        // Empty line - check if next line is indented (part of list item)
                        if i + 1 < lines.len() {
                            let next_info = &ctx.lines[i + 1];

                            // Check if next line is indented enough to be continuation
                            if !next_info.is_blank && next_info.indent >= min_continuation_indent {
                                // This blank line is between paragraphs/blocks in the list item
                                list_item_lines.push(LineType::Empty);
                                i += 1;
                                continue;
                            }
                        }
                        // No indented line after blank, end of list item
                        break;
                    }

                    // Use pre-computed indent from ctx
                    let indent = line_info.indent;

                    // Valid continuation must be indented at least content_continuation_indent.
                    // For non-blank continuation, use marker_len (e.g. 2 for "- ").
                    // MkDocs strict 4-space requirement applies only after blank lines.
                    if indent >= content_continuation_indent {
                        let trimmed = line_info.content(ctx.content).trim();

                        // Check for MkDocs admonition lines inside list items BEFORE
                        // checking in_code_block. Lines inside code blocks within
                        // admonitions have both in_admonition and in_code_block set;
                        // admonition membership takes priority so the entire admonition
                        // structure (including embedded code blocks) is preserved.
                        if line_info.in_admonition {
                            let raw_content = line_info.content(ctx.content);
                            if mkdocs_admonitions::is_admonition_start(raw_content) {
                                let header_text = raw_content[indent..].trim_end().to_string();
                                list_item_lines.push(LineType::AdmonitionHeader(header_text, indent));
                            } else {
                                let body_text = raw_content[indent..].trim_end().to_string();
                                list_item_lines.push(LineType::AdmonitionContent(body_text, indent));
                            }
                            i += 1;
                            continue;
                        }

                        // Use pre-computed in_code_block from ctx
                        if line_info.in_code_block {
                            list_item_lines.push(LineType::CodeBlock(
                                line_info.content(ctx.content)[indent..].to_string(),
                                indent,
                            ));
                            i += 1;
                            continue;
                        }

                        // A multi-line display-math block inside the item is verbatim:
                        // its line breaks carry meaning (see
                        // `line_in_multiline_math_block`), so reuse the code-block
                        // carrier to re-emit it unchanged.
                        if self.line_in_multiline_math_block(i + 1, ctx) {
                            list_item_lines.push(LineType::CodeBlock(
                                line_info.content(ctx.content)[indent..].to_string(),
                                indent,
                            ));
                            i += 1;
                            continue;
                        }

                        // A blockquote nested inside the list item is reflowed by the
                        // blockquote-aware path (it preserves the `>` prefix, including the
                        // list indent), not as list-item prose. Collecting it as Content
                        // would strip the markers and reflow `>` as words, collapsing the
                        // blank `>` line and dropping `>` from wrapped continuations. End
                        // the item here so the outer loop routes the blockquote line to
                        // generate_blockquote_paragraph_fix. Uncollect a pending blank so
                        // the separator between the list prose and the blockquote survives.
                        if line_info.blockquote.is_some() {
                            has_trailing_nested_structure = true;
                            if matches!(list_item_lines.last(), Some(LineType::Empty)) {
                                list_item_lines.pop();
                                i -= 1;
                            }
                            break;
                        }

                        // Check if this is a SIBLING list item (breaks parent)
                        // Nested lists are indented >= marker_len and are PART of the parent item
                        // Siblings are at indent < marker_len (at or before parent marker)
                        if is_list_item(trimmed) && indent < marker_len {
                            // This is a sibling item at same or higher level - end parent item
                            break;
                        }

                        // Nested list items are always processed independently
                        // by the outer loop, so break when we encounter one.
                        // If a blank line was collected before this, uncollect it
                        // so the outer loop preserves the blank between parent and nested.
                        if is_list_item(trimmed) && indent >= marker_len {
                            has_trailing_nested_structure = true;
                            if matches!(list_item_lines.last(), Some(LineType::Empty)) {
                                list_item_lines.pop();
                                i -= 1;
                            }
                            break;
                        }

                        // Normal continuation vs indented code block.
                        // Use min_continuation_indent for the threshold since
                        // code blocks start 4 spaces beyond the expected content
                        // level (which is min_continuation_indent for MkDocs).
                        if indent <= min_continuation_indent + 3 {
                            // Extract content (remove indentation and trailing whitespace)
                            // Preserve hard breaks (2 trailing spaces) while removing excessive whitespace
                            // See: https://github.com/rvben/rumdl/issues/76
                            let content = trim_preserving_hard_break(&line_info.content(ctx.content)[indent..]);

                            // Check if this is a div marker (::: opening or closing)
                            // These must be preserved on their own line, not merged into paragraphs
                            if line_info.is_div_marker {
                                list_item_lines.push(LineType::DivMarker(content));
                            }
                            // Check if this is a fence marker (opening or closing)
                            // These should be treated as code block lines, not paragraph content
                            else if is_fence_marker(&content) {
                                list_item_lines.push(LineType::CodeBlock(content, indent));
                            }
                            // Check if this is a semantic line (NOTE:, WARNING:, etc.)
                            else if is_semantic_line(&content) {
                                list_item_lines.push(LineType::SemanticLine(content));
                            }
                            // Check if this is a snippet block delimiter (-8<- or --8<--)
                            // These must be preserved on their own lines for MkDocs Snippets extension
                            else if is_snippet_block_delimiter(&content) {
                                list_item_lines.push(LineType::SnippetLine(content));
                            }
                            // Check if this is a GFM table row. Tables nested inside list
                            // items must be preserved verbatim — joining them with prose
                            // breaks the column structure.
                            //
                            // `is_potential_table_row` is intentionally permissive at the
                            // row level: any line with `|` and 2+ cells qualifies. To avoid
                            // misclassifying prose continuation lines that contain a literal
                            // pipe (e.g. "use grep | sort to ..."), require one of:
                            //   - the row is pipe-bordered (`| ... |`), the canonical form
                            //     for tables nested in lists; or
                            //   - the next line is a delimiter row (this is a header); or
                            //   - the previous classified line was already a Table (this is
                            //     a continuation row).
                            else if TableUtils::is_potential_table_row_with_flavor(&content, ctx.flavor) && {
                                let pipe_bordered = content.trim().starts_with('|') && content.trim().ends_with('|');
                                let next_is_delim = ctx
                                    .lines
                                    .get(i + 1)
                                    .is_some_and(|next| TableUtils::is_delimiter_row(next.content(ctx.content)));
                                let prev_was_table = matches!(list_item_lines.last(), Some(LineType::Table(..)));
                                pipe_bordered || next_is_delim || prev_was_table
                            } {
                                list_item_lines.push(LineType::Table(content, indent));
                            } else {
                                list_item_lines.push(LineType::Content(content, i + 1));
                            }
                            i += 1;
                        } else {
                            // indent >= min_continuation_indent + 4: indented code block
                            list_item_lines.push(LineType::CodeBlock(
                                line_info.content(ctx.content)[indent..].to_string(),
                                indent,
                            ));
                            i += 1;
                        }
                    } else {
                        // Not indented enough, end of list item
                        break;
                    }
                }

