rumdl_lib/rules/

md073_toc_validation.rs

//! MD073: Table of Contents validation rule
//!
//! Validates that TOC sections match the actual document headings.

use crate::lint_context::LintContext;
use crate::rule::{Fix, LintError, LintResult, LintWarning, Rule, RuleCategory, Severity};
use crate::utils::anchor_styles::AnchorStyle;
use crate::utils::header_id_utils::{extract_html_anchor_ids, is_backslash_escaped};
use percent_encoding::percent_decode_str;
use regex::Regex;
use std::borrow::Cow;
use std::collections::HashMap;
use std::sync::LazyLock;

/// Regex for TOC start marker: `<!-- toc -->` with optional whitespace variations
static TOC_START_MARKER: LazyLock<Regex> = LazyLock::new(|| Regex::new(r"(?i)<!--\s*toc\s*-->").unwrap());

/// Regex for TOC stop marker: `<!-- tocstop -->` or `<!-- /toc -->`
static TOC_STOP_MARKER: LazyLock<Regex> = LazyLock::new(|| Regex::new(r"(?i)<!--\s*(?:tocstop|/toc)\s*-->").unwrap());

/// Regex for extracting TOC entries: `- [text](#anchor)` or `* [text](#anchor)`
/// with optional leading whitespace for nested items
/// Handles nested brackets like `[`check [PATHS...]`](#check-paths)`
static TOC_ENTRY_PATTERN: LazyLock<Regex> =
    LazyLock::new(|| Regex::new(r"^(\s*)[-*]\s+\[([^\[\]]*(?:\[[^\[\]]*\][^\[\]]*)*)\]\(#([^)]+)\)").unwrap());

/// Represents a detected TOC region in the document
#[derive(Debug, Clone)]
struct TocRegion {
    /// 1-indexed start line of the TOC content (after the marker)
    start_line: usize,
    /// 1-indexed end line of the TOC content (before the stop marker)
    end_line: usize,
    /// Byte offset where TOC content starts
    content_start: usize,
    /// Byte offset where TOC content ends
    content_end: usize,
}

/// A parsed TOC entry from the existing TOC
#[derive(Debug, Clone)]
struct TocEntry {
    /// Display text of the link
    text: String,
    /// Anchor/fragment (without #)
    anchor: String,
    /// Number of leading whitespace characters (for indentation checking)
    indent_spaces: usize,
}

/// An expected TOC entry generated from document headings
#[derive(Debug, Clone)]
struct ExpectedTocEntry {
    /// 1-indexed line number of the heading's first line
    heading_line: usize,
    /// Heading level (1-6)
    level: u8,
    /// Heading text (for display)
    text: String,
    /// The fragment a generated entry links to
    anchor: String,
    /// Other fragments that also reach the heading. A heading with its own `<a id>`
    /// keeps the slug generated from its text, so a TOC written against either is
    /// right.
    aliases: Vec<String>,
}

impl ExpectedTocEntry {
    fn is_reached_by(&self, anchor: &str) -> bool {
        self.anchor == anchor || self.aliases.iter().any(|alias| alias == anchor)
    }
}

/// Types of mismatches between actual and expected TOC
#[derive(Debug)]
enum TocMismatch {
    /// Entry exists in TOC but heading doesn't exist
    StaleEntry { entry: TocEntry },
    /// Heading exists but no TOC entry for it
    MissingEntry { expected: ExpectedTocEntry },
    /// TOC entry text doesn't match heading text
    TextMismatch {
        entry: TocEntry,
        expected: ExpectedTocEntry,
    },
    /// TOC entries are in wrong order
    OrderMismatch { entry: TocEntry, expected_position: usize },
    /// TOC entry has wrong indentation level
    IndentationMismatch {
        entry: TocEntry,
        actual_indent: usize,
        expected_indent: usize,
    },
}

/// Regex patterns used by `strip_links_and_images`.
static MARKDOWN_LINK: LazyLock<Regex> = LazyLock::new(|| Regex::new(r"\[([^\]]+)\]\([^)]+\)").unwrap());
static MARKDOWN_REF_LINK: LazyLock<Regex> = LazyLock::new(|| Regex::new(r"\[([^\]]+)\]\[[^\]]*\]").unwrap());
static MARKDOWN_IMAGE: LazyLock<Regex> = LazyLock::new(|| Regex::new(r"!\[([^\]]*)\]\([^)]+\)").unwrap());

/// `anchor` written as the destination of a TOC link.
///
/// A bare link destination cannot hold whitespace, control characters or an
/// unbalanced parenthesis, and a `%` would read as the start of an escape, so
/// those ASCII characters are percent-encoded. Everything else is written as it
/// is, so a slug generated from heading text, non-ASCII letters included, is
/// unchanged.
fn encode_fragment(anchor: &str) -> Cow<'_, str> {
    fn needs_encoding(c: char) -> bool {
        c.is_ascii() && (c <= ' ' || c == '\x7f' || matches!(c, '(' | ')' | '%'))
    }

    if !anchor.contains(needs_encoding) {
        return Cow::Borrowed(anchor);
    }
    let mut encoded = String::with_capacity(anchor.len() + 6);
    for c in anchor.chars() {
        if needs_encoding(c) {
            encoded.push_str(&format!("%{:02X}", c as u32));
        } else {
            encoded.push(c);
        }
    }
    Cow::Owned(encoded)
}

/// The fragment a written TOC entry reaches, its percent-encoding undone, as a
/// browser decodes a fragment before matching it against ids. An escape sequence
/// that does not decode to UTF-8 leaves the fragment as written.
fn decode_fragment(anchor: &str) -> Cow<'_, str> {
    if !anchor.contains('%') {
        return Cow::Borrowed(anchor);
    }
    percent_decode_str(anchor)
        .decode_utf8()
        .unwrap_or(Cow::Borrowed(anchor))
}

/// Extract code-span byte ranges from `text` using the CommonMark rule:
/// a run of N backticks opens a span closed by exactly N backticks.
/// Returns a sorted list of `(start, end)` byte offsets that are inside code spans
/// (including the backtick delimiters themselves).
fn code_span_ranges(text: &str) -> Vec<(usize, usize)> {
    let chars: Vec<char> = text.chars().collect();
    let len = chars.len();
    let mut ranges = Vec::new();
    let mut i = 0;

    while i < len {
        if chars[i] == '`' {
            let span_start = i;
            while i < len && chars[i] == '`' {
                i += 1;
            }
            let n = i - span_start;

            // Search for the matching closing sequence of exactly n backticks
            let mut j = i;
            let mut found = false;
            while j < len {
                if chars[j] == '`' {
                    let close_start = j;
                    while j < len && chars[j] == '`' {
                        j += 1;
                    }
                    if j - close_start == n {
                        // Convert char indices to byte offsets
                        let byte_start: usize = text.char_indices().nth(span_start).map_or(0, |(b, _)| b);
                        let byte_end: usize = text.char_indices().nth(j).map_or(text.len(), |(b, _)| b);
                        ranges.push((byte_start, byte_end));
                        i = j;
                        found = true;
                        break;
                    }
                } else {
                    j += 1;
                }
            }
            if !found {
                // No matching close; skip past the opening backticks
                i = span_start + n;
            }
        } else {
            i += 1;
        }
    }

    ranges
}

/// Strip only links and images from `text`, preserving all other inline
/// formatting (code spans, bold, italic, etc.).
///
/// Links and images cannot appear inside a Markdown link label `[...]`, so they
/// must be removed when building TOC display text. Code spans and emphasis are
/// valid inside link labels and should be kept so the TOC entry faithfully
/// reflects the heading's visual appearance.
///
/// Code-span contents are protected: link-like syntax such as `[foo](bar)` that
/// appears inside backticks is left untouched.
///
/// Examples:
/// - `` `my header` `` → `` `my header` `` (code ticks preserved)
/// - `[terminal](url)` → `terminal` (link stripped)
/// - `![alt](img.png)` → `alt` (image stripped)
/// - `**bold**` → `**bold**` (emphasis preserved)
/// - `` `[foo](bar)` `` → `` `[foo](bar)` `` (link inside code span preserved)
/// - `Tool: [terminal](url)` → `Tool: terminal`
fn strip_links_and_images(text: &str) -> String {
    // Collect code-span byte ranges so we can protect their contents from
    // the link/image regex substitutions.
    let protected = code_span_ranges(text);

    // If there are no code spans the fast path avoids all the extra work.
    if protected.is_empty() {
        let mut result = text.to_string();
        result = MARKDOWN_IMAGE.replace_all(&result, "$1").to_string();
        result = MARKDOWN_LINK.replace_all(&result, "$1").to_string();
        result = MARKDOWN_REF_LINK.replace_all(&result, "$1").to_string();
        return result;
    }

    // Replace each code span with a unique placeholder that cannot be matched
    // by the link/image regexes, apply the regexes, then restore the originals.
    let mut placeholders: Vec<(&str, String)> = Vec::with_capacity(protected.len());
    let mut masked = text.to_string();
    // Process spans in reverse order so byte offsets remain valid after each replacement.
    for (i, &(start, end)) in protected.iter().enumerate().rev() {
        // Placeholder: a string containing no `[`, `]`, `(`, `)`, `!` characters.
        let placeholder = format!("\x00CODESPAN{i}\x00");
        let original = &text[start..end];
        placeholders.push((original, placeholder.clone()));
        masked.replace_range(start..end, &placeholder);
    }

    // Apply link/image stripping to the masked string
    masked = MARKDOWN_IMAGE.replace_all(&masked, "$1").to_string();
    masked = MARKDOWN_LINK.replace_all(&masked, "$1").to_string();
    masked = MARKDOWN_REF_LINK.replace_all(&masked, "$1").to_string();

    // Restore the original code-span text
    for (original, placeholder) in &placeholders {
        masked = masked.replace(placeholder.as_str(), original);
    }

    masked
}

/// MD073: Table of Contents Validation
///
/// This rule validates that TOC sections match the actual document headings.
/// It detects TOC regions via markers (`<!-- toc -->...<!-- tocstop -->`).
///
/// To opt into TOC validation, add markers to your document:
/// ```markdown
/// <!-- toc -->
/// - [Section](#section)
/// <!-- tocstop -->
/// ```
///
/// ## Configuration
///
/// ```toml
/// [MD073]
/// # Enable the rule (opt-in, disabled by default)
/// enabled = true
/// # Minimum heading level to include (default: 2)
/// min-level = 2
/// # Maximum heading level to include (default: 4)
/// max-level = 4
/// # Whether TOC order must match document order (default: true)
/// enforce-order = true
/// # Indent size per nesting level (default: from MD007 config, or 2)
/// indent = 2
/// ```
#[derive(Clone)]
pub struct MD073TocValidation {
    /// Whether this rule is enabled (default: false - opt-in rule)
    enabled: bool,
    /// Minimum heading level to include
    min_level: u8,
    /// Maximum heading level to include
    max_level: u8,
    /// Whether to enforce order matching
    enforce_order: bool,
    /// Indent size per nesting level (reads from MD007 config by default)
    pub indent: usize,
}

impl Default for MD073TocValidation {
    fn default() -> Self {
        Self {
            enabled: false, // Disabled by default - opt-in rule
            min_level: 2,
            max_level: 4,
            enforce_order: true,
            indent: 2, // Default indent, can be overridden by MD007 config
        }
    }
}

impl std::fmt::Debug for MD073TocValidation {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("MD073TocValidation")
            .field("enabled", &self.enabled)
            .field("min_level", &self.min_level)
            .field("max_level", &self.max_level)
            .field("enforce_order", &self.enforce_order)
            .field("indent", &self.indent)
            .finish()
    }
}

impl MD073TocValidation {
    /// Create a new rule with default settings
    pub fn new() -> Self {
        Self::default()
    }

