xbp 10.38.2

//! Walk a codebase and extract TODO/FIXME-style markers.

use crate::commands::text_util::normalize_whitespace;
use crate::utils::xbp_ignore::{XbpIgnoreSet, DEFAULT_DISCOVERY_SKIP_DIRS};
use crate::utils::find_xbp_config_upwards;
use once_cell::sync::Lazy;
use regex::Regex;
use serde::{Deserialize, Serialize};
use sha2::{Digest, Sha256};
use std::fs;
use std::path::Path;
use walkdir::WalkDir;

const MAX_FILE_BYTES: u64 = 1_500_000;
/// Default surrounding window when no construct (function/table/…) is detected.
const CONTEXT_RADIUS: usize = 10;
/// Hard cap so huge functions do not blow issue bodies.
const MAX_CONTEXT_LINES: usize = 120;

// C-style / SQL / Lisp line comments (may trail code).
// `///` / `//!` doc comments are rejected after the match (regex crate has no lookaround).
// `\b` after the kind avoids matching section headers like `// TODOs (`xbp todos`)`.
static TODO_SLASH_RE: Lazy<Regex> = Lazy::new(|| {
    Regex::new(r"(?i)(?://|--|;)\s*(TODO|FIXME|XXX|HACK)\b(?:\s*\([^)]*\))?\s*:?\s*(.*)$")
        .expect("todo slash regex")
});

// Shell/Python `#` comments only at line start; excludes markdown `##` headers.
static TODO_HASH_RE: Lazy<Regex> = Lazy::new(|| {
    Regex::new(r"(?i)^\s*#\s*(TODO|FIXME|XXX|HACK)\b(?:\s*\([^)]*\))?\s*:?\s*(.*)$")
        .expect("todo hash regex")
});

static TODO_BLOCK_RE: Lazy<Regex> = Lazy::new(|| {
    Regex::new(r"(?i)/\*\s*(TODO|FIXME|XXX|HACK)\b(?:\s*\([^)]*\))?\s*:?\s*(.*?)\*/")
        .expect("todo block regex")
});

#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
pub struct TodoHit {
    pub fingerprint: String,
    pub kind: String,
    pub path: String,
    pub line: usize,
    pub text: String,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub context: Option<String>,
}

pub fn scan_todos(root: &Path) -> Result<Vec<TodoHit>, String> {
    let ignore = build_ignore_set(root);
    let mut hits = Vec::new();

    for entry in WalkDir::new(root)
        .into_iter()
        .filter_entry(|e| {
            let name = e.file_name().to_string_lossy();
            if e.depth() > 0 && DEFAULT_DISCOVERY_SKIP_DIRS.iter().any(|d| *d == name) {
                return false;
            }
            if e.file_type().is_dir() {
                let rel = relative_path(root, e.path());
                if ignore.matches_dir(&rel) {
                    return false;
                }
            }
            true
        })
        .filter_map(|e| e.ok())
    {
        if !entry.file_type().is_file() {
            continue;
        }
        let path = entry.path();
        let rel = relative_path(root, path);
        if ignore.matches_file(&rel) {
            continue;
        }
        if should_skip_file(path) {
            continue;
        }
        let meta = match fs::metadata(path) {
            Ok(m) => m,
            Err(_) => continue,
        };
        if meta.len() > MAX_FILE_BYTES {
            continue;
        }
        let content = match fs::read_to_string(path) {
            Ok(c) => c,
            Err(_) => continue, // binary / non-utf8
        };
        hits.extend(extract_hits_from_content(&rel, &content));
    }

    hits.sort_by(|a, b| {
        a.path
            .cmp(&b.path)
            .then(a.line.cmp(&b.line))
            .then(a.kind.cmp(&b.kind))
    });
    // Dedupe identical fingerprints (keep first/earliest line).
    let mut seen = std::collections::HashSet::new();
    hits.retain(|h| seen.insert(h.fingerprint.clone()));
    Ok(hits)
}

pub fn extract_hits_from_content(rel_path: &str, content: &str) -> Vec<TodoHit> {
    let mut hits = Vec::new();
    let lines: Vec<&str> = content.lines().collect();
    for (idx, line) in lines.iter().enumerate() {
        let line_no = idx + 1;
        // Skip Rust/C++ doc-comment lines entirely (`///`, `//!`).
        let trimmed_line = line.trim_start();
        if trimmed_line.starts_with("///") || trimmed_line.starts_with("//!") {
            continue;
        }

