feat(translate): keep translated subtitles next to the video so they load on reopen (#14)

Co-authored-by: Tri Vo Khac <git@trivk.dev>

src/translate/translationservice.cpp

@@ -3,6 +3,7 @@
 #include "apikeystore.h"
 #include "qtpotatosettings.h"
 #include "translationpresets.h"
+#include "translationstore.h"
 
 #include <KLocalizedString>
 
@@ -39,28 +40,9 @@ constexpr int ParallelRequests = 3;
 // Bump when the prompt changes so old cached translations are not reused.
 constexpr auto CacheVersion = "llm-v2";
 
-// "Vietnamese" -> "vi", so the file ends in ".vi.srt" like other players expect.
-QString languageSuffix(const QString &language)
+QString cacheDir()
 {
-    for (int i = QLocale::C + 1; i <= QLocale::LastLanguage; ++i) {
-        const auto candidate = static_cast<QLocale::Language>(i);
-        if (QLocale::languageToString(candidate).compare(language, Qt::CaseInsensitive) == 0) {
-            const QString code = QLocale::languageToCode(candidate, QLocale::ISO639Part1);
-            return code.isEmpty() ? QLocale::languageToCode(candidate) : code;
-        }
-    }
-    QString suffix = language.toLower();
-    suffix.removeIf([](QChar c) {
-        return !c.isLetterOrNumber() && c != u'-';
-    });
-    return suffix.isEmpty() ? u"translated"_s : suffix;
-}
-
-QString cachePath(const QUrl &videoUrl, const QString &key, const QString &language)
-{
-    const QString dir = QStandardPaths::writableLocation(QStandardPaths::CacheLocation) + u"/translations"_s;
-    const QString baseName = videoUrl.isLocalFile() ? QFileInfo(videoUrl.toLocalFile()).completeBaseName() : u"subtitles"_s;
-    return u"%1/%2.%3.%4.srt"_s.arg(dir, baseName, key, languageSuffix(language));
+    return QStandardPaths::writableLocation(QStandardPaths::CacheLocation) + u"/translations"_s;
 }
 
 QList<GlossaryStore::Entry> glossaryEntriesFromVariants(const QVariantList &entries)
@@ -273,6 +255,12 @@ void TranslationService::applyGlossaryEdits(const GlossaryStore::Glossary &previ
     if (m_lastTranslationPath.isEmpty() || !QFileInfo::exists(m_lastTranslationPath)) {
         return;
     }
+    // A translation the user has edited since is theirs; leave it alone.
+    const QString registryFile = TranslationStore::registryFile();
+    TranslationStore::Registry registry = TranslationStore::load(registryFile);
+    if (!TranslationStore::isOwned(registry, m_lastTranslationPath)) {
+        return;
+    }
     QFile file(m_lastTranslationPath);
     if (!file.open(QIODevice::ReadOnly)) {
         return;
@@ -319,10 +307,14 @@ void TranslationService::applyGlossaryEdits(const GlossaryStore::Glossary &previ
         text.replace(markers.at(i), replacements.at(i).second);
     }
 
+    const QByteArray data = text.toUtf8();
     QSaveFile output(m_lastTranslationPath);
-    if (!output.open(QIODevice::WriteOnly) || output.write(text.toUtf8()) < 0 || !output.commit()) {
+    if (!output.open(QIODevice::WriteOnly) || output.write(data) < 0 || !output.commit()) {
         return;
     }
+    // Still QtPotato's file, so a later translation may replace it.
+    TranslationStore::recordRewrite(registry, m_lastTranslationPath, data);
+    TranslationStore::save(registryFile, registry);
     Q_EMIT translated(QUrl::fromLocalFile(m_lastTranslationPath), true);
 }
 
@@ -480,7 +472,8 @@ QCoro::Task<> TranslationService::run(QUrl videoUrl, QVariantMap track)
     const QString sourceSrt = Subtitles::toSrt(cues);
 