    /// Whether `content` contains a TOC marker that is Markdown syntax rather
    /// than literal text. A marker inside an inline code span is code, and one
    /// whose `<` is backslash-escaped renders as the characters themselves.
    fn has_active_marker(ctx: &LintContext, content: &str, line_byte_offset: usize, marker: &Regex) -> bool {
        marker.find_iter(content).any(|matched| {
            !ctx.is_in_code_span_byte(line_byte_offset + matched.start())
                && !is_backslash_escaped(content, matched.start())
        })
    }

    /// Detect TOC region using markers
    fn detect_by_markers(&self, ctx: &LintContext) -> Option<TocRegion> {
        let mut start_line = None;
        let mut start_byte = None;

        for (idx, line_info) in ctx.lines.iter().enumerate() {
            let line_num = idx + 1;
            let content = line_info.content(ctx.content);

            // Skip if in code block or front matter
            if line_info.in_code_block || line_info.in_front_matter {
                continue;
            }

            // Look for start marker or stop marker
            if let (Some(s_line), Some(s_byte)) = (start_line, start_byte) {
                // We have a start, now look for stop marker
                if Self::has_active_marker(ctx, content, line_info.byte_offset, &TOC_STOP_MARKER) {
                    let end_line = line_num - 1;
                    let content_end = line_info.byte_offset;

                    // Handle case where there's no content between markers
                    if end_line < s_line {
                        return Some(TocRegion {
                            start_line: s_line,
                            end_line: s_line,
                            content_start: s_byte,
                            content_end: s_byte,
                        });
                    }

                    return Some(TocRegion {
                        start_line: s_line,
                        end_line,
                        content_start: s_byte,
                        content_end,
                    });
                }
            } else if Self::has_active_marker(ctx, content, line_info.byte_offset, &TOC_START_MARKER) {
                // TOC content starts on the next line
                if idx + 1 < ctx.lines.len() {
                    start_line = Some(line_num + 1);
                    start_byte = Some(ctx.lines[idx + 1].byte_offset);
                }
            }
        }

        None
    }

    /// Detect TOC region using markers
    fn detect_toc_region(&self, ctx: &LintContext) -> Option<TocRegion> {
        self.detect_by_markers(ctx)
    }

    /// Remove `depth` blockquote markers, each with at most one following space,
    /// leaving any further indentation in place.
    fn strip_quote_markers(line: &str, depth: usize) -> &str {
        let mut rest = line;
        for _ in 0..depth {
            let Some(after) = rest.trim_start().strip_prefix('>') else {
                break;
            };
            rest = after.strip_prefix(' ').unwrap_or(after);
        }
        rest
    }

    /// Extract TOC entries from the detected region
    fn extract_toc_entries(&self, ctx: &LintContext, region: &TocRegion) -> Vec<TocEntry> {
        let mut entries = Vec::new();

        for idx in (region.start_line - 1)..region.end_line.min(ctx.lines.len()) {
            let line_info = &ctx.lines[idx];
            // A TOC inside a blockquote is matched on the text after the quote
            // markers, keeping the indentation of nested entries.
            let content = line_info.blockquote.as_ref().map_or_else(
                || line_info.content(ctx.content),
                |quote| Self::strip_quote_markers(line_info.content(ctx.content), quote.nesting_level),
            );

            if let Some(caps) = TOC_ENTRY_PATTERN.captures(content) {
                let indent_spaces = caps.get(1).map_or(0, |m| m.as_str().len());
                let text = caps.get(2).map_or("", |m| m.as_str()).to_string();
                let anchor = caps.get(3).map_or("", |m| m.as_str()).to_string();

                entries.push(TocEntry {
                    text,
                    anchor,
                    indent_spaces,
                });
            }
        }

        entries
    }

    /// Build expected TOC entries from document headings
    fn build_expected_toc(&self, ctx: &LintContext, toc_region: &TocRegion) -> Vec<ExpectedTocEntry> {
        let mut entries = Vec::new();
        let mut fragment_counts: HashMap<String, usize> = HashMap::new();

        // Every heading takes a number from the renderer, in document order, whether
        // or not it appears in the TOC, so the duplicate numbering runs over all of
        // them and only the entries are filtered.
        for parsed in ctx.headings() {
            let line_num = parsed.line_num;
            let line_info = parsed.line_info;
            let heading = parsed.heading;

            // Skip code blocks, front matter, HTML blocks
            if line_info.in_code_block || line_info.in_front_matter || line_info.in_html_block {
                continue;
            }

            // A custom ID ({#id}) replaces the generated slug. An HTML anchor
            // element beside the heading adds a target either way: the
            // renderer keeps the element's own id, and without a custom ID it
            // still generates the slug from the text, which still counts
            // towards the numbering of later duplicates.
            let (anchor, aliases) = match &heading.custom_id {
                Some(custom_id) => (
                    Self::deduplicate_fragment(&mut fragment_counts, custom_id),
                    extract_html_anchor_ids(&heading.raw_text),
                ),
                None => {
                    let slug = AnchorStyle::GitHub.generate_fragment(&heading.slug_text);
                    let generated = Self::deduplicate_fragment(&mut fragment_counts, &slug);
                    let mut explicit = extract_html_anchor_ids(&heading.raw_text);
                    match explicit.first().cloned() {
                        Some(preferred) => {
                            explicit.remove(0);
                            explicit.push(generated);
                            (preferred, explicit)
                        }
                        // A heading whose text slugs to nothing (an image, an
                        // emoji) gets no anchor from GitHub, only a number on a
                        // repeat, so without an element of its own no TOC entry
                        // can reach it. It still counts towards the numbering.
                        None if slug.is_empty() => continue,
                        None => (generated, Vec::new()),
                    }
                }
            };

            // Headings before/within the TOC region are not listed, and the
            // configured level range limits the rest.
            if line_num <= toc_region.end_line || heading.level < self.min_level || heading.level > self.max_level {
                continue;
            }

            entries.push(ExpectedTocEntry {
                // A Setext heading is recorded on the last line of its text,
                // and a reader looks for it where it starts.
                heading_line: parsed.first_line_num(),
                level: heading.level,
                text: heading.text.clone(),
                anchor,
                aliases,
            });
        }

        entries
    }

    /// The fragment a renderer gives a heading whose base fragment has already
    /// been seen `n` times: `base`, then `base-1`, `base-2`, ...
    fn deduplicate_fragment(fragment_counts: &mut HashMap<String, usize>, base: &str) -> String {
        match fragment_counts.get_mut(base) {
            Some(count) => {
                let suffix = *count;
                *count += 1;
                format!("{base}-{suffix}")
            }
            None => {
                fragment_counts.insert(base.to_string(), 1);
                base.to_string()
            }
        }
    }

    /// Pair each actual TOC entry with the first still-unclaimed heading its
    /// anchor reaches. `None` marks an entry that reaches no heading.
    fn match_entries(actual: &[TocEntry], expected: &[ExpectedTocEntry]) -> Vec<Option<usize>> {
        let mut claimed = vec![false; expected.len()];
        actual
            .iter()
            .map(|entry| {
                let anchor = decode_fragment(&entry.anchor);
                let found = expected
                    .iter()
                    .enumerate()
                    .position(|(idx, exp)| !claimed[idx] && exp.is_reached_by(&anchor));
                if let Some(idx) = found {
                    claimed[idx] = true;
                }
                found
            })
            .collect()
    }

    /// Compare actual TOC entries against expected and find mismatches.
    ///
    /// `matching` pairs each actual entry with the heading it reaches, as
    /// computed by [`Self::match_entries`].
    fn validate_toc(
        &self,
        actual: &[TocEntry],
        expected: &[ExpectedTocEntry],
        matching: &[Option<usize>],
    ) -> Vec<TocMismatch> {
        let mut mismatches = Vec::new();
        let pairs = || {
            actual
                .iter()
                .zip(matching)
                .enumerate()
                .filter_map(|(actual_idx, (entry, matched))| matched.map(|exp_idx| (actual_idx, entry, exp_idx)))
        };

        // Stale entries reach no heading.
        for (entry, matched) in actual.iter().zip(matching) {
            if matched.is_none() {
                mismatches.push(TocMismatch::StaleEntry { entry: entry.clone() });
            }
        }

        // Missing entries are headings no TOC entry reaches.
        for (exp_idx, exp) in expected.iter().enumerate() {
            if !matching.contains(&Some(exp_idx)) {
                mismatches.push(TocMismatch::MissingEntry { expected: exp.clone() });
            }
        }

        // Check for text mismatches. Compare with the same normalization used in
        // generate_toc: strip only links and images, preserve code spans and emphasis.
        // This ensures a correct user-written TOC entry like `` [`my header`](#anchor) ``
        // is not flagged against a heading `` `my header` ``.
        let mut text_mismatched = vec![false; actual.len()];
        for (actual_idx, entry, exp_idx) in pairs() {
            let exp = &expected[exp_idx];
            let actual_normalized = strip_links_and_images(entry.text.trim());
            let expected_normalized = strip_links_and_images(exp.text.trim());
            if actual_normalized != expected_normalized {
                text_mismatched[actual_idx] = true;
                mismatches.push(TocMismatch::TextMismatch {
                    entry: entry.clone(),
                    expected: exp.clone(),
                });
            }
        }

        // Check for indentation mismatches
        // Expected indentation is indent spaces per level difference from base level
        if !expected.is_empty() {
            let base_level = expected.iter().map(|e| e.level).min().unwrap_or(2);

            for (actual_idx, entry, exp_idx) in pairs() {
                let level_diff = expected[exp_idx].level.saturating_sub(base_level) as usize;
                let expected_indent = level_diff * self.indent;

                // An entry already reported for its text is not reported again
                if entry.indent_spaces != expected_indent && !text_mismatched[actual_idx] {
                    mismatches.push(TocMismatch::IndentationMismatch {
                        entry: entry.clone(),
                        actual_indent: entry.indent_spaces,
                        expected_indent,
                    });
                }
            }
        }