                // Determine the output continuation indent.
                // Normalize/Default modes canonicalize to min_continuation_indent
                // (fixing over-indented continuation). Semantic/SentencePerLine
                // modes preserve the user's actual indent since they only fix
                // line breaking, not indentation.
                let indent_size = match config.reflow_mode {
                    ReflowMode::SemanticLineBreaks | ReflowMode::SentencePerLine => {
                        // Find indent of the first plain text continuation line,
                        // skipping the marker line (index 0), nested list items,
                        // code blocks, and blank lines.
                        list_item_lines
                            .iter()
                            .enumerate()
                            .skip(1)
                            .find_map(|(k, lt)| {
                                if matches!(lt, LineType::Content(..)) {
                                    Some(ctx.lines[list_start + k].indent)
                                } else {
                                    None
                                }
                            })
                            .unwrap_or(min_continuation_indent)
                    }
                    _ => min_continuation_indent,
                };
                // For checkbox items in mkdocs flavor, enforce minimum indent so
                // continuation lines use the structural list indent (4), not the
                // content-aligned indent (6) which Python-Markdown doesn't support
                let has_checkbox = base_marker_len < marker_len;
                let indent_size = if has_checkbox && ctx.flavor.requires_strict_list_indent() {
                    indent_size.max(min_continuation_indent)
                } else {
                    indent_size
                };

                // Split list_item_lines into blocks (paragraphs, code blocks, nested lists, semantic lines, and HTML blocks)
                let mut builder = BlockBuilder::new_with_start_line(start_idx + 1);
                for line in &list_item_lines {
                    match line {
                        LineType::Empty => builder.feed_blank_line(),
                        LineType::Content(content, _) => builder.feed_content(content),
                        LineType::CodeBlock(content, indent) => builder.feed_code_line(content, *indent),
                        LineType::SemanticLine(content) => builder.feed_semantic_line(content),
                        LineType::SnippetLine(content) => builder.feed_snippet_line(content),
                        LineType::DivMarker(content) => builder.feed_div_marker(content),
                        LineType::AdmonitionHeader(header_text, indent) => {
                            builder.feed_admonition_header(header_text, *indent)
                        }
                        LineType::AdmonitionContent(content, indent) => {
                            builder.feed_admonition_content(content, *indent)
                        }
                        LineType::Table(content, indent) => builder.feed_table_line(content, *indent),
                    }
                }
                let blocks = builder.finalize();

                // Helper: check if a line (raw source or stripped content) is exempt
                // from line-length checks. Link reference definitions are always exempt;
                // standalone link/image lines are exempt when strict mode is off.
                // Also checks content after stripping list markers, since list item
                // continuation lines may contain link ref defs.
                let is_exempt_line = |raw_line: &str, line_num: usize| -> bool {
                    let trimmed = raw_line.trim();
                    // Link reference definitions: always exempt
                    if trimmed.starts_with('[') && trimmed.contains("]:") && LINK_REF_PATTERN.is_match(trimmed) {
                        return true;
                    }
                    // Also check after stripping list markers (for list item content)
                    if is_list_item(trimmed) {
                        let (_, content) = extract_list_marker_and_content(trimmed);
                        let content_trimmed = content.trim();
                        if content_trimmed.starts_with('[')
                            && content_trimmed.contains("]:")
                            && LINK_REF_PATTERN.is_match(content_trimmed)
                        {
                            return true;
                        }
                    }
                    // Standalone link/image lines: exempt when not strict
                    if !config.strict && is_standalone_link_or_image_line(ctx, line_num) {
                        return true;
                    }
                    // HTML-only lines: exempt when not strict
                    if !config.strict && is_html_only_line(raw_line) {
                        return true;
                    }
                    false
                };

                // Check if reflowing is needed (only for content paragraphs, not code blocks or nested lists)
                // Exclude link reference definitions and standalone link lines from content
                // so they don't pollute combined_content or trigger false reflow.
                let content_lines: Vec<String> = list_item_lines
                    .iter()
                    .filter_map(|line| {
                        if let LineType::Content(s, line_num) = line {
                            if is_exempt_line(s, *line_num) {
                                return None;
                            }
                            Some(s.clone())
                        } else {
                            None
                        }
                    })
                    .collect();

                // Check if we need to reflow this list item
                // We check the combined content to see if it exceeds length limits
                let combined_content = content_lines.join(" ").trim().to_string();

                // Helper to check if we should reflow in normalize mode
                let should_normalize = || {
                    // Don't normalize if the list item only contains nested lists, code blocks, or semantic lines
                    // DO normalize if it has plain text content that spans multiple lines
                    let has_code_blocks = blocks.iter().any(|b| matches!(b, Block::Code { .. }));
                    let has_semantic_lines = blocks.iter().any(|b| matches!(b, Block::SemanticLine(_)));
                    let has_snippet_lines = blocks.iter().any(|b| matches!(b, Block::SnippetLine(_)));
                    let has_div_markers = blocks.iter().any(|b| matches!(b, Block::DivMarker(_)));
                    let has_admonitions = blocks.iter().any(|b| matches!(b, Block::Admonition { .. }));
                    let has_tables = blocks.iter().any(|b| matches!(b, Block::Table { .. }));
                    let has_paragraphs = blocks.iter().any(|b| matches!(b, Block::Paragraph(_)));

                    // If we have structural blocks but no paragraphs, don't normalize
                    if (has_code_blocks
                        || has_semantic_lines
                        || has_snippet_lines
                        || has_div_markers
                        || has_admonitions
                        || has_tables)
                        && !has_paragraphs
                    {
                        return false;
                    }

                    // If we have paragraphs, check if they span multiple lines or there are multiple blocks
                    if has_paragraphs {
                        // Count only paragraphs that contain at least one non-exempt line.
                        // Paragraphs consisting entirely of link ref defs or standalone links
                        // should not trigger normalization.
                        let paragraph_count = blocks
                            .iter()
                            .filter(|b| {
                                if let Block::Paragraph(para_lines) = b {
                                    !para_lines
                                        .iter()
                                        .all(|(line, line_num)| is_exempt_line(line, *line_num))
                                } else {
                                    false
                                }
                            })
                            .count();
                        if paragraph_count > 1 {
                            // Multiple non-exempt paragraph blocks should be normalized
                            return true;
                        }

                        // Single paragraph block: normalize if it has multiple content lines
                        if content_lines.len() > 1 {
                            return true;
                        }
                    }

                    false
                };

                // Integrate MD030 list-marker spacing (and MD007's text-aligned
                // continuation). In Default/Normalize modes — the modes that already
                // canonicalize spacing/indent — derive the number of spaces after the
                // marker from the configured MD030 values instead of forcing a single
                // space, then align continuation lines to the resulting content column.
                // Sentence/Semantic modes only adjust line breaks, so they keep the
                // marker spacing and indentation already present in the source.
                //
                // With default MD030 (a single space everywhere) the rebuilt marker is
                // byte-identical to the source marker, so this is a no-op and existing
                // behaviour is preserved; only a non-default MD030 changes the output.
                // The MkDocs flavor enforces a rigid structural indent (4 spaces,
                // capped via max_list_continuation_indent) that Python-Markdown
                // requires; leave its specialized handling untouched.
                let (marker, indent_size, code_indent_shift) =
                    if matches!(config.reflow_mode, ReflowMode::Default | ReflowMode::Normalize)
                        && !ctx.flavor.requires_strict_list_indent()
                        && let Some(li) = ctx.lines[list_start].list_item.as_deref()
                    {
                        let bullet_len = li.marker.len();
                        // The checkbox (e.g. `[ ] `) is content, not part of the list
                        // marker MD030 governs; carry it over verbatim after the spacing.
                        let checkbox_tail = marker[base_marker_len..].to_string();
                        // Shift this item right by its ancestors' cumulative marker
                        // widening so a nested item stays under its parent's (widened)
                        // content column. Zero for top-level items and for the whole
                        // tree under default MD030, where the source indent is preserved
                        // verbatim (byte-identical output).
                        let ancestor_shift = list_shift_stack.last().map_or(0isize, |&(_, shift)| shift);
                        let shifted_indent = (item_indent as isize + ancestor_shift).max(0) as usize;
                        let indent_prefix = if ancestor_shift == 0 {
                            marker[..item_indent].to_string()
                        } else {
                            " ".repeat(shifted_indent)
                        };

