rumdl_lib/utils/

code_block_utils.rs

//!
//! Utility functions for detecting and handling code blocks and code spans in Markdown for rumdl.
//!
//! Code block detection is delegated to pulldown-cmark, which correctly implements the
//! CommonMark specification. This handles edge cases like:
//! - Backtick fences with backticks in the info string (invalid per spec)
//! - Nested fences (longer fence contains shorter fence as content)
//! - Mixed fence types (tilde fence contains backticks as content)
//! - Indented code blocks with proper list context handling

use pulldown_cmark::{CodeBlockKind, Event, Options, Parser, Tag, TagEnd};

use super::parser_options::rumdl_parser_options;

/// Detailed information about a code block captured during parsing
#[derive(Debug, Clone)]
pub struct CodeBlockDetail {
    /// Byte offset where this code block starts
    pub start: usize,
    /// Byte offset where this code block ends
    pub end: usize,
    /// Whether this is a fenced code block (true) or indented (false)
    pub is_fenced: bool,
    /// The info string from fenced blocks (e.g., "rust" from ```rust), empty for indented
    pub info_string: String,
}

/// A strong emphasis span captured during parsing
#[derive(Debug, Clone)]
pub struct StrongSpanDetail {
    /// Byte offset where the strong span starts (including **)
    pub start: usize,
    /// Byte offset where the strong span ends (including **)
    pub end: usize,
    /// Whether this uses asterisk (**) or underscore (__) markers
    pub is_asterisk: bool,
}

/// Text a definition holds directly, captured during parsing
///
/// One of the definition's paragraphs, or, for a tight definition, the run of
/// inline content the parser reports with no paragraph around it. Blocks nested
/// in the definition (lists, blockquotes, code) hold their own text and are not
/// part of one.
#[derive(Debug, Clone)]
pub struct DefinitionTextDetail {
    /// Byte offset where the text starts
    pub start: usize,
    /// Byte offset where the text ends
    pub end: usize,
    /// Byte offset where the definition holding the text starts: its `:`
    /// marker, or indentation before it inside a container
    pub definition_start: usize,
}

/// Ordered list membership: maps line number (1-indexed) to list ID
pub type LineToListMap = std::collections::HashMap<usize, usize>;
/// Ordered list start values: maps list ID to the start value
pub type ListStartValues = std::collections::HashMap<usize, u64>;

/// Result of the central pulldown-cmark parse, capturing all data needed by individual rules
pub struct ParseResult {
    /// Code block byte ranges (start, end)
    pub code_blocks: Vec<(usize, usize)>,
    /// Inline code span byte ranges (start, end)
    pub code_spans: Vec<(usize, usize)>,
    /// Detailed code block info (fenced vs indented, info string)
    pub code_block_details: Vec<CodeBlockDetail>,
    /// Strong emphasis span details
    pub strong_spans: Vec<StrongSpanDetail>,
    /// Ordered list membership: maps line number (1-indexed) to list ID
    pub line_to_list: LineToListMap,
    /// Ordered list start values: maps list ID to start value
    pub list_start_values: ListStartValues,
    /// HTML block byte ranges (start, end)
    ///
    /// Needed to know how far an unclosed `<!--` reaches. The block ends where
    /// CommonMark says it does, which is the end of the enclosing container
    /// rather than the end of the document: an unclosed comment in a blockquote
    /// stops at the quote, and one in a list item stops at the blank line.
    pub html_blocks: Vec<(usize, usize)>,
    /// Definition list item byte ranges (start, end) in document order: a
    /// definition together with the terms before it, blank lines after it left
    /// out, items of nested lists too. Only definitions passing
    /// `opens_definition`, and their terms, count.
    pub definition_items: Vec<(usize, usize)>,
    /// Definition list term byte ranges (start, end)
    pub definition_terms: Vec<(usize, usize)>,
    /// Text held directly by a definition, in document order
    pub definition_texts: Vec<DefinitionTextDetail>,
    /// Definitions failing `opens_definition`, with their terms, in document
    /// order, where reading them as the paragraph text rumdl takes them for
    /// changes which inline constructs the text holds. The parse the other
    /// fields come from reads a definition, which ends every code span, link
    /// and emphasis running into it.
    pub colon_definitions: Vec<ColonDefinitionDetail>,
}

/// A definition pulldown-cmark opens on a colon touching its text, together
/// with the terms before it
#[derive(Debug, Clone)]
pub struct ColonDefinitionDetail {
    /// Byte offset where the first term starts, or the definition when it has none
    pub start: usize,
    /// Byte offset where the definition ends, blank lines after it left out
    pub end: usize,
    /// Byte offset where the definition starts
    pub definition_start: usize,
}

/// Classification of code blocks relative to list contexts
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum CodeBlockContext {
    /// Code block that separates lists (root-level, with blank lines)
    Standalone,
    /// Code block that continues a list (properly indented)
    Indented,
    /// Code block adjacent to list content (edge case, defaults to non-breaking)
    Adjacent,
}

/// Whether a tag opens inline content rather than a block
fn is_inline_tag(tag: &Tag) -> bool {
    matches!(
        tag,
        Tag::Emphasis
            | Tag::Strong
            | Tag::Strikethrough
            | Tag::Superscript
            | Tag::Subscript
            | Tag::Link { .. }
            | Tag::Image { .. }
    )
}

