tiger_lib/data/

localization.rs

//! Validate `.yml` localization files

use std::borrow::Borrow;
use std::cmp::Ordering;
use std::collections::hash_map::Entry;
use std::ffi::OsStr;
use std::fs::read_to_string;
#[cfg(any(feature = "ck3", feature = "vic3", feature = "imperator"))]
use std::io::Cursor;
use std::path::{Path, PathBuf};
use std::sync::LazyLock;
use std::sync::atomic::AtomicBool;
use std::sync::atomic::Ordering::Relaxed;

use bitvec::order::Lsb0;
use bitvec::{BitArr, bitarr};
#[cfg(any(feature = "ck3", feature = "vic3", feature = "imperator"))]
use murmur3::murmur3_32;
use rayon::iter::{IntoParallelRefIterator, ParallelIterator};
use rayon::scope;
use strum::{EnumCount, IntoEnumIterator};
use strum_macros::{Display, EnumCount, EnumIter, EnumString, FromRepr, IntoStaticStr};

use crate::block::Block;
#[cfg(feature = "ck3")]
use crate::ck3::tables::localization::{BUILTIN_MACROS_CK3, COMPLEX_TOOLTIPS_CK3};
use crate::context::ScopeContext;
use crate::datacontext::DataContext;
use crate::datatype::{CodeChain, Datatype, validate_datatypes};
#[cfg(feature = "eu5")]
use crate::eu5::tables::localization::BUILTIN_MACROS_EU5;
use crate::everything::Everything;
use crate::fileset::{FileEntry, FileHandler, FileKind};
use crate::game::Game;
#[cfg(any(feature = "ck3", feature = "vic3", feature = "imperator"))]
use crate::helpers::TigerHashMapExt;
use crate::helpers::{TigerHashMap, dup_error, stringify_list};
#[cfg(feature = "hoi4")]
use crate::hoi4::tables::localization::BUILTIN_MACROS_HOI4;
#[cfg(feature = "imperator")]
use crate::imperator::tables::localization::BUILTIN_MACROS_IMPERATOR;
use crate::item::Item;
use crate::macros::{MACRO_MAP, MacroMapIndex};
use crate::parse::ParserMemory;
use crate::parse::localization::{ValueParser, parse_loca};
use crate::report::{ErrorKey, Severity, err, report, tips, warn};
use crate::scopes::Scopes;
use crate::token::Token;
#[cfg(feature = "vic3")]
use crate::vic3::tables::localization::BUILTIN_MACROS_VIC3;

#[derive(Debug)]
pub struct Languages([TigerHashMap<&'static str, LocaEntry>; Language::COUNT]);

impl core::ops::Index<Language> for Languages {
    type Output = TigerHashMap<&'static str, LocaEntry>;

    fn index(&self, index: Language) -> &Self::Output {
        &self.0[index.to_idx()]
    }
}

impl core::ops::IndexMut<Language> for Languages {
    fn index_mut(&mut self, index: Language) -> &mut Self::Output {
        &mut self.0[index.to_idx()]
    }
}

/// Database of all loaded localization keys and their values, for all supported languages.
#[derive(Debug)]
pub struct Localization {
    /// Which languages to check, according to the config file.
    check_langs: BitArr!(for Language::COUNT, in u16),
    /// Which languages also actually exist in the mod.
    /// This is used to not warn about missing loca when a mod doesn't have the language at all.
    /// (This saves them the effort of configuring `check_langs`).
    mod_langs: BitArr!(for Language::COUNT, in u16),
    /// Database of all localizations, indexed first by language and then by localization key.
    locas: Languages,
}

/// List of languages that are supported by the game engine.
// LAST UPDATED CK3 VERSION 1.15.0
// LAST UPDATED VIC3 VERSION 1.7.6
#[derive(
    Debug,
    PartialEq,
    Eq,
    Clone,
    Copy,
    EnumString,
    EnumCount,
    EnumIter,
    FromRepr,
    IntoStaticStr,
    Display,
)]
#[strum(serialize_all = "snake_case")]
#[repr(u8)]
pub enum Language {
    English,
    Spanish,
    French,
    German,
    Russian,
    #[cfg(any(feature = "ck3", feature = "vic3", feature = "eu5"))]
    Korean,
    SimpChinese,
    #[cfg(any(feature = "vic3", feature = "hoi4", feature = "eu5"))]
    BrazPor,
    #[cfg(any(feature = "ck3", feature = "vic3", feature = "hoi4", feature = "eu5"))]
    Japanese,
    #[cfg(any(feature = "ck3", feature = "vic3", feature = "hoi4", feature = "eu5"))]
    Polish,
    #[cfg(any(feature = "vic3", feature = "eu5"))]
    Turkish,
}

