rbt_weave

use std::collections::HashMap;
use std::path::{Path, PathBuf};

use rayon::prelude::*;
use unicode_normalization::UnicodeNormalization;

use crate::entity::{Entity, Label};
use crate::parser::ParseError;

/// Maximum length of entity filename stem (without `.md`).
const MAX_FILENAME_LEN: usize = 200;

/// A loaded entity with its source file path.
#[derive(Debug)]
pub struct RegistryEntry {
    pub entity: Entity,
    pub path: PathBuf,
    pub tags: Vec<String>,
}

/// Entity registry: holds all shared entities loaded from `people/` and
/// `organizations/` directories. Provides name-based lookup for cross-file
/// resolution.
#[derive(Debug)]
pub struct EntityRegistry {
    entries: Vec<RegistryEntry>,
    /// Name → index into `entries`. Names are case-sensitive.
    name_index: HashMap<String, usize>,
    /// Content root directory for computing file-path slugs.
    content_root: Option<PathBuf>,
}

impl EntityRegistry {
    /// Build a registry from a content root directory.
    ///
    /// Scans `{root}/people/**/*.md` and `{root}/organizations/**/*.md`, parses
    /// each file, validates for duplicates and filename mismatches. Supports both
    /// flat and nested (country-based) directory layouts.
    pub fn load(root: &Path) -> Result<Self, Vec<ParseError>> {
        let mut entries = Vec::new();
        let mut errors = Vec::new();

        let actor_dir = root.join("people");
        let institution_dir = root.join("organizations");

        load_directory(&actor_dir, Label::Person, &mut entries, &mut errors);
        load_directory(
            &institution_dir,
            Label::Organization,
            &mut entries,
            &mut errors,
        );

        // Build name index and detect duplicates
        let name_index = build_name_index(&entries, &mut errors);

        if errors.iter().any(|e| e.message.starts_with("duplicate")) {
            return Err(errors);
        }

        // Filename mismatch warnings are non-fatal, report via errors but don't fail
        // (caller can filter by message prefix if needed)

        if errors.iter().any(|e| !e.message.starts_with("warning:")) {
            return Err(errors);
        }

        // Warnings only -- attach them but succeed
        if !errors.is_empty() {
            for err in &errors {
                eprintln!("{err}");
            }
        }

        Ok(Self {
            entries,
            name_index,
            content_root: Some(root.to_path_buf()),
        })
    }

    /// Build a registry from pre-parsed entries.
    pub fn from_entries(entries: Vec<RegistryEntry>) -> Result<Self, Vec<ParseError>> {
        let mut errors = Vec::new();
        let name_index = build_name_index(&entries, &mut errors);

        let has_errors = errors.iter().any(|e| !e.message.starts_with("warning:"));
        if has_errors {
            return Err(errors);
        }

        Ok(Self {
            entries,
            name_index,
            content_root: None,
        })
    }

    /// Look up an entity by name. Returns None if not found.
    pub fn get_by_name(&self, name: &str) -> Option<&RegistryEntry> {
        self.name_index.get(name).map(|&idx| &self.entries[idx])
    }

    /// Number of entities in the registry.
    pub fn len(&self) -> usize {
        self.entries.len()
    }

    /// Whether the registry is empty.
    pub fn is_empty(&self) -> bool {
        self.entries.is_empty()
    }

    /// All entity names in the registry.
    pub fn names(&self) -> Vec<&str> {
        self.entries
            .iter()
            .map(|e| e.entity.name.as_str())
            .collect()
    }

    /// All registry entries.
    pub fn entries(&self) -> &[RegistryEntry] {
        &self.entries
    }

    /// Compute the file-path slug for an entry (path relative to content root, minus `.md`).
    /// Returns `None` if content root is not set.
    pub fn slug_for(&self, entry: &RegistryEntry) -> Option<String> {
        let root = self.content_root.as_ref()?;
        path_to_slug(&entry.path, root)
    }

    /// Content root directory, if set.
    pub fn content_root(&self) -> Option<&Path> {
        self.content_root.as_deref()
    }

    /// Check all entity filenames against expected naming convention.
    /// Returns warning messages for mismatches (same checks as `load()`,
    /// but accessible after loading for strict validation).
    pub fn check_filenames(&self) -> Vec<ParseError> {
        let mut warnings = Vec::new();
        for entry in &self.entries {
            validate_filename(&entry.path, &entry.entity, &mut warnings);
        }
        warnings
    }
}