                        // Decide single- vs multi-line spacing from the *rewritten* shape,
                        // not the source. A multi-line source is not enough: plain prose
                        // continuation collapses onto the marker line during reflow, so a
                        // two-line bullet that fits becomes a single physical line and must
                        // use MD030's single-line spacing (otherwise MD013 emits a result
                        // that MD030 immediately rewrites). The emitted item stays
                        // multi-line only when reflow cannot collapse it:
                        //   - the prose wraps past the line length, or
                        //   - a structural block remains (code, table, admonition, semantic
                        //     line, snippet, div marker, HTML) that is not joinable prose, or
                        //   - more than one paragraph remains (blank-separated), or
                        //   - a nested list/blockquote follows (reflowed independently, so it
                        //     is absent from `blocks` but still keeps the item multi-line).
                        // The wrap test uses the single-line content column so it is
                        // independent of the spacing we are about to choose (avoiding a
                        // circular result). `ol-align-column` ignores this flag entirely in
                        // expected_spaces(), so ordered lists are unaffected.
                        //
                        // This is the rewritten-shape counterpart of MD030's
                        // `is_multi_line_list_item` (which keys off the *source*). The two
                        // are related but technically distinct and intentionally separate;
                        // if the notion of "multi-line" changes in one, revisit the other.
                        let single_col = shifted_indent
                            + bullet_len
                            + self.list_spacing.expected_spaces(li.is_ordered, false, bullet_len)
                            + checkbox_tail.len();
                        let prose_wraps = !combined_content.is_empty()
                            && self
                                .calculate_effective_length(&format!("{}{combined_content}", " ".repeat(single_col)))
                                > config.line_length.effective_limit();
                        let has_structural_block = blocks.iter().any(|b| !matches!(b, Block::Paragraph(_)));
                        let multiple_paragraphs =
                            blocks.iter().filter(|b| matches!(b, Block::Paragraph(_))).count() > 1;
                        let is_multi =
                            prose_wraps || has_structural_block || multiple_paragraphs || has_trailing_nested_structure;

                        let spaces = self.list_spacing.expected_spaces(li.is_ordered, is_multi, bullet_len);
                        let new_marker = format!("{indent_prefix}{}{}{checkbox_tail}", li.marker, " ".repeat(spaces));
                        let new_col = new_marker.chars().count();
                        let shift = new_col as isize - source_content_col as isize;
                        (new_marker, new_col, shift)
                    } else {
                        // MkDocs enforces a rigid structural indent, so the item is emitted
                        // exactly as written and its nested blocks never move. Re-emitting
                        // the normalized marker here would narrow the content column while
                        // leaving those blocks behind.
                        let marker = source_marker.map_or(marker, |m| m.text);
                        (marker, indent_size, 0isize)
                    };
                let expected_indent = " ".repeat(indent_size);

                let needs_reflow = match config.reflow_mode {
                    ReflowMode::Normalize => {
                        // Only reflow if:
                        // 1. Any non-exempt paragraph, when joined, exceeds the limit, OR
                        // 2. Any admonition content line exceeds the limit, OR
                        // 3. The list item should be normalized (has multi-line plain text)
                        let any_paragraph_exceeds = blocks.iter().any(|block| match block {
                            Block::Paragraph(para_lines) => {
                                if para_lines
                                    .iter()
                                    .all(|(line, line_num)| is_exempt_line(line, *line_num))
                                {
                                    return false;
                                }
                                let joined = para_lines.iter().map(|(l, _)| l.as_str()).collect::<Vec<_>>().join(" ");
                                let with_marker = format!("{}{}", " ".repeat(indent_size), joined.trim());
                                self.calculate_effective_length(&with_marker) > config.line_length.get()
                            }
                            Block::Admonition {
                                content_lines,
                                header_indent,
                                ..
                            } => content_lines.iter().any(|(content, indent)| {
                                if content.is_empty() {
                                    return false;
                                }
                                let with_indent = format!("{}{}", " ".repeat(*indent.max(header_indent)), content);
                                self.calculate_effective_length(&with_indent) > config.line_length.get()
                            }),
                            _ => false,
                        });
                        if any_paragraph_exceeds {
                            true
                        } else {
                            should_normalize()
                        }
                    }
                    ReflowMode::SentencePerLine => {
                        // Check if list item has multiple sentences
                        let sentences = split_into_sentences(&combined_content, Some(&defined_references));
                        sentences.len() > 1
                    }
                    ReflowMode::SemanticLineBreaks => {
                        let sentences = split_into_sentences(&combined_content, Some(&defined_references));
                        sentences.len() > 1
                            || (list_start..i).any(|line_idx| {
                                let line = lines[line_idx];
                                let trimmed = line.trim();
                                if trimmed.is_empty() || is_exempt_line(line, line_idx + 1) {
                                    return false;
                                }
                                self.calculate_effective_length(line) > config.line_length.get()
                            })
                    }
                    ReflowMode::Default => {
                        // In default mode, only reflow if any individual non-exempt line exceeds limit
                        (list_start..i).any(|line_idx| {
                            let line = lines[line_idx];
                            let trimmed = line.trim();
                            // Skip blank lines and exempt lines
                            if trimmed.is_empty() || is_exempt_line(line, line_idx + 1) {
                                return false;
                            }
                            self.calculate_effective_length(line) > config.line_length.get()
                        })
                    }
                };

                // Record this item's frame so its nested children inherit the shift.
                // Only a reflowed item's marker actually moves; an unreflowed one keeps
                // its source position and so contributes no shift to its children. The
                // threshold that decides which following lines are inside this item is
                // the normalized marker width, NOT `source_content_col`: the collection
                // loop above gathers continuations by that same width, so the frame
                // boundary must match it or the two would disagree about ownership of
                // lines indented between the normalized and the source content column.
                list_shift_stack.push((marker_len, if needs_reflow { code_indent_shift } else { 0 }));

                if needs_reflow {
                    let start_range = ctx.whole_line_byte_range(list_start + 1);
                    let end_line = i - 1;
                    let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
                        ctx.line_text_byte_range(end_line + 1, 1, lines[end_line].len() + 1)
                    } else {
                        ctx.whole_line_byte_range(end_line + 1)
                    };
                    let byte_range = start_range.start..end_range.end;