static L_LANGS: LazyLock<Box<[Box<str>]>> =
    LazyLock::new(|| Language::iter().map(|l| format!("l_{l}").into_boxed_str()).collect());

static LANG_LIST: LazyLock<Box<str>> = LazyLock::new(|| {
    Language::iter().map(|l| l.to_string()).collect::<Vec<String>>().join(",").into_boxed_str()
});

impl Language {
    fn from_idx(idx: usize) -> Self {
        // SAFETY: This is safe to call assuming all indices were obtained from `to_idx`.
        #[allow(clippy::cast_possible_truncation)]
        Self::from_repr(idx as u8).unwrap()
    }
    fn to_idx(self) -> usize {
        self as usize
    }
}

/// List of known built-in keys used between `$...$` in any localization.
/// This list is used to avoid reporting false positives.
// TODO: maybe make the list more specific about which keys can contain which builtins
fn is_builtin_macro<S: Borrow<str>>(s: S) -> bool {
    let s = s.borrow();
    match Game::game() {
        #[cfg(feature = "ck3")]
        Game::Ck3 => BUILTIN_MACROS_CK3.contains(&s),
        #[cfg(feature = "vic3")]
        Game::Vic3 => BUILTIN_MACROS_VIC3.contains(&s),
        #[cfg(feature = "imperator")]
        Game::Imperator => BUILTIN_MACROS_IMPERATOR.contains(&s),
        #[cfg(feature = "eu5")]
        Game::Eu5 => BUILTIN_MACROS_EU5.contains(&s),
        #[cfg(feature = "hoi4")]
        Game::Hoi4 => BUILTIN_MACROS_HOI4.contains(&s),
    }
}

/// One parsed key: value line from the localization values.
#[derive(Debug)]
pub struct LocaEntry {
    key: Token,
    value: LocaValue,
    /// The original unparsed value, with enclosing `"` stripped.
    /// This is used for macro replacement.
    orig: Option<Token>,
    /// Whether this entry has been "used" (looked up) by anything in the mod
    used: AtomicBool,
    /// Whether this entry has been validated with a `ScopeContext`
    validated: AtomicBool,
}

impl PartialEq for LocaEntry {
    fn eq(&self, other: &LocaEntry) -> bool {
        self.key.loc == other.key.loc
    }
}

impl Eq for LocaEntry {}

impl PartialOrd for LocaEntry {
    fn partial_cmp(&self, other: &LocaEntry) -> Option<Ordering> {
        Some(self.cmp(other))
    }
}

impl Ord for LocaEntry {
    fn cmp(&self, other: &LocaEntry) -> Ordering {
        self.key.loc.cmp(&other.key.loc)
    }
}

impl LocaEntry {
    pub fn new(key: Token, value: LocaValue, orig: Option<Token>) -> Self {
        Self { key, value, orig, used: AtomicBool::new(false), validated: AtomicBool::new(false) }
    }

    // returns false to abort expansion in case of an error
    fn expand_macros<'a>(
        &'a self,
        vec: &mut Vec<Token>,
        from: &'a TigerHashMap<&'a str, LocaEntry>,
        count: &mut usize,
        sc: &mut ScopeContext,
        link: Option<MacroMapIndex>,
        data: &Everything,
    ) -> bool {
        // Are we (probably) stuck in a macro loop?
        if *count > 250 {
            return false;
        }
        *count += 1;

        if let LocaValue::Macro(v) = &self.value {
            for macrovalue in v {
                match macrovalue {
                    MacroValue::Text(token) => vec.push(token.clone().linked(link)),
                    MacroValue::Keyword(keyword) => {
                        if let Some(entry) = from.get(keyword.as_str()) {
                            entry.used.store(true, Relaxed);
                            entry.validated.store(true, Relaxed);
                            if !entry.expand_macros(
                                vec,
                                from,
                                count,
                                sc,
                                Some(MACRO_MAP.get_or_insert_loc(keyword.loc)),
                                data,
                            ) {
                                return false;
                            }
                        } else if is_builtin_macro(keyword) {
                            // we can't know what value it really has, so replace it with itself to
                            // at least get comprehensible error messages
                            vec.push(keyword.clone().linked(link));
                        } else if let Some(scopes) = sc.is_name_defined(keyword.as_str(), data) {
                            if scopes.contains(Scopes::Value) {
                                // same as above... we can't know what value it really has
                                vec.push(keyword.clone().linked(link));
                            } else {
                                let msg = &format!(
                                    "The substitution parameter ${keyword}$ is not defined anywhere as a key."
                                );
                                warn(ErrorKey::Localization).msg(msg).loc(keyword).push();
                            }
                        } else {
                            let msg = &format!(
                                "The substitution parameter ${keyword}$ is not defined anywhere as a key."
                            );
                            warn(ErrorKey::Localization).msg(msg).loc(keyword).push();
                        }
                    }
                }
            }
            true
        } else if let Some(orig) = &self.orig {
            vec.push(orig.clone().linked(link));
            true
        } else {
            false
        }
    }
}

