rskim-core 2.8.0

//! Minimal mode transformation
//!
//! ARCHITECTURE: Strip non-doc comments at module/class level while keeping all code intact.
//! Preserves doc comments, comments inside function bodies, and shebangs.
//!
//! Token reduction target: 15-30%

use crate::transform::utils::is_inside_function_body;
use crate::{Language, Result, SkimError, TransformConfig};
use tree_sitter::{Node, Tree};

/// Maximum AST recursion depth to prevent stack overflow attacks
pub(crate) const MAX_AST_DEPTH: usize = 500;

/// Maximum number of AST nodes to prevent memory exhaustion
pub(crate) const MAX_AST_NODES: usize = 100_000;

/// Bundled parameters for the recursive comment walker to avoid parameter explosion
pub(crate) struct CommentWalkContext<'a> {
    pub(crate) ranges: &'a mut Vec<(usize, usize)>,
    pub(crate) node_count: &'a mut usize,
}

/// Transform source by stripping non-doc comments and normalizing blank lines
///
/// Two-pass algorithm:
/// 1. Walk AST collecting byte ranges of non-doc comment nodes to remove
///    (skip doc comments, skip comments inside function bodies, skip shebangs)
/// 2. Adjust ranges for full-line removal, then remove from source
/// 3. Trim trailing whitespace and normalize blank lines (3+ consecutive -> 2)
pub(crate) fn transform_minimal(
    source: &str,
    tree: &Tree,
    language: Language,
    _config: &TransformConfig,
) -> Result<String> {
    let mut ranges_to_remove: Vec<(usize, usize)> = Vec::new();
    let mut node_count: usize = 0;
    let mut ctx = CommentWalkContext {
        ranges: &mut ranges_to_remove,
        node_count: &mut node_count,
    };
    collect_removable_comments(tree.root_node(), source, language, &mut ctx, 0)?;

    // Adjust ranges for full-line removal, sort, and dedup
    let mut final_ranges: Vec<(usize, usize)> = ctx
        .ranges
        .iter()
        .map(|&(start, end)| adjust_range_for_line_removal(source, start, end))
        .collect();
    final_ranges.sort_unstable_by_key(|&(start, _)| start);
    final_ranges.dedup();

    let after_removal = remove_ranges(source, &final_ranges)?;
    let normalized = trim_and_normalize(&after_removal);

    Ok(normalized)
}

/// Recursively collect byte ranges of comment nodes that should be removed
///
/// Collects raw (unadjusted) byte ranges. Line-level adjustment is applied
/// by the caller after collection, matching the pattern used by pseudo.rs.
///
/// # Security
/// - Enforces MAX_AST_DEPTH to prevent stack overflow
/// - Enforces MAX_AST_NODES to prevent memory exhaustion
pub(crate) fn collect_removable_comments(
    node: Node,
    source: &str,
    language: Language,
    ctx: &mut CommentWalkContext<'_>,
    depth: usize,
) -> Result<()> {
    // SECURITY: Prevent stack overflow from deeply nested AST
    if depth > MAX_AST_DEPTH {
        return Err(SkimError::ParseError(format!(
            "Maximum AST depth exceeded: {} (possible malicious input)",
            MAX_AST_DEPTH
        )));
    }

    // SECURITY: Prevent memory exhaustion from excessive nodes
    *ctx.node_count += 1;
    if *ctx.node_count > MAX_AST_NODES {
        return Err(SkimError::ParseError(format!(
            "Too many AST nodes: {} (max: {}). Possible malicious input.",
            *ctx.node_count, MAX_AST_NODES
        )));
    }

    if is_removable_comment(node, source, language) {
        ctx.ranges.push((node.start_byte(), node.end_byte()));
    }

    let mut cursor = node.walk();
    for child in node.children(&mut cursor) {
        collect_removable_comments(child, source, language, ctx, depth + 1)?;
    }