     // Same subtitles, model, language, prompt settings and glossary give the same file, so it is translated once.
-    const auto outputPathFor = [&](const GlossaryStore::Glossary &g) {
+    const QString suffix = TranslationStore::languageSuffix(config.targetLanguage);
+    const auto configKeyFor = [&](const GlossaryStore::Glossary &g) {
         QCryptographicHash hash(QCryptographicHash::Sha1);
         hash.addData(CacheVersion);
         hash.addData(config.model.toUtf8());
@@ -489,16 +482,13 @@ QCoro::Task<> TranslationService::run(QUrl videoUrl, QVariantMap track)
         hash.addData(config.customInstructions.toUtf8());
         hash.addData(GlossaryStore::serialize(g).toUtf8());
         hash.addData(sourceSrt.toUtf8());
-        return cachePath(videoUrl, QString::fromLatin1(hash.result().toHex().left(12)), config.targetLanguage);
+        return QString::fromLatin1(hash.result().toHex().left(12));
     };
-    QString outputPath = outputPathFor(glossary);
+    QString configKey = configKeyFor(glossary);
+    QString outputPath = TranslationStore::cachePath(cacheDir(), videoUrl, configKey, suffix);
 
     if (QFileInfo::exists(outputPath)) {
-        m_lastTranslationPath = outputPath;
-        m_job.reset();
-        setBusy(false);
-        setStatusText({});
-        Q_EMIT translated(QUrl::fromLocalFile(outputPath), true);
+        finish(job, videoUrl, suffix, configKey, outputPath, true);
         co_return;
     }
 
@@ -545,13 +535,10 @@ QCoro::Task<> TranslationService::run(QUrl videoUrl, QVariantMap track)
         } else {
             setStatusText(i18n("Glossary unavailable; translating without it."));
         }
-        outputPath = outputPathFor(glossary);
+        configKey = configKeyFor(glossary);
+        outputPath = TranslationStore::cachePath(cacheDir(), videoUrl, configKey, suffix);
         if (QFileInfo::exists(outputPath)) {
-            m_lastTranslationPath = outputPath;
-            m_job.reset();
-            setBusy(false);
-            setStatusText({});
-            Q_EMIT translated(QUrl::fromLocalFile(outputPath), true);
+            finish(job, videoUrl, suffix, configKey, outputPath, true);
             co_return;
         }
     }
@@ -610,11 +597,38 @@ QCoro::Task<> TranslationService::run(QUrl videoUrl, QVariantMap track)
         co_return;
     }
 
+    finish(job, videoUrl, suffix, configKey, outputPath, false);
+}
+
+void TranslationService::finish(const std::shared_ptr<Job> &job,
+                                const QUrl &videoUrl,
+                                const QString &suffix,
+                                const QString &configKey,
+                                const QString &cacheFile,
+                                bool fromCache)
+{
+    QFile file(cacheFile);
+    if (!file.open(QIODevice::ReadOnly)) {
+        fail(job, i18n("Could not read the translation from %1: %2", cacheFile, file.errorString()));
+        return;
+    }
+    // The cache keeps every version; the published copy is the one players load.
+    const QString registryFile = TranslationStore::registryFile();
+    TranslationStore::Registry registry = TranslationStore::load(registryFile);
+    const TranslationStore::PublishResult published = TranslationStore::publish(registry, videoUrl, suffix, configKey, file.readAll(), cacheFile);
+    TranslationStore::save(registryFile, registry);
+
     m_job.reset();
-    m_lastTranslationPath = outputPath;
+    m_lastTranslationPath = published.path;
     setBusy(false);
     setStatusText({});
-    Q_EMIT translated(QUrl::fromLocalFile(outputPath), false);
+    Q_EMIT translated(QUrl::fromLocalFile(published.path), fromCache);
+}
+
+QString TranslationService::savedTranslation(const QUrl &videoUrl) const
+{
+    const QString suffix = TranslationStore::languageSuffix(QtPotatoSettings::self()->targetLanguage().trimmed());
+    return TranslationStore::savedTranslation(TranslationStore::load(TranslationStore::registryFile()), videoUrl, suffix);
 }
 