#[derive(Clone, Debug, Default)]
pub enum LocaValue {
    // If the LocaValue is a Macro type, then it should be re-parsed after the macro values
    // have been filled in. Some macro values are supplied at runtime and we'll have to guess
    // at those.
    Macro(Vec<MacroValue>),
    Concat(Vec<LocaValue>),
    #[allow(dead_code)] // the Token is only used for ck3
    Text(Token),
    Markup,
    MarkupEnd,
    Tooltip(Token),
    // Tag, key, value. Tag can influence how tooltip is looked up. If tag is `GAME_TRAIT`,
    // tooltip is a trait name and value is a character id. Any of the tokens may be a datatype
    // expression, which is passed through unparsed here.
    // The value is not stored in the enum because we don't validate it.
    // TODO: instead of Token here, maybe need Box<LocaValue> or a Vec<LocaValue>, or maybe a type
    // that's specifically "Token or CodeChain"
    ComplexTooltip(Box<Token>, Token),
    // The optional token is the formatting
    Code(CodeChain, Option<Token>),
    Icon(Token),
    // An Icon with an [ ] expression inside it
    CalculatedIcon(Vec<LocaValue>),
    Flag(Token),
    #[default]
    Error,
}

#[derive(Clone, Debug)]
pub enum MacroValue {
    Text(Token),
    // The formatting is not stored in the enum because it's not validated.
    Keyword(Token),
}

fn get_file_lang(filename: &OsStr) -> Option<Language> {
    // Deliberate discrepancy here between the check and the error msg below.
    // `l_{}` anywhere in the filename works, but `_l_{}.yml` is still recommended.
    //
    // Using to_string_lossy is ok here because non-unicode sequences will
    // never match the suffix anyway.
    let filename = filename.to_string_lossy();
    L_LANGS.iter().position(|l| filename.contains(l.as_ref())).map(Language::from_idx)
}

impl Localization {
    fn iter_lang(&self) -> impl Iterator<Item = Language> {
        Language::iter().filter(|i| self.mod_langs[i.to_idx()])
    }

    pub fn exists(&self, key: &str) -> bool {
        for lang in self.iter_lang() {
            if !self.locas[lang].contains_key(key) {
                return false;
            }
        }
        true
    }

    // Undocumented; the hash algorithm was revealed by inspecting error.log and reverse
    // engineering of CK3 binary through magic numbers. CK3 and VIC3 are supported.
    #[cfg(any(feature = "ck3", feature = "vic3", feature = "imperator"))]
    fn all_collision_keys(&self, lang: Language) -> TigerHashMap<u32, Vec<&LocaEntry>> {
        let loca_hashes: Vec<_> = self.locas[lang]
            .par_iter()
            .map(|(_, loca)| (loca, murmur3_32(&mut Cursor::new(loca.key.as_str()), 0).unwrap()))
            .collect();
        let mut result: TigerHashMap<u32, Vec<&LocaEntry>> =
            TigerHashMap::with_capacity(loca_hashes.len());

        for (l, h) in loca_hashes {
            result.entry(h).or_default().push(l);
        }
        result.retain(|_, locas| locas.len() > 1);
        result
    }

    pub fn iter_keys(&self) -> impl Iterator<Item = &Token> {
        self.iter_lang()
            .map(|i| &self.locas[i])
            .flat_map(|hash| hash.values().map(|item| &item.key))
    }

    pub fn verify_exists_implied(&self, key: &str, token: &Token, max_sev: Severity) {
        if key.is_empty() {
            return;
        }
        let langs_missing = self.mark_used_return_missing(key);
        if !langs_missing.is_empty() {
            let msg = format!("missing {} localization key {key}", stringify_list(&langs_missing));
            // TODO: get confidence level from caller
            report(ErrorKey::MissingLocalization, Item::Localization.severity().at_most(max_sev))
                .msg(msg)
                .loc(token)
                .push();
        }
    }