        // Check order if enforce_order is enabled: walk the headings alongside the
        // matched entries, and an entry whose heading lies behind the walk is out
        // of order.
        if self.enforce_order && !actual.is_empty() && !expected.is_empty() {
            let mut expected_idx = 0;
            for (actual_idx, entry, exp_idx) in pairs() {
                if exp_idx >= expected_idx {
                    expected_idx = exp_idx + 1;
                } else if !text_mismatched[actual_idx] {
                    mismatches.push(TocMismatch::OrderMismatch {
                        entry: entry.clone(),
                        expected_position: exp_idx + 1,
                    });
                }
            }
        }

        mismatches
    }

    /// Expected entries as they should be regenerated: a TOC entry that already
    /// reaches its heading keeps the anchor it was written with. The anchor is
    /// stored decoded, as every expected anchor is, and encoded again on output.
    fn keep_reached_anchors(
        actual: &[TocEntry],
        expected: &[ExpectedTocEntry],
        matching: &[Option<usize>],
    ) -> Vec<ExpectedTocEntry> {
        let mut regenerated = expected.to_vec();
        for (entry, exp_idx) in actual
            .iter()
            .zip(matching)
            .filter_map(|(entry, m)| m.map(|idx| (entry, idx)))
        {
            regenerated[exp_idx].anchor = decode_fragment(&entry.anchor).into_owned();
        }
        regenerated
    }

    /// Generate a new TOC from expected entries (always uses nested indentation)
    fn generate_toc(&self, expected: &[ExpectedTocEntry]) -> String {
        if expected.is_empty() {
            return String::new();
        }

        let mut result = String::new();
        let base_level = expected.iter().map(|e| e.level).min().unwrap_or(2);
        let indent_str = " ".repeat(self.indent);

        for entry in expected {
            let level_diff = entry.level.saturating_sub(base_level) as usize;
            let indent = indent_str.repeat(level_diff);

            // Build display text: strip only links and images (which would create invalid
            // nested-link syntax inside `[...]`), but preserve code spans and emphasis so
            // the TOC entry reflects the heading's visual appearance.
            let display_text = strip_links_and_images(&entry.text);
            let fragment = encode_fragment(&entry.anchor);
            result.push_str(&format!("{indent}- [{display_text}](#{fragment})\n"));
        }

        result
    }
}

impl Rule for MD073TocValidation {
    fn name(&self) -> &'static str {
        "MD073"
    }

    fn description(&self) -> &'static str {
        "Table of Contents should match document headings"
    }

    fn should_skip(&self, ctx: &LintContext) -> bool {
        // Quick check: skip if no TOC markers. detect_toc_region() is
        // case-insensitive, so use a case-insensitive containment check here
        // to avoid skipping fix() on documents with uppercase markers like
        // `<!-- TOC -->`.
        let lower = ctx.content.to_ascii_lowercase();
        !(lower.contains("<!-- toc") || lower.contains("<!--toc"))
    }

    fn check(&self, ctx: &LintContext) -> LintResult {
        let mut warnings = Vec::new();

        // Detect TOC region
        let Some(region) = self.detect_toc_region(ctx) else {
            // No TOC found - nothing to validate
            return Ok(warnings);
        };

        // Extract actual TOC entries
        let actual_entries = self.extract_toc_entries(ctx, &region);

        // Build expected TOC from headings
        let expected_entries = self.build_expected_toc(ctx, &region);

        // If no expected entries and no actual entries, nothing to validate
        if expected_entries.is_empty() && actual_entries.is_empty() {
            return Ok(warnings);
        }

        // Validate
        let matching = Self::match_entries(&actual_entries, &expected_entries);
        let mismatches = self.validate_toc(&actual_entries, &expected_entries, &matching);

        if !mismatches.is_empty() {
            // Generate a single warning at the TOC region with details
            let mut details = Vec::new();

            for mismatch in &mismatches {
                match mismatch {
                    TocMismatch::StaleEntry { entry } => {
                        details.push(format!("Stale entry: '{}' (heading no longer exists)", entry.text));
                    }
                    TocMismatch::MissingEntry { expected } => {
                        details.push(format!(
                            "Missing entry: '{}' (line {})",
                            expected.text, expected.heading_line
                        ));
                    }
                    TocMismatch::TextMismatch { entry, expected } => {
                        details.push(format!(
                            "Text mismatch: TOC has '{}', heading is '{}'",
                            entry.text, expected.text
                        ));
                    }
                    TocMismatch::OrderMismatch {
                        entry,
                        expected_position,
                    } => {
                        details.push(format!(
                            "Order mismatch: '{}' should be at position {}",
                            entry.text, expected_position
                        ));
                    }
                    TocMismatch::IndentationMismatch {
                        entry,
                        actual_indent,
                        expected_indent,
                        ..
                    } => {
                        details.push(format!(
                            "Indentation mismatch: '{}' has {} spaces, expected {} spaces",
                            entry.text, actual_indent, expected_indent
                        ));
                    }
                }
            }

            let message = format!(
                "Table of Contents does not match document headings: {}",
                details.join("; ")
            );

            // Generate fix: replace entire TOC content
            let regenerated = Self::keep_reached_anchors(&actual_entries, &expected_entries, &matching);
            let new_toc = self.generate_toc(&regenerated);
            let fix_range = region.content_start..region.content_end;
            // Generated entries carry no container prefix, so a TOC whose markers
            // sit in a blockquote is reported but not rewritten.
            let (stop_marker_line, _) = ctx.offset_to_line_col(region.content_end);
            let in_blockquote = [region.start_line - 1, stop_marker_line]
                .into_iter()
                .any(|line| ctx.line_info(line).is_some_and(|info| info.blockquote.is_some()));

            warnings.push(LintWarning {
                rule_name: Some(self.name().to_string()),
                message,
                line: region.start_line,
                column: 1,
                end_line: region.end_line,
                end_column: 1,
                severity: Severity::Warning,
                fix: (!in_blockquote).then(|| Fix::new(fix_range, new_toc)),
            });
        }

        Ok(warnings)
    }

    fn fixes_from_diagnostics(&self) -> bool {
        true
    }

    fn fix(&self, ctx: &LintContext) -> Result<String, LintError> {
        if self.should_skip(ctx) {
            return Ok(ctx.content.to_string());
        }
        let warnings = self.check(ctx)?;
        if warnings.is_empty() {
            return Ok(ctx.content.to_string());
        }
        let warnings =
            crate::utils::fix_utils::filter_warnings_by_inline_config(warnings, ctx.inline_config(), self.name());
        crate::utils::fix_utils::apply_warning_fixes(ctx.content, &warnings).map_err(LintError::InvalidInput)
    }

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

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

    fn default_config_section(&self) -> Option<(String, toml::Value)> {
        let value: toml::Value = toml::from_str(
            r#"
# Whether this rule is enabled (opt-in, disabled by default)
enabled = false
# Minimum heading level to include
min-level = 2
# Maximum heading level to include
max-level = 4
# Whether TOC order must match document order
enforce-order = true
# Indentation per nesting level (defaults to MD007's indent value)
indent = 2
"#,
        )
        .ok()?;
        Some(("MD073".to_string(), value))
    }

    fn from_config(config: &crate::config::Config) -> Box<dyn Rule>
    where
        Self: Sized,
    {
        let mut rule = MD073TocValidation::default();
        let mut indent_from_md073 = false;

        if let Some(rule_config) = config.rules.get("MD073") {
            // Parse enabled (opt-in rule, defaults to false)
            if let Some(enabled) = rule_config.values.get("enabled").and_then(toml::Value::as_bool) {
                rule.enabled = enabled;
            }

            // Parse min-level
            if let Some(min_level) = rule_config.values.get("min-level").and_then(toml::Value::as_integer) {
                rule.min_level = (min_level.clamp(1, 6)) as u8;
            }

            // Parse max-level
            if let Some(max_level) = rule_config.values.get("max-level").and_then(toml::Value::as_integer) {
                rule.max_level = (max_level.clamp(1, 6)) as u8;
            }

            // Parse enforce-order
            if let Some(enforce_order) = rule_config.values.get("enforce-order").and_then(toml::Value::as_bool) {
                rule.enforce_order = enforce_order;
            }

            // Parse indent (MD073-specific override)
            if let Some(indent) = rule_config.values.get("indent").and_then(toml::Value::as_integer) {
                rule.indent = (indent.clamp(1, 8)) as usize;
                indent_from_md073 = true;
            }
        }

        // If indent not explicitly set in MD073, read from MD007 config
        if !indent_from_md073
            && let Some(md007_config) = config.rules.get("MD007")
            && let Some(indent) = md007_config.values.get("indent").and_then(toml::Value::as_integer)
        {
            rule.indent = (indent.clamp(1, 8)) as usize;
        }

        Box::new(rule)
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::config::MarkdownFlavor;
    use regex::Regex;
    use std::sync::LazyLock;

    // ---- Test-only helpers for stripping all inline formatting ----
    // These are not used in production code; they exist only to test
    // the individual stripping primitives in isolation.

    /// Strip code spans from text, handling multi-backtick spans per CommonMark spec.
    fn strip_code_spans(text: &str) -> String {
        let chars: Vec<char> = text.chars().collect();
        let len = chars.len();
        let mut result = String::with_capacity(text.len());
        let mut i = 0;

        while i < len {
            if chars[i] == '`' {
                let open_start = i;
                while i < len && chars[i] == '`' {
                    i += 1;
                }
                let backtick_count = i - open_start;

                let content_start = i;
                let mut found_close = false;
                while i < len {
                    if chars[i] == '`' {
                        let close_start = i;
                        while i < len && chars[i] == '`' {
                            i += 1;
                        }
                        if i - close_start == backtick_count {
                            let content: String = chars[content_start..close_start].iter().collect();
                            let stripped = if content.starts_with(' ') && content.ends_with(' ') && content.len() > 1 {
                                content[1..content.len() - 1].to_string()
                            } else {
                                content
                            };
                            result.push_str(&stripped);
                            found_close = true;
                            break;
                        }
                    } else {
                        i += 1;
                    }
                }
                if !found_close {
                    for _ in 0..backtick_count {
                        result.push('`');
                    }
                    let remaining: String = chars[content_start..].iter().collect();
                    result.push_str(&remaining);
                    break;
                }
            } else {
                result.push(chars[i]);
                i += 1;
            }
        }

        result
    }

    static TEST_BOLD_ASTERISK: LazyLock<Regex> = LazyLock::new(|| Regex::new(r"\*\*([^*]+)\*\*").unwrap());
    static TEST_BOLD_UNDERSCORE: LazyLock<Regex> = LazyLock::new(|| Regex::new(r"__([^_]+)__").unwrap());
    static TEST_ITALIC_ASTERISK: LazyLock<Regex> = LazyLock::new(|| Regex::new(r"\*([^*]+)\*").unwrap());
    static TEST_ITALIC_UNDERSCORE: LazyLock<Regex> =
        LazyLock::new(|| Regex::new(r"(^|[^a-zA-Z0-9])_([^_]+)_([^a-zA-Z0-9]|$)").unwrap());