/// Compute file-path slug from an absolute path relative to content root.
/// Returns the path minus the `.md` extension, e.g. `people/id/harvey-moeis`.
/// Strips diacritics from the filename stem for ASCII-safe URLs.
pub fn path_to_slug(path: &Path, content_root: &Path) -> Option<String> {
    let relative = path.strip_prefix(content_root).ok()?;
    let s = relative.to_str()?;
    // NFC-normalize to handle macOS NFD filesystem paths, then strip
    // diacritics from the filename portion for ASCII-safe URL slugs.
    let nfc: String = s.nfc().collect();
    let without_ext = nfc.strip_suffix(".md").unwrap_or(&nfc);
    // Only transliterate the filename, not directory segments (country codes etc.)
    if let Some((dir, stem)) = without_ext.rsplit_once('/') {
        let ascii_stem = strip_diacritics(stem);
        Some(format!("{dir}/{ascii_stem}"))
    } else {
        Some(strip_diacritics(without_ext))
    }
}

/// Strip diacritics from a string by NFD-decomposing and removing combining marks.
/// Also transliterates non-decomposable Latin variants (ł→l, đ→d, ı→i, ß→ss, etc.).
fn strip_diacritics(s: &str) -> String {
    let mut out = String::with_capacity(s.len());
    for c in s.nfd().filter(|c| !is_combining_mark(*c)) {
        match transliterate_latin(c) {
            Some(replacement) => out.push_str(replacement),
            None => out.push(c),
        }
    }
    out
}

/// Transliterate non-decomposable Latin characters to ASCII equivalents.
/// NFD decomposition handles most diacritics (é→e+combining), but some
/// characters like ł, đ, ı are atomic and need explicit mapping.
/// Returns `Some` for characters that need transliteration, `None` to keep as-is.
fn transliterate_latin(c: char) -> Option<&'static str> {
    match c {
        'ł' => Some("l"),
        'Ł' => Some("L"),
        'đ' | 'ð' => Some("d"),
        'Đ' | 'Ð' => Some("D"),
        'ı' => Some("i"), // Turkish dotless i
        'ø' => Some("o"),
        'Ø' => Some("O"),
        'æ' => Some("ae"),
        'Æ' => Some("AE"),
        'œ' => Some("oe"),
        'Œ' => Some("OE"),
        'ß' => Some("ss"), // German sharp s
        'þ' => Some("th"), // Icelandic thorn
        'Þ' => Some("TH"),
        _ => None,
    }
}

/// Load all `.md` files from a directory tree, parsing each as an entity file.
/// Supports both flat (`people/*.md`) and nested (`people/<country>/*.md`)
/// layouts. Uses rayon to parse files in parallel.
fn load_directory(
    dir: &Path,
    label: Label,
    entries: &mut Vec<RegistryEntry>,
    errors: &mut Vec<ParseError>,
) {
    let mut paths = Vec::new();
    collect_md_files(dir, &mut paths, 0);

    // Sort for deterministic ordering
    paths.sort();

    // Parse all files in parallel, collect results
    let results: Vec<ParseResult> = paths
        .par_iter()
        .map(|path| parse_entity_file(path, label))
        .collect();

    // Merge results sequentially to preserve deterministic order
    for result in results {
        if let Some(entry) = result.entry {
            entries.push(entry);
        }
        errors.extend(result.errors);
    }
}

/// Recursively collect `.md` files from a directory tree.
/// Max depth 2 supports `people/<country>/file.md` layout.
fn collect_md_files(dir: &Path, paths: &mut Vec<PathBuf>, depth: usize) {
    const MAX_DEPTH: usize = 2;
    if depth > MAX_DEPTH {
        return;
    }

    let Ok(read_dir) = std::fs::read_dir(dir) else {
        return;
    };

    let mut dir_entries: Vec<_> = read_dir.filter_map(Result::ok).collect();
    dir_entries.sort_by_key(std::fs::DirEntry::file_name);

    for entry in dir_entries {
        let path = entry.path();
        if path.is_dir() {
            collect_md_files(&path, paths, depth + 1);
        } else if path.extension().and_then(|e| e.to_str()) == Some("md") {
            paths.push(path);
        }
    }
}

/// Result of parsing a single entity file.
struct ParseResult {
    entry: Option<RegistryEntry>,
    errors: Vec<ParseError>,
}

/// Parse a single entity file, returning the entry and any errors/warnings.
fn parse_entity_file(path: &Path, label: Label) -> ParseResult {
    let content = match std::fs::read_to_string(path) {
        Ok(c) => c,
        Err(e) => {
            return ParseResult {
                entry: None,
                errors: vec![ParseError {
                    line: 0,
                    message: format!("{}: error reading file: {e}", path.display()),
                }],
            };
        }
    };

    let parsed = match crate::parser::parse_entity_file(&content) {
        Ok(p) => p,
        Err(parse_errors) => {
            return ParseResult {
                entry: None,
                errors: parse_errors
                    .into_iter()
                    .map(|err| ParseError {
                        line: err.line,
                        message: format!("{}: {}", path.display(), err.message),
                    })
                    .collect(),
            };
        }
    };

    let mut field_errors = Vec::new();
    let mut entity = crate::entity::parse_entity_file_body(
        &parsed.name,
        &parsed.body,
        label,
        parsed.id,
        parsed.title_line,
        &mut field_errors,
    );
    entity.tags.clone_from(&parsed.tags);

    let mut errors: Vec<ParseError> = field_errors
        .into_iter()
        .map(|err| ParseError {
            line: err.line,
            message: format!("{}: {}", path.display(), err.message),
        })
        .collect();

