code_kb_core/

workspace.rs

use std::path::{Path, PathBuf};
use thiserror::Error;

#[derive(Debug, Error)]
pub enum WorkspaceError {
    #[error("Failed to canonicalize path {path}: {source}")]
    CanonicalizationFailed {
        path: PathBuf,
        #[source]
        source: std::io::Error,
    },
    #[error("Path '{0}' is outside workspace root '{1}'")]
    PathOutsideWorkspace(PathBuf, PathBuf),
    #[error("Could not discover workspace root from '{0}'")]
    DiscoveryFailed(PathBuf),
    #[error("Database artifact not found at '{0}'")]
    ArtifactNotFound(PathBuf),
}

/// Lexically clean a path by collapsing `.` and `..` components.
pub fn clean_path(path: &Path) -> PathBuf {
    use std::path::Component;
    let s = path.to_string_lossy();
    let norm = if cfg!(not(windows)) && s.contains('\\') {
        std::borrow::Cow::Owned(PathBuf::from(s.replace('\\', "/")))
    } else {
        std::borrow::Cow::Borrowed(path)
    };
    let mut stack = Vec::new();
    for comp in norm.components() {
        match comp {
            Component::CurDir => {}
            Component::ParentDir => {
                if let Some(Component::Normal(_)) = stack.last() {
                    stack.pop();
                } else {
                    stack.push(comp);
                }
            }
            _ => stack.push(comp),
        }
    }
    stack.into_iter().collect()
}

fn percent_decode(input: &str) -> String {
    let mut bytes = Vec::with_capacity(input.len());
    let input_bytes = input.as_bytes();
    let mut i = 0;
    while i < input_bytes.len() {
        if input_bytes[i] == b'%'
            && i + 2 < input_bytes.len()
            && let Ok(hex) = std::str::from_utf8(&input_bytes[i + 1..i + 3])
            && let Ok(byte) = u8::from_str_radix(hex, 16)
        {
            bytes.push(byte);
            i += 3;
            continue;
        }
        bytes.push(input_bytes[i]);
        i += 1;
    }
    String::from_utf8_lossy(&bytes).into_owned()
}

/// Extract drive letter and remainder if the string begins with a drive specification
/// delimited by ':', '|', or percent-encoded "%7C" / "%3A".
fn extract_drive_letter_and_remainder(s: &str) -> Option<(char, &str)> {
    let bytes = s.as_bytes();
    if bytes.is_empty() || !bytes[0].is_ascii_alphabetic() {
        return None;
    }
    let drive = bytes[0] as char;

    // Single-byte delimiters: ':' or '|'
    if bytes.len() >= 2
        && (bytes[1] == b':' || bytes[1] == b'|')
        && (bytes.len() == 2
            || bytes[2] == b'/'
            || bytes[2] == b'\\'
            || bytes[2] == b'?'
            || bytes[2] == b'#')
    {
        return Some((drive, &s[2..]));
    }

    // Three-byte percent-encoded delimiters: "%7C", "%7c", "%3A", "%3a"
    if bytes.len() >= 4 {
        let delim = &bytes[1..4];
        if (delim.eq_ignore_ascii_case(b"%7c") || delim.eq_ignore_ascii_case(b"%3a"))
            && (bytes.len() == 4
                || bytes[4] == b'/'
                || bytes[4] == b'\\'
                || bytes[4] == b'?'
                || bytes[4] == b'#')
        {
            return Some((drive, &s[4..]));
        }
    }

    None
}

/// Strip an optional "localhost/" or "localhost\" prefix (with or without a leading slash).
fn strip_localhost_prefix(s: &str) -> &str {
    let without_slash = s.strip_prefix('/').unwrap_or(s);
    let bytes = without_slash.as_bytes();
    if bytes.len() >= 10
        && bytes[..9].eq_ignore_ascii_case(b"localhost")
        && (bytes[9] == b'/' || bytes[9] == b'\\')
    {
        &without_slash[10..]
    } else {
        s
    }
}

/// Convert a path string starting with a pipe drive specification (e.g. "C|/..." or "/C|/...")
/// to use a standard colon ':' delimiter (e.g. "C:/...").
fn normalize_drive_pipe_str(s: &str) -> String {
    let clean = strip_localhost_prefix(s);
    let target = clean.strip_prefix('/').unwrap_or(clean);
    let target = strip_localhost_prefix(target);
    if let Some((drive, remainder)) = extract_drive_letter_and_remainder(target) {
        if remainder.is_empty() || remainder.starts_with('?') || remainder.starts_with('#') {
            format!("{}:/{}", drive, remainder)
        } else {
            format!("{}:{}", drive, remainder)
        }
    } else {
        s.to_string()
    }
}

