Files
rustdesk/src/lang.rs
Ricardo Simões c6a358afbe Update lang.rs - Splits ambiguous pt into pt-pt and pt-br (#16155)
* Update lang.rs - Splits ambiguous pt into pt-pt and pt-br

Splits the ambiguous "Português" entry into two explicit options: Português (Portugal) [pt-pt] and Português (Brasil) [pt-br]. The pt_PT.rs translation file already existed in the repo (fully translated, key-complete) but was never wired into lang.rs. Kept the old "pt" and "br" locale codes mapped to ptbr::T for backward compatibility with existing user configs.

* Preserve pt-PT/pt-BR region in automatic locale resolution

Preserve pt-PT/pt-BR region in automatic locale resolution

* Fix pt-pt detection: check region suffix instead of redundant "pt" substring

Fix pt-pt detection: check region suffix instead of redundant "pt" substring

* Normalize legacy pt/br saved lang option to pt-br

Normalize legacy pt/br saved lang option to pt-br

* Normalize legacy pt/br lang option on Web client

* Add regression tests for pt-pt/pt-br locale resolution

* Follow CLDR inheritance for Portuguese regional locales (pt-AO, pt-MZ, etc.)
2026-09-22 16:31:23 +08:00

387 lines
11 KiB
Rust

use hbb_common::regex::Regex;
use std::ops::Deref;
mod ar;
mod az;
mod be;
mod bg;
mod ca;
mod cn;
mod cs;
mod da;
mod de;
mod el;
mod en;
mod eo;
mod es;
mod et;
mod eu;
mod fa;
mod gu;
mod fr;
mod he;
mod hi;
mod hr;
mod hu;
mod id;
mod it;
mod ja;
mod ko;
mod kz;
mod lt;
mod lv;
mod nb;
mod nl;
mod pl;
mod ptbr;
mod pt_PT;
mod ro;
mod ru;
mod sc;
mod sk;
mod sl;
mod sq;
mod sr;
mod sv;
mod th;
mod tr;
mod tw;
mod uk;
mod ur;
mod vi;
mod ta;
mod ge;
mod fi;
mod ml;
mod gl;
pub const LANGS: &[(&str, &str)] = &[
("en", "English"),
("it", "Italiano"),
("fr", "Français"),
("de", "Deutsch"),
("nl", "Nederlands"),
("nb", "Norsk bokmål"),
("zh-cn", "简体中文"),
("zh-tw", "繁體中文"),
("pt-pt", "Português (Portugal)"),
("pt-br", "Português (Brasil)"),
("es", "Español"),
("et", "Eesti keel"),
("eu", "Euskara"),
("hu", "Magyar"),
("bg", "Български"),
("be", "Беларуская"),
("ru", "Русский"),
("sk", "Slovenčina"),
("id", "Indonesia"),
("cs", "Čeština"),
("da", "Dansk"),
("eo", "Esperanto"),
("tr", "Türkçe"),
("vi", "Tiếng Việt"),
("pl", "Polski"),
("ja", "日本語"),
("ko", "한국어"),
("kz", "Қазақ"),
("uk", "Українська"),
("ur", "اردو"),
("fa", "فارسی"),
("ca", "Català"),
("gl", "Galego"),
("el", "Ελληνικά"),
("sv", "Svenska"),
("sq", "Shqip"),
("sr", "Srpski"),
("th", "ภาษาไทย"),
("sl", "Slovenščina"),
("ro", "Română"),
("lt", "Lietuvių"),
("lv", "Latviešu"),
("ar", "العربية"),
("he", "עברית"),
("hr", "Hrvatski"),
("sc", "Sardu"),
("ta", "தமிழ்"),
("ge", "ქართული"),
("fi", "Suomi"),
("ml", "മലയാളം"),
("hi", "हिंदी"),
("gu", "ગુજરાતી"),
("az", "Azərbaycan dili"),
];
pub(crate) fn cjk_ui_unavailable() -> bool {
cfg!(all(
target_os = "linux",
target_arch = "aarch64",
feature = "flutter"
))
}
pub(crate) fn is_cjk_lang(lang_or_locale: &str) -> bool {
let lang = lang_or_locale
.split(|c| c == '-' || c == '_')
.next()
.unwrap_or_default()
.to_lowercase();
matches!(lang.as_str(), "zh" | "ja" | "ko")
}
fn resolve_lang(saved_lang: &str, locale: &str, cjk_fallback: bool) -> String {
let locale = locale.to_lowercase();