    // Validate filename matches content
    validate_filename(path, &entity, &mut errors);

    ParseResult {
        entry: Some(RegistryEntry {
            entity,
            path: path.to_path_buf(),
            tags: parsed.tags,
        }),
        errors,
    }
}

/// Build name → index map, detecting duplicate names.
fn build_name_index(
    entries: &[RegistryEntry],
    errors: &mut Vec<ParseError>,
) -> HashMap<String, usize> {
    let mut index = HashMap::new();

    for (i, entry) in entries.iter().enumerate() {
        let name = &entry.entity.name;
        if let Some(&existing_idx) = index.get(name.as_str()) {
            let existing: &RegistryEntry = &entries[existing_idx];
            errors.push(ParseError {
                line: entry.entity.line,
                message: format!(
                    "duplicate entity name {name:?} in {} (first defined in {})",
                    entry.path.display(),
                    existing.path.display(),
                ),
            });
        } else {
            index.insert(name.clone(), i);
        }
    }

    index
}

/// Warn if entity filename doesn't match content.
/// Expected: `<name>--<qualifier>.md` in kebab-case.
fn validate_filename(path: &Path, entity: &Entity, errors: &mut Vec<ParseError>) {
    let Some(raw_stem) = path.file_stem().and_then(|s| s.to_str()) else {
        return;
    };
    // macOS APFS/HFS+ returns NFD filenames; normalize to NFC for comparison.
    let stem: String = raw_stem.nfc().collect();

    if stem.len() > MAX_FILENAME_LEN {
        errors.push(ParseError {
            line: 0,
            message: format!(
                "warning: {}: filename stem exceeds {MAX_FILENAME_LEN} chars",
                path.display()
            ),
        });
    }

    let expected_name = to_kebab_case(&entity.name);
    let qualifier = entity
        .fields
        .iter()
        .find(|(k, _)| k == "qualifier")
        .and_then(|(_, v)| match v {
            crate::entity::FieldValue::Single(s) => Some(s.as_str()),
            crate::entity::FieldValue::List(_) => None,
        });

    let expected_stem = match qualifier {
        Some(q) => format!("{expected_name}--{}", to_kebab_case(q)),
        None => expected_name,
    };

    if stem != expected_stem {
        errors.push(ParseError {
            line: 0,
            message: format!(
                "warning: {}: filename {stem:?} doesn't match expected {expected_stem:?}",
                path.display()
            ),
        });
    }
}

/// Convert a display name to kebab-case for filename comparison.
/// Strips diacritics by NFD-decomposing and removing combining marks,
/// producing ASCII-safe slugs (e.g. "Prlić" → "prlic", "José García" → "jose-garcia").
fn to_kebab_case(s: &str) -> String {
    use unicode_normalization::UnicodeNormalization;
    let mut transliterated = String::with_capacity(s.len());
    for c in s.nfd().filter(|c| !is_combining_mark(*c)) {
        match transliterate_latin(c) {
            Some(replacement) => transliterated.push_str(replacement),
            None => transliterated.push(c),
        }
    }
    transliterated
        .chars()
        .map(|c| {
            if c.is_ascii_alphanumeric() {
                c.to_ascii_lowercase()
            } else if c.is_alphanumeric() {
                // Non-ASCII alphanumeric after stripping diacritics (e.g. CJK, Arabic)
                // — keep as-is in lowercase
                c.to_lowercase().next().unwrap_or(c)
            } else {
                '-'
            }
        })
        .collect::<String>()
        .split('-')
        .filter(|p| !p.is_empty())
        .collect::<Vec<_>>()
        .join("-")
}

/// Check if a character is a Unicode combining mark (category M).
fn is_combining_mark(c: char) -> bool {
    // Combining marks: Mn (nonspacing), Mc (spacing combining), Me (enclosing)
    // After NFD decomposition, diacritics become separate combining characters.
    matches!(c as u32,
        0x0300..=0x036F    // Combining Diacritical Marks
        | 0x1AB0..=0x1AFF  // Combining Diacritical Marks Extended
        | 0x1DC0..=0x1DFF  // Combining Diacritical Marks Supplement
        | 0x20D0..=0x20FF  // Combining Diacritical Marks for Symbols
        | 0xFE20..=0xFE2F  // Combining Half Marks
    )
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::entity::{Entity, FieldValue, Label};

    fn make_entry(name: &str, label: Label, path: &str) -> RegistryEntry {
        RegistryEntry {
            entity: Entity {
                name: name.to_string(),
                label,
                fields: Vec::new(),
                id: None,
                line: 1,
                tags: Vec::new(),
                slug: None,
            },
            path: PathBuf::from(path),
            tags: Vec::new(),
        }
    }