        let caps = TODO_SLASH_RE
            .captures(line)
            .filter(|caps| !is_doc_comment_slash_match(line, caps))
            .filter(|caps| !is_inside_double_quotes(line, caps.get(0).map(|m| m.start()).unwrap_or(0)))
            .or_else(|| {
                // Reject markdown headings like `## TODO`
                let trimmed = line.trim_start();
                if trimmed.starts_with("##") {
                    return None;
                }
                TODO_HASH_RE.captures(line)
            })
            .or_else(|| {
                TODO_BLOCK_RE.captures(line).filter(|caps| {
                    !is_inside_double_quotes(
                        line,
                        caps.get(0).map(|m| m.start()).unwrap_or(0),
                    )
                })
            });
        let Some(caps) = caps else {
            continue;
        };
        let kind = caps.get(1).map(|m| m.as_str()).unwrap_or("TODO");
        let text = caps.get(2).map(|m| m.as_str()).unwrap_or("").trim();
        if text.is_empty() && kind.eq_ignore_ascii_case("XXX") {
            continue;
        }
        let text = if text.is_empty() {
            format!("({kind} without text)")
        } else {
            text.to_string()
        };
        let context = rich_context(rel_path, &lines, idx);
        hits.push(TodoHit {
            fingerprint: fingerprint(rel_path, kind, &text),
            kind: kind.to_ascii_uppercase(),
            path: rel_path.replace('\\', "/"),
            line: line_no,
            text,
            context: Some(context),
        });
    }
    hits
}

pub fn fingerprint(path: &str, kind: &str, text: &str) -> String {
    let normalized_path = path.replace('\\', "/");
    let normalized_text = normalize_whitespace(text);
    let kind = kind.trim().to_ascii_uppercase();
    let mut hasher = Sha256::new();
    hasher.update(normalized_path.as_bytes());
    hasher.update([0]);
    hasher.update(kind.as_bytes());
    hasher.update([0]);
    hasher.update(normalized_text.as_bytes());
    format!("{:x}", hasher.finalize())
}

/// True when a `// TODO` match is actually inside `///` or `////` (doc comment).
fn is_doc_comment_slash_match(line: &str, caps: &regex::Captures<'_>) -> bool {
    let Some(whole) = caps.get(0) else {
        return false;
    };
    let start = whole.start();
    let bytes = line.as_bytes();
    // Match began at a `//` that is preceded by `/` → part of `///`…
    if start > 0 && bytes[start - 1] == b'/' {
        return true;
    }
    // Or `//` is immediately followed by another `/` before the marker (///TODO).
    if bytes.get(start + 2) == Some(&b'/') {
        return true;
    }
    false
}

/// Rough string-literal check so clap `about = "…// TODO…"` is not filed as a TODO.
fn is_inside_double_quotes(line: &str, index: usize) -> bool {
    let mut in_string = false;
    let mut escaped = false;
    for (i, ch) in line.char_indices() {
        if i >= index {
            break;
        }
        if escaped {
            escaped = false;
            continue;
        }
        if in_string && ch == '\\' {
            escaped = true;
            continue;
        }
        if ch == '"' {
            in_string = !in_string;
        }
    }
    in_string
}

fn context_snippet(lines: &[&str], idx: usize, radius: usize) -> String {
    let start = idx.saturating_sub(radius);
    let end = (idx + radius + 1).min(lines.len());
    format_context_slice(lines, start, end)
}

/// Prefer a full surrounding construct (function / table / rule) when the TODO
/// sits above or inside one; otherwise fall back to ±`CONTEXT_RADIUS` lines.
fn rich_context(rel_path: &str, lines: &[&str], idx: usize) -> String {
    let lang = ContextLang::from_path(rel_path);
    if let Some((start, end)) = expand_construct_range(lines, idx, lang) {
        return format_context_slice(lines, start, end);
    }
    context_snippet(lines, idx, CONTEXT_RADIUS)
}

fn format_context_slice(lines: &[&str], start: usize, end: usize) -> String {
    let start = start.min(lines.len());
    let mut end = end.min(lines.len()).max(start);
    // Truncate from the bottom if the construct is enormous.
    if end.saturating_sub(start) > MAX_CONTEXT_LINES {
        end = start + MAX_CONTEXT_LINES;
    }
    let last_line_no = end.max(1);
    let width = ((last_line_no as f64).log10().floor() as usize) + 1;
    lines[start..end]
        .iter()
        .enumerate()
        .map(|(offset, line)| {
            let line_no = start + offset + 1;
            format!("{line_no:>width$} | {line}")
        })
        .collect::<Vec<_>>()
        .join("\n")
}