/// Whether a tag end closes inline content rather than a block
fn is_inline_tag_end(tag_end: TagEnd) -> bool {
    matches!(
        tag_end,
        TagEnd::Emphasis
            | TagEnd::Strong
            | TagEnd::Strikethrough
            | TagEnd::Superscript
            | TagEnd::Subscript
            | TagEnd::Link
            | TagEnd::Image
    )
}

/// Whether the definition starting at `start` opens with a colon followed by
/// whitespace or the end of its line.
///
/// That is the marker the definition-list extensions agree on (PHP Markdown
/// Extra, Python-Markdown, Pandoc). pulldown-cmark also opens a definition on a
/// colon touching its text, as in `:warning:` or a `:::` fence closing a div,
/// which those read as prose or as the fence it is.
fn opens_definition(content: &str, start: usize) -> bool {
    content[start..]
        .trim_start_matches([' ', '\t'])
        .strip_prefix(':')
        .is_some_and(|rest| rest.is_empty() || rest.starts_with([' ', '\t', '\n', '\r']))
}

/// The byte ranges of the inline constructs a parse with `options` finds in
/// `content`, in document order: code spans, math, inline HTML, and the
/// emphasis, link and image spans holding text.
fn inline_ranges(content: &str, options: Options) -> Vec<(usize, usize)> {
    Parser::new_ext(content, options)
        .into_offset_iter()
        .filter_map(|(event, range)| match event {
            Event::Start(tag) if is_inline_tag(&tag) => Some((range.start, range.end)),
            Event::Code(_) | Event::InlineHtml(_) | Event::InlineMath(_) | Event::DisplayMath(_) => {
                Some((range.start, range.end))
            }
            _ => None,
        })
        .collect()
}

/// Read source-ordered inline ranges for source-ordered definition queries.
/// Keep only earlier ranges that still overlap the current definition, and
/// inspect later ranges only when they actually fall inside that definition.
/// This also preserves nested ranges and spans crossing several definitions.
struct InlineRangeCursor<'a> {
    ranges: &'a [(usize, usize)],
    next: usize,
    active: Vec<(usize, usize)>,
}

impl<'a> InlineRangeCursor<'a> {
    fn new(ranges: &'a [(usize, usize)]) -> Self {
        Self {
            ranges,
            next: 0,
            active: Vec::new(),
        }
    }

    fn overlapping(&mut self, detail: &ColonDefinitionDetail) -> impl Iterator<Item = (usize, usize)> + '_ {
        while self.next < self.ranges.len() && self.ranges[self.next].0 <= detail.start {
            self.active.push(self.ranges[self.next]);
            self.next += 1;
        }
        self.active.retain(|&(_, end)| end > detail.start);
        let end = detail.end;
        self.active.iter().copied().chain(
            self.ranges[self.next..]
                .iter()
                .take_while(move |&&(start, _)| start < end)
                .copied(),
        )
    }
}

/// Utility functions for detecting and handling code blocks in Markdown
pub struct CodeBlockUtils;

impl CodeBlockUtils {
    /// Detect all code blocks in the content (NOT including inline code spans)
    ///
    /// Uses pulldown-cmark for spec-compliant CommonMark parsing. This correctly handles:
    /// - Fenced code blocks (``` and ~~~)
    /// - Indented code blocks (4 spaces or tab)
    /// - Code blocks inside lists, blockquotes, and other containers
    /// - Edge cases like backticks in info strings (which invalidate the fence)
    ///
    /// Returns a sorted vector of (start, end) byte offset tuples.
    pub fn detect_code_blocks(content: &str) -> Vec<(usize, usize)> {
        Self::detect_code_blocks_and_spans(content).code_blocks
    }

    /// Returns code block ranges, inline code span ranges, and detailed code block info
    /// in a single pulldown-cmark pass.
    pub fn detect_code_blocks_and_spans(content: &str) -> ParseResult {
        let mut blocks = Vec::new();
        let mut spans = Vec::new();
        let mut details = Vec::new();
        let mut strong_spans = Vec::new();
        let mut html_blocks = Vec::new();
        let mut code_block_start: Option<(usize, bool, String)> = None;

        // Definition structure. `block_stack` holds one entry per open block,
        // the definition's start for a definition and `None` for anything else,
        // so the innermost entry says whether content belongs to a definition
        // directly. Inline tags never touch it.
        //
        // Only a definition passing `opens_definition` counts, and a term
        // waits in `pending_terms` until a definition following it does. Extents
        // are recorded per item rather than per list: the parser's range for a
        // list can run over the paragraph after it, and a list can hold items
        // that do not count between items that do.
        let mut definition_items: Vec<(usize, usize)> = Vec::new();
        let mut pending_terms: Vec<(usize, usize)> = Vec::new();
        let mut definition_terms = Vec::new();
        let mut definition_texts = Vec::new();
        let mut colon_definitions = Vec::new();
        let mut block_stack: Vec<Option<usize>> = Vec::new();
        let mut definition_paragraph: Option<(usize, usize)> = None;
        let mut tight_run: Option<DefinitionTextDetail> = None;