    /// Strip all inline markdown formatting from text, reducing it to plain text.
    /// Builds on `strip_links_and_images` and additionally removes code spans,
    /// bold, and italic markers. Used in tests only.
    fn strip_markdown_formatting(text: &str) -> String {
        let mut result = strip_links_and_images(text);
        result = strip_code_spans(&result);
        result = TEST_BOLD_ASTERISK.replace_all(&result, "$1").to_string();
        result = TEST_BOLD_UNDERSCORE.replace_all(&result, "$1").to_string();
        result = TEST_ITALIC_ASTERISK.replace_all(&result, "$1").to_string();
        result = TEST_ITALIC_UNDERSCORE.replace_all(&result, "$1$2$3").to_string();
        result
    }

    fn create_ctx(content: &str) -> LintContext<'_> {
        LintContext::new(content, MarkdownFlavor::Standard, None)
    }

    /// Create rule with enabled=true for tests that call check() directly
    fn create_enabled_rule() -> MD073TocValidation {
        MD073TocValidation {
            enabled: true,
            ..MD073TocValidation::default()
        }
    }

    // ========== Detection Tests ==========

    #[test]
    fn test_detect_markers_basic() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

- [Heading 1](#heading-1)

<!-- tocstop -->

## Heading 1

Content here.
"#;
        let ctx = create_ctx(content);
        let region = rule.detect_by_markers(&ctx);
        assert!(region.is_some());
        let region = region.unwrap();
        // Verify region boundaries are detected correctly
        assert_eq!(region.start_line, 4);
        assert_eq!(region.end_line, 6);
    }

    #[test]
    fn test_detect_markers_variations() {
        let rule = MD073TocValidation::new();

        // Test <!--toc--> (no spaces)
        let content1 = "<!--toc-->\n- [A](#a)\n<!--tocstop-->\n";
        let ctx1 = create_ctx(content1);
        assert!(rule.detect_by_markers(&ctx1).is_some());

        // Test <!-- TOC --> (uppercase)
        let content2 = "<!-- TOC -->\n- [A](#a)\n<!-- TOCSTOP -->\n";
        let ctx2 = create_ctx(content2);
        assert!(rule.detect_by_markers(&ctx2).is_some());

        // Test <!-- /toc --> (alternative stop marker)
        let content3 = "<!-- toc -->\n- [A](#a)\n<!-- /toc -->\n";
        let ctx3 = create_ctx(content3);
        assert!(rule.detect_by_markers(&ctx3).is_some());
    }

    #[test]
    fn test_no_toc_region() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

## Heading 1

Content here.

## Heading 2

More content.
"#;
        let ctx = create_ctx(content);
        let region = rule.detect_toc_region(&ctx);
        assert!(region.is_none());
    }

    // ========== Validation Tests ==========

    #[test]
    fn test_toc_matches_headings() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [Heading 1](#heading-1)
- [Heading 2](#heading-2)

<!-- tocstop -->

## Heading 1

Content.

## Heading 2

More content.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert!(result.is_empty(), "Expected no warnings for matching TOC");
    }

    #[test]
    fn test_missing_entry() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [Heading 1](#heading-1)

<!-- tocstop -->

## Heading 1

Content.

## Heading 2

New heading not in TOC.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(result[0].message.contains("Missing entry"));
        assert!(result[0].message.contains("Heading 2"));
    }

    #[test]
    fn test_stale_entry() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [Heading 1](#heading-1)
- [Deleted Heading](#deleted-heading)

<!-- tocstop -->

## Heading 1

Content.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(result[0].message.contains("Stale entry"));
        assert!(result[0].message.contains("Deleted Heading"));
    }

    #[test]
    fn test_text_mismatch() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [Old Name](#heading-1)

<!-- tocstop -->

## Heading 1

Content.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(result[0].message.contains("Text mismatch"));
    }

    // ========== Level Filtering Tests ==========

    #[test]
    fn test_min_level_excludes_h1() {
        let mut rule = MD073TocValidation::new();
        rule.min_level = 2;

        let content = r#"<!-- toc -->

<!-- tocstop -->

# Should Be Excluded

## Should Be Included

Content.
"#;
        let ctx = create_ctx(content);
        let region = rule.detect_toc_region(&ctx).unwrap();
        let expected = rule.build_expected_toc(&ctx, &region);

        assert_eq!(expected.len(), 1);
        assert_eq!(expected[0].text, "Should Be Included");
    }

    #[test]
    fn test_max_level_excludes_h5_h6() {
        let mut rule = MD073TocValidation::new();
        rule.max_level = 4;

        let content = r#"<!-- toc -->

<!-- tocstop -->

## Level 2

### Level 3

#### Level 4

##### Level 5 Should Be Excluded

###### Level 6 Should Be Excluded
"#;
        let ctx = create_ctx(content);
        let region = rule.detect_toc_region(&ctx).unwrap();
        let expected = rule.build_expected_toc(&ctx, &region);

        assert_eq!(expected.len(), 3);
        assert!(expected.iter().all(|e| e.level <= 4));
    }

    // ========== Fix Tests ==========

    #[test]
    fn test_fix_adds_missing_entry() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

- [Heading 1](#heading-1)

<!-- tocstop -->

## Heading 1

Content.

## Heading 2

New heading.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();
        assert!(fixed.contains("- [Heading 2](#heading-2)"));
    }

    #[test]
    fn test_fix_removes_stale_entry() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

- [Heading 1](#heading-1)
- [Deleted](#deleted)

<!-- tocstop -->

## Heading 1

Content.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();
        assert!(fixed.contains("- [Heading 1](#heading-1)"));
        assert!(!fixed.contains("Deleted"));
    }

    #[test]
    fn test_fix_idempotent() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

- [Heading 1](#heading-1)
- [Heading 2](#heading-2)

<!-- tocstop -->

## Heading 1

Content.

## Heading 2

More.
"#;
        let ctx = create_ctx(content);
        let fixed1 = rule.fix(&ctx).unwrap();
        let ctx2 = create_ctx(&fixed1);
        let fixed2 = rule.fix(&ctx2).unwrap();

        // Second fix should produce same output
        assert_eq!(fixed1, fixed2);
    }

    #[test]
    fn test_fix_preserves_markers() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

Old TOC content.

<!-- tocstop -->

## New Heading

Content.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();

        // Markers should still be present
        assert!(fixed.contains("<!-- toc -->"));
        assert!(fixed.contains("<!-- tocstop -->"));
        // New content should be generated
        assert!(fixed.contains("- [New Heading](#new-heading)"));
    }

    #[test]
    fn test_fix_requires_markers() {
        let rule = create_enabled_rule();

        // Document without markers - no TOC detected, no changes
        let content_no_markers = r#"# Title

## Heading 1

Content.
"#;
        let ctx = create_ctx(content_no_markers);
        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(fixed, content_no_markers);

        // Document with markers - TOC detected and fixed
        let content_markers = r#"# Title

<!-- toc -->

- [Old Entry](#old-entry)

<!-- tocstop -->

## Heading 1

Content.
"#;
        let ctx = create_ctx(content_markers);
        let fixed = rule.fix(&ctx).unwrap();
        assert!(fixed.contains("- [Heading 1](#heading-1)"));
        assert!(!fixed.contains("Old Entry"));
    }

    // ========== Anchor Tests ==========

    #[test]
    fn test_duplicate_heading_anchors() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## Duplicate

Content.

## Duplicate

More content.

## Duplicate

Even more.
"#;
        let ctx = create_ctx(content);
        let region = rule.detect_toc_region(&ctx).unwrap();
        let expected = rule.build_expected_toc(&ctx, &region);

        assert_eq!(expected.len(), 3);
        assert_eq!(expected[0].anchor, "duplicate");
        assert_eq!(expected[1].anchor, "duplicate-1");
        assert_eq!(expected[2].anchor, "duplicate-2");
    }

    // ========== Edge Cases ==========

    #[test]
    fn test_headings_in_code_blocks_ignored() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [Real Heading](#real-heading)

<!-- tocstop -->

## Real Heading

```markdown
## Fake Heading In Code
```

Content.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert!(result.is_empty(), "Should not report fake heading in code block");
    }

    #[test]
    fn test_empty_toc_region() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->
<!-- tocstop -->

## Heading 1

Content.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1);
        assert!(result[0].message.contains("Missing entry"));
    }

    #[test]
    fn test_nested_indentation() {
        let rule = create_enabled_rule();

        let content = r#"<!-- toc -->

<!-- tocstop -->

## Level 2

### Level 3

#### Level 4

## Another Level 2
"#;
        let ctx = create_ctx(content);
        let region = rule.detect_toc_region(&ctx).unwrap();
        let expected = rule.build_expected_toc(&ctx, &region);
        let toc = rule.generate_toc(&expected);

        // Check indentation (always nested)
        assert!(toc.contains("- [Level 2](#level-2)"));
        assert!(toc.contains("  - [Level 3](#level-3)"));
        assert!(toc.contains("    - [Level 4](#level-4)"));
        assert!(toc.contains("- [Another Level 2](#another-level-2)"));
    }

    // ========== Indentation Mismatch Tests ==========

    #[test]
    fn test_indentation_mismatch_detected() {
        let rule = create_enabled_rule();
        // TOC entries are all at same indentation level, but headings have different levels
        let content = r#"<!-- toc -->
- [Hello](#hello)
- [Another](#another)
- [Heading](#heading)
<!-- tocstop -->

## Hello

### Another

## Heading
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        // Should detect indentation mismatch - "Another" is level 3 but has no indent
        assert_eq!(result.len(), 1, "Should report indentation mismatch: {result:?}");
        assert!(
            result[0].message.contains("Indentation mismatch"),
            "Message should mention indentation: {}",
            result[0].message
        );
        assert!(
            result[0].message.contains("Another"),
            "Message should mention the entry: {}",
            result[0].message
        );
    }

    #[test]
    fn test_indentation_mismatch_fixed() {
        let rule = create_enabled_rule();
        // TOC entries are all at same indentation level, but headings have different levels
        let content = r#"<!-- toc -->
- [Hello](#hello)
- [Another](#another)
- [Heading](#heading)
<!-- tocstop -->

## Hello

### Another

## Heading
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();
        // After fix, "Another" should be indented
        assert!(fixed.contains("- [Hello](#hello)"));
        assert!(fixed.contains("  - [Another](#another)")); // Indented with 2 spaces
        assert!(fixed.contains("- [Heading](#heading)"));
    }

    #[test]
    fn test_no_indentation_mismatch_when_correct() {
        let rule = create_enabled_rule();
        // TOC has correct indentation
        let content = r#"<!-- toc -->
- [Hello](#hello)
  - [Another](#another)
- [Heading](#heading)
<!-- tocstop -->

## Hello

### Another

## Heading
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        // Should not report any issues - indentation is correct
        assert!(result.is_empty(), "Should not report issues: {result:?}");
    }

    // ========== Order Mismatch Tests ==========

    #[test]
    fn test_order_mismatch_detected() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [Section B](#section-b)
- [Section A](#section-a)

<!-- tocstop -->

## Section A

Content A.