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum ContextLang {
    Rust,
    JsTs,
    Python,
    Sql,
    Css,
    Go,
    Generic,
}

impl ContextLang {
    fn from_path(path: &str) -> Self {
        let lower = path.replace('\\', "/").to_ascii_lowercase();
        let ext = lower.rsplit('.').next().unwrap_or("");
        match ext {
            "rs" => Self::Rust,
            "js" | "jsx" | "mjs" | "cjs" | "ts" | "tsx" | "mts" | "cts" => Self::JsTs,
            "py" | "pyi" => Self::Python,
            "sql" | "pgsql" | "psql" => Self::Sql,
            "css" | "scss" | "sass" | "less" => Self::Css,
            "go" => Self::Go,
            _ => Self::Generic,
        }
    }
}

/// Expand to the full construct the TODO is above or inside.
/// Returns 0-based [start, end) line indices.
fn expand_construct_range(
    lines: &[&str],
    hit_idx: usize,
    lang: ContextLang,
) -> Option<(usize, usize)> {
    // 1) Prefer a construct that *starts* shortly after the TODO (comment above fn/table).
    if let Some(decl) = find_following_declaration(lines, hit_idx, lang) {
        if let Some(end) = find_construct_end(lines, decl, lang) {
            let start = hit_idx; // include the TODO line itself
            if end > start {
                return Some((start, (end + 1).min(lines.len())));
            }
        }
    }

    // 2) Else, if the TODO is *inside* a construct, expand to that whole body.
    if let Some(decl) = find_enclosing_declaration(lines, hit_idx, lang) {
        if let Some(end) = find_construct_end(lines, decl, lang) {
            if end >= hit_idx {
                // Include a few leading comment/doc lines above the declaration.
                let start = expand_leading_comments(lines, decl);
                return Some((start, (end + 1).min(lines.len())));
            }
        }
    }

    None
}

fn find_following_declaration(
    lines: &[&str],
    hit_idx: usize,
    lang: ContextLang,
) -> Option<usize> {
    // Only treat as "above a function" when the next real code arrives soon.
    let search_limit = (hit_idx + 1 + 12).min(lines.len());
    let mut i = hit_idx + 1;
    while i < search_limit {
        let trimmed = lines[i].trim();
        if trimmed.is_empty()
            || is_comment_only_line(trimmed, lang)
            || is_attribute_or_decorator(trimmed, lang)
        {
            i += 1;
            continue;
        }
        if is_declaration_start(trimmed, lang) {
            return Some(i);
        }
        // Hit other real code → not sitting above a construct.
        return None;
    }
    None
}

fn is_attribute_or_decorator(trimmed: &str, lang: ContextLang) -> bool {
    match lang {
        ContextLang::Rust => trimmed.starts_with("#[") || trimmed.starts_with("#!["),
        ContextLang::Python | ContextLang::JsTs => trimmed.starts_with('@'),
        _ => false,
    }
}

fn find_enclosing_declaration(
    lines: &[&str],
    hit_idx: usize,
    lang: ContextLang,
) -> Option<usize> {
    // Walk upward for a declaration whose body still covers hit_idx.
    let mut i = hit_idx;
    let mut scanned = 0usize;
    while i > 0 && scanned < MAX_CONTEXT_LINES {
        i -= 1;
        scanned += 1;
        let trimmed = lines[i].trim();
        if !is_declaration_start(trimmed, lang) {
            continue;
        }
        if let Some(end) = find_construct_end(lines, i, lang) {
            if end >= hit_idx {
                return Some(i);
            }
        }
    }
    None
}