        // List membership tracking for ordered lists
        let mut line_to_list = LineToListMap::new();
        let mut list_start_values = ListStartValues::new();
        let mut list_stack: Vec<(usize, bool, u64)> = Vec::new(); // (list_id, is_ordered, start_value)
        let mut next_list_id: usize = 0;

        // Pre-compute line start offsets for byte-to-line conversion
        let line_starts: Vec<usize> = std::iter::once(0)
            .chain(content.match_indices('\n').map(|(i, _)| i + 1))
            .collect();

        let byte_to_line = |byte_offset: usize| -> usize { line_starts.partition_point(|&start| start <= byte_offset) };

        let options = rumdl_parser_options();
        let parser = Parser::new_ext(content, options).into_offset_iter();

        for (event, range) in parser {
            match &event {
                Event::Start(tag) if !is_inline_tag(tag) => {
                    definition_texts.extend(tight_run.take());
                    if let (Tag::Paragraph, Some(Some(definition_start))) = (tag, block_stack.last()) {
                        definition_paragraph = Some((range.start, *definition_start));
                    }
                    let mut definition_start = None;
                    match tag {
                        Tag::DefinitionList => pending_terms.clear(),
                        Tag::DefinitionListTitle => pending_terms.push((range.start, range.end)),
                        Tag::DefinitionListDefinition => {
                            // The definition's range takes in the blank lines after
                            // it, quoted ones too, which belong to no block.
                            let item_start = pending_terms.first().map_or(range.start, |&(start, _)| start);
                            let item_end = range.start
                                + content[range.clone()]
                                    .trim_end_matches([' ', '\t', '\n', '\r', '>'])
                                    .len();
                            if opens_definition(content, range.start) {
                                definition_items.push((item_start, item_end));
                                definition_terms.append(&mut pending_terms);
                                definition_start = Some(range.start);
                            } else {
                                colon_definitions.push(ColonDefinitionDetail {
                                    start: item_start,
                                    end: item_end,
                                    definition_start: range.start,
                                });
                                pending_terms.clear();
                            }
                        }
                        _ => {}
                    }
                    block_stack.push(definition_start);
                }
                Event::End(tag_end) if !is_inline_tag_end(*tag_end) => {
                    definition_texts.extend(tight_run.take());
                    if let (TagEnd::Paragraph, Some((start, definition_start))) = (tag_end, definition_paragraph.take())
                    {
                        definition_texts.push(DefinitionTextDetail {
                            start,
                            end: range.end,
                            definition_start,
                        });
                    }
                    if let TagEnd::DefinitionList = tag_end {
                        pending_terms.clear();
                    }
                    block_stack.pop();
                }
                // Inline content directly under a definition is a tight
                // definition's text, which the parser does not wrap in a paragraph.
                _ => {
                    if let Some(Some(definition_start)) = block_stack.last() {
                        match &mut tight_run {
                            Some(run) => run.end = run.end.max(range.end),
                            None => {
                                tight_run = Some(DefinitionTextDetail {
                                    start: range.start,
                                    end: range.end,
                                    definition_start: *definition_start,
                                });
                            }
                        }
                    }
                }
            }
            match event {
                Event::Start(Tag::CodeBlock(kind)) => {
                    let (is_fenced, info_string) = match &kind {
                        CodeBlockKind::Fenced(info) => (true, info.to_string()),
                        CodeBlockKind::Indented => (false, String::new()),
                    };
                    code_block_start = Some((range.start, is_fenced, info_string));
                }
                Event::End(TagEnd::CodeBlock) => {
                    if let Some((start, is_fenced, info_string)) = code_block_start.take() {
                        blocks.push((start, range.end));
                        details.push(CodeBlockDetail {
                            start,
                            end: range.end,
                            is_fenced,
                            info_string,
                        });
                    }
                }
                Event::Start(Tag::Strong) => {
                    if range.start + 2 <= content.len() {
                        let is_asterisk = &content[range.start..range.start + 2] == "**";
                        strong_spans.push(StrongSpanDetail {
                            start: range.start,
                            end: range.end,
                            is_asterisk,
                        });
                    }
                }
                Event::Start(Tag::List(start_num)) => {
                    let is_ordered = start_num.is_some();
                    let start_value = start_num.unwrap_or(1);
                    list_stack.push((next_list_id, is_ordered, start_value));
                    if is_ordered {
                        list_start_values.insert(next_list_id, start_value);
                    }
                    next_list_id += 1;
                }
                Event::End(TagEnd::List(_)) => {
                    list_stack.pop();
                }
                Event::Start(Tag::Item) => {
                    if let Some(&(list_id, is_ordered, _)) = list_stack.last()
                        && is_ordered
                    {
                        let line_num = byte_to_line(range.start);
                        line_to_list.insert(line_num, list_id);
                    }
                }
                Event::Start(Tag::HtmlBlock) => {
                    // The start event's range already spans the whole block.
                    html_blocks.push((range.start, range.end));
                }
                Event::Code(_) => {
                    spans.push((range.start, range.end));
                }
                _ => {}
            }
        }

        // Handle edge case: unclosed code block at end of content
        // pulldown-cmark should handle this, but be defensive
        if let Some((start, is_fenced, info_string)) = code_block_start {
            blocks.push((start, content.len()));
            details.push(CodeBlockDetail {
                start,
                end: content.len(),
                is_fenced,
                info_string,
            });
        }