## Section B

Content B.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        // Should detect order mismatch - Section B appears before Section A in TOC
        // but Section A comes first in document
        assert!(!result.is_empty(), "Should detect order mismatch");
    }

    #[test]
    fn test_order_mismatch_ignored_when_disabled() {
        let mut rule = create_enabled_rule();
        rule.enforce_order = false;
        let content = r#"# Title

<!-- toc -->

- [Section B](#section-b)
- [Section A](#section-a)

<!-- tocstop -->

## Section A

Content A.

## Section B

Content B.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        // With enforce_order=false, order mismatches should be ignored
        assert!(result.is_empty(), "Should not report order mismatch when disabled");
    }

    // ========== Unicode and Special Characters Tests ==========

    #[test]
    fn test_unicode_headings() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [日本語の見出し](#日本語の見出し)
- [Émojis 🎉](#émojis-)

<!-- tocstop -->

## 日本語の見出し

Japanese content.

## Émojis 🎉

Content with emojis.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        // Should handle unicode correctly
        assert!(result.is_empty(), "Should handle unicode headings");
    }

    #[test]
    fn test_special_characters_in_headings() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [What's New?](#whats-new)
- [C++ Guide](#c-guide)

<!-- tocstop -->

## What's New?

News content.

## C++ Guide

C++ content.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert!(result.is_empty(), "Should handle special characters");
    }

    #[test]
    fn test_code_spans_in_headings() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [`check [PATHS...]`](#check-paths)

<!-- tocstop -->

## `check [PATHS...]`

Command documentation.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert!(result.is_empty(), "Should handle code spans in headings with brackets");
    }

    // ========== Config Tests ==========

    #[test]
    fn test_from_config_defaults() {
        let config = crate::config::Config::default();
        let rule = MD073TocValidation::from_config(&config);
        let rule = rule.as_any().downcast_ref::<MD073TocValidation>().unwrap();

        assert_eq!(rule.min_level, 2);
        assert_eq!(rule.max_level, 4);
        assert!(rule.enforce_order);
        assert_eq!(rule.indent, 2);
    }

    #[test]
    fn test_indent_from_md007_config() {
        use crate::config::{Config, RuleConfig};
        use std::collections::BTreeMap;

        let mut config = Config::default();

        // Set MD007 indent to 4
        let mut md007_values = BTreeMap::new();
        md007_values.insert("indent".to_string(), toml::Value::Integer(4));
        config.rules.insert(
            "MD007".to_string(),
            RuleConfig {
                severity: None,
                values: md007_values,
            },
        );

        let rule = MD073TocValidation::from_config(&config);
        let rule = rule.as_any().downcast_ref::<MD073TocValidation>().unwrap();

        assert_eq!(rule.indent, 4, "Should read indent from MD007 config");
    }

    #[test]
    fn test_indent_md073_overrides_md007() {
        use crate::config::{Config, RuleConfig};
        use std::collections::BTreeMap;

        let mut config = Config::default();

        // Set MD007 indent to 4
        let mut md007_values = BTreeMap::new();
        md007_values.insert("indent".to_string(), toml::Value::Integer(4));
        config.rules.insert(
            "MD007".to_string(),
            RuleConfig {
                severity: None,
                values: md007_values,
            },
        );

        // Set MD073 indent to 3 (should override MD007)
        let mut md073_values = BTreeMap::new();
        md073_values.insert("enabled".to_string(), toml::Value::Boolean(true));
        md073_values.insert("indent".to_string(), toml::Value::Integer(3));
        config.rules.insert(
            "MD073".to_string(),
            RuleConfig {
                severity: None,
                values: md073_values,
            },
        );

        let rule = MD073TocValidation::from_config(&config);
        let rule = rule.as_any().downcast_ref::<MD073TocValidation>().unwrap();

        assert_eq!(rule.indent, 3, "MD073 indent should override MD007");
    }

    #[test]
    fn test_generate_toc_with_4_space_indent() {
        let mut rule = create_enabled_rule();
        rule.indent = 4;

        let content = r#"<!-- toc -->

<!-- tocstop -->

## Level 2

### Level 3

#### Level 4

## Another Level 2
"#;
        let ctx = create_ctx(content);
        let region = rule.detect_toc_region(&ctx).unwrap();
        let expected = rule.build_expected_toc(&ctx, &region);
        let toc = rule.generate_toc(&expected);

        // With 4-space indent:
        // Level 2 = 0 spaces (base level)
        // Level 3 = 4 spaces
        // Level 4 = 8 spaces
        assert!(toc.contains("- [Level 2](#level-2)"), "Level 2 should have no indent");
        assert!(
            toc.contains("    - [Level 3](#level-3)"),
            "Level 3 should have 4-space indent"
        );
        assert!(
            toc.contains("        - [Level 4](#level-4)"),
            "Level 4 should have 8-space indent"
        );
        assert!(toc.contains("- [Another Level 2](#another-level-2)"));
    }

    #[test]
    fn test_validate_toc_with_4_space_indent() {
        let mut rule = create_enabled_rule();
        rule.indent = 4;

        // TOC with correct 4-space indentation
        let content = r#"<!-- toc -->
- [Hello](#hello)
    - [Another](#another)
- [Heading](#heading)
<!-- tocstop -->

## Hello

### Another

## Heading
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "Should accept 4-space indent when configured: {result:?}"
        );
    }

    #[test]
    fn test_validate_toc_wrong_indent_with_4_space_config() {
        let mut rule = create_enabled_rule();
        rule.indent = 4;

        // TOC with 2-space indentation (wrong when 4-space is configured)
        let content = r#"<!-- toc -->
- [Hello](#hello)
  - [Another](#another)
- [Heading](#heading)
<!-- tocstop -->

## Hello

### Another

## Heading
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should detect wrong indent");
        assert!(
            result[0].message.contains("Indentation mismatch"),
            "Should report indentation mismatch: {}",
            result[0].message
        );
        assert!(
            result[0].message.contains("expected 4 spaces"),
            "Should mention expected 4 spaces: {}",
            result[0].message
        );
    }

    // ========== Markdown Stripping Tests ==========

    #[test]
    fn test_strip_markdown_formatting_link() {
        let result = strip_markdown_formatting("Tool: [terminal](https://example.com)");
        assert_eq!(result, "Tool: terminal");
    }

    #[test]
    fn test_strip_markdown_formatting_bold() {
        let result = strip_markdown_formatting("This is **bold** text");
        assert_eq!(result, "This is bold text");

        let result = strip_markdown_formatting("This is __bold__ text");
        assert_eq!(result, "This is bold text");
    }

    #[test]
    fn test_strip_markdown_formatting_italic() {
        let result = strip_markdown_formatting("This is *italic* text");
        assert_eq!(result, "This is italic text");

        let result = strip_markdown_formatting("This is _italic_ text");
        assert_eq!(result, "This is italic text");
    }

    #[test]
    fn test_strip_markdown_formatting_code_span() {
        let result = strip_markdown_formatting("Use the `format` function");
        assert_eq!(result, "Use the format function");
    }

    #[test]
    fn test_strip_markdown_formatting_image() {
        let result = strip_markdown_formatting("See ![logo](image.png) for details");
        assert_eq!(result, "See logo for details");
    }

    #[test]
    fn test_strip_markdown_formatting_reference_link() {
        let result = strip_markdown_formatting("See [documentation][docs] for details");
        assert_eq!(result, "See documentation for details");
    }

    #[test]
    fn test_strip_markdown_formatting_combined() {
        // Link is stripped first, leaving bold, then bold is stripped
        let result = strip_markdown_formatting("Tool: [**terminal**](https://example.com)");
        assert_eq!(result, "Tool: terminal");
    }

    #[test]
    fn test_toc_with_link_in_heading_matches_stripped_text() {
        let rule = create_enabled_rule();

        // TOC entry text matches the stripped heading text
        let content = r#"# Title

<!-- toc -->

- [Tool: terminal](#tool-terminal)

<!-- tocstop -->

## Tool: [terminal](https://example.com)

Content here.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "Stripped heading text should match TOC entry: {result:?}"
        );
    }

    #[test]
    fn test_toc_with_simplified_text_still_mismatches() {
        let rule = create_enabled_rule();

        // TOC entry "terminal" does NOT match stripped heading "Tool: terminal"
        let content = r#"# Title

<!-- toc -->

- [terminal](#tool-terminal)

<!-- tocstop -->

## Tool: [terminal](https://example.com)

Content here.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert_eq!(result.len(), 1, "Should report text mismatch");
        assert!(result[0].message.contains("Text mismatch"));
    }

    #[test]
    fn test_fix_generates_stripped_toc_entries() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## Tool: [busybox](https://www.busybox.net/)

Content.

## Tool: [mount](https://en.wikipedia.org/wiki/Mount)

More content.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();

        // Generated TOC should have stripped text (links removed)
        assert!(
            fixed.contains("- [Tool: busybox](#tool-busybox)"),
            "TOC entry should have stripped link text"
        );
        assert!(
            fixed.contains("- [Tool: mount](#tool-mount)"),
            "TOC entry should have stripped link text"
        );
        // TOC entries should NOT contain the URL (the actual headings in the document still will)
        // Check only within the TOC region (between toc markers)
        let toc_start = fixed.find("<!-- toc -->").unwrap();
        let toc_end = fixed.find("<!-- tocstop -->").unwrap();
        let toc_content = &fixed[toc_start..toc_end];
        assert!(
            !toc_content.contains("busybox.net"),
            "TOC should not contain URLs: {toc_content}"
        );
        assert!(
            !toc_content.contains("wikipedia.org"),
            "TOC should not contain URLs: {toc_content}"
        );
    }

    #[test]
    fn test_fix_with_bold_in_heading() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## **Important** Section

Content.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();

        // Generated TOC preserves bold markers in display text; anchor strips them.
        assert!(fixed.contains("- [**Important** Section](#important-section)"));
    }

    #[test]
    fn test_fix_with_code_in_heading() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## Using `async` Functions

Content.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();

        // Generated TOC preserves code ticks in display text; anchor strips them.
        assert!(fixed.contains("- [Using `async` Functions](#using-async-functions)"));
    }

    // ========== Custom Anchor Tests ==========

    #[test]
    fn test_custom_anchor_id_respected() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [My Section](#my-custom-anchor)

<!-- tocstop -->

## My Section {#my-custom-anchor}

Content here.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert!(result.is_empty(), "Should respect custom anchor IDs: {result:?}");
    }

    #[test]
    fn test_custom_anchor_id_in_generated_toc() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## First Section {#custom-first}

Content.

## Second Section {#another-custom}

More content.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();
        assert!(fixed.contains("- [First Section](#custom-first)"));
        assert!(fixed.contains("- [Second Section](#another-custom)"));
    }

    #[test]
    fn test_mixed_custom_and_generated_anchors() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [Custom Section](#my-id)
- [Normal Section](#normal-section)

<!-- tocstop -->

## Custom Section {#my-id}

Content.