 QCoro::Task<LlmClient::BatchResult> TranslationService::translateLines(std::shared_ptr<Job> job, QStringList lines, QStringList context)

src/translate/translationservice.h

@@ -18,8 +18,8 @@
 
 class QNetworkAccessManager;
 
-// Translates a whole subtitle track with an LLM, batch by batch, and caches the result
-// as an .srt file that mpv can load as a subtitle track.
+// Translates a whole subtitle track with an LLM, batch by batch, caches the result as an
+// .srt file and saves a copy next to the video (see TranslationStore).
 class TranslationService : public QObject
 {
     Q_OBJECT
@@ -62,6 +62,9 @@ public:
     Q_INVOKABLE void fetchModels();
     Q_INVOKABLE void clearError();
     Q_INVOKABLE void forgetTranslation();
+    // Path of the translation of \a videoUrl into the target language set in Settings,
+    // or an empty string when there is none.
+    Q_INVOKABLE QString savedTranslation(const QUrl &videoUrl) const;
     Q_INVOKABLE void loadGlossary(const QUrl &videoUrl);
     Q_INVOKABLE bool saveGlossary(const QVariantList &entries, const QStringList &episodes = {});
     Q_INVOKABLE void clearGlossary();
@@ -97,6 +100,8 @@ private:
     // something is missing.
     QCoro::Task<std::pair<LlmClient::Config, QString>> loadConfig();
     void fail(const std::shared_ptr<Job> &job, const QString &error);
+    // Saves the translation in \a cacheFile where players find it and announces it.
+    void finish(const std::shared_ptr<Job> &job, const QUrl &videoUrl, const QString &suffix, const QString &configKey, const QString &cacheFile, bool fromCache);
 
     void setBusy(bool busy);
     void setProgress(double progress);