        // A colon-led definition whose text holds the same inline constructs
        // either way is read the same by both, and drops out.
        if !colon_definitions.is_empty() {
            let as_definitions = inline_ranges(content, options);
            let as_text = inline_ranges(content, options - Options::ENABLE_DEFINITION_LIST);
            let mut definition_ranges = InlineRangeCursor::new(&as_definitions);
            let mut text_ranges = InlineRangeCursor::new(&as_text);
            colon_definitions.retain(|detail| {
                !definition_ranges
                    .overlapping(detail)
                    .eq(text_ranges.overlapping(detail))
            });
        }

        // Sort by start position (should already be sorted, but ensure consistency)
        blocks.sort_by_key(|&(start, _)| start);
        spans.sort_by_key(|&(start, _)| start);
        details.sort_by_key(|d| d.start);
        strong_spans.sort_by_key(|s| s.start);
        html_blocks.sort_by_key(|&(start, _)| start);
        ParseResult {
            definition_items,
            definition_terms,
            definition_texts,
            colon_definitions,
            code_blocks: blocks,
            code_spans: spans,
            code_block_details: details,
            strong_spans,
            line_to_list,
            list_start_values,
            html_blocks,
        }
    }

    /// Check if a position is within a code block (for compatibility)
    pub fn is_in_code_block_or_span(blocks: &[(usize, usize)], pos: usize) -> bool {
        Self::is_in_code_block(blocks, pos)
    }

    /// Check if a byte position falls within any of the given sorted, non-overlapping ranges.
    ///
    /// Uses binary search on the sorted block ranges for O(log n) lookup.
    /// The blocks slice must be sorted by start position (as returned by
    /// `detect_code_blocks` and `detect_code_blocks_and_spans`).
    pub fn is_in_code_block(blocks: &[(usize, usize)], pos: usize) -> bool {
        // Binary search: find the last block whose start <= pos
        let idx = blocks.partition_point(|&(start, _)| start <= pos);
        // partition_point returns the first index where start > pos,
        // so the candidate is at idx - 1
        idx > 0 && pos < blocks[idx - 1].1
    }

    /// Analyze code block context relative to list parsing
    /// This is the core function implementing Design #3's three-tier classification
    pub fn analyze_code_block_context(
        lines: &[crate::lint_context::LineInfo],
        line_idx: usize,
        min_continuation_indent: usize,
    ) -> CodeBlockContext {
        if let Some(line_info) = lines.get(line_idx) {
            // Rule 1: Indentation Analysis - Is it sufficiently indented for list continuation?
            if line_info.indent >= min_continuation_indent {
                return CodeBlockContext::Indented;
            }

            // Rule 2: Blank Line Context - Check for structural separation indicators
            let (prev_blanks, next_blanks) = Self::count_surrounding_blank_lines(lines, line_idx);

            // Rule 3: Standalone Detection - Insufficient indentation + blank line separation
            // This is the key fix: root-level code blocks with blank lines separate lists
            if prev_blanks > 0 || next_blanks > 0 {
                return CodeBlockContext::Standalone;
            }

            // Rule 4: Default - Adjacent (conservative, non-breaking for edge cases)
            CodeBlockContext::Adjacent
        } else {
            // Fallback for invalid line index
            CodeBlockContext::Adjacent
        }
    }

    /// Count blank lines before and after the given line index
    fn count_surrounding_blank_lines(lines: &[crate::lint_context::LineInfo], line_idx: usize) -> (usize, usize) {
        let mut prev_blanks = 0;
        let mut next_blanks = 0;

        // Count blank lines before (look backwards)
        for i in (0..line_idx).rev() {
            if let Some(line) = lines.get(i) {
                if line.is_blank {
                    prev_blanks += 1;
                } else {
                    break;
                }
            } else {
                break;
            }
        }

        // Count blank lines after (look forwards)
        for i in (line_idx + 1)..lines.len() {
            if let Some(line) = lines.get(i) {
                if line.is_blank {
                    next_blanks += 1;
                } else {
                    break;
                }
            } else {
                break;
            }
        }

        (prev_blanks, next_blanks)
    }

    /// Calculate minimum indentation required for code block to continue a list
    /// Based on the most recent list item's marker width
    pub fn calculate_min_continuation_indent(
        content: &str,
        lines: &[crate::lint_context::LineInfo],
        current_line_idx: usize,
    ) -> usize {
        // Look backwards to find the most recent list item
        for i in (0..current_line_idx).rev() {
            if let Some(line_info) = lines.get(i) {
                if let Some(list_item) = &line_info.list_item {
                    // Calculate minimum continuation indent for this list item
                    return if list_item.is_ordered {
                        list_item.marker_column + list_item.marker.len() + 1 // +1 for space after marker
                    } else {
                        list_item.marker_column + 2 // Unordered lists need marker + space (min 2)
                    };
                }

                // Stop at structural separators that would break list context
                if line_info.heading.is_some() || Self::is_structural_separator(line_info.content(content)) {
                    break;
                }
            }
        }

        0 // No list context found
    }

    /// Check if content is a structural separator (headings, horizontal rules, etc.)
    fn is_structural_separator(content: &str) -> bool {
        let trimmed = content.trim();
        trimmed.starts_with("---")
            || trimmed.starts_with("***")
            || trimmed.starts_with("___")
            || crate::utils::skip_context::is_table_line(trimmed)
            || trimmed.starts_with('>') // Blockquotes
    }