let mut lang = saved_lang.to_lowercase();
if cjk_fallback && is_cjk_lang(&lang) {
return "en".to_owned();
}
if lang.is_empty() {
// zh_CN on Linux, zh-Hans-CN on mac, zh_CN_#Hans on Android
if locale.starts_with("zh") {
lang = (if locale.contains("tw") {
"zh-tw"
} else {
"zh-cn"
})
.to_owned();
}
}
if lang.is_empty() {
// pt_PT on Linux, pt-PT on mac, pt_PT on Android.
// Per CLDR locale inheritance, every Portuguese-speaking locale
// besides bare "pt" and Brazil's own variants (pt-BR/pt_BR) has
// pt-PT as its parent (Angola, Mozambique, Cape Verde, etc.),
// so it should resolve to European Portuguese.
if locale.starts_with("pt") {
lang = (if locale == "pt" || locale.starts_with("pt-br") || locale.starts_with("pt_br") {
"pt-br"
} else {
"pt-pt"
})
.to_owned();
}
}
if lang.is_empty() {
lang = locale
.split("-")
.next()
.map(|x| x.split("_").next().unwrap_or_default())
.unwrap_or_default()
.to_owned();
}
if cjk_fallback && is_cjk_lang(&lang) {
"en".to_owned()
} else {
lang
}
}
#[cfg(not(any(target_os = "android", target_os = "ios")))]
pub fn translate(name: String) -> String {
let locale = sys_locale::get_locale().unwrap_or_default();
translate_locale(name, &locale)
}
pub fn translate_locale(name: String, locale: &str) -> String {
let lang = resolve_lang(
&hbb_common::config::LocalConfig::get_option("lang"),
locale,
cjk_ui_unavailable(),
);
let m = match lang.as_str() {
"fr" => fr::T.deref(),
"zh-cn" => cn::T.deref(),
"it" => it::T.deref(),
"zh-tw" => tw::T.deref(),
"de" => de::T.deref(),
"nb" => nb::T.deref(),
"nl" => nl::T.deref(),
"es" => es::T.deref(),
"et" => et::T.deref(),
"eu" => eu::T.deref(),
"hu" => hu::T.deref(),
"ru" => ru::T.deref(),
"eo" => eo::T.deref(),
"id" => id::T.deref(),
"br" => ptbr::T.deref(),
"pt" => ptbr::T.deref(),
"pt-br" => ptbr::T.deref(),
"pt-pt" => pt_PT::T.deref(),
"tr" => tr::T.deref(),
"cs" => cs::T.deref(),
"da" => da::T.deref(),
"sk" => sk::T.deref(),
"vi" => vi::T.deref(),
"pl" => pl::T.deref(),
"ja" => ja::T.deref(),
"ko" => ko::T.deref(),
"kz" => kz::T.deref(),
"uk" => uk::T.deref(),
"fa" => fa::T.deref(),
"fi" => fi::T.deref(),
"ca" => ca::T.deref(),
"el" => el::T.deref(),
"sv" => sv::T.deref(),
"sq" => sq::T.deref(),
"sr" => sr::T.deref(),
"th" => th::T.deref(),
"sl" => sl::T.deref(),
"ro" => ro::T.deref(),
"lt" => lt::T.deref(),
"lv" => lv::T.deref(),
"ar" => ar::T.deref(),
"bg" => bg::T.deref(),
"be" => be::T.deref(),
"he" => he::T.deref(),
"hr" => hr::T.deref(),
"ur" => ur::T.deref(),
"sc" => sc::T.deref(),
"ta" => ta::T.deref(),
"ge" => ge::T.deref(),
"ml" => ml::T.deref(),
"hi" => hi::T.deref(),
"gu" => gu::T.deref(),
"gl" => gl::T.deref(),
"az" => az::T.deref(),
_ => en::T.deref(),
};
let (name, placeholder_value) = extract_placeholder(&name);
let replace = |s: &&str| {
let mut s = s.to_string();
if let Some(value) = placeholder_value.as_ref() {
s = s.replace("{}", &value);
}
if !crate::is_rustdesk() {
if s.contains("RustDesk")
&& !name.starts_with("upgrade_rustdesk_server_pro")
&& name != "powered_by_me"
{
let app_name = crate::get_app_name();
if !app_name.contains("RustDesk") {
s = s.replace("RustDesk", &app_name);
} else {
// https://github.com/rustdesk/rustdesk-server-pro/issues/845
// If app_name contains "RustDesk" (e.g., "RustDesk-Admin"), we need to avoid