## Normal Section

More content.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert!(result.is_empty(), "Should handle mixed custom and generated anchors");
    }

    // ========== Anchor Generation Tests ==========

    #[test]
    fn test_github_anchor_style() {
        let rule = create_enabled_rule();

        let content = r#"<!-- toc -->

<!-- tocstop -->

## Test_With_Underscores

Content.
"#;
        let ctx = create_ctx(content);
        let region = rule.detect_toc_region(&ctx).unwrap();
        let expected = rule.build_expected_toc(&ctx, &region);

        // GitHub-style anchors preserve underscores
        assert_eq!(expected[0].anchor, "test_with_underscores");
    }

    // ========== Stress Tests ==========

    #[test]
    fn test_stress_many_headings() {
        let rule = create_enabled_rule();

        // Generate a document with 150 headings
        let mut content = String::from("# Title\n\n<!-- toc -->\n\n<!-- tocstop -->\n\n");

        for i in 1..=150 {
            content.push_str(&format!("## Heading Number {i}\n\nContent for section {i}.\n\n"));
        }

        let ctx = create_ctx(&content);

        // Should not panic or timeout
        let result = rule.check(&ctx).unwrap();

        // Should report missing entries for all 150 headings
        assert_eq!(result.len(), 1, "Should report single warning for TOC");
        assert!(result[0].message.contains("Missing entry"));

        // Fix should generate TOC with 150 entries
        let fixed = rule.fix(&ctx).unwrap();
        assert!(fixed.contains("- [Heading Number 1](#heading-number-1)"));
        assert!(fixed.contains("- [Heading Number 100](#heading-number-100)"));
        assert!(fixed.contains("- [Heading Number 150](#heading-number-150)"));
    }

    #[test]
    fn test_stress_deeply_nested() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## Level 2 A

### Level 3 A

#### Level 4 A

## Level 2 B

### Level 3 B

#### Level 4 B

## Level 2 C

### Level 3 C

#### Level 4 C

## Level 2 D

### Level 3 D

#### Level 4 D
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();

        // Check nested indentation is correct
        assert!(fixed.contains("- [Level 2 A](#level-2-a)"));
        assert!(fixed.contains("  - [Level 3 A](#level-3-a)"));
        assert!(fixed.contains("    - [Level 4 A](#level-4-a)"));
        assert!(fixed.contains("- [Level 2 D](#level-2-d)"));
        assert!(fixed.contains("  - [Level 3 D](#level-3-d)"));
        assert!(fixed.contains("    - [Level 4 D](#level-4-d)"));
    }

    // ==================== Duplicate TOC anchors ====================

    #[test]
    fn test_duplicate_toc_anchors_produce_correct_diagnostics() {
        let rule = create_enabled_rule();
        // Document has headings "Example", "Another", "Example" which produce anchors:
        // "example", "another", "example-1"
        // TOC incorrectly uses #example twice instead of #example and #example-1
        let content = r#"# Document

<!-- toc -->

- [Example](#example)
- [Another](#another)
- [Example](#example)

<!-- tocstop -->

## Example
First.

## Another
Middle.

## Example
Second.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();

        // The TOC has #example twice but expected has #example and #example-1.
        // Should report that #example-1 is missing from the TOC.
        assert!(!result.is_empty(), "Should detect mismatch with duplicate TOC anchors");
        assert!(
            result[0].message.contains("Missing entry") || result[0].message.contains("Stale entry"),
            "Should report missing or stale entries for duplicate anchors. Got: {}",
            result[0].message
        );
    }

    // ==================== Multi-backtick code spans ====================

    #[test]
    fn test_strip_double_backtick_code_span() {
        // Double-backtick code spans should be stripped
        let result = strip_markdown_formatting("Using ``code with ` backtick``");
        assert_eq!(
            result, "Using code with ` backtick",
            "Should strip double-backtick code spans"
        );
    }

    #[test]
    fn test_strip_triple_backtick_code_span() {
        // Triple-backtick code spans should be stripped
        let result = strip_markdown_formatting("Using ```code with `` backticks```");
        assert_eq!(
            result, "Using code with `` backticks",
            "Should strip triple-backtick code spans"
        );
    }

    #[test]
    fn test_toc_with_double_backtick_heading() {
        let rule = create_enabled_rule();
        // Use fix() to generate the correct TOC (including anchor), then check()
        // should produce no warnings on the fixed output.
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## Using ``code with ` backtick``

Content here.
"#;
        let ctx = create_ctx(content);
        // The heading uses double-backtick code span: ``code with ` backtick``
        // TOC display text preserves the code span; anchor is derived from raw text.
        let fixed = rule.fix(&ctx).unwrap();

        // Verify that the generated TOC entry preserves the double-backtick code span
        // in the display text.
        let toc_start = fixed.find("<!-- toc -->").unwrap();
        let toc_end = fixed.find("<!-- tocstop -->").unwrap();
        let toc_content = &fixed[toc_start..toc_end];
        assert!(
            toc_content.contains("``code with ` backtick``"),
            "Fix should preserve double-backtick code span in TOC display text. Got: {toc_content}"
        );

        // After fix, check() must produce no warnings (idempotency check)
        let ctx2 = create_ctx(&fixed);
        let result = rule.check(&ctx2).unwrap();
        assert!(
            result.is_empty(),
            "check() should not warn on fixed output. Warnings: {result:?}"
        );
    }

    #[test]
    fn test_stress_many_duplicates() {
        let rule = create_enabled_rule();

        // Generate 50 headings with the same text
        let mut content = String::from("# Title\n\n<!-- toc -->\n\n<!-- tocstop -->\n\n");
        for _ in 0..50 {
            content.push_str("## FAQ\n\nContent.\n\n");
        }

        let ctx = create_ctx(&content);
        let region = rule.detect_toc_region(&ctx).unwrap();
        let expected = rule.build_expected_toc(&ctx, &region);

        // Should generate unique anchors for all 50
        assert_eq!(expected.len(), 50);
        assert_eq!(expected[0].anchor, "faq");
        assert_eq!(expected[1].anchor, "faq-1");
        assert_eq!(expected[49].anchor, "faq-49");
    }

    /// Core invariant: for every warning with a Fix, fix() must produce
    /// output consistent with applying that fix directly.
    #[test]
    fn test_roundtrip_check_and_fix_alignment() {
        let rule = create_enabled_rule();

        let inputs = [
            // Stale entry
            "# Title\n\n<!-- toc -->\n- [Old Section](#old-section)\n<!-- tocstop -->\n\n## New Section\n",
            // Missing entry
            "# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## One\n\n## Two\n",
            // Text mismatch
            "# Title\n\n<!-- toc -->\n- [Wrong Text](#real-section)\n<!-- tocstop -->\n\n## Real Section\n",
            // Already correct (no warnings, no change)
            "# Title\n\n<!-- toc -->\n- [One](#one)\n- [Two](#two)\n<!-- tocstop -->\n\n## One\n\n## Two\n",
        ];

        for input in &inputs {
            let ctx = create_ctx(input);
            let fixed = rule.fix(&ctx).unwrap();

            // Idempotency: fix(fix(x)) == fix(x)
            let ctx2 = create_ctx(&fixed);
            let fixed_twice = rule.fix(&ctx2).unwrap();
            assert_eq!(
                fixed, fixed_twice,
                "fix() is not idempotent for input: {input:?}\nfirst:  {fixed:?}\nsecond: {fixed_twice:?}"
            );

            // After fix, check() should produce no warnings
            let warnings_after = rule.check(&ctx2).unwrap();
            assert!(
                warnings_after.is_empty(),
                "check() should return no warnings after fix() for input: {input:?}\nfixed: {fixed:?}\nwarnings: {warnings_after:?}"
            );
        }
    }

    /// If a TOC has no mismatches, check() emits no warnings and fix()
    /// returns content unchanged.
    #[test]
    fn test_no_mismatch_preserves_content() {
        let rule = create_enabled_rule();

        let content = "# Title\n\n<!-- toc -->\n- [First Section](#first-section)\n- [Second Section](#second-section)\n<!-- tocstop -->\n\n## First Section\n\ntext\n\n## Second Section\n\ntext\n";
        let ctx = create_ctx(content);

        let warnings = rule.check(&ctx).unwrap();
        assert!(warnings.is_empty(), "No mismatches should emit no warnings");

        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(fixed, content, "Content should be unchanged when TOC matches headings");
    }

    /// Inline-disabled TOC should not be modified by fix().
    #[test]
    fn test_inline_disable_preserves_toc() {
        let rule = create_enabled_rule();

        // TOC with a stale entry, but MD073 disabled for the TOC region
        let content = "# Title\n\n<!-- rumdl-disable MD073 -->\n<!-- toc -->\n- [Stale](#stale)\n<!-- tocstop -->\n<!-- rumdl-enable MD073 -->\n\n## Real\n";
        let ctx = create_ctx(content);

        let fixed = rule.fix(&ctx).unwrap();
        assert_eq!(fixed, content, "TOC in a disabled region should be preserved exactly");
    }

    // ========== Inline Formatting Preservation Tests (#634) ==========

    /// Backticks in a heading must be preserved in the TOC display text.
    /// The anchor is generated from the raw heading text (which includes backticks)
    /// and must still use the stripped form.
    #[test]
    fn test_fix_code_ticks_preserved_in_toc_display_text() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

### `my header`

Content.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();

        assert!(
            fixed.contains("- [`my header`](#my-header)"),
            "Code ticks must be preserved in TOC display text. Got: {fixed}"
        );
    }

    /// A correct user-written TOC entry with code ticks must not be re-flagged.
    #[test]
    fn test_validate_toc_with_code_ticks_is_valid() {
        let rule = create_enabled_rule();
        let content = r#"# Title

<!-- toc -->

- [`my header`](#my-header)

<!-- tocstop -->

## `my header`

Content.
"#;
        let ctx = create_ctx(content);
        let result = rule.check(&ctx).unwrap();
        assert!(
            result.is_empty(),
            "A TOC entry with preserved code ticks should be accepted as valid: {result:?}"
        );
    }

    /// A heading with bold/italic preserves emphasis markers in the TOC display text;
    /// the anchor is generated from the raw (formatted) heading text and still uses
    /// the stripped form.
    #[test]
    fn test_fix_emphasis_preserved_in_toc_display_text() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## **bold** and *italic*

Content.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();

        assert!(
            fixed.contains("- [**bold** and *italic*](#bold-and-italic)"),
            "Emphasis markers must be preserved in TOC display text. Got: {fixed}"
        );
    }

    /// A heading containing a link must have the link stripped from the TOC display
    /// text (nested links are invalid in Markdown).
    #[test]
    fn test_fix_link_in_heading_is_stripped() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## See [docs](http://example.com) for details