    /// Detect fenced code blocks with markdown/md language tag.
    ///
    /// Returns a vector of `MarkdownCodeBlock` containing byte ranges for the
    /// content between the fences (excluding the fence lines themselves).
    ///
    /// Only detects fenced code blocks (``` or ~~~), not indented code blocks,
    /// since indented blocks don't have a language tag.
    pub fn detect_markdown_code_blocks(content: &str) -> Vec<MarkdownCodeBlock> {
        use pulldown_cmark::{CodeBlockKind, Event, Parser, Tag, TagEnd};

        let mut blocks = Vec::new();
        let mut current_block: Option<MarkdownCodeBlockBuilder> = None;

        let options = rumdl_parser_options();
        let parser = Parser::new_ext(content, options).into_offset_iter();

        for (event, range) in parser {
            match event {
                Event::Start(Tag::CodeBlock(CodeBlockKind::Fenced(info))) => {
                    // Check if language is markdown or md (first word of info string)
                    let language = info.split_whitespace().next().unwrap_or("");
                    if language.eq_ignore_ascii_case("markdown") || language.eq_ignore_ascii_case("md") {
                        // Find where content starts (after the opening fence line)
                        let block_start = range.start;
                        let content_start = content[block_start..]
                            .find('\n')
                            .map_or(content.len(), |i| block_start + i + 1);

                        current_block = Some(MarkdownCodeBlockBuilder { content_start });
                    }
                }
                Event::End(TagEnd::CodeBlock) => {
                    if let Some(builder) = current_block.take() {
                        // Find where content ends (before the closing fence line)
                        let block_end = range.end;

                        // Validate range before slicing
                        if builder.content_start > block_end || builder.content_start > content.len() {
                            continue;
                        }

                        let search_range = &content[builder.content_start..block_end.min(content.len())];
                        let content_end = search_range
                            .rfind('\n')
                            .map_or(builder.content_start, |i| builder.content_start + i);

                        // Only add block if it has valid content range
                        if content_end >= builder.content_start {
                            blocks.push(MarkdownCodeBlock {
                                content_start: builder.content_start,
                                content_end,
                            });
                        }
                    }
                }
                _ => {}
            }
        }

        blocks
    }
}

/// Information about a markdown code block for recursive formatting
#[derive(Debug, Clone)]
pub struct MarkdownCodeBlock {
    /// Byte offset where the content starts (after opening fence line)
    pub content_start: usize,
    /// Byte offset where the content ends (before closing fence line)
    pub content_end: usize,
}

/// Builder for MarkdownCodeBlock during parsing
struct MarkdownCodeBlockBuilder {
    content_start: usize,
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_detect_fenced_code_blocks() {
        // The function detects fenced blocks and inline code spans
        // Fence markers (``` at line start) are now skipped in inline span detection

        // Basic fenced code block with backticks
        let content = "Some text\n```\ncode here\n```\nMore text";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // Should find: 1 fenced block (fences are no longer detected as inline spans)
        assert_eq!(blocks.len(), 1);

        // Check that we have the fenced block
        let fenced_block = blocks
            .iter()
            .find(|(start, end)| end - start > 10 && content[*start..*end].contains("code here"));
        assert!(fenced_block.is_some());

        // Fenced code block with tildes (no inline code detection for ~)
        let content = "Some text\n~~~\ncode here\n~~~\nMore text";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 1);
        assert_eq!(&content[blocks[0].0..blocks[0].1], "~~~\ncode here\n~~~");

        // Multiple code blocks
        let content = "Text\n```\ncode1\n```\nMiddle\n~~~\ncode2\n~~~\nEnd";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // 2 fenced blocks (fence markers no longer detected as inline spans)
        assert_eq!(blocks.len(), 2);
    }

    #[test]
    fn test_detect_code_blocks_with_language() {
        // Code block with language identifier
        let content = "Text\n```rust\nfn main() {}\n```\nMore";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // 1 fenced block (fence markers no longer detected as inline spans)
        assert_eq!(blocks.len(), 1);
        // Check we have the full fenced block
        let fenced = blocks.iter().find(|(s, e)| content[*s..*e].contains("fn main"));
        assert!(fenced.is_some());
    }

    #[test]
    fn test_unclosed_code_block() {
        // Unclosed code block should extend to end of content
        let content = "Text\n```\ncode here\nno closing fence";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 1);
        assert_eq!(blocks[0].1, content.len());
    }

    #[test]
    fn test_indented_code_blocks() {
        // Basic indented code block
        let content = "Paragraph\n\n    code line 1\n    code line 2\n\nMore text";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 1);
        assert!(content[blocks[0].0..blocks[0].1].contains("code line 1"));
        assert!(content[blocks[0].0..blocks[0].1].contains("code line 2"));

        // Indented code with tabs
        let content = "Paragraph\n\n\tcode with tab\n\tanother line\n\nText";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 1);
    }

    #[test]
    fn test_indented_code_requires_blank_line() {
        // Indented lines without preceding blank line are not code blocks
        let content = "Paragraph\n    indented but not code\nMore text";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 0);

        // With blank line, it becomes a code block
        let content = "Paragraph\n\n    now it's code\nMore text";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 1);
    }