// replacing "RustDesk" within the already-substituted app_name, which would
// cause duplication like "RustDesk-Admin" -> "RustDesk-Admin-Admin".
//
// app_name only contains alphanumeric and hyphen.
const PLACEHOLDER: &str = "#A-P-P-N-A-M-E#";
if !s.contains(PLACEHOLDER) {
s = s.replace(&app_name, PLACEHOLDER);
s = s.replace("RustDesk", &app_name);
s = s.replace(PLACEHOLDER, &app_name);
} else {
// It's very unlikely to reach here.
// Skip replacement to avoid incorrect result.
}
}
}
}
s
};
if let Some(v) = m.get(&name as &str) {
if !v.is_empty() {
return replace(v);
}
}
if lang != "en" {
if let Some(v) = en::T.get(&name as &str) {
if !v.is_empty() {
return replace(v);
}
}
}
replace(&name.as_str())
}
// Matching pattern is {}
// Write {value} in the UI and {} in the translation file
//
// Example:
// Write in the UI: translate("There are {24} hours in a day")
// Write in the translation file: ("There are {} hours in a day", "{} hours make up a day")
fn extract_placeholder(input: &str) -> (String, Option<String>) {
if let Ok(re) = Regex::new(r#"\{(.*?)\}"#) {
if let Some(captures) = re.captures(input) {
if let Some(inner_match) = captures.get(1) {
let name = re.replace(input, "{}").to_string();
let value = inner_match.as_str().to_string();
return (name, Some(value));
}
}
}
(input.to_string(), None)
}
mod test {
#[test]
fn test_extract_placeholders() {
use super::extract_placeholder as f;
assert_eq!(f(""), ("".to_string(), None));
assert_eq!(
f("{3} sessions"),
("{} sessions".to_string(), Some("3".to_string()))
);
assert_eq!(f(" } { "), (" } { ".to_string(), None));
// Allow empty value
assert_eq!(
f("{} sessions"),
("{} sessions".to_string(), Some("".to_string()))
);
// Match only the first one
assert_eq!(
f("{2} times {4} makes {8}"),
("{} times {4} makes {8}".to_string(), Some("2".to_string()))
);
}
#[test]
fn test_resolve_lang_forces_english_for_saved_cjk_when_target_disables_cjk() {
use super::resolve_lang as f;
assert_eq!(f("zh-cn", "en-US", true), "en");
assert_eq!(f("zh-tw", "en-US", true), "en");
assert_eq!(f("ja", "en-US", true), "en");
assert_eq!(f("ko", "en-US", true), "en");
}
#[test]
fn test_resolve_lang_forces_english_for_cjk_locale_when_target_disables_cjk() {
use super::resolve_lang as f;
assert_eq!(f("", "zh_CN", true), "en");
assert_eq!(f("", "ja-JP", true), "en");
assert_eq!(f("", "ko_KR", true), "en");
}
#[test]
fn test_resolve_lang_preserves_cjk_when_target_allows_cjk() {
use super::resolve_lang as f;
assert_eq!(f("zh-cn", "en-US", false), "zh-cn");
assert_eq!(f("", "zh_TW", false), "zh-tw");
assert_eq!(f("", "ja-JP", false), "ja");
}
#[test]
fn test_resolve_lang_detects_pt_pt_and_pt_br_from_locale() {
use super::resolve_lang as f;
assert_eq!(f("", "pt-PT", false), "pt-pt");
assert_eq!(f("", "pt_PT", false), "pt-pt");
assert_eq!(f("", "pt-BR", false), "pt-br");
assert_eq!(f("", "pt_BR", false), "pt-br");
assert_eq!(f("", "pt-AO", false), "pt-pt");
assert_eq!(f("", "pt-MZ", false), "pt-pt");
}
#[test]
fn test_resolve_lang_saved_lang_takes_precedence_over_locale() {
use super::resolve_lang as f;
assert_eq!(f("pt-br", "pt-PT", false), "pt-br");
assert_eq!(f("pt-pt", "pt-BR", false), "pt-pt");
}
}