/// Parse an MCP file URI or plain path into a normalized PathBuf.
/// Handles standard file URIs (`file:///path`), two-slash drive letter URIs (`file://C:/...`),
/// pipe drive delimiters (`file:///C|/...`, `file://C|/...`), percent-encoding (`%20`, `%7C`), and plain paths.
pub fn parse_file_uri(cand: &str) -> Option<PathBuf> {
    if let Some(rest) = cand.strip_prefix("file://") {
        let path_part = rest.strip_prefix('/').unwrap_or(rest);
        let path_part = strip_localhost_prefix(path_part);
        let normalized_cand = if let Some((drive, remainder)) =
            extract_drive_letter_and_remainder(path_part)
        {
            if remainder.is_empty() || remainder.starts_with('?') || remainder.starts_with('#') {
                format!("file:///{}:/{}", drive, remainder)
            } else if remainder.starts_with('/') || remainder.starts_with('\\') {
                format!("file:///{}:{}", drive, remainder)
            } else {
                format!("file:///{}:/{}", drive, remainder)
            }
        } else {
            cand.to_string()
        };

        let file_path = std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| {
            url::Url::parse(&normalized_cand)
                .ok()
                .and_then(|url| url.to_file_path().ok())
        }))
        .ok()
        .flatten();

        if let Some(path) = file_path {
            return Some(normalize_path(&path));
        }
        // Fallback for non-standard file:// patterns with percent decoding
        if let Some(s) = cand.strip_prefix("file:///") {
            let decoded = percent_decode(s);
            let normalized = normalize_drive_pipe_str(&decoded);
            if cfg!(windows) {
                Some(normalize_path(Path::new(&normalized)))
            } else {
                Some(normalize_path(&PathBuf::from(format!("/{}", normalized))))
            }
        } else {
            let s = cand.strip_prefix("file://").unwrap_or(cand);
            let decoded = percent_decode(s);
            let normalized = normalize_drive_pipe_str(&decoded);
            Some(normalize_path(Path::new(&normalized)))
        }
    } else {
        let normalized = normalize_drive_pipe_str(cand);
        Some(normalize_path(Path::new(&normalized)))
    }
}

/// Directory that holds the workspace's rolling `code-kb.log.*` files.
pub fn log_dir(workspace_root: &Path) -> PathBuf {
    workspace_root.join(".code-kb").join("logs")
}

/// The workspace's `code-kb.log.*` files, newest first by modification time. Other files in
/// the log directory are ignored so a stray file never enters a bug report.
pub fn log_files_newest_first(workspace_root: &Path) -> Vec<PathBuf> {
    let mut files: Vec<(std::time::SystemTime, PathBuf)> =
        std::fs::read_dir(log_dir(workspace_root))
            .into_iter()
            .flatten()
            .flatten()
            .filter(|entry| {
                entry
                    .file_name()
                    .to_string_lossy()
                    .starts_with("code-kb.log")
            })
            .filter(|entry| entry.path().is_file())
            .filter_map(|entry| {
                let modified = entry.metadata().ok()?.modified().ok()?;
                Some((modified, entry.path()))
            })
            .collect();
    files.sort_by(|a, b| b.cmp(a));
    files.into_iter().map(|(_, path)| path).collect()
}

/// Most recently modified `code-kb.log.*` file, if any.
pub fn latest_log_file(workspace_root: &Path) -> Option<PathBuf> {
    log_files_newest_first(workspace_root).into_iter().next()
}

/// Strip Windows verbatim prefix (\\?\, \\?\UNC\) using dunce.
pub fn normalize_path(path: &Path) -> PathBuf {
    let s = path.to_string_lossy();
    if let Some(rest) = s.strip_prefix(r"\\?\UNC\") {
        let unc = format!(r"\\{rest}");
        return dunce::simplified(Path::new(&unc)).to_path_buf();
    }
    if let Some(rest) = s.strip_prefix(r"\\?\") {
        return dunce::simplified(Path::new(rest)).to_path_buf();
    }
    dunce::simplified(path).to_path_buf()
}

/// Convert a path to forward-slash string representation for stable relative paths.
pub fn to_forward_slash(path: &Path) -> String {
    let s = path.to_string_lossy();
    s.replace('\\', "/")
}