    #[test]
    fn test_indented_content_with_list_markers_is_code_block() {
        // Per CommonMark spec: 4-space indented content after blank line IS a code block,
        // even if the content looks like list markers. The indentation takes precedence.
        // Verified with: echo 'List:\n\n    - Item 1' | npx commonmark
        // Output: <pre><code>- Item 1</code></pre>
        let content = "List:\n\n    - Item 1\n    - Item 2\n    * Item 3\n    + Item 4";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 1); // This IS a code block per spec

        // Same for numbered list markers
        let content = "List:\n\n    1. First\n    2. Second";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 1); // This IS a code block per spec
    }

    #[test]
    fn test_actual_list_items_not_code_blocks() {
        // Actual list items (no preceding blank line + 4 spaces) are NOT code blocks
        let content = "- Item 1\n- Item 2\n* Item 3";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 0);

        // Nested list items
        let content = "- Item 1\n  - Nested item\n- Item 2";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 0);
    }

    #[test]
    fn test_inline_code_spans_not_detected() {
        // Inline code spans should NOT be detected as code blocks
        let content = "Text with `inline code` here";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 0); // No blocks, only inline spans

        // Multiple backtick code span
        let content = "Text with ``code with ` backtick`` here";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 0); // No blocks, only inline spans

        // Multiple code spans
        let content = "Has `code1` and `code2` spans";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 0); // No blocks, only inline spans
    }

    #[test]
    fn test_unclosed_code_span() {
        // Unclosed code span should not be detected
        let content = "Text with `unclosed code span";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 0);

        // Mismatched backticks
        let content = "Text with ``one style` different close";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 0);
    }

    #[test]
    fn test_mixed_code_blocks_and_spans() {
        let content = "Has `span1` text\n```\nblock\n```\nand `span2`";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // Should only detect the fenced block, NOT the inline spans
        assert_eq!(blocks.len(), 1);

        // Check we have the fenced block only
        assert!(blocks.iter().any(|(s, e)| content[*s..*e].contains("block")));
        // Should NOT detect inline spans
        assert!(!blocks.iter().any(|(s, e)| &content[*s..*e] == "`span1`"));
        assert!(!blocks.iter().any(|(s, e)| &content[*s..*e] == "`span2`"));
    }

    #[test]
    fn test_is_in_code_block_or_span() {
        let blocks = vec![(10, 20), (30, 40), (50, 60)];

        // Test positions inside blocks
        assert!(CodeBlockUtils::is_in_code_block_or_span(&blocks, 15));
        assert!(CodeBlockUtils::is_in_code_block_or_span(&blocks, 35));
        assert!(CodeBlockUtils::is_in_code_block_or_span(&blocks, 55));

        // Test positions at boundaries
        assert!(CodeBlockUtils::is_in_code_block_or_span(&blocks, 10)); // Start is inclusive
        assert!(!CodeBlockUtils::is_in_code_block_or_span(&blocks, 20)); // End is exclusive

        // Test positions outside blocks
        assert!(!CodeBlockUtils::is_in_code_block_or_span(&blocks, 5));
        assert!(!CodeBlockUtils::is_in_code_block_or_span(&blocks, 25));
        assert!(!CodeBlockUtils::is_in_code_block_or_span(&blocks, 65));
    }

    #[test]
    fn test_empty_content() {
        let blocks = CodeBlockUtils::detect_code_blocks("");
        assert_eq!(blocks.len(), 0);
    }

    #[test]
    fn test_code_block_at_start() {
        let content = "```\ncode\n```\nText after";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // 1 fenced block (fence markers no longer detected as inline spans)
        assert_eq!(blocks.len(), 1);
        assert_eq!(blocks[0].0, 0); // Fenced block starts at 0
    }

    #[test]
    fn test_code_block_at_end() {
        let content = "Text before\n```\ncode\n```";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // 1 fenced block (fence markers no longer detected as inline spans)
        assert_eq!(blocks.len(), 1);
        // Check we have the fenced block
        let fenced = blocks.iter().find(|(s, e)| content[*s..*e].contains("code"));
        assert!(fenced.is_some());
    }

    #[test]
    fn test_nested_fence_markers() {
        // Code block containing fence markers as content
        let content = "Text\n````\n```\nnested\n```\n````\nAfter";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // Should detect: outer block, inner ```, outer ````
        assert!(!blocks.is_empty());
        // Check we have the outer block
        let outer = blocks.iter().find(|(s, e)| content[*s..*e].contains("nested"));
        assert!(outer.is_some());
    }

    #[test]
    fn test_indented_code_with_blank_lines() {
        // Indented code blocks can contain blank lines
        let content = "Text\n\n    line1\n\n    line2\n\nAfter";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // May have multiple blocks due to blank line handling
        assert!(!blocks.is_empty());
        // Check that we captured the indented code
        let all_content: String = blocks
            .iter()
            .map(|(s, e)| &content[*s..*e])
            .collect::<Vec<_>>()
            .join("");
        assert!(all_content.contains("line1") || content[blocks[0].0..blocks[0].1].contains("line1"));
    }

    #[test]
    fn test_code_span_with_spaces() {
        // Code spans should NOT be detected as code blocks
        let content = "Text ` code with spaces ` more";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 0); // No blocks, only inline span
    }