    #[cfg(feature = "ck3")]
    pub fn verify_name_exists(&self, name: &Token, max_sev: Severity) {
        if name.as_str().is_empty() {
            report(ErrorKey::MissingLocalization, Severity::Warning.at_most(max_sev))
                .msg("empty name")
                .loc(name)
                .push();
            return;
        }

        let langs_missing = self.mark_used_return_missing(name.as_str());
        if !langs_missing.is_empty() {
            // It's merely untidy if the name is only missing in latin-script languages and the
            // name doesn't have indicators that it really needs to be localized (such as underscores
            // or extra uppercase letters). In all other cases it's a warning.
            //
            // TODO: this logic assumes the input name is in English and it doesn't consider for example
            // a Russian mod that only supports Russian localization and has names in Cyrillic.
            let sev = if only_latin_script(&langs_missing)
                && !name.as_str().contains('_')
                && normal_capitalization_for_name(name.as_str())
            {
                Severity::Untidy
            } else {
                Severity::Warning
            };

            let msg =
                format!("missing {} localization for name {name}", stringify_list(&langs_missing));
            report(ErrorKey::MissingLocalization, sev.at_most(max_sev))
                .strong()
                .msg(msg)
                .loc(name)
                .push();
        }
    }

    #[allow(dead_code)]
    pub fn exists_lang(&self, key: &str, lang: Language) -> bool {
        if !self.locas[lang].contains_key(key) {
            return false;
        }
        true
    }

    pub fn verify_exists_lang(&self, token: &Token, lang: Option<Language>) {
        self.verify_exists_implied_lang(token.as_str(), token, lang);
    }

    pub fn verify_exists_implied_lang(&self, key: &str, token: &Token, lang: Option<Language>) {
        if key.is_empty() {
            return;
        }
        if let Some(lang) = lang {
            if !self.mark_used_lang_return_exists(key, lang) {
                let msg = format!("missing {lang} localization key {key}");
                // TODO: get confidence level from caller
                warn(ErrorKey::MissingLocalization).msg(msg).loc(token).push();
            }
        } else {
            self.verify_exists_implied(key, token, Severity::Warning);
        }
    }

    /// Marks a localization key as used for all languages.
    /// Returns whether the key exists for any language (same as [`Localization::exists`]).
    pub fn mark_used_return_exists(&self, key: &str) -> bool {
        let mut exists = false;
        for lang in self.iter_lang() {
            exists |= self.mark_used_lang_return_exists(key, lang);
        }
        exists
    }

    /// Marks a localization key as used for all languages.
    /// Returns a [`Vec<&str>`] containing the languages for which the key does not exist.
    fn mark_used_return_missing(&self, key: &str) -> Vec<&'static str> {
        let mut langs_missing = Vec::new();
        for lang in self.iter_lang() {
            if !self.mark_used_lang_return_exists(key, lang) {
                langs_missing.push(lang.into());
            }
        }
        langs_missing
    }

    /// Marks a localization key as used for one language.
    /// Returns whether the key exists for this language (same as [`Localization::exists_lang`]).
    fn mark_used_lang_return_exists(&self, key: &str, lang: Language) -> bool {
        if let Some(entry) = self.locas[lang].get(key) {
            entry.used.store(true, Relaxed);
            return true;
        }
        false
    }

    #[allow(dead_code)]
    pub fn suggest(&self, key: &str, token: &Token) {
        if key.is_empty() {
            return;
        }
        let langs_missing = self.mark_used_return_missing(key);
        // They're all missing
        if langs_missing.len() == self.iter_lang().count() {
            let msg = format!("you can define localization `{key}`");
            tips(ErrorKey::SuggestLocalization).msg(msg).loc(token).push();
        }
        // The loca is defined for some languages but not others.
        // This inconsistency is worth warning about.
        else if !langs_missing.is_empty() {
            let msg = format!("missing {} localization key {key}", stringify_list(&langs_missing));
            report(ErrorKey::MissingLocalization, Item::Localization.severity())
                .msg(msg)
                .loc(token)
                .push();
        }
    }

    /// Return whether any language uses the given macro in its loca entry for this key.
    /// Only a macro at the top level of this entry counts; ones hidden recursively in
    /// other macros do not.
    #[allow(dead_code)]
    pub fn uses_macro(&self, key: &str, look_for: &str) -> bool {
        let look_for = format!("${look_for}$");
        for lang in self.iter_lang() {
            if let Some(entry) = self.locas[lang].get(key) {
                if let Some(orig) = &entry.orig {
                    if orig.as_str().contains(&look_for) {
                        return true;
                    }
                }
            }
        }
        false
    }