                    // Reflow each block (paragraphs only, preserve code blocks)
                    // When line_length = 0 (no limit), use a very large value for reflow
                    let reflow_line_length = if config.line_length.is_unlimited() {
                        usize::MAX
                    } else {
                        config.line_length.get().saturating_sub(indent_size).max(1)
                    };
                    let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);

                    let mut result: Vec<String> = Vec::new();
                    let mut is_first_block = true;

                    for (block_idx, block) in blocks.iter().enumerate() {
                        match block {
                            Block::Paragraph(para_lines) => {
                                // If every line in this paragraph is exempt (link ref defs,
                                // standalone links), preserve the paragraph verbatim instead
                                // of reflowing it. Reflowing would corrupt link ref defs.
                                let all_exempt = para_lines
                                    .iter()
                                    .all(|(line, line_num)| is_exempt_line(line, *line_num));

                                if all_exempt {
                                    for (idx, (line, _)) in para_lines.iter().enumerate() {
                                        if is_first_block && idx == 0 {
                                            result.push(format!("{marker}{line}"));
                                            is_first_block = false;
                                        } else {
                                            result.push(format!("{expected_indent}{line}"));
                                        }
                                    }
                                } else {
                                    // Split the paragraph into segments at hard break boundaries
                                    // Each segment can be reflowed independently
                                    let segments = split_into_segments(para_lines);

                                    for (segment_idx, segment) in segments.iter().enumerate() {
                                        // Check if this segment ends with a hard break and what type
                                        let hard_break_type = segment.last().and_then(|(line, _)| {
                                            let line = line.strip_suffix('\r').unwrap_or(line);
                                            if line.ends_with('\\') {
                                                Some("\\")
                                            } else if line.ends_with("  ") {
                                                Some("  ")
                                            } else {
                                                None
                                            }
                                        });

                                        // Join and reflow the segment (removing the hard break marker for processing)
                                        let segment_for_reflow: Vec<String> = segment
                                            .iter()
                                            .map(|(line, _)| {
                                                // Strip hard break marker (2 spaces or backslash) for reflow processing
                                                if line.ends_with('\\') {
                                                    line[..line.len() - 1].trim_end().to_string()
                                                } else if line.ends_with("  ") {
                                                    line[..line.len() - 2].trim_end().to_string()
                                                } else {
                                                    line.clone()
                                                }
                                            })
                                            .collect();

                                        let segment_text = segment_for_reflow.join(" ").trim().to_string();
                                        if !segment_text.is_empty() {
                                            let reflowed =
                                                crate::utils::text_reflow::reflow_line(&segment_text, &reflow_options);

                                            if is_first_block && segment_idx == 0 {
                                                // First segment of first block starts with marker
                                                result.push(format!("{marker}{}", reflowed[0]));
                                                for line in reflowed.iter().skip(1) {
                                                    result.push(format!("{expected_indent}{line}"));
                                                }
                                                is_first_block = false;
                                            } else {
                                                // Subsequent segments
                                                for line in reflowed {
                                                    result.push(format!("{expected_indent}{line}"));
                                                }
                                            }

                                            // If this segment had a hard break, add it back to the last line
                                            // Preserve the original hard break format (backslash or two spaces)
                                            if let Some(break_marker) = hard_break_type
                                                && let Some(last_line) = result.last_mut()
                                            {
                                                last_line.push_str(break_marker);
                                            }
                                        }
                                    }
                                }

                                // Add blank line after paragraph block if there's a next block.
                                // Check if next block is a code block that doesn't want a preceding blank.
                                // Also don't add blank lines before snippet lines (they should stay tight).
                                // Only add if not already ending with one (avoids double blanks).
                                if block_idx < blocks.len() - 1 {
                                    let next_block = &blocks[block_idx + 1];
                                    let should_add_blank = match next_block {
                                        Block::Code {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::Table {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::SnippetLine(_) | Block::DivMarker(_) => false,
                                        _ => true, // For all other blocks, add blank line
                                    };
                                    if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
                                        result.push(String::new());
                                    }
                                }
                            }
                            Block::Code {
                                lines: code_lines,
                                has_preceding_blank: _,
                            } => {
                                // Preserve code blocks as-is with original indentation
                                // NOTE: Blank line before code block is handled by the previous block
                                // (see paragraph block's logic above)

                                for (idx, (content, orig_indent)) in code_lines.iter().enumerate() {
                                    if is_first_block && idx == 0 {
                                        // First line of first block gets marker
                                        result.push(format!(
                                            "{marker}{}",
                                            " ".repeat(orig_indent - marker_len) + content.as_str()
                                        ));
                                        is_first_block = false;
                                    } else if content.is_empty() {
                                        result.push(String::new());
                                    } else {
                                        // Shift nested code with the marker so it stays
                                        // aligned under content when MD030 widens spacing.
                                        result.push(format!(
                                            "{}{}",
                                            " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
                                            content
                                        ));
                                    }
                                }
                            }
                            Block::SemanticLine(content) => {
                                // Preserve semantic lines (NOTE:, WARNING:, etc.) as-is on their own line.
                                // Only add blank before if not already ending with one.
                                if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
                                    result.push(String::new());
                                }

                                if is_first_block {
                                    // First block starts with marker
                                    result.push(format!("{marker}{content}"));
                                    is_first_block = false;
                                } else {
                                    // Subsequent blocks use expected indent
                                    result.push(format!("{expected_indent}{content}"));
                                }

                                // Add blank line after semantic line if there's a next block.
                                // Only add if not already ending with one.
                                if block_idx < blocks.len() - 1 {
                                    let next_block = &blocks[block_idx + 1];
                                    let should_add_blank = match next_block {
                                        Block::Code {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::Table {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::SnippetLine(_) | Block::DivMarker(_) => false,
                                        _ => true, // For all other blocks, add blank line
                                    };
                                    if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
                                        result.push(String::new());
                                    }
                                }
                            }
                            Block::SnippetLine(content) => {
                                // Preserve snippet delimiters (-8<-) as-is on their own line
                                // Unlike semantic lines, snippet lines don't add extra blank lines
                                if is_first_block {
                                    // First block starts with marker
                                    result.push(format!("{marker}{content}"));
                                    is_first_block = false;
                                } else {
                                    // Subsequent blocks use expected indent
                                    result.push(format!("{expected_indent}{content}"));
                                }
                                // No blank lines added before or after snippet delimiters
                            }
                            Block::DivMarker(content) => {
                                // Preserve div markers (::: opening or closing) as-is on their own line
                                if is_first_block {
                                    result.push(format!("{marker}{content}"));
                                    is_first_block = false;
                                } else {
                                    result.push(format!("{expected_indent}{content}"));
                                }
                            }
                            Block::Html {
                                lines: html_lines,
                                has_preceding_blank: _,
                            } => {
                                // Preserve HTML blocks exactly as-is with original indentation
                                // NOTE: Blank line before HTML block is handled by the previous block

                                for (idx, line) in html_lines.iter().enumerate() {
                                    if is_first_block && idx == 0 {
                                        // First line of first block gets marker
                                        result.push(format!("{marker}{line}"));
                                        is_first_block = false;
                                    } else if line.is_empty() {
                                        // Preserve blank lines inside HTML blocks
                                        result.push(String::new());
                                    } else {
                                        // Preserve lines with their original content (already includes indentation)
                                        result.push(format!("{expected_indent}{line}"));
                                    }
                                }