fn expand_leading_comments(lines: &[&str], decl_idx: usize) -> usize {
    let mut start = decl_idx;
    while start > 0 {
        let prev = lines[start - 1].trim();
        if prev.is_empty()
            || prev.starts_with("//")
            || prev.starts_with('#')
            || prev.starts_with("/*")
            || prev.starts_with('*')
            || prev.starts_with("///")
            || prev.starts_with("//!")
            || prev.starts_with("--")
        {
            start -= 1;
            // Stop after collecting a reasonable doc block.
            if decl_idx.saturating_sub(start) > 20 {
                break;
            }
            continue;
        }
        break;
    }
    start
}

fn is_comment_only_line(trimmed: &str, lang: ContextLang) -> bool {
    match lang {
        ContextLang::Python => trimmed.starts_with('#'),
        ContextLang::Sql => trimmed.starts_with("--") || trimmed.starts_with("/*"),
        ContextLang::Css => trimmed.starts_with("/*") || trimmed.starts_with("//"),
        _ => {
            trimmed.starts_with("//")
                || trimmed.starts_with("/*")
                || trimmed.starts_with('*')
                || trimmed.starts_with('#')
                || trimmed.starts_with("--")
        }
    }
}

fn is_declaration_start(trimmed: &str, lang: ContextLang) -> bool {
    if trimmed.is_empty() {
        return false;
    }
    match lang {
        ContextLang::Rust => is_rust_declaration(trimmed),
        ContextLang::JsTs => is_js_declaration(trimmed),
        ContextLang::Python => is_python_declaration(trimmed),
        ContextLang::Sql => is_sql_declaration(trimmed),
        ContextLang::Css => is_css_rule_start(trimmed),
        ContextLang::Go => is_go_declaration(trimmed),
        ContextLang::Generic => {
            is_rust_declaration(trimmed)
                || is_js_declaration(trimmed)
                || is_python_declaration(trimmed)
                || is_sql_declaration(trimmed)
                || is_go_declaration(trimmed)
        }
    }
}

fn is_rust_declaration(trimmed: &str) -> bool {
    // Strip common visibility / qualifiers prefixes loosely.
    let t = trimmed.trim_start_matches("pub(crate) ").trim_start();
    let t = t.trim_start_matches("pub ").trim_start();
    let t = t.trim_start_matches("async ").trim_start();
    let t = t.trim_start_matches("const ").trim_start();
    let t = t.trim_start_matches("unsafe ").trim_start();
    t.starts_with("fn ")
        || t.starts_with("fn(") // unlikely but
        || t.starts_with("struct ")
        || t.starts_with("enum ")
        || t.starts_with("impl ")
        || t.starts_with("trait ")
        || t.starts_with("mod ")
        || t.starts_with("type ")
}

fn is_js_declaration(trimmed: &str) -> bool {
    let t = strip_js_export_prefix(trimmed.trim_start());
    if t.starts_with("function ")
        || t.starts_with("function*")
        || t.starts_with("class ")
        || t.starts_with("interface ")
        || t.starts_with("type ")
        || t.starts_with("enum ")
        || t.starts_with("namespace ")
    {
        return true;
    }
    // const foo = (…) => / function / async
    if (t.starts_with("const ") || t.starts_with("let ") || t.starts_with("var "))
        && (t.contains("=>")
            || t.contains("function")
            || t.contains("= (")
            || t.contains("=("))
    {
        return true;
    }
    // method / constructor-ish: name(...) {  on its own line
    is_js_method_header(t)
}

fn strip_js_export_prefix(mut t: &str) -> &str {
    loop {
        let next = t
            .strip_prefix("export ")
            .or_else(|| t.strip_prefix("default "))
            .or_else(|| t.strip_prefix("async "))
            .or_else(|| t.strip_prefix("declare "))
            .or_else(|| t.strip_prefix("public "))
            .or_else(|| t.strip_prefix("private "))
            .or_else(|| t.strip_prefix("protected "))
            .or_else(|| t.strip_prefix("static "))
            .or_else(|| t.strip_prefix("readonly "));
        match next {
            Some(rest) => t = rest.trim_start(),
            None => break,
        }
    }
    t
}