/// Compare two path components for equality.
/// On Windows, compares `Component::Normal` case-insensitively and drive letters in `Component::Prefix` case-insensitively.
#[cfg(windows)]
fn components_equal(c1: &std::path::Component, c2: &std::path::Component) -> bool {
    if c1 == c2 {
        return true;
    }
    {
        use std::path::Component;
        match (c1, c2) {
            (Component::Normal(s1), Component::Normal(s2)) => s1
                .to_string_lossy()
                .eq_ignore_ascii_case(&s2.to_string_lossy()),
            (Component::Prefix(p1), Component::Prefix(p2)) => {
                use std::path::Prefix;
                match (p1.kind(), p2.kind()) {
                    (Prefix::Disk(d1), Prefix::Disk(d2))
                    | (Prefix::VerbatimDisk(d1), Prefix::VerbatimDisk(d2))
                    | (Prefix::Disk(d1), Prefix::VerbatimDisk(d2))
                    | (Prefix::VerbatimDisk(d1), Prefix::Disk(d2)) => d1.eq_ignore_ascii_case(&d2),
                    (Prefix::UNC(s1, sh1), Prefix::UNC(s2, sh2))
                    | (Prefix::VerbatimUNC(s1, sh1), Prefix::VerbatimUNC(s2, sh2))
                    | (Prefix::UNC(s1, sh1), Prefix::VerbatimUNC(s2, sh2))
                    | (Prefix::VerbatimUNC(s1, sh1), Prefix::UNC(s2, sh2)) => {
                        s1.to_string_lossy()
                            .eq_ignore_ascii_case(&s2.to_string_lossy())
                            && sh1
                                .to_string_lossy()
                                .eq_ignore_ascii_case(&sh2.to_string_lossy())
                    }
                    (Prefix::DeviceNS(d1), Prefix::DeviceNS(d2))
                    | (Prefix::Verbatim(d1), Prefix::Verbatim(d2)) => d1
                        .to_string_lossy()
                        .eq_ignore_ascii_case(&d2.to_string_lossy()),
                    _ => false,
                }
            }
            _ => false,
        }
    }
}

/// Strips `base` from `path`. On Windows, if standard `strip_prefix` fails,
/// performs case-insensitive component comparison to support Windows case-preserving filesystems.
pub fn strip_prefix_lossy<'a>(path: &'a Path, base: &Path) -> Option<&'a Path> {
    if let Ok(rel) = path.strip_prefix(base) {
        return Some(rel);
    }

    #[cfg(windows)]
    {
        let mut path_comps = path.components();
        for base_comp in base.components() {
            let path_comp = path_comps.next()?;
            if !components_equal(&base_comp, &path_comp) {
                return None;
            }
        }
        Some(path_comps.as_path())
    }
    #[cfg(not(windows))]
    {
        None
    }
}

/// Compare two paths for logical identity.
/// On Windows, normalizes verbatim prefixes via dunce and compares disk prefixes and components case-insensitively.
/// On non-Windows, compares paths directly.
pub fn paths_equal(p1: &Path, p2: &Path) -> bool {
    let p1_norm = normalize_path(p1);
    let p2_norm = normalize_path(p2);
    if p1_norm == p2_norm {
        return true;
    }
    if to_forward_slash(&p1_norm) == to_forward_slash(&p2_norm) {
        return true;
    }
    if let (Ok(c1), Ok(c2)) = (dunce::canonicalize(p1), dunce::canonicalize(p2)) {
        let c1_norm = normalize_path(&c1);
        let c2_norm = normalize_path(&c2);
        if c1_norm == c2_norm || to_forward_slash(&c1_norm) == to_forward_slash(&c2_norm) {
            return true;
        }
    }
    #[cfg(windows)]
    {
        let mut c1 = p1_norm.components();
        let mut c2 = p2_norm.components();
        loop {
            match (c1.next(), c2.next()) {
                (None, None) => return true,
                (Some(comp1), Some(comp2)) => {
                    if !components_equal(&comp1, &comp2) {
                        return false;
                    }
                }
                _ => return false,
            }
        }
    }
    #[cfg(not(windows))]
    {
        false
    }
}

/// Check if a relative path contains directories or file patterns that must never be indexed or watched.
pub fn is_hard_excluded(rel_path: &str) -> bool {
    let p = rel_path.replace('\\', "/");
    let has_excluded_dir = p.split('/').any(|component| {
        matches!(
            component,
            ".git"
                | ".hg"
                | ".svn"
                | ".julie"
                | ".code-kb"
                | ".memories"
                | ".agents"
                | ".razorback"
                | ".worktrees"
                | "worktrees"
                | ".claude"
                | ".venv"
                | "venv"
                | ".env"
                | ".tox"
                | ".vs"
                | "node_modules"
                | "vendor"
                | "target"
                | "dist"
                | "build"
                | ".cache"
                | "obj"
                | "TestResults"
                | ".idea"
                | ".vscode"
        )
    });

    if has_excluded_dir {
        return true;
    }

    const EXCLUDED_SUFFIXES: &[&str] = &[
        ".min.js",
        ".bundle.js",
        ".generated.js",
        ".generated.jsx",
        ".generated.ts",
        ".generated.tsx",
        ".generated.d.ts",
        ".tmp",
        ".swp",
        "~",
    ];