    Ok(())
}

/// Check if a comment node is a shebang line (e.g., `#!/usr/bin/env python3`)
///
/// Shebangs must start at byte 0 and begin with `#!`.
fn is_shebang(node: Node, source: &str) -> bool {
    if node.start_byte() != 0 {
        return false;
    }
    node.utf8_text(source.as_bytes())
        .map(|text| text.starts_with("#!"))
        .unwrap_or(false)
}

/// Check if a node kind represents a comment in the given language
pub(crate) fn is_comment_node(kind: &str, language: Language) -> bool {
    match language {
        Language::TypeScript
        | Language::JavaScript
        | Language::Python
        | Language::Go
        | Language::C
        | Language::Cpp
        | Language::CSharp
        | Language::Ruby
        | Language::Sql => kind == "comment",
        Language::Rust | Language::Java | Language::Kotlin => {
            kind == "line_comment" || kind == "block_comment"
        }
        Language::Swift => kind == "comment" || kind == "multiline_comment",
        // Markdown, JSON, YAML, TOML don't have comment nodes to strip
        Language::Markdown | Language::Json | Language::Yaml | Language::Toml => false,
    }
}

/// Check if a comment node should be removed (not a doc comment, shebang, or in-body comment)
///
/// Combines `is_comment_node` with doc-comment filtering, shebang detection, and
/// function-body detection. Returns true if the node is a comment that should be stripped.
///
/// Used by both minimal mode (via `collect_removable_comments`) and pseudo mode
/// (inlined into `collect_noise_ranges` for single-pass processing).
pub(crate) fn is_removable_comment(node: Node, source: &str, language: Language) -> bool {
    if !is_comment_node(node.kind(), language) {
        return false;
    }
    let should_preserve = is_shebang(node, source)
        || is_inside_function_body(node, language)
        || is_doc_comment(node, source, language);
    !should_preserve
}

/// Check if a comment node is a doc comment that should be preserved
///
/// Language-specific doc comment detection. See match arms below for
/// per-language rules covering all supported tree-sitter languages.
fn is_doc_comment(node: Node, source: &str, language: Language) -> bool {
    let text = match node.utf8_text(source.as_bytes()) {
        Ok(t) => t,
        Err(_) => return false,
    };

    match language {
        Language::TypeScript | Language::JavaScript => {
            // JSDoc comments start with /**
            text.starts_with("/**")
        }
        Language::Python => {
            // Python docstrings are expression_statement > string nodes, NOT comment nodes.
            // All Python `comment` nodes (starting with #) at module level are regular comments.
            false
        }
        Language::Rust => {
            // Rust doc comments: ///, //!, /**, /*!
            text.starts_with("///")
                || text.starts_with("//!")
                || text.starts_with("/**")
                || text.starts_with("/*!")
        }
        Language::Go => {
            // Go doc comments are comments that are adjacent to a declaration.
            // Walk forward through siblings to find next non-comment named sibling.
            is_go_doc_comment(node, source)
        }
        Language::Java => {
            // Javadoc comments start with /**
            text.starts_with("/**")
        }
        Language::C | Language::Cpp => {
            // Doxygen comments: /** or ///
            text.starts_with("/**") || text.starts_with("///")
        }
        Language::CSharp => {
            // C# XML doc comments: ///
            text.starts_with("///") || text.starts_with("/**")
        }
        Language::Ruby => {
            // Ruby doesn't have doc comments — all `#` comments are regular.
            // RDoc and YARD conventions use `#` but there's no syntactic distinction.
            false
        }
        Language::Kotlin => {
            // Kotlin doc comments: /** */ (KDoc)
            text.starts_with("/**")
        }
        Language::Swift => {
            // Swift doc comments: /// or /** */
            text.starts_with("///") || text.starts_with("/**")
        }
        Language::Sql => {
            // SQL `--` comments have no doc comment convention
            false
        }
        // Markdown, JSON, YAML, TOML don't reach here
        Language::Markdown | Language::Json | Language::Yaml | Language::Toml => false,
    }
}