    // Does every `[concept|E]` reference have a defined game concept?
    // Does every other `[code]` block have valid promotes and functions?
    // Does every $key$ in a macro have a corresponding loca key or named scope?
    fn check_loca_code(
        value: &LocaValue,
        data: &Everything,
        sc: &mut ScopeContext,
        lang: Language,
    ) {
        match value {
            LocaValue::Concat(v) | LocaValue::CalculatedIcon(v) => {
                for value in v {
                    Self::check_loca_code(value, data, sc, lang);
                }
            }
            // TODO: validate the formatting codes
            LocaValue::Code(chain, format) => {
                // |E is the formatting used for game concepts in ck3
                #[cfg(feature = "ck3")]
                if Game::is_ck3() {
                    if let Some(format) = format {
                        if format.as_str().contains('E') || format.as_str().contains('e') {
                            if let Some(name) = chain.as_gameconcept() {
                                if !is_builtin_macro(name) {
                                    data.verify_exists(Item::GameConcept, name);
                                }
                                return;
                            }
                        }
                    }
                }

                // TODO: datatype is not really Unknown here, it should be a CString or CFixedPoint or some kind of number.
                // But we can't express that yet.
                validate_datatypes(
                    chain,
                    data,
                    sc,
                    &DataContext::new(),
                    Datatype::Unknown,
                    Some(lang),
                    format.as_ref(),
                    false,
                );
            }
            LocaValue::Tooltip(token) => {
                // TODO: should this be validated with validate_localization_sc ? (remember to avoid infinite loops)
                if !(Game::is_vic3() && token.is("BREAKDOWN_TAG")) {
                    data.localization.verify_exists_lang(token, Some(lang));
                }
            }
            #[allow(unused_variables)] // tag only used by ck3
            LocaValue::ComplexTooltip(tag, token) => {
                // TODO: if any of the three are datatype expressions, validate them.
                #[cfg(feature = "ck3")]
                if Game::is_ck3() && !token.starts_with("[") && !is_builtin_macro(token) {
                    match COMPLEX_TOOLTIPS_CK3.get(&*tag.as_str().to_lowercase()).copied() {
                        None => {
                            // TODO: should this be validated with validate_localization_sc ? (remember to avoid infinite loops)
                            data.localization.verify_exists_lang(token, Some(lang));
                        }
                        Some(None) => (), // token is a runtime id
                        Some(Some(itype)) => data.verify_exists(itype, token),
                    }
                }
                #[cfg(feature = "vic3")]
                if Game::is_vic3() && !token.starts_with("[") && !is_builtin_macro(token) {
                    data.localization.verify_exists_lang(token, Some(lang));
                }
                // TODO: - imperator -
            }
            LocaValue::Icon(token) => {
                if !is_builtin_macro(token) && !token.is("ICONKEY_icon") && !token.is("KEY_icon") {
                    data.verify_exists(Item::TextIcon, token);
                }
            }
            #[allow(unused_variables)]
            LocaValue::Flag(token) => {
                // TODO: Instead of this awkward 'contains TAG' heuristic, mark macros in the text
                // somehow.
                #[cfg(feature = "hoi4")]
                if !is_builtin_macro(token) && !token.as_str().contains("TAG") {
                    data.verify_exists(Item::CountryTag, token);
                    let pathname = format!("gfx/flags/{token}.tga");
                    data.verify_exists_implied(Item::File, &pathname, token);
                }
            }
            _ => (),
        }
    }

    #[cfg(feature = "ck3")]
    pub fn verify_key_has_options(&self, loca: &str, key: &Token, n: i64, prefix: &str) {
        for lang in self.iter_lang() {
            if let Some(entry) = self.locas[lang].get(loca) {
                if let Some(ref orig) = entry.orig {
                    for i in 1..=n {
                        let find = format!("${prefix}{i}$");
                        let find2 = format!("${prefix}{i}|");
                        if !orig.as_str().contains(&find) && !orig.as_str().contains(&find2) {
                            warn(ErrorKey::Validation)
                                .msg(format!("localization is missing {find}"))
                                .loc(key)
                                .loc_msg(&entry.key, "here")
                                .push();
                        }
                    }
                    let find = format!("${prefix}{}$", n + 1);
                    let find2 = format!("${prefix}{}|", n + 1);
                    if orig.as_str().contains(&find) && !orig.as_str().contains(&find2) {
                        warn(ErrorKey::Validation)
                            .msg("localization has too many options")
                            .loc(key)
                            .loc_msg(&entry.key, "here")
                            .push();
                    }
                } else if n > 0 {
                    let msg = format!("localization is missing ${prefix}1$");
                    warn(ErrorKey::Validation).msg(msg).loc(key).loc_msg(&entry.key, "here").push();
                }
            }
        }
    }