    EXCLUDED_SUFFIXES.iter().any(|suffix| p.ends_with(suffix))
}

/// Represents a bound workspace session.
#[derive(Debug, Clone)]
pub struct Workspace {
    pub root: PathBuf,
    pub canonical_root: PathBuf,
    pub repo_name: String,
}

fn trim_trailing_slash(p: &Path) -> PathBuf {
    let s = p.to_string_lossy();
    if s.len() > 1 && (s.ends_with('/') || s.ends_with('\\')) {
        let trimmed = s.trim_end_matches(['/', '\\']);
        if trimmed.is_empty() {
            return PathBuf::from(if cfg!(windows) && s.starts_with('\\') {
                "\\"
            } else {
                "/"
            });
        }
        if cfg!(windows)
            && trimmed.len() == 2
            && trimmed.as_bytes()[0].is_ascii_alphabetic()
            && trimmed.as_bytes()[1] == b':'
        {
            return PathBuf::from(format!("{}\\", trimmed));
        }
        return PathBuf::from(trimmed);
    }
    p.to_path_buf()
}

/// True when `root` carries a repository or language project marker.
pub fn is_project_root(root: &Path) -> bool {
    [
        ".git",
        "Cargo.toml",
        "package.json",
        "go.mod",
        "pyproject.toml",
    ]
    .iter()
    .any(|marker| root.join(marker).exists())
}

impl Workspace {
    /// Discover and bind a workspace from an optional path, falling back to CWD and upward traversal.
    pub fn discover(start_path: Option<&Path>) -> Result<Self, WorkspaceError> {
        let current = match start_path {
            Some(p) => p.to_path_buf(),
            None => std::env::current_dir().unwrap_or_else(|_| PathBuf::from(".")),
        };

        let root = Self::find_workspace_root(&current)?;
        Ok(Self::new(root))
    }

    /// Create workspace binding directly for a known root directory.
    pub fn new(root: PathBuf) -> Self {
        let root_str = root.to_string_lossy();
        let root = if root_str.starts_with("file://") {
            parse_file_uri(&root_str).unwrap_or_else(|| normalize_path(&root))
        } else {
            normalize_path(&root)
        };
        let root = trim_trailing_slash(&root);
        let canonical_root =
            normalize_path(&dunce::canonicalize(&root).unwrap_or_else(|_| root.clone()));
        let repo_name = canonical_root
            .file_name()
            .map(|n| n.to_string_lossy().to_string())
            .unwrap_or_else(|| "repo".to_string());

        Self {
            root,
            canonical_root,
            repo_name,
        }
    }

    /// Find root by searching upwards for .git, .code-kb, or workspace markers.
    pub fn find_workspace_root(start: &Path) -> Result<PathBuf, WorkspaceError> {
        let raw = start.to_string_lossy();
        let parsed = if raw.starts_with("file://") {
            parse_file_uri(&raw).unwrap_or_else(|| start.to_path_buf())
        } else {
            start.to_path_buf()
        };
        let parsed = trim_trailing_slash(&parsed);
        let curr = if parsed.is_file() {
            parsed.parent().unwrap_or(&parsed).to_path_buf()
        } else {
            parsed.clone()
        };

        // Pass 1: Look for .git or an existing index all the way up. A bare `.code-kb`
        // directory is not a marker: `~/.code-kb` holds telemetry and plugin downloads.
        let mut probe = curr.clone();
        loop {
            if probe.join(".code-kb").join("artifact.db").exists() || probe.join(".git").exists() {
                let canon = dunce::canonicalize(&probe).unwrap_or(probe);
                return Ok(normalize_path(&canon));
            }
            if let Some(name) = probe.file_name().and_then(|n| n.to_str())
                && is_hard_excluded(name)
            {
                break;
            }
            if let Some(parent) = probe.parent() {
                if parent == probe {
                    break;
                }
                probe = parent.to_path_buf();
            } else {
                break;
            }
        }

        // Pass 2: Look for language project markers
        let mut curr_marker = curr.clone();
        loop {
            if curr_marker.join("Cargo.toml").exists()
                || curr_marker.join("package.json").exists()
                || curr_marker.join("go.mod").exists()
                || curr_marker.join("pyproject.toml").exists()
            {
                let canon = dunce::canonicalize(&curr_marker).unwrap_or(curr_marker);
                return Ok(normalize_path(&canon));
            }
            if let Some(name) = curr_marker.file_name().and_then(|n| n.to_str())
                && is_hard_excluded(name)
            {
                break;
            }

            if let Some(parent) = curr_marker.parent() {
                if parent == curr_marker {
                    break;
                }
                curr_marker = parent.to_path_buf();
            } else {
                break;
            }
        }

        // Default to start directory if no markers found
        let start_dir = if parsed.is_file() {
            parsed.parent().unwrap_or(&parsed).to_path_buf()
        } else {
            parsed
        };
        let canon = dunce::canonicalize(&start_dir).unwrap_or(start_dir);
        Ok(normalize_path(&canon))
    }