/// Check if a Go comment is a doc comment (adjacent to a declaration)
///
/// Go doc comments are comments that immediately precede a declaration
/// (function, type, var, const) with no blank lines between them.
/// Walks forward through siblings to find the end of the contiguous comment
/// block and checks whether a declaration immediately follows.
fn is_go_doc_comment(node: Node, source: &str) -> bool {
    let mut current_end = node.end_byte();
    let mut sibling = node.next_named_sibling();

    while let Some(sib) = sibling {
        let sib_start = sib.start_byte();

        if current_end <= sib_start && sib_start <= source.len() {
            let between = &source[current_end..sib_start];
            let newline_count = between.chars().filter(|&c| c == '\n').count();
            if newline_count > 1 {
                return false;
            }
        }

        if is_comment_node(sib.kind(), Language::Go) {
            current_end = sib.end_byte();
            sibling = sib.next_named_sibling();
            continue;
        }

        return is_go_declaration(sib.kind());
    }

    false
}

/// Check if a Go node kind is a declaration type
fn is_go_declaration(kind: &str) -> bool {
    matches!(
        kind,
        "function_declaration"
            | "method_declaration"
            | "type_declaration"
            | "var_declaration"
            | "const_declaration"
            | "type_spec"
    )
}

/// Adjust a range to remove the entire line if the range is the only
/// non-whitespace content on that line.
///
/// If the range occupies the full line (only whitespace before/after on same line),
/// remove the entire line including the newline. Otherwise, just remove the range
/// and any leading whitespace before it on the same line (for inline trailing content).
///
/// Used by both minimal mode (comment removal) and pseudo mode (noise removal).
pub(crate) fn adjust_range_for_line_removal(
    source: &str,
    start: usize,
    end: usize,
) -> (usize, usize) {
    // Find the start of the line containing this range
    let line_start = source[..start].rfind('\n').map(|pos| pos + 1).unwrap_or(0);

    // Find the end of the line containing this range
    let line_end = source[end..]
        .find('\n')
        .map(|pos| end + pos + 1)
        .unwrap_or(source.len());

    // Check if the range is the only non-whitespace content on the line
    let before_range = &source[line_start..start];
    let after_range = if end < line_end {
        let after_end = if line_end > 0 && source.as_bytes().get(line_end - 1) == Some(&b'\n') {
            line_end - 1
        } else {
            line_end
        };
        &source[end..after_end]
    } else {
        ""
    };

    let only_whitespace_before = before_range.chars().all(|c| c.is_whitespace());
    let only_whitespace_after = after_range.chars().all(|c| c.is_whitespace());

    if only_whitespace_before && only_whitespace_after {
        // Range is the only content on this line - remove the entire line
        (line_start, line_end)
    } else if only_whitespace_after {
        // Inline trailing range: remove leading whitespace before the range too
        let trimmed_start = source[line_start..start].trim_end().len() + line_start;
        (trimmed_start, end)
    } else {
        // Range is in the middle or start of a line with other content - just remove the range
        (start, end)
    }
}

/// Remove collected byte ranges from source
///
/// Builds a new string by copying everything except the removed ranges.
pub(crate) fn remove_ranges(source: &str, ranges: &[(usize, usize)]) -> Result<String> {
    if ranges.is_empty() {
        return Ok(source.to_string());
    }

    let mut result = String::with_capacity(source.len());
    let mut last_pos = 0;

    for &(start, end) in ranges {
        if end < start {
            return Err(SkimError::ParseError(format!(
                "Invalid range: start={} end={}",
                start, end
            )));
        }
        if end > source.len() {
            return Err(SkimError::ParseError(format!(
                "Range exceeds source length: end={} len={}",
                end,
                source.len()
            )));
        }