                                // Add blank line after HTML block if there's a next block.
                                // Only add if not already ending with one (avoids double blanks
                                // when the HTML block itself contained a trailing blank line).
                                if block_idx < blocks.len() - 1 {
                                    let next_block = &blocks[block_idx + 1];
                                    let should_add_blank = match next_block {
                                        Block::Code {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::Html {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::Table {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::SnippetLine(_) | Block::DivMarker(_) => false,
                                        _ => true, // For all other blocks, add blank line
                                    };
                                    if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
                                        result.push(String::new());
                                    }
                                }
                            }
                            Block::Table {
                                lines: table_lines,
                                has_preceding_blank: _,
                            } => {
                                // Preserve table rows verbatim with their original indentation.
                                // Reflowing rows would corrupt column alignment and inject `|`
                                // characters mid-paragraph (issue #590).
                                // The leading blank line is emitted by the previous block.
                                for (idx, (content, orig_indent)) in table_lines.iter().enumerate() {
                                    if is_first_block && idx == 0 {
                                        // First line of first block gets the list marker
                                        result.push(format!(
                                            "{marker}{}",
                                            " ".repeat(orig_indent.saturating_sub(marker_len)) + content.as_str()
                                        ));
                                        is_first_block = false;
                                    } else {
                                        // Shift nested table rows with the marker so they
                                        // stay aligned when MD030 widens marker spacing.
                                        result.push(format!(
                                            "{}{}",
                                            " ".repeat((*orig_indent as isize + code_indent_shift).max(0) as usize),
                                            content
                                        ));
                                    }
                                }

                                // Add blank line after table block if there's a next block.
                                if block_idx < blocks.len() - 1 {
                                    let next_block = &blocks[block_idx + 1];
                                    let should_add_blank = match next_block {
                                        Block::Code {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::Html {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::Table {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::SnippetLine(_) | Block::DivMarker(_) => false,
                                        _ => true,
                                    };
                                    if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
                                        result.push(String::new());
                                    }
                                }
                            }
                            Block::Admonition {
                                header,
                                header_indent,
                                content_lines: admon_lines,
                            } => {
                                // Reconstruct admonition block with header at original indent
                                // and body content reflowed to fit within the line length limit

                                // Add blank line before admonition if not first block
                                if !is_first_block && result.last().is_none_or(|s: &String| !s.is_empty()) {
                                    result.push(String::new());
                                }

                                // Output the header at its original indent
                                let header_indent_str = " ".repeat(*header_indent);
                                if is_first_block {
                                    result.push(format!(
                                        "{marker}{}",
                                        " ".repeat(header_indent.saturating_sub(marker_len)) + header.as_str()
                                    ));
                                    is_first_block = false;
                                } else {
                                    result.push(format!("{header_indent_str}{header}"));
                                }

                                // Derive body indent from the first non-empty content line's
                                // stored indent, falling back to header_indent + 4 for
                                // empty-body admonitions
                                let body_indent = admon_lines
                                    .iter()
                                    .find(|(content, _)| !content.is_empty())
                                    .map_or(header_indent + 4, |(_, indent)| *indent);
                                let body_indent_str = " ".repeat(body_indent);

                                // Segment body content into code blocks (verbatim) and
                                // text paragraphs (reflowable), separated by blank lines.
                                // Code lines store (content, orig_indent) to reconstruct
                                // internal indentation relative to body_indent.
                                enum AdmonSegment {
                                    Text(Vec<String>),
                                    Code(Vec<(String, usize)>),
                                }

                                let mut segments: Vec<AdmonSegment> = Vec::new();
                                let mut current_text: Vec<String> = Vec::new();
                                let mut current_code: Vec<(String, usize)> = Vec::new();
                                let mut in_admon_code = false;
                                // Track the opening fence character so closing fences
                                // must match (backticks close backticks, tildes close tildes)
                                let mut fence_char: char = '`';

                                // Opening fences: ``` or ~~~ followed by optional info string
                                let get_opening_fence = |s: &str| -> Option<(char, usize)> {
                                    let t = s.trim_start();
                                    if t.starts_with("```") {
                                        Some(('`', t.bytes().take_while(|&b| b == b'`').count()))
                                    } else if t.starts_with("~~~") {
                                        Some(('~', t.bytes().take_while(|&b| b == b'~').count()))
                                    } else {
                                        None
                                    }
                                };
                                // Closing fences: ONLY fence chars + optional trailing spaces
                                let get_closing_fence = |s: &str| -> Option<(char, usize)> {
                                    let t = s.trim();
                                    if t.starts_with("```") && t.bytes().all(|b| b == b'`') {
                                        Some(('`', t.len()))
                                    } else if t.starts_with("~~~") && t.bytes().all(|b| b == b'~') {
                                        Some(('~', t.len()))
                                    } else {
                                        None
                                    }
                                };
                                let mut fence_len: usize = 3;

                                for (content, orig_indent) in admon_lines {
                                    if in_admon_code {
                                        // Closing fence must use the same character, be
                                        // at least as long, and have no info string
                                        if let Some((ch, len)) = get_closing_fence(content)
                                            && ch == fence_char
                                            && len >= fence_len
                                        {
                                            current_code.push((content.clone(), *orig_indent));
                                            in_admon_code = false;
                                            segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
                                            continue;
                                        }
                                        current_code.push((content.clone(), *orig_indent));
                                    } else if let Some((ch, len)) = get_opening_fence(content) {
                                        if !current_text.is_empty() {
                                            segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
                                        }
                                        in_admon_code = true;
                                        fence_char = ch;
                                        fence_len = len;
                                        current_code.push((content.clone(), *orig_indent));
                                    } else if content.is_empty() {
                                        if !current_text.is_empty() {
                                            segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
                                        }
                                    } else {
                                        current_text.push(content.clone());
                                    }
                                }
                                if in_admon_code && !current_code.is_empty() {
                                    segments.push(AdmonSegment::Code(std::mem::take(&mut current_code)));
                                }
                                if !current_text.is_empty() {
                                    segments.push(AdmonSegment::Text(std::mem::take(&mut current_text)));
                                }

                                // Build reflow options once for all text segments
                                let admon_reflow_length = if config.line_length.is_unlimited() {
                                    usize::MAX
                                } else {
                                    config.line_length.get().saturating_sub(body_indent).max(1)
                                };

                                let admon_reflow_options = Self::reflow_options(ctx, config, admon_reflow_length);

                                // Output each segment
                                for segment in &segments {
                                    // Blank line before each segment (after the header or previous segment)
                                    result.push(String::new());

                                    match segment {
                                        AdmonSegment::Code(lines) => {
                                            for (line, orig_indent) in lines {
                                                if line.is_empty() {
                                                    // Preserve blank lines inside code blocks
                                                    result.push(String::new());
                                                } else {
                                                    // Reconstruct with body_indent + any extra
                                                    // indentation the line had beyond body_indent
                                                    let extra = orig_indent.saturating_sub(body_indent);
                                                    let indent_str = " ".repeat(body_indent + extra);
                                                    result.push(format!("{indent_str}{line}"));
                                                }
                                            }
                                        }
                                        AdmonSegment::Text(lines) => {
                                            let paragraph_text = lines.join(" ").trim().to_string();
                                            if paragraph_text.is_empty() {
                                                continue;
                                            }
                                            let reflowed = crate::utils::text_reflow::reflow_line(
                                                &paragraph_text,
                                                &admon_reflow_options,
                                            );
                                            for line in &reflowed {
                                                result.push(format!("{body_indent_str}{line}"));
                                            }
                                        }
                                    }
                                }