fn is_js_method_header(t: &str) -> bool {
    // Rough: ident(args) {   or   ident(args)
    // Avoid if/for/while/switch.
    let t = t.trim();
    if t.starts_with("if ")
        || t.starts_with("if(")
        || t.starts_with("for ")
        || t.starts_with("for(")
        || t.starts_with("while ")
        || t.starts_with("while(")
        || t.starts_with("switch ")
        || t.starts_with("switch(")
        || t.starts_with("catch ")
        || t.starts_with("catch(")
    {
        return false;
    }
    let Some(open) = t.find('(') else {
        return false;
    };
    let name = t[..open].trim();
    if name.is_empty()
        || !name
            .chars()
            .next()
            .is_some_and(|c| c.is_ascii_alphabetic() || c == '_' || c == '$')
    {
        return false;
    }
    if name.contains('=') || name.contains(':') {
        return false;
    }
    // Prefer lines that open a body soon.
    t.contains('{') || t.ends_with(')') || t.ends_with('{')
}

fn is_python_declaration(trimmed: &str) -> bool {
    let t = trimmed.trim_start();
    t.starts_with("def ")
        || t.starts_with("async def ")
        || t.starts_with("class ")
        || t.starts_with("@") // decorator leading into def/class
}

fn is_sql_declaration(trimmed: &str) -> bool {
    let upper = trimmed.to_ascii_uppercase();
    upper.starts_with("CREATE TABLE")
        || upper.starts_with("CREATE OR REPLACE")
        || upper.starts_with("CREATE FUNCTION")
        || upper.starts_with("CREATE PROCEDURE")
        || upper.starts_with("CREATE VIEW")
        || upper.starts_with("CREATE INDEX")
        || upper.starts_with("CREATE UNIQUE INDEX")
        || upper.starts_with("CREATE TYPE")
        || upper.starts_with("ALTER TABLE")
        || upper.starts_with("CREATE SCHEMA")
}

fn is_css_rule_start(trimmed: &str) -> bool {
    if trimmed.starts_with('}') || trimmed.starts_with("//") || trimmed.starts_with("/*") {
        return false;
    }
    // @media / @keyframes / @supports
    if trimmed.starts_with('@') {
        return true;
    }
    if trimmed.contains('{') {
        return true;
    }
    // Multi-line selector list
    if trimmed.ends_with(',') && !trimmed.contains(':') {
        return true;
    }
    // Bare selector line (class/id/element) without a property semicolon.
    let looks_like_selector = trimmed.starts_with('.')
        || trimmed.starts_with('#')
        || trimmed.starts_with('[')
        || trimmed
            .chars()
            .next()
            .is_some_and(|c| c.is_ascii_alphabetic());
    looks_like_selector && !trimmed.contains(';')
}

fn is_go_declaration(trimmed: &str) -> bool {
    let t = trimmed.trim_start();
    t.starts_with("func ") || t.starts_with("type ") || t.starts_with("package ")
}

fn find_construct_end(lines: &[&str], decl_idx: usize, lang: ContextLang) -> Option<usize> {
    match lang {
        ContextLang::Python => find_python_block_end(lines, decl_idx),
        ContextLang::Sql => find_sql_statement_end(lines, decl_idx),
        ContextLang::Css => find_brace_block_end(lines, decl_idx),
        ContextLang::Rust | ContextLang::JsTs | ContextLang::Go | ContextLang::Generic => {
            // Prefer braces; if the declaration line has no `{` soon, scan a few lines.
            if let Some(end) = find_brace_block_end(lines, decl_idx) {
                return Some(end);
            }
            // Single-line / semicolon terminated (e.g. `type Foo = Bar;`)
            find_semicolon_end(lines, decl_idx)
        }
    }
}

fn find_brace_block_end(lines: &[&str], decl_idx: usize) -> Option<usize> {
    let mut depth = 0i32;
    let mut seen_open = false;
    let limit = (decl_idx + MAX_CONTEXT_LINES).min(lines.len());
    for i in decl_idx..limit {
        let line = lines[i];
        for ch in line.chars() {
            match ch {
                '{' => {
                    depth += 1;
                    seen_open = true;
                }
                '}' => {
                    depth -= 1;
                    if seen_open && depth == 0 {
                        return Some(i);
                    }
                }
                _ => {}
            }
        }
    }
    None
}

fn find_semicolon_end(lines: &[&str], decl_idx: usize) -> Option<usize> {
    let limit = (decl_idx + 30).min(lines.len());
    for i in decl_idx..limit {
        if lines[i].contains(';') {
            return Some(i);
        }
    }
    None
}