Content.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();

        assert!(
            fixed.contains("- [See docs for details](#see-docs-for-details)"),
            "Link must be stripped from TOC display text. Got: {fixed}"
        );
        // Ensure no URL leaks into TOC entry
        let toc_start = fixed.find("<!-- toc -->").unwrap();
        let toc_end = fixed.find("<!-- tocstop -->").unwrap();
        let toc_content = &fixed[toc_start..toc_end];
        assert!(
            !toc_content.contains("http://example.com"),
            "TOC should not contain link URL: {toc_content}"
        );
    }

    /// An image in a heading must still be stripped from the TOC display text.
    #[test]
    fn test_fix_image_in_heading_is_stripped() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## Section ![icon](icon.png) Title

Content.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();

        assert!(
            fixed.contains("- [Section icon Title](#section-icon-title)"),
            "Image must be stripped from TOC display text. Got: {fixed}"
        );
    }

    /// Running fix() twice on a document with inline-formatted headings must
    /// produce stable output (idempotency).
    #[test]
    fn test_fix_idempotent_with_inline_formatting() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## `code` heading

### **bold** heading

## See [link](http://x.com)

"#;
        let ctx = create_ctx(content);
        let fixed1 = rule.fix(&ctx).unwrap();
        let ctx2 = create_ctx(&fixed1);
        let fixed2 = rule.fix(&ctx2).unwrap();

        assert_eq!(fixed1, fixed2, "fix() must be idempotent for inline-formatted headings");

        // After fix, check() must produce no warnings
        let warnings = rule.check(&ctx2).unwrap();
        assert!(
            warnings.is_empty(),
            "check() must not warn after fix() for inline-formatted headings: {warnings:?}"
        );
    }

    /// Link-like syntax inside a code span must not be stripped, because it is
    /// literal content of the code span and not a real Markdown link.
    #[test]
    fn test_link_inside_code_span_preserved_in_toc() {
        let rule = MD073TocValidation::new();
        let content = r#"# Title

<!-- toc -->

<!-- tocstop -->

## Use `[foo](bar)` syntax

Content.
"#;
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();

        // The code span `[foo](bar)` must survive intact in the TOC display text.
        // The anchor is generated from the raw heading text by the GitHub algorithm,
        // which strips backtick, bracket, and paren characters. Verify only the
        // display-text preservation, not the exact anchor (which depends on the anchor
        // generation algorithm's treatment of non-alphanumeric chars in code spans).
        let toc_start = fixed.find("<!-- toc -->").unwrap();
        let toc_end = fixed.find("<!-- tocstop -->").unwrap();
        let toc_content = &fixed[toc_start..toc_end];
        assert!(
            toc_content.contains("Use `[foo](bar)` syntax"),
            "Link-like text inside code span must be preserved in TOC display text. Got: {toc_content}"
        );
        // Also ensure the real link stripping (outside code spans) still works
        assert!(
            !toc_content.contains("http://"),
            "Real links (outside code spans) should be stripped: {toc_content}"
        );
    }

    // ========== HTML anchor targets ==========

    /// The TOC region after `fix`, without the markers and surrounding blank lines.
    fn generated_toc(content: &str) -> String {
        let rule = create_enabled_rule();
        let ctx = create_ctx(content);
        let fixed = rule.fix(&ctx).unwrap();
        let start = fixed.find("<!-- toc -->").unwrap() + "<!-- toc -->".len();
        let end = fixed.find("<!-- tocstop -->").unwrap();
        fixed[start..end].trim().to_string()
    }

    fn check_toc(content: &str) -> Vec<LintWarning> {
        let rule = create_enabled_rule();
        let ctx = create_ctx(content);
        rule.check(&ctx).unwrap()
    }

    #[test]
    fn test_toc_may_target_the_generated_slug_of_a_heading_with_an_html_anchor() {
        // GitHub gives the heading both `#cheat-sheets` (generated) and `#cheatsheets`
        // (the element's own id), so a TOC written against either reaches it.
        for anchor in ["cheat-sheets", "cheatsheets"] {
            let content = format!(
                "# Title\n\n<!-- toc -->\n\n- [Cheat Sheets](#{anchor})\n\n<!-- tocstop -->\n\n## <a name=\"cheatsheets\"></a>Cheat Sheets\n\nContent.\n"
            );
            let result = check_toc(&content);
            assert!(result.is_empty(), "#{anchor} reaches the heading, got {result:?}");
        }
    }

    #[test]
    fn test_every_anchor_element_on_a_heading_is_a_valid_target() {
        for anchor in ["old", "older", "foo"] {
            let content = format!(
                "# Title\n\n<!-- toc -->\n\n- [Foo](#{anchor})\n\n<!-- tocstop -->\n\n## <a name=\"old\"></a><a id=\"older\"></a>Foo\n"
            );
            let result = check_toc(&content);
            assert!(result.is_empty(), "#{anchor} reaches the heading, got {result:?}");
        }
    }

    #[test]
    fn test_generated_toc_prefers_the_explicit_html_anchor() {
        let toc =
            generated_toc("# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## <a name=\"cheatsheets\"></a>Cheat Sheets\n");
        assert_eq!(toc, "- [Cheat Sheets](#cheatsheets)");
    }

    #[test]
    fn test_regeneration_keeps_an_existing_anchor_that_reaches_its_heading() {
        // The unrelated missing entry forces a rewrite; the entry already reaching its
        // heading is not switched to the other valid spelling on the way.
        let content = "# Title\n\n<!-- toc -->\n\n- [Cheat Sheets](#cheat-sheets)\n\n<!-- tocstop -->\n\n## <a name=\"cheatsheets\"></a>Cheat Sheets\n\n## Other\n";
        assert_eq!(
            generated_toc(content),
            "- [Cheat Sheets](#cheat-sheets)\n- [Other](#other)"
        );
    }

    #[test]
    fn test_an_attribute_merely_ending_in_id_or_name_is_not_an_anchor() {
        let toc = generated_toc(
            "# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## Foo<a data-id=\"tracking\" data-name=\"pixel\"></a>\n",
        );
        assert_eq!(toc, "- [Foo](#foo)");

        let result = check_toc(
            "# Title\n\n<!-- toc -->\n\n- [Foo](#tracking)\n\n<!-- tocstop -->\n\n## Foo<a data-id=\"tracking\"></a>\n",
        );
        assert_eq!(result.len(), 1, "{result:?}");
        assert!(
            result[0].message.contains("Stale entry: 'Foo'"),
            "{}",
            result[0].message
        );
        assert!(
            result[0].message.contains("Missing entry: 'Foo'"),
            "{}",
            result[0].message
        );
    }

    #[test]
    fn test_an_empty_id_is_not_an_anchor() {
        let toc = generated_toc("# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## Foo<a id=\"\"></a>\n");
        assert_eq!(toc, "- [Foo](#foo)");
    }

    #[test]
    fn test_an_explicit_anchor_still_consumes_the_generated_slug() {
        // GitHub numbers duplicate generated slugs whether or not the first heading also
        // carries its own id, so the second `Same` is `same-1`, never `same`.
        let toc =
            generated_toc("# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## Same<a id=\"stable\"></a>\n\n## Same\n");
        assert_eq!(toc, "- [Same](#stable)\n- [Same](#same-1)");
    }

    #[test]
    fn test_anchor_markup_inside_a_code_span_is_heading_text() {
        let heading = "Showing `<a id=\"literal\"></a>` syntax";
        let toc = generated_toc(&format!("# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## {heading}\n"));
        let slug = AnchorStyle::GitHub.generate_fragment(heading);
        assert_ne!(slug, "literal");
        assert_eq!(toc, format!("- [{heading}](#{slug})"));
    }

    #[test]
    fn test_anchor_markup_inside_an_html_comment_is_not_a_target() {
        let heading = "Foo <!-- <a id=\"hidden\"></a> -->";
        let toc = generated_toc(&format!("# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## {heading}\n"));
        let slug = AnchorStyle::GitHub.generate_fragment(heading);
        assert_ne!(slug, "hidden");
        assert_eq!(toc, format!("- [{heading}](#{slug})"));
    }

    #[test]
    fn test_an_explicit_anchor_is_written_as_a_valid_link_destination() {
        // A bare link destination cannot hold `)` or a space, so the anchor is
        // percent-encoded on the way out, and the encoded entry reaches the heading
        // again on the next check, so the generated TOC is stable.
        let headings = "## Alpha<a id=\"foo)bar\"></a>\n\n## Beta<a id=\"has space\"></a>\n";
        let toc = generated_toc(&format!("# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n{headings}"));
        assert_eq!(toc, "- [Alpha](#foo%29bar)\n- [Beta](#has%20space)");

        let result = check_toc(&format!(
            "# Title\n\n<!-- toc -->\n\n{toc}\n\n<!-- tocstop -->\n\n{headings}"
        ));
        assert!(result.is_empty(), "{result:?}");
    }

    #[test]
    fn test_regeneration_does_not_encode_an_encoded_entry_twice() {
        // The entry reaching `foo)bar` is kept as written; the stale sibling forces
        // the rewrite.
        let content = "# Title\n\n<!-- toc -->\n\n- [Alpha](#foo%29bar)\n- [Gone](#gone)\n\n<!-- tocstop -->\n\n## Alpha<a id=\"foo)bar\"></a>\n";
        assert_eq!(generated_toc(content), "- [Alpha](#foo%29bar)");
    }

    #[test]
    fn test_a_backslash_escaped_anchor_element_defines_no_target() {
        // `\<a ...>` renders as literal text, so `#example` reaches nothing.
        let heading = "Show \\<a id=\"example\"></a> syntax";
        let result = check_toc(&format!(
            "# Title\n\n<!-- toc -->\n\n- [{heading}](#example)\n\n<!-- tocstop -->\n\n## {heading}\n"
        ));
        assert_eq!(result.len(), 1, "{result:?}");
        assert!(result[0].message.contains("Stale entry"), "{}", result[0].message);

        let toc = generated_toc(&format!("# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## {heading}\n"));
        let slug = AnchorStyle::GitHub.generate_fragment(heading);
        assert_ne!(slug, "example");
        assert_eq!(toc, format!("- [{heading}](#{slug})"));
    }

    #[test]
    fn test_a_backslash_escaped_marker_is_text() {
        // `\<!-- toc -->` renders as the literal text `<!-- toc -->`, so the pair
        // below delimits no TOC region and the prose between them is left alone.
        let content = "# Title\n\nWrite these lines:\n\n\\<!-- toc -->\n\nProse that must survive.\n\n\\<!-- tocstop -->\n\n## Alpha\n";
        let result = check_toc(content);
        assert!(result.is_empty(), "{result:?}");
        assert_eq!(create_enabled_rule().fix(&create_ctx(content)).unwrap(), content);

        // An even run of backslashes leaves the `<` itself unescaped, so the comment
        // is rendered and the marker is real.
        let toc = generated_toc("# Title\n\n\\\\<!-- toc -->\n<!-- tocstop -->\n\n## Alpha\n");
        assert_eq!(toc, "- [Alpha](#alpha)");
    }