    fn validate_loca<'b>(
        entry: &LocaEntry,
        from: &'b TigerHashMap<&'b str, LocaEntry>,
        data: &Everything,
        sc: &mut ScopeContext,
        lang: Language,
    ) {
        if matches!(entry.value, LocaValue::Macro(_)) {
            let mut new_line = Vec::new();
            let mut count = 0;
            if entry.expand_macros(&mut new_line, from, &mut count, sc, None, data) {
                // re-parse after macro expansion
                let new_line_as_ref = new_line.iter().collect();
                let value = ValueParser::new(new_line_as_ref).parse();
                Self::check_loca_code(&value, data, sc, lang);
            }
        } else {
            Self::check_loca_code(&entry.value, data, sc, lang);
        }
    }

    pub fn validate_use(&self, key: &str, data: &Everything, sc: &mut ScopeContext) {
        for lang in self.iter_lang() {
            let loca = &self.locas[lang];
            if let Some(entry) = loca.get(key) {
                entry.used.store(true, Relaxed);
                entry.validated.store(true, Relaxed);
                Self::validate_loca(entry, loca, data, sc, lang);
            }
        }
    }

    #[cfg(any(feature = "ck3", feature = "vic3", feature = "imperator"))]
    fn check_collisions(&self, lang: Language) {
        for (k, v) in self.all_collision_keys(lang) {
            let mut rep = report(ErrorKey::LocalizationKeyCollision, Severity::Error)
                .strong()
                .msg(format!(
                    "localization keys '{}' have same MURMUR3A hash '0x{k:08X}'",
                    stringify_list(&v.iter().map(|loca| loca.key.as_str()).collect::<Vec<&str>>())
                ))
                .info("localization keys hash collision will cause some of them fail to load")
                .loc(&v[0].key);
            for loc in v.iter().skip(1) {
                rep = rep.loc_msg(&loc.key, "here");
            }
            rep.push();
        }
    }

    // This is in pass2 to make sure all `validated` entries have been marked.
    pub fn validate_pass2(&self, data: &Everything) {
        #[allow(unused_variables)]
        scope(|s| {
            for lang in self.iter_lang() {
                let loca = &self.locas[lang];
                // Check localization key collisions
                #[cfg(any(feature = "ck3", feature = "vic3", feature = "imperator"))]
                s.spawn(move |_| self.check_collisions(lang));

                // Collect and sort the entries before looping, to create more stable output
                let mut unvalidated_entries: Vec<&LocaEntry> =
                    loca.values().filter(|e| !e.validated.load(Relaxed)).collect();
                unvalidated_entries.sort_unstable();
                unvalidated_entries.par_iter().for_each(|entry| {
                    // Technically we can now store true in entry.validated,
                    // but the value is not needed anymore after this.
                    let mut sc = ScopeContext::new_unrooted(Scopes::all(), &entry.key);
                    sc.set_strict_scopes(false);
                    Self::validate_loca(entry, loca, data, &mut sc, lang);
                });
            }
        });
    }

    pub fn mark_category_used(&self, prefix: &str) {
        let mut i = 0;
        loop {
            let loca = format!("{prefix}{i}");
            if !self.mark_used_return_exists(&loca) {
                break;
            }
            i += 1;
        }
    }

    pub fn check_unused(&self, _data: &Everything) {
        self.mark_category_used("LOADING_TIP_");
        self.mark_category_used("HYBRID_NAME_FORMAT_");
        self.mark_category_used("DIVERGE_NAME_FORMAT_");

        for lang in self.iter_lang() {
            let mut vec = Vec::new();
            for entry in self.locas[lang].values() {
                if !entry.used.load(Relaxed) {
                    vec.push(entry);
                }
            }
            vec.sort_unstable_by_key(|entry| &entry.key.loc);
            for entry in vec {
                report(ErrorKey::UnusedLocalization, Severity::Untidy)
                    .msg("Unused localization")
                    .abbreviated(&entry.key)
                    .push();
            }
        }
    }