                                // Add blank line after admonition if there's a next block
                                if block_idx < blocks.len() - 1 {
                                    let next_block = &blocks[block_idx + 1];
                                    let should_add_blank = match next_block {
                                        Block::Code {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::Table {
                                            has_preceding_blank, ..
                                        } => *has_preceding_blank,
                                        Block::SnippetLine(_) | Block::DivMarker(_) => false,
                                        _ => true,
                                    };
                                    if should_add_blank && result.last().is_none_or(|s: &String| !s.is_empty()) {
                                        result.push(String::new());
                                    }
                                }
                            }
                        }
                    }

                    let reflowed_text = result.join(line_ending);

                    // Preserve trailing newline
                    let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
                        format!("{reflowed_text}{line_ending}")
                    } else {
                        reflowed_text
                    };

                    // Get the original text to compare
                    let original_text = &ctx.content[byte_range.clone()];

                    // Physical-line-length scan, shared by the Normalize-mode gate and its
                    // message. The list-item reflow preserves code blocks, HTML blocks,
                    // admonition headers, fence markers, semantic markers, and snippet/div
                    // markers verbatim; only paragraph content and admonition bodies are
                    // restructured. Only those lines drive the length warning, so that
                    // preserved-but-overlong content does not keep the paragraph-level
                    // warning alive when the reflow would not fix that line.
                    let should_count_for_length = |line_idx: usize| -> bool {
                        let line = lines[line_idx];
                        let trimmed = line.trim();
                        if trimmed.is_empty() || is_exempt_line(line, line_idx + 1) {
                            return false;
                        }
                        let info = &ctx.lines[line_idx];
                        if info.in_code_block || info.in_html_block {
                            return false;
                        }
                        if info.in_admonition && mkdocs_admonitions::is_admonition_start(line) {
                            return false;
                        }
                        if is_fence_marker(line) || is_semantic_line(line) {
                            return false;
                        }
                        if is_snippet_block_delimiter(line) {
                            return false;
                        }
                        if line.trim_start().starts_with(":::") {
                            return false;
                        }
                        true
                    };
                    let max_physical_length = (list_start..i)
                        .filter(|&idx| should_count_for_length(idx))
                        .map(|idx| self.calculate_effective_length(lines[idx]))
                        .max()
                        .unwrap_or(0);
                    // `line-length = 0` means "no limit", so no physical line can be
                    // "over"; the message below then describes a structural join rather
                    // than a length violation.
                    let any_paragraph_line_over =
                        !config.line_length.is_unlimited() && max_physical_length > config.line_length.get();

                    // Normalize mode reflows list-item prose just like paragraphs:
                    // joining continuation lines and re-wrapping to `line-length`.
                    // `prose_changed` is true only when the reflow alters the words or
                    // line breaks, not when it would merely re-indent continuation
                    // lines or trim trailing whitespace. Comparing the texts with each
                    // line's leading and trailing whitespace removed isolates "did the
                    // words/line breaks change" from "did the surrounding whitespace
                    // change". Continuation indentation is MD077's responsibility and
                    // trailing whitespace is MD009's; an MD013 warning for either would
                    // both duplicate those rules and resurface a persistent advisory on
                    // already-fitting items that users disable MD013 fixing to avoid.
                    let prose_changed = {
                        let stripped = |text: &str| text.lines().map(str::trim).collect::<Vec<_>>().join("\n");
                        stripped(original_text) != stripped(&replacement)
                    };
                    // Warn when the reflow rewraps prose (the normalize feature for
                    // list items), or when a physical line genuinely exceeds the limit
                    // and the reflow can change something (a true length violation,
                    // even if all that changes is the continuation indent). A line that
                    // is already optimal in both respects produces no warning.
                    let gate_ok = prose_changed || (any_paragraph_line_over && original_text != replacement);
                    if gate_ok {
                        // Generate an appropriate message based on why reflow is needed
                        let message = match config.reflow_mode {
                            ReflowMode::SentencePerLine => {
                                let num_sentences =
                                    split_into_sentences(&combined_content, Some(&defined_references)).len();
                                let num_lines = content_lines.len();
                                if num_lines == 1 {
                                    // Single line with multiple sentences
                                    format!("Line contains {num_sentences} sentences (one sentence per line required)")
                                } else {
                                    // Multiple lines - could be split sentences or mixed
                                    format!(
                                        "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
                                    )
                                }
                            }
                            ReflowMode::SemanticLineBreaks => {
                                let num_sentences =
                                    split_into_sentences(&combined_content, Some(&defined_references)).len();
                                format!("Paragraph should use semantic line breaks ({num_sentences} sentences)")
                            }
                            ReflowMode::Normalize => {
                                // When a physical line genuinely exceeds the limit, report
                                // it as a length violation. Otherwise the reflow is a
                                // structural normalization (joining/re-wrapping multi-line
                                // content that already fits), mirroring the paragraph path.
                                if any_paragraph_line_over {
                                    format!(
                                        "Line length {} exceeds {} characters",
                                        max_physical_length,
                                        config.line_length.get()
                                    )
                                } else {
                                    format!(
                                        "List item could be normalized to use line length of {} characters",
                                        config.line_length.get()
                                    )
                                }
                            }
                            ReflowMode::Default => {
                                // Report the actual longest non-exempt line, not the combined content
                                let max_length = (list_start..i)
                                    .filter(|&line_idx| {
                                        let line = lines[line_idx];
                                        let trimmed = line.trim();
                                        !trimmed.is_empty() && !is_exempt_line(line, line_idx + 1)
                                    })
                                    .map(|line_idx| self.calculate_effective_length(lines[line_idx]))
                                    .max()
                                    .unwrap_or(0);
                                format!(
                                    "Line length {} exceeds {} characters",
                                    max_length,
                                    config.line_length.get()
                                )
                            }
                        };

                        warnings.push(LintWarning {
                            rule_name: Some(self.name().to_string()),
                            message,
                            line: list_start + 1,
                            column: 1,
                            end_line: end_line + 1,
                            end_column: lines[end_line].chars().count() + 1,
                            severity: Severity::Warning,
                            fix: Some(crate::rule::Fix::new(byte_range, replacement)),
                        });
                    }
                }
                continue;
            }

            // Found start of a paragraph - collect all lines in it
            let paragraph_start = i;
            let mut paragraph_lines = vec![lines[i]];
            i += 1;

            while i < lines.len() {
                let next_line = lines[i];
                let next_line_num = i + 1;
                let next_trimmed = next_line.trim();