fn find_python_block_end(lines: &[&str], decl_idx: usize) -> Option<usize> {
    // Skip decorators to the actual def/class when starting on @.
    let mut start = decl_idx;
    while start < lines.len() {
        let t = lines[start].trim_start();
        if t.starts_with('@') {
            start += 1;
            continue;
        }
        break;
    }
    if start >= lines.len() {
        return None;
    }
    let header = lines[start];
    let base_indent = leading_indent(header);
    if !header.trim_end().ends_with(':')
        && !lines
            .get(start)
            .map(|l| l.contains("def ") || l.contains("class "))
            .unwrap_or(false)
    {
        // Still accept multi-line headers; fall through.
    }
    let limit = (start + MAX_CONTEXT_LINES).min(lines.len());
    let mut last = start;
    for i in (start + 1)..limit {
        let line = lines[i];
        if line.trim().is_empty() {
            last = i;
            continue;
        }
        let indent = leading_indent(line);
        if indent <= base_indent && !line.trim_start().starts_with('#') {
            // Dedented to same or outer level → block ended on previous non-empty.
            return Some(last);
        }
        last = i;
    }
    Some(last)
}

fn leading_indent(line: &str) -> usize {
    line.chars()
        .take_while(|c| *c == ' ' || *c == '\t')
        .map(|c| if c == '\t' { 4 } else { 1 })
        .sum()
}

fn find_sql_statement_end(lines: &[&str], decl_idx: usize) -> Option<usize> {
    let mut paren = 0i32;
    let limit = (decl_idx + MAX_CONTEXT_LINES).min(lines.len());
    for i in decl_idx..limit {
        for ch in lines[i].chars() {
            match ch {
                '(' => paren += 1,
                ')' => paren -= 1,
                ';' if paren <= 0 => return Some(i),
                _ => {}
            }
        }
    }
    // No semicolon — take until blank line after content or max window.
    let mut last = decl_idx;
    for i in decl_idx..limit {
        if lines[i].trim().is_empty() && i > decl_idx {
            return Some(last);
        }
        last = i;
    }
    Some(last)
}

fn relative_path(root: &Path, path: &Path) -> String {
    path.strip_prefix(root)
        .unwrap_or(path)
        .to_string_lossy()
        .replace('\\', "/")
}

fn should_skip_file(path: &Path) -> bool {
    let name = path
        .file_name()
        .and_then(|n| n.to_str())
        .unwrap_or("")
        .to_ascii_lowercase();
    if name.ends_with(".lock")
        || name.ends_with(".min.js")
        || name.ends_with(".min.css")
        || name.ends_with(".map")
        || name.ends_with(".png")
        || name.ends_with(".jpg")
        || name.ends_with(".jpeg")
        || name.ends_with(".gif")
        || name.ends_with(".webp")
        || name.ends_with(".ico")
        || name.ends_with(".woff")
        || name.ends_with(".woff2")
        || name.ends_with(".ttf")
        || name.ends_with(".pdf")
        || name.ends_with(".zip")
        || name == "cargo.lock"
        || name == "pnpm-lock.yaml"
        || name == "package-lock.json"
        || name == "yarn.lock"
        || name == "todo-issues.json"
        || name == "catalog.json"
        // Docs / trackers — not actionable code TODOs
        || name == "todo.md"
        || name == "readme.md"
        || name == "changelog.md"
        || name == "agents.md"
        || name.ends_with(".mdx")
    {
        return true;
    }
    // Skip generated MCP catalog path segments
    let rel = path.to_string_lossy().replace('\\', "/").to_ascii_lowercase();
    if rel.contains("/generated/catalog.json") || rel.contains("/target/") {
        return true;
    }
    false
}