    #[cfg(feature = "ck3")]
    pub fn check_pod_loca(&self, data: &Everything) {
        for lang in self.iter_lang() {
            for key in data.database.iter_keys(Item::PerkTree) {
                let loca = format!("{key}_name");
                if let Some(entry) = self.locas[lang].get(loca.as_str()) {
                    if let LocaValue::Text(token) = &entry.value {
                        if token.as_str().ends_with("_visible") {
                            data.verify_exists(Item::ScriptedGui, token);
                            data.verify_exists(Item::Localization, token);
                        }
                        continue;
                    }
                }
                let msg = format!("missing loca `{key}_name: \"{key}_visible\"`");
                let info = "this is needed for the `window_character_lifestyle.gui` code";
                err(ErrorKey::PrincesOfDarkness).msg(msg).info(info).loc(key).push();
            }
        }
    }
}

impl FileHandler<(Language, Vec<LocaEntry>)> for Localization {
    fn config(&mut self, config: &Block) {
        if let Some(block) = config.get_field_block("languages") {
            // By default, self.check_langs is all true.
            // If a languages block exists in the config, then check_langs
            // should contain only the configured languages, so langs is
            // initialized to all false here.
            let mut langs = bitarr![u16, Lsb0; 0; Language::COUNT];

            // TODO: warn if there are unknown languages in check or skip?
            let check = block.get_field_values("check");
            let skip = block.get_field_values("skip");

            // If check is used, then check only those languages.
            // If instead skip is used, then check all languages except the skipped ones.
            for lang in Language::iter() {
                let lang_str = lang.into();
                if check.iter().any(|t| t.is(lang_str))
                    || (check.is_empty() && skip.iter().all(|t| !t.is(lang_str)))
                {
                    langs.set(lang.to_idx(), true);
                }
            }
            self.check_langs = langs;
        }
    }

    fn subpath(&self) -> PathBuf {
        if Game::is_hoi4() { PathBuf::from("localisation") } else { PathBuf::from("localization") }
    }

    fn load_file(
        &self,
        entry: &FileEntry,
        _parser: &ParserMemory,
    ) -> Option<(Language, Vec<LocaEntry>)> {
        if !entry.filename().to_string_lossy().ends_with(".yml") {
            return None;
        }

        // unwrap is safe here because we're only handed files under localization/
        // to_string_lossy is ok because we compare lang against a set of known strings.
        let lang_str = entry.path().components().nth(1).unwrap().as_os_str().to_string_lossy();

        // special case for this file
        if lang_str == "languages.yml" {
            return None;
        }

        if let Some(filelang) = get_file_lang(entry.filename()) {
            if !self.check_langs[filelang.to_idx()] {
                return None;
            }
            // Localization files don't have to be in a subdirectory corresponding to their language.
            // However, if there's one in a subdirectory for a *different* language than the one in its name,
            // then something is probably wrong.
            if let Ok(lang) = Language::try_from(lang_str.as_ref()) {
                if filelang != lang {
                    let msg = "localization file with wrong name or in wrong directory";
                    let info = "A localization file should be in a subdirectory corresponding to its language.";
                    warn(ErrorKey::Filename).msg(msg).info(info).loc(entry).push();
                }
            }
            match read_to_string(entry.fullpath()) {
                Ok(content) => {
                    return Some((filelang, parse_loca(entry, content, filelang).collect()));
                }
                Err(e) => {
                    let msg = "could not read file";
                    let info = &format!("{e:#}");
                    err(ErrorKey::ReadError).msg(msg).info(info).loc(entry).push();
                }
            }
        } else if entry.kind() >= FileKind::Vanilla {
            // Check for `FileKind::Vanilla` because Jomini and Clausewitz support more languages
            let msg = "could not determine language from filename";
            let info = format!(
                "Localization filenames should end in _l_language.yml, where language is one of {}",
                *LANG_LIST
            );
            err(ErrorKey::Filename).msg(msg).info(info).loc(entry).push();
        }
        None
    }

    fn handle_file(&mut self, entry: &FileEntry, loaded: (Language, Vec<LocaEntry>)) {
        let (filelang, vec) = loaded;
        let hash = &mut self.locas[filelang];
        if hash.is_empty() {
            // empirically ~290k for each lang of ck3
            hash.reserve(300_000);
        }

        if entry.kind() == FileKind::Mod {
            self.mod_langs.set(filelang.to_idx(), true);
        }

        for loca in vec {
            match hash.entry(loca.key.as_str()) {
                Entry::Occupied(mut occupied_entry) => {
                    let other = occupied_entry.get();
                    // other.key and loca.key are in the other order than usual here,
                    // because in loca the older definition overrides the later one.
                    if is_replace_path(entry.path()) {
                        occupied_entry.insert(loca);
                    } else if other.key.loc.kind == entry.kind() && other.orig != loca.orig {
                        dup_error(&other.key, &loca.key, "localization");
                    }
                }
                Entry::Vacant(vacant_entry) => {
                    vacant_entry.insert(loca);
                }
            }
        }
    }
}

impl Default for Localization {
    fn default() -> Self {
        Localization {
            check_langs: bitarr![u16, Lsb0; 1; Language::COUNT],
            mod_langs: bitarr![u16, Lsb0; 0; Language::COUNT],
            locas: Languages(std::array::from_fn(|_| TigerHashMap::default())),
        }
    }
}