                // Stop at paragraph boundaries
                if next_trimmed.is_empty()
                    || ctx.line_info(next_line_num).is_some_and(|info| info.in_code_block)
                    || ctx.line_info(next_line_num).is_some_and(|info| info.in_front_matter)
                    || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_block)
                    || ctx.line_info(next_line_num).is_some_and(|info| info.in_html_comment)
                    || ctx.line_info(next_line_num).is_some_and(|info| info.in_esm_block)
                    || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_expression)
                    || ctx.line_info(next_line_num).is_some_and(|info| info.in_jsx_block)
                    || ctx.line_info(next_line_num).is_some_and(|info| info.in_mdx_comment)
                    || ctx
                        .line_info(next_line_num)
                        .is_some_and(super::super::lint_context::types::LineInfo::in_mkdocs_container)
                    || (next_line_num > 0
                        && next_line_num <= ctx.lines.len()
                        && ctx.lines[next_line_num - 1].blockquote.is_some())
                    || next_trimmed.starts_with('#')
                    // A setext heading ends the paragraph before it. Stopping on
                    // the text line also protects the underline, which can only
                    // follow it, so absorbing the pair and joining it into prose
                    // is unreachable from here. `is_horizontal_rule` below catches
                    // a `---` underline only by coincidence (3+ dashes are also a
                    // thematic break); `=` and short `-` runs have no such overlap.
                    || is_setext_heading_text_line(ctx, next_line_num)
                    || TableUtils::is_potential_table_row_with_flavor(next_line, ctx.flavor)
                    || is_list_item(next_trimmed)
                    || is_horizontal_rule(next_line)
                    || (next_trimmed.starts_with('[') && next_line.contains("]:"))
                    || is_template_directive_only(next_line)
                    || is_standalone_attr_list(next_line)
                    || is_snippet_block_delimiter(next_line)
                    || ctx.line_info(next_line_num).is_some_and(|info| info.is_div_marker)
                    || is_html_only_line(next_line)
                    || self.line_in_multiline_math_block(next_line_num, ctx)
                    || (!config.strict && is_standalone_link_or_image_line(ctx, next_line_num))
                {
                    break;
                }

                // Check if the previous line ends with a hard break (2+ spaces or backslash)
                if i > 0 && has_hard_break(lines[i - 1]) {
                    // Don't include lines after hard breaks in the same paragraph
                    break;
                }

                paragraph_lines.push(next_line);
                i += 1;
            }

            // Compute the common leading indent of all non-empty paragraph lines,
            // but only when those lines are structurally inside a list block.
            // Indented continuation lines that follow a nested list arrive here
            // with their structural indentation intact (e.g. 2 spaces for a
            // top-level list item). Stripping the indent before reflow and
            // re-applying it afterward prevents the fixer from moving those
            // lines to column 0.
            //
            // The list-block guard is essential: top-level paragraphs that happen
            // to start with spaces (insignificant in Markdown) must NOT have those
            // spaces preserved or injected by the fixer.
            let common_indent: String = if ctx.is_in_list_block(paragraph_start + 1) {
                let min_len = paragraph_lines
                    .iter()
                    .filter(|l| !l.trim().is_empty())
                    .map(|l| l.len() - l.trim_start().len())
                    .min()
                    .unwrap_or(0);
                paragraph_lines
                    .iter()
                    .find(|l| !l.trim().is_empty())
                    .map(|l| l[..min_len].to_string())
                    .unwrap_or_default()
            } else {
                String::new()
            };

            // Combine paragraph lines into a single string for processing.
            // This must be done BEFORE the needs_reflow check for sentence-per-line mode.
            let paragraph_text = if common_indent.is_empty() {
                paragraph_lines.join(" ")
            } else {
                paragraph_lines
                    .iter()
                    .map(|l| {
                        if l.starts_with(common_indent.as_str()) {
                            &l[common_indent.len()..]
                        } else {
                            l.trim_start()
                        }
                    })
                    .collect::<Vec<_>>()
                    .join(" ")
            };

            // Skip reflowing if this paragraph contains definition list items
            // Definition lists are multi-line structures that should not be joined
            let contains_definition_list = paragraph_lines
                .iter()
                .any(|line| crate::utils::is_definition_list_item(line));

            if contains_definition_list {
                // Don't reflow definition lists - skip this paragraph
                i = paragraph_start + paragraph_lines.len();
                continue;
            }

            // Skip reflowing if this paragraph contains MkDocs Snippets markers
            // Snippets blocks (-8<- ... -8<-) should be preserved exactly
            let contains_snippets = paragraph_lines.iter().any(|line| is_snippet_block_delimiter(line));

            if contains_snippets {
                // Don't reflow Snippets blocks - skip this paragraph
                i = paragraph_start + paragraph_lines.len();
                continue;
            }

            // Leave a line of a multi-line display-math block as it is, where
            // joining lines would corrupt the equation (see
            // `line_in_multiline_math_block`). Only the first line has to be
            // asked about: such a line ends the paragraph above it, so a
            // paragraph reaching here holds one only when it starts on one.
            //
            // Only that line is passed over, not the rest of what was collected
            // with it: prose written directly under the closing delimiter is an
            // ordinary paragraph and still reflows.
            if self.line_in_multiline_math_block(paragraph_start + 1, ctx) {
                i = paragraph_start + 1;
                continue;
            }

            // Check if this paragraph needs reflowing
            let needs_reflow = match config.reflow_mode {
                ReflowMode::Normalize => self.normalize_mode_needs_reflow(paragraph_lines.iter().copied(), config),
                ReflowMode::SentencePerLine => {
                    // In sentence-per-line mode, check if the JOINED paragraph has multiple sentences
                    // Note: we check the joined text because sentences can span multiple lines
                    let sentences = split_into_sentences(&paragraph_text, Some(&defined_references));

                    // Always reflow if multiple sentences on one line
                    if sentences.len() > 1 {
                        true
                    } else if paragraph_lines.len() > 1 {
                        // For single-sentence paragraphs spanning multiple lines:
                        // Reflow if they COULD fit on one line (respecting line-length constraint)
                        if config.line_length.is_unlimited() {
                            // No line-length constraint - always join single sentences
                            true
                        } else {
                            // Only join if it fits within line-length.
                            // paragraph_text has the common indent stripped, so add it
                            // back to get the true output length before comparing.
                            let effective_length =
                                self.calculate_effective_length(&paragraph_text) + common_indent.len();
                            effective_length <= config.line_length.get()
                        }
                    } else {
                        false
                    }
                }
                ReflowMode::SemanticLineBreaks => {
                    let sentences = split_into_sentences(&paragraph_text, Some(&defined_references));
                    // Reflow if multiple sentences, multiple lines, or any line exceeds limit
                    sentences.len() > 1
                        || paragraph_lines.len() > 1
                        || paragraph_lines
                            .iter()
                            .any(|line| self.calculate_effective_length(line) > config.line_length.get())
                }
                ReflowMode::Default => {
                    // In default mode, only reflow if lines exceed limit
                    paragraph_lines
                        .iter()
                        .any(|line| self.calculate_effective_length(line) > config.line_length.get())
                }
            };

            if needs_reflow {
                // Calculate byte range for this paragraph
                // Use whole_line_range for each line and combine
                let start_range = ctx.whole_line_byte_range(paragraph_start + 1);
                let end_line = paragraph_start + paragraph_lines.len() - 1;

                // For the last line, we want to preserve any trailing newline
                let end_range = if end_line == lines.len() - 1 && !ctx.content.ends_with('\n') {
                    // Last line without trailing newline - use line_text_range
                    ctx.line_text_byte_range(end_line + 1, 1, lines[end_line].len() + 1)
                } else {
                    // Not the last line or has trailing newline - use whole_line_range
                    ctx.whole_line_byte_range(end_line + 1)
                };

                let byte_range = start_range.start..end_range.end;

                // Check if the paragraph ends with a hard break and what type
                let hard_break_type = paragraph_lines.last().and_then(|line| {
                    let line = line.strip_suffix('\r').unwrap_or(line);
                    if line.ends_with('\\') {
                        Some("\\")
                    } else if line.ends_with("  ") {
                        Some("  ")
                    } else {
                        None
                    }
                });