    /// Resolves an input path (relative, absolute, or file:// URI) to a canonical absolute path and relative path.
    pub fn resolve_path(&self, input: &Path) -> Result<(PathBuf, String), WorkspaceError> {
        let raw_str = input.to_string_lossy();
        let path = if raw_str.starts_with("file://") {
            parse_file_uri(&raw_str).unwrap_or_else(|| input.to_path_buf())
        } else {
            input.to_path_buf()
        };
        let path = normalize_path(&path);

        let is_abs = path.is_absolute()
            || (cfg!(windows) && (path.to_string_lossy().chars().nth(1) == Some(':')));

        let joined = if is_abs {
            path
        } else {
            let rel_str = if cfg!(not(windows)) && path.to_string_lossy().contains('\\') {
                path.to_string_lossy().replace('\\', "/")
            } else {
                path.to_string_lossy().to_string()
            };
            self.canonical_root.join(Path::new(&rel_str))
        };

        // Lexically clean the path to collapse `.` and `..` components
        let cleaned = clean_path(&joined);
        let abs_path = normalize_path(&cleaned);

        // If file exists, canonicalize to resolve any symlinks
        let effective_abs = if abs_path.exists() {
            dunce::canonicalize(&abs_path)
                .map(|p| normalize_path(&p))
                .unwrap_or_else(|_| abs_path.clone())
        } else {
            abs_path.clone()
        };

        let norm_root = dunce::canonicalize(&self.canonical_root)
            .map(|p| normalize_path(&p))
            .unwrap_or_else(|_| self.canonical_root.clone());

        // Check if within canonical root (trying multiple normalization variants with case-insensitivity on Windows)
        let rel = match strip_prefix_lossy(&effective_abs, &norm_root)
            .or_else(|| strip_prefix_lossy(&effective_abs, &self.canonical_root))
            .or_else(|| {
                // Only fall back to uncanonicalized abs_path if the file does not exist yet (e.g. filters or uncreated files)
                if !abs_path.exists() {
                    strip_prefix_lossy(&abs_path, &norm_root)
                        .or_else(|| strip_prefix_lossy(&abs_path, &self.canonical_root))
                } else {
                    None
                }
            }) {
            Some(r) => {
                let forward = to_forward_slash(r);
                if forward.starts_with("../") || forward == ".." {
                    return Err(WorkspaceError::PathOutsideWorkspace(
                        abs_path,
                        self.canonical_root.clone(),
                    ));
                }
                forward
            }
            None => {
                return Err(WorkspaceError::PathOutsideWorkspace(
                    abs_path,
                    self.canonical_root.clone(),
                ));
            }
        };

        Ok((effective_abs, rel))
    }

    /// Relativizes a path filter string (which may be absolute, file:// URI, or relative)
    /// against this workspace root into a forward-slash relative path suitable for SQLite queries.
    pub fn relativize_filter(&self, filter: &str) -> String {
        let trimmed = filter.trim();
        if trimmed.is_empty() {
            return String::new();
        }

        // Handle file:// URI
        let path_str = if trimmed.starts_with("file://") {
            parse_file_uri(trimmed)
                .map(|p| p.to_string_lossy().to_string())
                .unwrap_or_else(|| trimmed.to_string())
        } else {
            trimmed.to_string()
        };

        let raw_path = Path::new(&path_str);
        let simplified = dunce::simplified(raw_path);

        if simplified.is_absolute() {
            if let Ok((_, rel)) = self.resolve_path(simplified) {
                return rel;
            }
            // If resolve_path failed (e.g. non-existent path), try prefix stripping on normalized strings
            let norm_simplified = normalize_path(simplified);
            let norm_root = normalize_path(&self.canonical_root);
            if let Some(rel) = strip_prefix_lossy(&norm_simplified, &norm_root)
                .or_else(|| strip_prefix_lossy(&norm_simplified, &self.root))
            {
                let forward = to_forward_slash(rel);
                if !forward.starts_with("../") && forward != ".." {
                    return forward.trim_matches('/').to_string();
                }
            }
        }

        // Relative path: pass through clean_path to collapse `.` and `..`, normalize slashes, and trim leading ./ or /
        let cleaned = clean_path(Path::new(&path_str));
        let forward = to_forward_slash(&cleaned);
        let trimmed = forward.trim_start_matches("./").trim_matches('/');
        if trimmed == "." {
            String::new()
        } else {
            trimmed.to_string()
        }
    }

    /// Resolve candidate database paths for this workspace:
    /// 1. Explicit override path (if provided)
    /// 2. In-tree `.code-kb/artifact.db` or `.code-kb/store.db`
    pub fn candidate_db_paths(&self, explicit_db: Option<&Path>) -> Vec<PathBuf> {
        let mut candidates = Vec::new();

        if let Some(p) = explicit_db {
            candidates.push(normalize_path(p));
        }