fn build_ignore_set(root: &Path) -> XbpIgnoreSet {
    let mut set = XbpIgnoreSet::from_patterns(
        DEFAULT_DISCOVERY_SKIP_DIRS
            .iter()
            .map(|d| format!("{d}/")),
    );
    for candidate in [
        root.join(".xbpignore"),
        root.join(".xbp").join(".xbpignore"),
        root.join(".gitignore"),
    ] {
        if let Ok(content) = fs::read_to_string(candidate) {
            let other = XbpIgnoreSet::from_file_content(&content);
            set.merge(&other);
        }
    }
    if let Some(found) = find_xbp_config_upwards(root) {
        if let Ok(content) = fs::read_to_string(&found.config_path) {
            if let Ok(cfg) = serde_yaml::from_str::<serde_yaml::Value>(&content) {
                if let Some(paths) = cfg
                    .get("ignore_paths")
                    .or_else(|| cfg.get("ignored_paths"))
                    .and_then(|v| v.as_sequence())
                {
                    let extra: Vec<String> = paths
                        .iter()
                        .filter_map(|v| v.as_str().map(str::to_string))
                        .collect();
                    let other = XbpIgnoreSet::from_patterns(extra);
                    set.merge(&other);
                }
            }
        }
    }
    set
}

trait IgnoreMatch {
    fn matches_dir(&self, rel: &str) -> bool;
    fn matches_file(&self, rel: &str) -> bool;
}

impl IgnoreMatch for XbpIgnoreSet {
    fn matches_dir(&self, rel: &str) -> bool {
        self.is_ignored_relative(rel, true)
    }
    fn matches_file(&self, rel: &str) -> bool {
        self.is_ignored_relative(rel, false)
    }
}

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

    #[test]
    fn extracts_rust_and_python_todos() {
        // Build markers at runtime so this source file is not itself a scan hit.
        let content = [
            format!("// {}: wire secrets fallback", "TODO"),
            "fn main() {}".to_string(),
            format!("# {}: handle empty list", "FIXME"),
            format!("/* {}: temporary */", "HACK"),
        ]
        .join("\n");
        let hits = extract_hits_from_content("src/main.rs", &content);
        assert!(hits
            .iter()
            .any(|h| h.kind == "TODO" && h.text.contains("wire secrets")));
        assert!(hits.iter().any(|h| h.kind == "FIXME"));
        assert!(hits.iter().any(|h| h.kind == "HACK"));
    }

    #[test]
    fn fingerprint_stable_across_line_moves() {
        let a = fingerprint("src/a.rs", "TODO", "fix the thing");
        let b = fingerprint("src/a.rs", "TODO", "  fix the   thing ");
        assert_eq!(a, b);
        let c = fingerprint("src\\a.rs", "todo", "fix the thing");
        assert_eq!(a, c);
    }

    #[test]
    fn fingerprint_differs_for_different_text() {
        let a = fingerprint("src/a.rs", "TODO", "one");
        let b = fingerprint("src/a.rs", "TODO", "two");
        assert_ne!(a, b);
    }

    #[test]
    fn ignores_markdown_headers_and_string_literals() {
        let content = [
            "        body.push_str(\"## TODO section\");".to_string(),
            "## TODO should not match as a heading either".to_string(),
            format!("// {}: real one", "TODO"),
        ]
        .join("\n");
        let hits = extract_hits_from_content("src/x.rs", &content);
        assert_eq!(hits.len(), 1);
        assert!(hits[0].text.contains("real one"));
    }

    #[test]
    fn ignores_rust_doc_comments() {
        let content = "///     FIXME: 1\n//! TODO: module docs\n// TODO: real code todo\n";
        let hits = extract_hits_from_content("src/lib.rs", content);
        assert_eq!(hits.len(), 1);
        assert!(hits[0].text.contains("real code todo"));
    }

    #[test]
    fn ignores_todos_inside_string_literals() {
        // Build the false-positive line without embedding a live marker comment in this file.
        let about = format!(
            "            about = \"Scan {} markers and file issues\",",
            concat!("// ", "TODO")
        );
        let content = [about, format!("            // {}: actual work item", "TODO")].join("\n");
        let hits = extract_hits_from_content("src/cli.rs", &content);
        assert_eq!(hits.len(), 1);
        assert!(hits[0].text.contains("actual work item"));
    }