                // Reflow the paragraph
                // When line_length = 0 (no limit), use a very large value for reflow
                let reflow_line_length = if config.line_length.is_unlimited() {
                    usize::MAX
                } else {
                    config.line_length.get().saturating_sub(common_indent.len()).max(1)
                };
                let reflow_options = Self::reflow_options(ctx, config, reflow_line_length);
                let mut reflowed = crate::utils::text_reflow::reflow_line(&paragraph_text, &reflow_options);

                // Re-apply the common indent to each non-empty reflowed line so
                // that the replacement preserves the original structural indentation.
                if !common_indent.is_empty() {
                    for line in &mut reflowed {
                        if !line.is_empty() {
                            *line = format!("{common_indent}{line}");
                        }
                    }
                }

                // If the original paragraph ended with a hard break, preserve it
                // Preserve the original hard break format (backslash or two spaces)
                if let Some(break_marker) = hard_break_type
                    && !reflowed.is_empty()
                {
                    let last_idx = reflowed.len() - 1;
                    if !has_hard_break(&reflowed[last_idx]) {
                        reflowed[last_idx].push_str(break_marker);
                    }
                }

                let reflowed_text = reflowed.join(line_ending);

                // Preserve trailing newline if the original paragraph had one
                let replacement = if end_line < lines.len() - 1 || ctx.content.ends_with('\n') {
                    format!("{reflowed_text}{line_ending}")
                } else {
                    reflowed_text
                };

                // Get the original text to compare
                let original_text = &ctx.content[byte_range.clone()];

                // Only generate a warning if the replacement is different from the original
                if original_text != replacement {
                    // Determine which line ranges and messages to report based on the reflow mode.
                    let warnings_to_report: Vec<(usize, usize, String)> = match config.reflow_mode {
                        ReflowMode::Default => {
                            // In default mode, report a warning for *every* line in the paragraph
                            // that exceeds the limit. Each warning will carry the same paragraph-level
                            // fix, making all of them auto-fixable.
                            paragraph_lines
                                .iter()
                                .enumerate()
                                .filter(|(_, line)| self.calculate_effective_length(line) > config.line_length.get())
                                .map(|(idx, _)| {
                                    let violating_line = paragraph_start + idx + 1;
                                    (
                                        violating_line,
                                        violating_line,
                                        format!("Line length exceeds {} characters", config.line_length.get()),
                                    )
                                })
                                .collect()
                        }
                        ReflowMode::Normalize => {
                            // In normalize mode, report the whole paragraph as needing normalization.
                            vec![(
                                paragraph_start + 1,
                                end_line + 1,
                                format!(
                                    "Paragraph could be normalized to use line length of {} characters",
                                    config.line_length.get()
                                ),
                            )]
                        }
                        ReflowMode::SentencePerLine => {
                            // In sentence-per-line mode, highlight the entire paragraph that needs reformatting.
                            let num_sentences = split_into_sentences(&paragraph_text, Some(&defined_references)).len();
                            let message = if paragraph_lines.len() == 1 {
                                // Single line with multiple sentences
                                format!("Line contains {num_sentences} sentences (one sentence per line required)")
                            } else {
                                // Multiple lines - could be split sentences or mixed
                                let num_lines = paragraph_lines.len();
                                format!(
                                    "Paragraph should have one sentence per line (found {num_sentences} sentences across {num_lines} lines)"
                                )
                            };
                            vec![(paragraph_start + 1, paragraph_start + paragraph_lines.len(), message)]
                        }
                        ReflowMode::SemanticLineBreaks => {
                            // In semantic-line-breaks mode, highlight the entire paragraph.
                            let num_sentences = split_into_sentences(&paragraph_text, Some(&defined_references)).len();
                            vec![(
                                paragraph_start + 1,
                                paragraph_start + paragraph_lines.len(),
                                format!("Paragraph should use semantic line breaks ({num_sentences} sentences)"),
                            )]
                        }
                    };

                    // Generate the actual lint warnings. All warnings for this paragraph
                    // share the same paragraph-level fix.
                    for (w_start, w_end, msg) in warnings_to_report {
                        warnings.push(LintWarning {
                            rule_name: Some(self.name().to_string()),
                            message: msg,
                            line: w_start,
                            column: 1,
                            end_line: w_end,
                            end_column: lines[w_end.saturating_sub(1)].chars().count() + 1,
                            severity: Severity::Warning,
                            fix: Some(crate::rule::Fix::new(byte_range.clone(), replacement.clone())),
                        });
                    }
                }
            }
        }

        warnings
    }

    /// Calculate string length based on the configured length mode
    fn calculate_string_length(&self, s: &str) -> usize {
        match self.config.length_mode {
            LengthMode::Chars => s.chars().count(),
            LengthMode::Visual => s.width(),
            LengthMode::Bytes => s.len(),
        }
    }

    /// Calculate effective line length
    ///
    /// Returns the actual display length of the line using the configured length mode.
    fn calculate_effective_length(&self, line: &str) -> usize {
        self.calculate_string_length(line)
    }

    /// Calculate line length with inline link/image URLs removed.
    ///
    /// For each inline link `[text](url)` or image `![alt](url)` on the line,
    /// computes the "savings" from removing the URL portion (keeping only `[text]`
    /// or `![alt]`). Returns `effective_length - total_savings`.
    ///
    /// Handles nested constructs (e.g., `[![img](url)](url)`) by only counting the
    /// outermost construct to avoid double-counting.
    fn length_without_inline_link_urls(
        &self,
        effective_length: usize,
        line_number: usize,
        ctx: &crate::lint_context::LintContext,
    ) -> usize {
        let line_range = ctx.line_content_byte_range(line_number);
        let line_byte_end = line_range.end;

        // Collect inline links/images on this line: (byte_offset, byte_end, text_only_display_len)
        let mut constructs: Vec<(usize, usize, usize)> = Vec::new();

        for link in ctx.links_on_line(line_number) {
            if link.is_reference {
                continue;
            }
            if !matches!(link.link_type, LinkType::Inline) {
                continue;
            }
            if link.byte_end > line_byte_end {
                continue;
            }
            let text_only_len = 2 + self.calculate_string_length(&link.text);
            constructs.push((link.byte_offset, link.byte_end, text_only_len));
        }

        for image in ctx.images_on_line(line_number) {
            if image.is_reference {
                continue;
            }
            if !matches!(image.link_type, LinkType::Inline) {
                continue;
            }
            if image.byte_end > line_byte_end {
                continue;
            }
            let text_only_len = 3 + self.calculate_string_length(&image.alt_text);
            constructs.push((image.byte_offset, image.byte_end, text_only_len));
        }

        if constructs.is_empty() {
            return effective_length;
        }

        // Sort by byte offset to handle overlapping/nested constructs
        constructs.sort_by_key(|&(start, _, _)| start);

        let mut total_savings: usize = 0;
        let mut last_end: usize = 0;

        for (start, end, text_only_len) in &constructs {
            // Skip constructs nested inside a previously counted one
            if *start < last_end {
                continue;
            }
            // Full construct length in configured length mode
            let full_source = &ctx.content[*start..*end];
            let full_len = self.calculate_string_length(full_source);
            total_savings += full_len.saturating_sub(*text_only_len);
            last_end = *end;
        }

        effective_length.saturating_sub(total_savings)
    }
}