    #[test]
    fn test_fenced_block_with_info_string() {
        // Fenced code blocks with complex info strings
        let content = "```rust,no_run,should_panic\ncode\n```";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // 1 fenced block (fence markers no longer detected as inline spans)
        assert_eq!(blocks.len(), 1);
        assert_eq!(blocks[0].0, 0);
    }

    #[test]
    fn test_indented_fences_not_code_blocks() {
        // Indented fence markers should still work as fences
        let content = "Text\n  ```\n  code\n  ```\nAfter";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // Only 1 fenced block (indented fences still work)
        assert_eq!(blocks.len(), 1);
    }

    // Issue #175: Backticks in info string invalidate the fence
    #[test]
    fn test_backticks_in_info_string_not_code_block() {
        // Per CommonMark spec: "If the info string comes after a backtick fence,
        // it may not contain any backtick characters."
        // So ```something``` is NOT a valid fence - the backticks are treated as inline code.
        // Verified with: echo '```something```' | npx commonmark
        // Output: <p><code>something</code></p>
        let content = "```something```\n\n```bash\n# comment\n```";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // Should find only the valid ```bash block, NOT the invalid ```something```
        assert_eq!(blocks.len(), 1);
        // The valid block should contain "# comment"
        assert!(content[blocks[0].0..blocks[0].1].contains("# comment"));
    }

    #[test]
    fn test_issue_175_reproduction() {
        // Full reproduction of issue #175
        let content = "```something```\n\n```bash\n# Have a parrot\necho \"🦜\"\n```";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // Only the bash block is a code block
        assert_eq!(blocks.len(), 1);
        assert!(content[blocks[0].0..blocks[0].1].contains("Have a parrot"));
    }

    #[test]
    fn test_tilde_fence_allows_tildes_in_info_string() {
        // Tilde fences CAN have tildes in info string (only backtick restriction exists)
        // ~~~abc~~~ opens an unclosed code block with info string "abc~~~"
        let content = "~~~abc~~~\ncode content\n~~~";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // This is a valid tilde fence that opens and closes
        assert_eq!(blocks.len(), 1);
    }

    #[test]
    fn test_nested_longer_fence_contains_shorter() {
        // Longer fence (````) can contain shorter fence (```) as content
        let content = "````\n```\nnested content\n```\n````";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 1);
        assert!(content[blocks[0].0..blocks[0].1].contains("nested content"));
    }

    #[test]
    fn test_mixed_fence_types() {
        // Tilde fence contains backtick markers as content
        let content = "~~~\n```\nmixed content\n~~~";
        let blocks = CodeBlockUtils::detect_code_blocks(content);
        assert_eq!(blocks.len(), 1);
        assert!(content[blocks[0].0..blocks[0].1].contains("mixed content"));
    }

    #[test]
    fn test_indented_code_in_list_issue_276() {
        // Issue #276: Indented code block inside a list should be detected by pulldown-cmark
        let content = r#"1. First item
2. Second item with code:

        # This is a code block in a list
        print("Hello, world!")

4. Third item"#;

        let blocks = CodeBlockUtils::detect_code_blocks(content);
        // pulldown-cmark SHOULD detect this indented code block inside the list
        assert!(!blocks.is_empty(), "Should detect indented code block inside list");

        // Verify the detected block contains our code
        let all_content: String = blocks
            .iter()
            .map(|(s, e)| &content[*s..*e])
            .collect::<Vec<_>>()
            .join("");
        assert!(
            all_content.contains("code block in a list") || all_content.contains("print"),
            "Detected block should contain the code content: {all_content:?}"
        );
    }

    #[test]
    fn test_detect_markdown_code_blocks() {
        let content = r#"# Example

```markdown
# Heading
Content here
```

```md
Another heading
More content
```

```rust
// Not markdown
fn main() {}
```
"#;

        let blocks = CodeBlockUtils::detect_markdown_code_blocks(content);

        // Should detect 2 blocks (markdown and md, not rust)
        assert_eq!(
            blocks.len(),
            2,
            "Should detect exactly 2 markdown blocks, got {blocks:?}"
        );

        // First block should be the ```markdown block
        let first = &blocks[0];
        let first_content = &content[first.content_start..first.content_end];
        assert!(
            first_content.contains("# Heading"),
            "First block should contain '# Heading', got: {first_content:?}"
        );

        // Second block should be the ```md block
        let second = &blocks[1];
        let second_content = &content[second.content_start..second.content_end];
        assert!(
            second_content.contains("Another heading"),
            "Second block should contain 'Another heading', got: {second_content:?}"
        );
    }

    #[test]
    fn test_detect_markdown_code_blocks_empty() {
        let content = "# Just a heading\n\nNo code blocks here\n";
        let blocks = CodeBlockUtils::detect_markdown_code_blocks(content);
        assert_eq!(blocks.len(), 0);
    }

    #[test]
    fn test_detect_markdown_code_blocks_case_insensitive() {
        let content = "```MARKDOWN\nContent\n```\n";
        let blocks = CodeBlockUtils::detect_markdown_code_blocks(content);
        assert_eq!(blocks.len(), 1);
    }