        // In-tree options
        candidates.push(normalize_path(
            &self.canonical_root.join(".code-kb").join("artifact.db"),
        ));
        candidates.push(normalize_path(
            &self.canonical_root.join(".code-kb").join("store.db"),
        ));
        candidates.push(normalize_path(&self.canonical_root.join("artifact.db")));

        candidates
    }

    /// Finds the first existing database file, or returns the default target location.
    pub fn locate_db(&self, explicit_db: Option<&Path>) -> Result<PathBuf, WorkspaceError> {
        if let Some(p) = explicit_db {
            return Ok(normalize_path(p));
        }

        let candidates = self.candidate_db_paths(None);
        for candidate in &candidates {
            if candidate.exists() && candidate.is_file() {
                return Ok(normalize_path(candidate));
            }
        }

        Ok(normalize_path(
            &self.canonical_root.join(".code-kb").join("artifact.db"),
        ))
    }
}

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

    #[test]
    #[cfg(windows)]
    fn test_normalize_path() {
        let p = PathBuf::from(r"\\?\C:\source\code-kb\src\main.rs");
        let norm = normalize_path(&p);
        assert!(!norm.to_string_lossy().starts_with(r"\\?\"));
    }

    #[test]
    fn test_find_workspace_root_ignores_ancestor_code_kb_without_index() {
        let temp = crate::safe_tempdir();
        let home = temp.path();
        std::fs::create_dir_all(home.join(".code-kb")).unwrap();
        std::fs::write(home.join(".code-kb").join("telemetry.db"), b"").unwrap();
        let project = home.join("project");
        std::fs::create_dir_all(&project).unwrap();
        std::fs::write(project.join("Cargo.toml"), "[package]\n").unwrap();

        let root = Workspace::find_workspace_root(&project).unwrap();

        assert!(paths_equal(&root, &project), "{}", root.display());
    }

    #[test]
    fn test_find_workspace_root_uses_ancestor_index() {
        let temp = crate::safe_tempdir();
        let repo = temp.path().join("repo");
        std::fs::create_dir_all(repo.join(".code-kb")).unwrap();
        std::fs::write(repo.join(".code-kb").join("artifact.db"), b"").unwrap();
        let nested = repo.join("src").join("deep");
        std::fs::create_dir_all(&nested).unwrap();

        let root = Workspace::find_workspace_root(&nested).unwrap();

        assert!(paths_equal(&root, &repo), "{}", root.display());
    }

    #[test]
    fn test_to_forward_slash() {
        let p = PathBuf::from(r"src\models\mod.rs");
        assert_eq!(to_forward_slash(&p), "src/models/mod.rs");
    }

    #[test]
    fn test_workspace_resolve_path() {
        let ws = Workspace::new(PathBuf::from("C:/source/test-project"));
        let (abs, rel) = ws.resolve_path(Path::new("src/lib.rs")).unwrap();
        assert_eq!(rel, "src/lib.rs");
        assert!(abs.to_string_lossy().contains("test-project"));

        #[cfg(windows)]
        {
            // Lowercase drive letter
            let (_abs2, rel2) = ws
                .resolve_path(Path::new("c:/source/test-project/src/lib.rs"))
                .unwrap();
            assert_eq!(rel2, "src/lib.rs");

            // Case-insensitive directory on Windows
            let (_abs3, rel3) = ws
                .resolve_path(Path::new("C:/SOURCE/test-project/src/lib.rs"))
                .unwrap();
            assert_eq!(rel3, "src/lib.rs");

            // file:// URI
            let (_abs4, rel4) = ws
                .resolve_path(Path::new("file:///C:/source/test-project/src/lib.rs"))
                .unwrap();
            assert_eq!(rel4, "src/lib.rs");

            // file:// URI with lowercase drive letter
            let (_abs5, rel5) = ws
                .resolve_path(Path::new("file:///c:/source/test-project/src/lib.rs"))
                .unwrap();
            assert_eq!(rel5, "src/lib.rs");
        }
    }

    #[test]
    fn test_workspace_resolve_path_traversal_escape() {
        let temp = crate::safe_tempdir();
        let ws = Workspace::new(temp.path().to_path_buf());
        let res = ws.resolve_path(Path::new("sub/../../outside.rs"));
        assert!(
            matches!(res, Err(WorkspaceError::PathOutsideWorkspace(..))),
            "Expected PathOutsideWorkspace error, but got: {:?}",
            res
        );
    }

    #[test]
    fn test_parse_file_uri() {
        #[cfg(windows)]
        let (uri, expected) = ("file:///C:/my%20folder/project", "C:/my folder/project");
        #[cfg(not(windows))]
        let (uri, expected) = ("file:///tmp/my%20folder/project", "/tmp/my folder/project");

        let p1 = parse_file_uri(uri).unwrap();
        assert_eq!(p1, normalize_path(Path::new(expected)));