    #[test]
    fn registry_from_entries_lookup() {
        let entries = vec![
            make_entry("Alice", Label::Person, "people/alice.md"),
            make_entry("Corp Inc", Label::Organization, "organizations/corp-inc.md"),
        ];

        let registry = EntityRegistry::from_entries(entries).unwrap();
        assert_eq!(registry.len(), 2);
        assert!(registry.get_by_name("Alice").is_some());
        assert!(registry.get_by_name("Corp Inc").is_some());
        assert!(registry.get_by_name("Bob").is_none());
    }

    #[test]
    fn registry_detects_duplicate_names() {
        let entries = vec![
            make_entry("Alice", Label::Person, "people/alice-a.md"),
            make_entry("Alice", Label::Person, "people/alice-b.md"),
        ];

        let errors = EntityRegistry::from_entries(entries).unwrap_err();
        assert!(errors.iter().any(|e| e.message.contains("duplicate")));
    }

    #[test]
    fn registry_names_list() {
        let entries = vec![
            make_entry("Alice", Label::Person, "people/alice.md"),
            make_entry("Bob", Label::Person, "people/bob.md"),
        ];

        let registry = EntityRegistry::from_entries(entries).unwrap();
        let names = registry.names();
        assert!(names.contains(&"Alice"));
        assert!(names.contains(&"Bob"));
    }

    #[test]
    fn to_kebab_case_conversion() {
        assert_eq!(to_kebab_case("Mark Bonnick"), "mark-bonnick");
        assert_eq!(to_kebab_case("Arsenal FC"), "arsenal-fc");
        assert_eq!(
            to_kebab_case("English Football Club"),
            "english-football-club"
        );
        assert_eq!(to_kebab_case("Bob"), "bob");
    }

    #[test]
    fn to_kebab_case_strips_diacritics() {
        assert_eq!(to_kebab_case("Prlić"), "prlic");
        assert_eq!(to_kebab_case("José García"), "jose-garcia");
        assert_eq!(
            to_kebab_case("Médecins Sans Frontières"),
            "medecins-sans-frontieres"
        );
        assert_eq!(to_kebab_case("Túpac Amaru"), "tupac-amaru");
        assert_eq!(
            to_kebab_case("Mészáros és Mészáros"),
            "meszaros-es-meszaros"
        );
        assert_eq!(to_kebab_case("Sûreté du Québec"), "surete-du-quebec");
        // Non-decomposable Latin characters
        assert_eq!(to_kebab_case("Wałęsa"), "walesa");
        assert_eq!(to_kebab_case("Đinđić"), "dindic");
        assert_eq!(to_kebab_case("İbrahim Fırtına"), "ibrahim-firtina");
        // German umlauts
        assert_eq!(to_kebab_case("Müller Über Straße"), "muller-uber-strasse");
        assert_eq!(to_kebab_case("Köln Düsseldorf"), "koln-dusseldorf");
    }

    #[test]
    fn validate_filename_matching() {
        let entity = Entity {
            name: "Mark Bonnick".to_string(),
            label: Label::Person,
            fields: vec![(
                "qualifier".to_string(),
                FieldValue::Single("Arsenal Kit Manager".to_string()),
            )],
            id: None,
            line: 1,
            tags: Vec::new(),
            slug: None,
        };

        let mut errors = Vec::new();

        // Correct filename
        validate_filename(
            Path::new("people/mark-bonnick--arsenal-kit-manager.md"),
            &entity,
            &mut errors,
        );
        assert!(errors.is_empty(), "errors: {errors:?}");

        // Wrong filename
        validate_filename(Path::new("people/wrong-name.md"), &entity, &mut errors);
        assert!(errors.iter().any(|e| e.message.contains("warning:")));
    }

    #[test]
    fn validate_filename_no_qualifier() {
        let entity = Entity {
            name: "Bob".to_string(),
            label: Label::Person,
            fields: Vec::new(),
            id: None,
            line: 1,
            tags: Vec::new(),
            slug: None,
        };

        let mut errors = Vec::new();
        validate_filename(Path::new("people/bob.md"), &entity, &mut errors);
        assert!(errors.is_empty(), "errors: {errors:?}");
    }

    #[test]
    fn empty_registry() {
        let registry = EntityRegistry::from_entries(Vec::new()).unwrap();
        assert!(registry.is_empty());
        assert_eq!(registry.len(), 0);
        assert!(registry.get_by_name("anything").is_none());
    }
}