        // Skip overlapping ranges, extending the removal window if needed
        if start < last_pos {
            last_pos = last_pos.max(end);
            continue;
        }

        if !source.is_char_boundary(start) || !source.is_char_boundary(end) {
            return Err(SkimError::ParseError(format!(
                "Invalid UTF-8 boundary at range [{}, {})",
                start, end
            )));
        }

        result.push_str(&source[last_pos..start]);
        last_pos = end;
    }

    if !source.is_char_boundary(last_pos) {
        return Err(SkimError::ParseError(format!(
            "Invalid UTF-8 boundary at position {}",
            last_pos
        )));
    }

    result.push_str(&source[last_pos..]);

    Ok(result)
}

/// Trim trailing whitespace from each line and normalize blank lines in a single pass
///
/// Combines two operations to avoid an extra allocation:
/// 1. Trims trailing whitespace from each line
/// 2. Normalizes blank lines: 3+ consecutive blank lines become 2
pub(crate) fn trim_and_normalize(source: &str) -> String {
    let mut result = String::with_capacity(source.len());
    let mut consecutive_blanks: usize = 0;

    for line in source.lines() {
        let trimmed = line.trim_end();
        if trimmed.is_empty() {
            consecutive_blanks += 1;
            if consecutive_blanks > 2 {
                continue;
            }
        } else {
            consecutive_blanks = 0;
        }

        if !result.is_empty() {
            result.push('\n');
        }
        result.push_str(trimmed);
    }

    if source.ends_with('\n') {
        result.push('\n');
    }

    result
}

#[cfg(test)]
#[allow(clippy::unwrap_used)] // Unwrapping is acceptable in tests
mod tests {
    use super::*;

    #[test]
    fn test_trim_and_normalize_preserves_two_blanks() {
        let input = "a\n\n\nb\n";
        let result = trim_and_normalize(input);
        assert_eq!(result, "a\n\n\nb\n");
    }

    #[test]
    fn test_trim_and_normalize_reduces_four_blanks_to_two() {
        let input = "a\n\n\n\n\nb\n";
        let result = trim_and_normalize(input);
        assert_eq!(result, "a\n\n\nb\n");
    }

    #[test]
    fn test_trim_and_normalize_no_change_needed() {
        let input = "a\n\nb\n";
        let result = trim_and_normalize(input);
        assert_eq!(result, "a\n\nb\n");
    }

    #[test]
    fn test_trim_and_normalize_trims_trailing_whitespace() {
        let input = "hello   \nworld  \n";
        let result = trim_and_normalize(input);
        assert_eq!(result, "hello\nworld\n");
    }

    #[test]
    fn test_trim_and_normalize_combined() {
        // Verify both trimming and normalization happen in one pass
        let input = "hello   \n\n\n\n\nworld  \n";
        let result = trim_and_normalize(input);
        assert_eq!(result, "hello\n\n\nworld\n");
    }

    #[test]
    fn test_adjust_range_full_line_comment() {
        let source = "code\n// comment\nmore code\n";
        // "// comment" starts at byte 5, ends at byte 15
        let (start, end) = adjust_range_for_line_removal(source, 5, 15);
        // Should remove the entire line including newline
        assert_eq!(start, 5);
        assert_eq!(end, 16); // includes the newline
    }

    // ========================================================================
    // Issue 5: adjust_range_for_line_removal trailing/inline comment branches
    // ========================================================================

    #[test]
    fn test_adjust_range_trailing_comment() {
        let source = "let x = 1; // trailing\nmore code\n";
        // "// trailing" starts at byte 11, ends at byte 22
        let (start, end) = adjust_range_for_line_removal(source, 11, 22);
        // Should remove " // trailing" (the trailing whitespace + comment) but keep "let x = 1;"
        // The function trims whitespace before the comment on the same line
        assert!(start <= 11, "start should be at or before comment start");
        assert_eq!(end, 22);
        // Verify the remaining text makes sense
        let remaining = format!("{}{}", &source[..start], &source[end..]);
        assert!(
            remaining.starts_with("let x = 1;"),
            "should preserve code before trailing comment, got: {:?}",
            remaining
        );
    }