/// It's been tested that localization/replace/english and localization/english/replace both work
fn is_replace_path(path: &Path) -> bool {
    for element in path {
        if element.to_string_lossy() == "replace" {
            return true;
        }
    }
    false
}

/// These are the languages in which it's reasonable to present an ascii name unchanged.
#[cfg(feature = "ck3")]
const LATIN_SCRIPT_LANGS: &[&str] =
    &["english", "french", "german", "spanish", "braz_por", "polish", "turkish"];

/// Return true iff `langs` only contains languages in which it's reasonable to present an ascii
/// name unchanged.
#[cfg(feature = "ck3")]
fn only_latin_script(langs: &[&str]) -> bool {
    langs.iter().all(|lang| LATIN_SCRIPT_LANGS.contains(lang))
}

/// Check that the string only has capital letters at the start or after a space or hyphen
#[cfg(feature = "ck3")]
fn normal_capitalization_for_name(name: &str) -> bool {
    let mut expect_cap = true;
    for ch in name.chars() {
        if ch.is_uppercase() && !expect_cap {
            return false;
        }
        expect_cap = ch == ' ' || ch == '-';
    }
    true
}

#[cfg(all(test, feature = "ck3"))]
mod tests {
    use super::*;
    use crate::fileset::{FileKind, FileStage};
    use crate::token::{Loc, Token};
    use std::path::PathBuf;

    #[test]
    fn test_only_latin_script() {
        let mut langs = vec!["english", "french", "german"];
        assert!(only_latin_script(&langs));
        langs.push("korean");
        assert!(!only_latin_script(&langs));
        langs.clear();
        assert!(only_latin_script(&langs));
    }

    #[test]
    fn test_normal_capitalization_for_name() {
        assert!(normal_capitalization_for_name("George"));
        assert!(normal_capitalization_for_name("george"));
        assert!(!normal_capitalization_for_name("BjOrn"));
        assert!(normal_capitalization_for_name("Jean-Claude"));
        assert!(normal_capitalization_for_name("Abu-l-Fadl al-Malik"));
        assert!(normal_capitalization_for_name("Abu Abdallah Muhammad"));
        assert!(!normal_capitalization_for_name("AbuAbdallahMuhammad"));
    }

    #[test]
    fn test_collision_detection() {
        // build a localization database containing known colliding keys
        let mut loc = Localization::default();
        let lang = Language::English;
        // dummy location for tokens
        let dummy_loc =
            Loc::for_file(PathBuf::new(), FileStage::NoStage, FileKind::Mod, PathBuf::new());

        let pairs = [
            // CK3 examples
            ("Mallobald", "laamp_base_contract_schemes.2541.e.tt.employer_has_trait.paranoid"),
            ("dynn_Hkeng", "debug_min_popular_opinion_modifier"),
            ("b_hinggan_adj", "grand_wedding_completed_guest"),
            // Imperator examples
            ("carthage_mission_trade_metropolis_west", "me_diadochi_empire_events.316.at"),
            ("Azdumani", "me_patauion_02.43.b_tt"),
            ("PROV7234_hellenic", "me_kush_15_desc"),
        ];

        for &(k1, k2) in &pairs {
            let t1 = Token::from_static_str(k1, dummy_loc);
            let t2 = Token::from_static_str(k2, dummy_loc);
            let e1 = LocaEntry::new(t1.clone(), LocaValue::Text(t1.clone()), None);
            let e2 = LocaEntry::new(t2.clone(), LocaValue::Text(t2.clone()), None);
            loc.locas[lang].insert(k1, e1);
            loc.locas[lang].insert(k2, e2);
        }

        let collisions = loc.all_collision_keys(lang);
        for &(k1, k2) in &pairs {
            assert!(
                collisions.values().any(|vec| {
                    vec.iter().any(|e| e.key.as_str() == k1)
                        && vec.iter().any(|e| e.key.as_str() == k2)
                }),
                "expected collision between {k1} and {k2}"
            );
        }
    }
}