        // Plain path fallback
        let p2 = parse_file_uri("C:/direct/path").unwrap();
        assert_eq!(p2, normalize_path(Path::new("C:/direct/path")));
    }

    #[test]
    fn test_relativize_filter() {
        let temp = crate::safe_tempdir();
        let ws = Workspace::new(temp.path().to_path_buf());

        // Relative path
        assert_eq!(ws.relativize_filter("."), "");
        assert_eq!(ws.relativize_filter("./"), "");
        assert_eq!(ws.relativize_filter("src/models"), "src/models");
        assert_eq!(ws.relativize_filter("./src/models/"), "src/models");
        assert_eq!(
            ws.relativize_filter(r"src\models\mod.rs"),
            "src/models/mod.rs"
        );

        // Relative path with ..
        assert_eq!(
            ws.relativize_filter("src/../src/models/mod.rs"),
            "src/models/mod.rs"
        );

        // Absolute path inside workspace
        let abs_file = temp.path().join("src").join("lib.rs");
        std::fs::create_dir_all(abs_file.parent().unwrap()).unwrap();
        std::fs::write(&abs_file, "").unwrap();

        assert_eq!(
            ws.relativize_filter(&abs_file.to_string_lossy()),
            "src/lib.rs"
        );

        // File URI
        let uri = format!("file://{}", abs_file.to_string_lossy().replace('\\', "/"));
        assert_eq!(ws.relativize_filter(&uri), "src/lib.rs");

        #[cfg(windows)]
        {
            // Case-insensitive absolute path for existing file resolves to canonical disk casing
            let upper_abs = abs_file.to_string_lossy().to_uppercase();
            assert_eq!(ws.relativize_filter(&upper_abs), "src/lib.rs");

            // File URI with alternate case
            let uri_cased = format!(
                "file:///{}",
                abs_file.to_string_lossy().replace('\\', "/").to_lowercase()
            );
            assert_eq!(ws.relativize_filter(&uri_cased), "src/lib.rs");
        }
    }

    #[test]
    fn test_paths_equal() {
        assert!(paths_equal(
            Path::new("src/lib.rs"),
            Path::new("src/lib.rs")
        ));
        assert!(!paths_equal(
            Path::new("src/lib.rs"),
            Path::new("src/main.rs")
        ));

        #[cfg(windows)]
        {
            // Case-insensitive drive letters and paths
            assert!(paths_equal(
                Path::new(r"C:\source\code-kb\src\lib.rs"),
                Path::new(r"c:\source\code-kb\src\lib.rs")
            ));
            assert!(paths_equal(
                Path::new(r"C:\source\code-kb\src\lib.rs"),
                Path::new(r"c:\SOURCE\CODE-KB\SRC\LIB.RS")
            ));
            // Verbatim prefixes
            assert!(paths_equal(
                Path::new(r"\\?\C:\source\code-kb\src\lib.rs"),
                Path::new(r"C:\source\code-kb\src\lib.rs")
            ));
            assert!(paths_equal(
                Path::new(r"\\?\c:\source\code-kb\src\lib.rs"),
                Path::new(r"C:\source\code-kb\src\lib.rs")
            ));
            // UNC paths
            assert!(paths_equal(
                Path::new(r"\\server\share\file"),
                Path::new(r"\\SERVER\SHARE\file")
            ));
            assert!(paths_equal(
                Path::new(r"\\server\share\file"),
                Path::new(r"\\server\share\file")
            ));
            assert!(!paths_equal(
                Path::new(r"\\server\share1\file"),
                Path::new(r"\\server\share2\file")
            ));
        }
    }

    #[test]
    fn test_strip_prefix_lossy() {
        let base = Path::new("src");
        assert_eq!(
            strip_prefix_lossy(Path::new("src/lib.rs"), base),
            Some(Path::new("lib.rs"))
        );
        assert_eq!(strip_prefix_lossy(Path::new("tests/foo.rs"), base), None);

        #[cfg(windows)]
        {
            let base_win = Path::new(r"C:\source\code-kb");
            // Standard path
            assert_eq!(
                strip_prefix_lossy(Path::new(r"C:\source\code-kb\src\lib.rs"), base_win),
                Some(Path::new(r"src\lib.rs"))
            );
            // Disk prefix casing
            assert_eq!(
                strip_prefix_lossy(Path::new(r"c:\source\code-kb\src\lib.rs"), base_win),
                Some(Path::new(r"src\lib.rs"))
            );
            assert_eq!(
                strip_prefix_lossy(Path::new(r"c:\SOURCE\CODE-KB\src\lib.rs"), base_win),
                Some(Path::new(r"src\lib.rs"))
            );
            // Verbatim prefixes
            assert_eq!(
                strip_prefix_lossy(Path::new(r"\\?\C:\source\code-kb\src\lib.rs"), base_win),
                Some(Path::new(r"src\lib.rs"))
            );
            assert_eq!(
                strip_prefix_lossy(Path::new(r"\\?\c:\source\code-kb\src\lib.rs"), base_win),
                Some(Path::new(r"src\lib.rs"))
            );
            // Negative non-matching paths
            assert_eq!(
                strip_prefix_lossy(Path::new(r"C:\other\code-kb\src\lib.rs"), base_win),
                None
            );
            assert_eq!(
                strip_prefix_lossy(Path::new(r"D:\source\code-kb\src\lib.rs"), base_win),
                None
            );
        }
    }