    #[test]
    fn test_adjust_range_inline_comment_with_code_after() {
        // Comment at start of line with code after it -- the "middle" branch
        let source = "/* comment */ let x = 1;\n";
        // "/* comment */" starts at byte 0, ends at byte 13
        let (start, end) = adjust_range_for_line_removal(source, 0, 13);
        // There is non-whitespace after the comment, so just remove the comment itself
        assert_eq!(start, 0);
        assert_eq!(end, 13);
    }

    // ========================================================================
    // Issue 4: remove_ranges error-path tests
    // ========================================================================

    #[test]
    fn test_remove_ranges_end_before_start() {
        let source = "hello world";
        let ranges = vec![(5, 3)]; // end < start
        let result = remove_ranges(source, &ranges);
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("Invalid range"),
            "Expected 'Invalid range' error, got: {}",
            err_msg
        );
    }

    #[test]
    fn test_remove_ranges_end_exceeds_source_length() {
        let source = "hello";
        let ranges = vec![(0, 100)]; // end > source.len()
        let result = remove_ranges(source, &ranges);
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("Range exceeds source length"),
            "Expected 'Range exceeds source length' error, got: {}",
            err_msg
        );
    }

    #[test]
    fn test_remove_ranges_non_char_boundary() {
        // Multi-byte UTF-8 character: the euro sign takes 3 bytes
        let source = "a\u{20AC}b"; // "a" + euro sign (3 bytes) + "b" = 5 bytes total
                                   // Byte 2 is in the middle of the euro sign (bytes 1..4)
        let ranges = vec![(2, 4)];
        let result = remove_ranges(source, &ranges);
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("Invalid UTF-8 boundary"),
            "Expected 'Invalid UTF-8 boundary' error, got: {}",
            err_msg
        );
    }

    #[test]
    fn test_remove_ranges_empty_ranges() {
        let source = "hello world";
        let ranges = vec![];
        let result = remove_ranges(source, &ranges).unwrap();
        assert_eq!(result, "hello world");
    }

    #[test]
    fn test_remove_ranges_valid_removal() {
        let source = "hello beautiful world";
        let ranges = vec![(5, 15)]; // remove " beautiful"
        let result = remove_ranges(source, &ranges).unwrap();
        assert_eq!(result, "hello world");
    }

    // ========================================================================
    // Issue 3: Security limit error-path tests
    // ========================================================================

    #[test]
    fn test_max_ast_nodes_limit() {
        // Generate Python source with many expressions to exceed MAX_AST_NODES (100,000).
        // Each line `x = 0 + 1 + 2 + ... + 19` generates ~25 AST nodes (identifiers,
        // operators, integers, expression_statement wrappers), so 4500 lines is enough.
        let mut source = String::new();
        for i in 0..4500 {
            source.push_str("x = ");
            for j in 0..20 {
                if j > 0 {
                    source.push_str(" + ");
                }
                source.push_str(&(i * 20 + j).to_string());
            }
            source.push('\n');
        }

        let mut parser = crate::Parser::new(Language::Python).unwrap();
        let tree = parser.parse(&source).unwrap();
        let config = TransformConfig::default();

        let result = transform_minimal(&source, &tree, Language::Python, &config);
        assert!(
            result.is_err(),
            "Expected error when exceeding MAX_AST_NODES"
        );
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("Too many AST nodes"),
            "Expected 'Too many AST nodes' error, got: {}",
            err_msg
        );
    }

    // NOTE: MAX_AST_DEPTH (500) is not tested because tree-sitter grammars impose
    // their own nesting limits that are well below 500 levels. Even deeply nested
    // expressions like `(((((...))))` do not produce 500 levels of AST depth in
    // practice. The depth guard exists as a defense-in-depth measure against
    // hypothetical malicious grammars or future grammar changes.
}