    #[test]
    fn test_detect_markdown_code_blocks_at_eof_no_trailing_newline() {
        // Block at end of file without trailing newline after closing fence
        let content = "# Doc\n\n```markdown\nContent\n```";
        let blocks = CodeBlockUtils::detect_markdown_code_blocks(content);
        assert_eq!(blocks.len(), 1);
        // Content should be extractable without panic
        let block_content = &content[blocks[0].content_start..blocks[0].content_end];
        assert!(block_content.contains("Content"));
    }

    #[test]
    fn test_detect_markdown_code_blocks_single_line_content() {
        // Single line of content, no extra newlines
        let content = "```markdown\nX\n```\n";
        let blocks = CodeBlockUtils::detect_markdown_code_blocks(content);
        assert_eq!(blocks.len(), 1);
        let block_content = &content[blocks[0].content_start..blocks[0].content_end];
        assert_eq!(block_content, "X");
    }

    #[test]
    fn test_detect_markdown_code_blocks_empty_content() {
        // Block with no content between fences
        let content = "```markdown\n```\n";
        let blocks = CodeBlockUtils::detect_markdown_code_blocks(content);
        // Should detect block but with empty range or not at all
        // Either behavior is acceptable as long as no panic
        if !blocks.is_empty() {
            // If detected, content range should be valid
            assert!(blocks[0].content_start <= blocks[0].content_end);
        }
    }

    #[test]
    fn test_detect_markdown_code_blocks_validates_ranges() {
        // Ensure no panic on various edge cases
        let test_cases = [
            "",                             // Empty content
            "```markdown",                  // Unclosed block
            "```markdown\n",                // Unclosed block with newline
            "```\n```",                     // Non-markdown block
            "```markdown\n```",             // Empty markdown block
            "   ```markdown\n   X\n   ```", // Indented block
        ];

        for content in test_cases {
            // Should not panic
            let blocks = CodeBlockUtils::detect_markdown_code_blocks(content);
            // All detected blocks should have valid ranges
            for block in &blocks {
                assert!(
                    block.content_start <= block.content_end,
                    "Invalid range in content: {content:?}"
                );
                assert!(
                    block.content_end <= content.len(),
                    "Range exceeds content length in: {content:?}"
                );
            }
        }
    }

    // ── is_in_code_block binary search tests ─────────────────────────────

    #[test]
    fn test_is_in_code_block_empty_blocks() {
        assert!(!CodeBlockUtils::is_in_code_block(&[], 0));
        assert!(!CodeBlockUtils::is_in_code_block(&[], 100));
        assert!(!CodeBlockUtils::is_in_code_block(&[], usize::MAX));
    }

    #[test]
    fn test_is_in_code_block_single_range() {
        let blocks = [(10, 20)];
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 0));
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 9));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 10));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 15));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 19));
        // end is exclusive
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 20));
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 21));
    }

    #[test]
    fn test_is_in_code_block_multiple_ranges() {
        let blocks = [(5, 10), (20, 30), (50, 60)];
        // Before all
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 0));
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 4));
        // In first
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 5));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 9));
        // Gap between first and second
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 10));
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 15));
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 19));
        // In second
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 20));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 29));
        // Gap between second and third
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 30));
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 49));
        // In third
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 50));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 59));
        // After all
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 60));
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 1000));
    }

    #[test]
    fn test_is_in_code_block_adjacent_ranges() {
        // Ranges that are exactly adjacent (end of one == start of next)
        let blocks = [(0, 10), (10, 20), (20, 30)];
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 0));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 9));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 10));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 19));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 20));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 29));
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 30));
    }

    #[test]
    fn test_is_in_code_block_single_byte_range() {
        let blocks = [(5, 6)];
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 4));
        assert!(CodeBlockUtils::is_in_code_block(&blocks, 5));
        assert!(!CodeBlockUtils::is_in_code_block(&blocks, 6));
    }

    #[test]
    fn test_is_in_code_block_matches_linear_scan() {
        // Verify binary search produces identical results to linear scan
        // for a realistic document layout
        let content = "# Heading\n\n```rust\nlet x = 1;\nlet y = 2;\n```\n\nSome text\n\n```\nmore code\n```\n\nEnd\n";
        let blocks = CodeBlockUtils::detect_code_blocks(content);

        for pos in 0..content.len() {
            let binary = CodeBlockUtils::is_in_code_block(&blocks, pos);
            let linear = blocks.iter().any(|&(s, e)| pos >= s && pos < e);
            assert_eq!(
                binary, linear,
                "Mismatch at pos {pos}: binary={binary}, linear={linear}, blocks={blocks:?}"
            );
        }
    }

    #[test]
    fn test_is_in_code_block_at_range_boundaries() {
        // Exhaustive boundary testing for every block start/end
        let blocks = [(100, 200), (300, 400), (500, 600)];
        for &(start, end) in &blocks {
            assert!(
                !CodeBlockUtils::is_in_code_block(&blocks, start - 1),
                "pos={} should be outside",
                start - 1
            );
            assert!(
                CodeBlockUtils::is_in_code_block(&blocks, start),
                "pos={start} should be inside"
            );
            assert!(
                CodeBlockUtils::is_in_code_block(&blocks, end - 1),
                "pos={} should be inside",
                end - 1
            );
            assert!(
                !CodeBlockUtils::is_in_code_block(&blocks, end),
                "pos={end} should be outside"
            );
        }
    }
}