    #[test]
    fn context_includes_ten_line_window_with_line_numbers() {
        let mut lines = Vec::new();
        for i in 1..=25 {
            lines.push(format!("line_{i}"));
        }
        // Put marker on line 12 (index 11).
        lines[11] = format!("// {}: mid", "TODO");
        let content = lines.join("\n");
        let hits = extract_hits_from_content("notes.txt", &content);
        assert_eq!(hits.len(), 1);
        let ctx = hits[0].context.as_deref().unwrap_or("");
        // ±10 around line 12 → lines 2..=22
        assert!(ctx.contains("2 | line_2"), "{ctx}");
        assert!(ctx.contains("12 | // TODO: mid") || ctx.contains("TODO: mid"), "{ctx}");
        assert!(ctx.contains("22 | line_22"), "{ctx}");
        assert!(!ctx.contains("1 | line_1\n") || ctx.contains("1 |"), "{ctx}");
        assert!(!ctx.contains("line_25"), "{ctx}");
    }

    #[test]
    fn context_captures_entire_rust_function_when_todo_is_above() {
        let content = [
            "use std::io;".to_string(),
            format!("// {}: handle empty path", "TODO"),
            "pub fn load_config(path: &str) -> Result<String, String> {".to_string(),
            "    if path.is_empty() {".to_string(),
            "        return Err(\"empty\".into());".to_string(),
            "    }".to_string(),
            "    std::fs::read_to_string(path).map_err(|e| e.to_string())".to_string(),
            "}".to_string(),
            "fn other() {}".to_string(),
        ]
        .join("\n");
        let hits = extract_hits_from_content("src/config.rs", &content);
        assert_eq!(hits.len(), 1);
        let ctx = hits[0].context.as_deref().unwrap_or("");
        assert!(ctx.contains("load_config"), "{ctx}");
        assert!(ctx.contains("read_to_string"), "{ctx}");
        assert!(ctx.contains("}"), "{ctx}");
        assert!(
            !ctx.contains("fn other"),
            "should stop at function end: {ctx}"
        );
    }

    #[test]
    fn context_captures_sql_create_table_when_fixme_is_above() {
        let content = [
            format!(
                "-- {}: existing drift is user_permission_scopes table, we migrate to grants",
                "FIXME"
            ),
            "CREATE TABLE IF NOT EXISTS athena.user_permission_scopes (".to_string(),
            "  id bigint NOT NULL,".to_string(),
            "  user_id uuid NOT NULL,".to_string(),
            "  scope text NOT NULL".to_string(),
            ");".to_string(),
            "CREATE TABLE other (id int);".to_string(),
        ]
        .join("\n");
        let hits = extract_hits_from_content("sql/001_scopes.sql", &content);
        assert_eq!(hits.len(), 1);
        let ctx = hits[0].context.as_deref().unwrap_or("");
        assert!(ctx.contains("CREATE TABLE IF NOT EXISTS athena.user_permission_scopes"), "{ctx}");
        assert!(ctx.contains("scope text NOT NULL"), "{ctx}");
        assert!(
            !ctx.contains("CREATE TABLE other"),
            "should stop at first statement: {ctx}"
        );
    }

    #[test]
    fn context_captures_python_function_body() {
        let content = [
            format!("# {}: validate input", "TODO"),
            "def parse_row(raw: str) -> dict:".to_string(),
            "    if not raw:".to_string(),
            "        return {}".to_string(),
            "    return {\"v\": raw}".to_string(),
            "".to_string(),
            "def other():".to_string(),
            "    pass".to_string(),
        ]
        .join("\n");
        let hits = extract_hits_from_content("app/parse.py", &content);
        assert_eq!(hits.len(), 1);
        let ctx = hits[0].context.as_deref().unwrap_or("");
        assert!(ctx.contains("def parse_row"), "{ctx}");
        assert!(ctx.contains("return {\"v\": raw}"), "{ctx}");
        assert!(!ctx.contains("def other"), "{ctx}");
    }

    #[test]
    fn context_captures_ts_function_when_todo_inside() {
        let content = [
            "export function hydrate(userId: string) {".to_string(),
            "  const scope = { userId };".to_string(),
            format!("  // {}: cache miss path", "FIXME"),
            "  return fetchUser(scope);".to_string(),
            "}".to_string(),
            "export function other() { return 1; }".to_string(),
        ]
        .join("\n");
        let hits = extract_hits_from_content("src/user.ts", &content);
        assert_eq!(hits.len(), 1);
        let ctx = hits[0].context.as_deref().unwrap_or("");
        assert!(ctx.contains("export function hydrate"), "{ctx}");
        assert!(ctx.contains("fetchUser"), "{ctx}");
        assert!(!ctx.contains("function other"), "{ctx}");
    }
}