    #[test]
    fn test_the_whitespace_beside_an_anchor_element_stays_in_the_heading_text() {
        // `Foo<a id="alias"></a> Bar` renders as "Foo Bar", so an entry written
        // that way is up to date and a regenerated entry reads the same.
        let content =
            "# Title\n\n<!-- toc -->\n- [Foo Bar](#foo-bar)\n<!-- tocstop -->\n\n## Foo<a id=\"alias\"></a> Bar\n";
        let warnings = check_toc(content);
        assert!(warnings.is_empty(), "{warnings:?}");

        let toc = generated_toc("# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## Foo<a id=\"alias\"></a> Bar\n");
        assert_eq!(toc, "- [Foo Bar](#alias)");
    }

    #[test]
    fn test_the_whitespace_an_anchor_element_leaves_at_either_end_is_part_of_the_slug() {
        // `## Alpha <a id="x"></a>` renders as "Alpha", but GitHub slugs the text
        // content with the space the element leaves, so the heading answers to
        // `#alpha-` and `#x` and not to `#alpha`. A leading element likewise
        // leaves `#-beta`.
        let document = "# Title\n\n<!-- toc -->\n- [Alpha](#alpha-)\n- [Beta](#-beta)\n<!-- tocstop -->\n\n\
                        ## Alpha <a id=\"x\"></a>\n\n## <a id=\"y\"></a> Beta\n";
        let warnings = check_toc(document);
        assert!(warnings.is_empty(), "{warnings:?}");

        let trimmed = document.replace("#alpha-", "#alpha").replace("#-beta", "#beta");
        assert_eq!(
            check_toc(&trimmed).len(),
            1,
            "a TOC written against trimmed slugs is stale"
        );

        let toc = generated_toc(
            "# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## Alpha <a id=\"x\"></a>\n\n## <a id=\"y\"></a> Beta\n",
        );
        assert_eq!(toc, "- [Alpha](#x)\n- [Beta](#y)");
    }

    #[test]
    fn test_a_heading_whose_text_slugs_to_nothing_gets_no_toc_entry() {
        // An image's description becomes its alt text, in which markup is
        // escaped, so `<a id="img">` inside it is no element and no target, and
        // the heading's text content is empty: GitHub gives it no anchor at all.
        // A heading no fragment can reach gets no entry, so the TOC converges
        // instead of regenerating an `[](#)` entry on every run.
        let body = "## ![<a id=\"img\"></a>](img.png)\n\n## ![](img.png)\n\n## Plain\n";
        let toc = generated_toc(&format!("# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n{body}"));
        assert_eq!(toc, "- [Plain](#plain)");

        let warnings = check_toc(&format!(
            "# Title\n\n<!-- toc -->\n- [Plain](#plain)\n<!-- tocstop -->\n\n{body}"
        ));
        assert!(warnings.is_empty(), "{warnings:?}");

        let warnings = check_toc(&format!(
            "# Title\n\n<!-- toc -->\n- [img](#img)\n- [Plain](#plain)\n<!-- tocstop -->\n\n{body}"
        ));
        assert_eq!(
            warnings.len(),
            1,
            "an entry for the image's anchor is stale: {warnings:?}"
        );
    }

    #[test]
    fn test_an_anchor_inside_another_tags_attribute_value_is_heading_text() {
        // The `<a>` sits in the span's `title` value, so the browser makes no
        // element from it: the heading answers to `#foo` and keeps its bytes.
        let toc = generated_toc(
            "# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## <span title='<a id=\"fake\"></a>'>Foo</span>\n",
        );
        assert_eq!(toc, "- [<span title='<a id=\"fake\"></a>'>Foo</span>](#foo)");
    }

    #[test]
    fn test_a_heading_with_a_custom_id_keeps_its_html_anchor_as_a_target() {
        // `{#new}` replaces the generated slug, but the element's own `name` is still
        // an anchor in the rendered page, so a TOC entry may use either of the two.
        let heading = "<a name=\"old\"></a>My Section {#new}";
        let result = check_toc(&format!(
            "# Title\n\n<!-- toc -->\n\n- [My Section](#old)\n\n<!-- tocstop -->\n\n## {heading}\n"
        ));
        assert!(result.is_empty(), "{result:?}");

        let toc = generated_toc(&format!("# Title\n\n<!-- toc -->\n<!-- tocstop -->\n\n## {heading}\n"));
        assert_eq!(toc, "- [My Section](#new)");

        let result = check_toc(&format!(
            "# Title\n\n<!-- toc -->\n\n- [My Section](#my-section)\n\n<!-- tocstop -->\n\n## {heading}\n"
        ));
        assert_eq!(result.len(), 1, "the generated slug is replaced, got {result:?}");
    }

    #[test]
    fn test_toc_entry_matches_a_multi_line_setext_heading() {
        // A setext underline makes a heading of the whole paragraph above it, so
        // the entry a TOC needs holds the joined text and the slug of that text.
        let content = "# Title\n\n<!-- toc -->\n\n- [First line second line](#first-line-second-line)\n\n<!-- tocstop -->\n\nFirst line\nsecond line\n---\n\nContent.\n";
        assert!(check_toc(content).is_empty(), "{:?}", check_toc(content));
        assert_eq!(
            generated_toc(content),
            "- [First line second line](#first-line-second-line)"
        );
    }

    #[test]
    fn test_missing_entry_names_the_first_line_of_a_multi_line_setext_heading() {
        // The heading a missing entry points at starts on the first line of the
        // paragraph its underline ends, which is where a reader looks for it.
        let content = "# Title\n\n<!-- toc -->\n\n<!-- tocstop -->\n\nFirst line\nsecond line\n---\n\nContent.\n";
        let result = check_toc(content);
        assert_eq!(result.len(), 1, "the TOC misses the heading, got {result:?}");
        assert!(
            result[0]
                .message
                .contains("Missing entry: 'First line second line' (line 7)"),
            "the heading starts on line 7: {}",
            result[0].message
        );
    }
    mod quoted_headings {
        use super::*;

        fn warnings(content: &str) -> usize {
            let rule = create_enabled_rule();
            rule.check(&create_ctx(content)).unwrap().len()
        }

        #[test]
        fn quoted_heading_belongs_in_the_toc() {
            let toc = |entries: &str| format!("# T\n\n<!-- toc -->\n\n{entries}\n<!-- tocstop -->\n\n## A\n\n> ## Q\n");
            assert_eq!(warnings(&toc("- [A](#a)\n- [Q](#q)\n")), 0);
            assert!(
                warnings(&toc("- [A](#a)\n")) > 0,
                "a quoted heading missing from the TOC is reported"
            );
        }

        #[test]
        fn quoted_heading_in_every_container_shape_is_listed() {
            for body in ["> ## Q\n", "> > ## Q\n", "- x\n\n  > ## Q\n", "> - x\n>\n>   ## Q\n"] {
                let content = format!("# T\n\n<!-- toc -->\n\n<!-- tocstop -->\n\n{body}");
                let rule = create_enabled_rule();
                let fixed = rule.fix(&create_ctx(&content)).unwrap();
                assert!(fixed.contains("[Q](#q)"), "{body:?}: {fixed:?}");
            }
        }

        #[test]
        fn a_toc_inside_a_blockquote_is_reported_without_a_fix() {
            // Regenerated entries carry no quote prefix, so splicing them in
            // would end the quote.
            let content = "# T\n\n> <!-- toc -->\n>\n> <!-- tocstop -->\n\n## A\n";
            let rule = create_enabled_rule();
            let warnings = rule.check(&create_ctx(content)).unwrap();
            assert_eq!(warnings.len(), 1);
            assert!(warnings[0].fix.is_none());
            assert_eq!(rule.fix(&create_ctx(content)).unwrap(), content);
        }

        #[test]
        fn a_correct_toc_inside_a_blockquote_is_valid() {
            let content =
                "# T\n\n> <!-- toc -->\n>\n> - [A](#a)\n> - [Q](#q)\n>\n> <!-- tocstop -->\n\n## A\n\n> ## Q\n";
            assert_eq!(warnings(content), 0);
            let stale = content.replace("> - [Q](#q)\n", "");
            assert_eq!(warnings(&stale), 1);
        }

        #[test]
        fn a_quoted_toc_keeps_the_indentation_of_nested_entries() {
            for quote in [">", ">>"] {
                let mut toc = String::new();
                for line in ["<!-- toc -->", "", "- [A](#a)", "  - [B](#b)", "", "<!-- tocstop -->"] {
                    toc.push_str(quote);
                    if !line.is_empty() {
                        toc.push(' ');
                        toc.push_str(line);
                    }
                    toc.push('\n');
                }
                let content = format!("# T\n\n{toc}\n## A\n\nx\n\n### B\n");
                assert_eq!(warnings(&content), 0, "{content:?}");
                let flat = content.replace("  - [B]", "- [B]");
                assert_eq!(warnings(&flat), 1, "{flat:?}");
            }
        }

        #[test]
        fn a_quote_next_to_an_unquoted_toc_does_not_withhold_its_fix() {
            let rule = create_enabled_rule();
            for content in [
                "# T\n\n<!-- toc -->\n<!-- tocstop -->\n> Intro\n\n## A\n",
                "> Intro\n<!-- toc -->\n<!-- tocstop -->\n\n## A\n",
                "> Intro\n\n<!-- toc -->\n\n<!-- tocstop -->\n\n> Outro\n\n## A\n",
            ] {
                let warnings = rule.check(&create_ctx(content)).unwrap();
                assert_eq!(warnings.len(), 1, "{content:?}");
                assert!(warnings[0].fix.is_some(), "{content:?}");
            }
        }

        #[test]
        fn duplicates_are_numbered_across_quoted_and_unquoted_headings() {
            let content = "# T\n\n<!-- toc -->\n\n- [Q](#q)\n- [Q](#q-1)\n\n<!-- tocstop -->\n\n> ## Q\n\n## Q\n";
            assert_eq!(warnings(content), 0);
        }

        #[test]
        fn headings_outside_the_level_range_still_take_a_number() {
            // The title is level 1, outside the default range, yet GitHub numbers the
            // later `## Repeat` as `repeat-1`.
            let content = "# Repeat\n\n<!-- toc -->\n\n- [Repeat](#repeat-1)\n\n<!-- tocstop -->\n\n## Repeat\n";
            assert_eq!(warnings(content), 0);
            let quoted = "<!-- toc -->\n\n- [Repeat](#repeat-1)\n\n<!-- tocstop -->\n\n> # Repeat\n\n## Repeat\n";
            assert_eq!(warnings(quoted), 0);
        }

        #[test]
        fn quoted_heading_in_a_quoted_code_block_is_not_listed() {
            let content = "# T\n\n<!-- toc -->\n\n<!-- tocstop -->\n\n> ```\n> ## Q\n> ```\n";
            assert_eq!(warnings(content), 0);
        }
    }
}