    #[test]
    fn test_parse_file_uri_two_slash_and_percent() {
        #[cfg(windows)]
        {
            let p1 = parse_file_uri("file://C:/my%20folder/lib.rs").unwrap();
            assert_eq!(p1, normalize_path(Path::new("C:/my folder/lib.rs")));

            let p2 = parse_file_uri("file://c:/my%20folder/lib.rs").unwrap();
            assert_eq!(p2, normalize_path(Path::new("c:/my folder/lib.rs")));

            let p3 = parse_file_uri("file:///C:/my%20folder/lib.rs").unwrap();
            assert_eq!(p3, normalize_path(Path::new("C:/my folder/lib.rs")));
        }
        #[cfg(not(windows))]
        {
            let p1 = parse_file_uri("file:///my%20folder/lib.rs").unwrap();
            assert_eq!(p1, normalize_path(Path::new("/my folder/lib.rs")));
        }
    }

    #[test]
    fn test_workspace_verbatim_root_and_db_cleanup() {
        let temp = crate::safe_tempdir();
        let verbatim_path = format!(r"\\?\{}", temp.path().display());
        let ws = Workspace::new(PathBuf::from(&verbatim_path));
        assert!(!ws.root.to_string_lossy().starts_with(r"\\?\"));
        assert!(!ws.canonical_root.to_string_lossy().starts_with(r"\\?\"));

        let explicit = PathBuf::from(format!(r"\\?\{}\test.db", temp.path().display()));
        let located = ws.locate_db(Some(&explicit)).unwrap();
        assert!(!located.to_string_lossy().starts_with(r"\\?\"));
    }

    #[test]
    fn test_trim_trailing_slash_edge_cases() {
        assert_eq!(trim_trailing_slash(Path::new("/")), PathBuf::from("/"));
        assert_eq!(trim_trailing_slash(Path::new("///")), PathBuf::from("/"));
        assert_eq!(
            trim_trailing_slash(Path::new("/a/b/")),
            PathBuf::from("/a/b")
        );
        assert_eq!(
            trim_trailing_slash(Path::new("foo/bar/")),
            PathBuf::from("foo/bar")
        );

        #[cfg(windows)]
        {
            assert_eq!(
                trim_trailing_slash(Path::new("C:\\")),
                PathBuf::from("C:\\")
            );
            assert_eq!(trim_trailing_slash(Path::new("C:/")), PathBuf::from("C:\\"));
            assert_eq!(
                trim_trailing_slash(Path::new("C://")),
                PathBuf::from("C:\\")
            );
            assert_eq!(
                trim_trailing_slash(Path::new("C:\\\\")),
                PathBuf::from("C:\\")
            );
            assert_eq!(
                trim_trailing_slash(Path::new("C:/foo/")),
                PathBuf::from("C:/foo")
            );
        }
    }

    #[test]
    fn test_unicode_and_emoji_uri_safety() {
        // Must not panic on non-ASCII character boundaries
        let p1 = parse_file_uri("file:///a😀/x");
        assert!(p1.is_some());

        let p2 = parse_file_uri("file:///c😀/x");
        assert!(p2.is_some());

        let p3 = parse_file_uri("file:///localhost😀/x");
        assert!(p3.is_some());

        let p4 = parse_file_uri("file://C:/😀😀/main.rs");
        assert!(p4.is_some());
    }

    #[test]
    #[cfg(unix)]
    fn test_escaping_symlink_rejected() {
        let ws_dir = crate::safe_tempdir();
        let ext_dir = crate::safe_tempdir();

        let ext_file = ext_dir.path().join("secret.txt");
        std::fs::write(&ext_file, "secret").unwrap();

        let symlink_path = ws_dir.path().join("link.txt");
        std::os::unix::fs::symlink(&ext_file, &symlink_path).unwrap();
        let ws = Workspace::new(ws_dir.path().to_path_buf());
        let res = ws.resolve_path(&symlink_path);
        assert!(
            matches!(res, Err(WorkspaceError::PathOutsideWorkspace(..))),
            "Expected PathOutsideWorkspace, got: {res:?}"
        );
    }
}