src/translate/translationstore.cpp

@@ -0,0 +1,231 @@
+#include "translationstore.h"
+
+#include <QCryptographicHash>
+#include <QDir>
+#include <QFile>
+#include <QFileInfo>
+#include <QJsonArray>
+#include <QJsonDocument>
+#include <QJsonObject>
+#include <QLocale>
+#include <QSaveFile>
+#include <QStandardPaths>
+
+using namespace Qt::Literals::StringLiterals;
+
+namespace
+{
+constexpr int RegistryVersion = 1;
+
+QString normalized(const QString &path)
+{
+    return path.isEmpty() ? QString() : QFileInfo(path).absoluteFilePath();
+}
+
+std::optional<QByteArray> readFile(const QString &path)
+{
+    QFile file(path);
+    if (!file.open(QIODevice::ReadOnly)) {
+        return std::nullopt;
+    }
+    return file.readAll();
+}
+
+bool writeFile(const QString &path, const QByteArray &data)
+{
+    // No direct-write fallback: a folder that can't take a temporary file is treated
+    // as read-only, and the translation stays in the cache.
+    QSaveFile file(path);
+    return file.open(QIODevice::WriteOnly) && file.write(data) == data.size() && file.commit();
+}
+}
+
+namespace TranslationStore
+{
+QString languageSuffix(const QString &language)
+{
+    for (int i = QLocale::C + 1; i <= QLocale::LastLanguage; ++i) {
+        const auto candidate = static_cast<QLocale::Language>(i);
+        if (QLocale::languageToString(candidate).compare(language, Qt::CaseInsensitive) == 0) {
+            const QString code = QLocale::languageToCode(candidate, QLocale::ISO639Part1);
+            return code.isEmpty() ? QLocale::languageToCode(candidate) : code;
+        }
+    }
+    QString suffix = language.toLower();
+    suffix.removeIf([](QChar c) {
+        return !c.isLetterOrNumber() && c != u'-';
+    });
+    return suffix.isEmpty() ? u"translated"_s : suffix;
+}
+
+QString videoKey(const QUrl &videoUrl)
+{
+    return videoUrl.isLocalFile() ? normalized(videoUrl.toLocalFile()) : videoUrl.toString();
+}
+
+QString cachePath(const QString &cacheDir, const QUrl &videoUrl, const QString &configKey, const QString &suffix)
+{
+    const QString baseName = videoUrl.isLocalFile() ? QFileInfo(videoUrl.toLocalFile()).completeBaseName() : u"subtitles"_s;
+    return u"%1/%2.%3.%4.srt"_s.arg(cacheDir, baseName, configKey, suffix);
+}
+
+QStringList sidecarPaths(const QString &videoPath, const QString &suffix)
+{
+    const QFileInfo info(videoPath);
+    const QString stem = info.absolutePath() + u'/' + info.completeBaseName();
+    return {u"%1.%2.srt"_s.arg(stem, suffix), u"%1.qtpotato.%2.srt"_s.arg(stem, suffix)};
+}
+
+QByteArray contentHash(const QByteArray &data)
+{
+    return QCryptographicHash::hash(data, QCryptographicHash::Sha256).toHex();
+}
+
+std::optional<Record> Registry::find(const QString &video, const QString &language) const
+{
+    for (const Record &record : records) {
+        if (record.video == video && record.language == language) {
+            return record;
+        }
+    }
+    return std::nullopt;
+}
+
+std::optional<Record> Registry::findPath(const QString &path) const
+{
+    const QString wanted = normalized(path);
+    for (const Record &record : records) {
+        if (record.path == wanted) {
+            return record;
+        }
+    }
+    return std::nullopt;
+}
+
+void Registry::put(const Record &record)
+{
+    Record stored = record;
+    stored.path = normalized(record.path);
+    for (Record &existing : records) {
+        if (existing.video == stored.video && existing.language == stored.language) {
+            existing = stored;
+            return;
+        }
+    }
+    records.append(stored);
+}
+
+Registry load(const QString &file)
+{
+    Registry registry;
+    const auto data = readFile(file);
+    if (!data) {
+        return registry;
+    }
+    const QJsonObject root = QJsonDocument::fromJson(*data).object();
+    for (const QJsonValue &value : root.value(u"translations"_s).toArray()) {
+        const QJsonObject object = value.toObject();
+        Record record{object.value(u"video"_s).toString(),
+                      object.value(u"language"_s).toString(),
+                      object.value(u"path"_s).toString(),
+                      object.value(u"configKey"_s).toString(),
+                      object.value(u"contentHash"_s).toString()};
+        if (!record.video.isEmpty() && !record.language.isEmpty() && !record.path.isEmpty()) {
+            registry.put(record);
+        }
+    }
+    return registry;
+}
+
+bool save(const QString &file, const Registry &registry, QString *error)
+{
+    QJsonArray translations;
+    for (const Record &record : registry.records) {
+        if (!QFileInfo::exists(record.path)) {
+            continue;
+        }
+        translations.append(QJsonObject{{u"video"_s, record.video},
+                                        {u"language"_s, record.language},
+                                        {u"path"_s, record.path},
+                                        {u"configKey"_s, record.configKey},
+                                        {u"contentHash"_s, record.contentHash}});
+    }
+    const QJsonObject root{{u"version"_s, RegistryVersion}, {u"translations"_s, translations}};
+
+    QDir().mkpath(QFileInfo(file).absolutePath());
+    QSaveFile output(file);
+    if (!output.open(QIODevice::WriteOnly) || output.write(QJsonDocument(root).toJson()) < 0 || !output.commit()) {
+        if (error) {
+            *error = output.errorString();
+        }
+        return false;
+    }
+    return true;
+}
+
+QString registryFile()
+{
+    return QStandardPaths::writableLocation(QStandardPaths::AppDataLocation) + u"/translations.json"_s;
+}
+
+bool isOwned(const Registry &registry, const QString &path)
+{
+    const auto record = registry.findPath(path);
+    if (!record) {
+        return false;
+    }
+    const auto data = readFile(path);
+    return data && contentHash(*data) == record->contentHash.toLatin1();
+}
+
+PublishResult publish(Registry &registry,
+                      const QUrl &videoUrl,
+                      const QString &suffix,
+                      const QString &configKey,
+                      const QByteArray &data,
+                      const QString &cacheFile)
+{
+    const QString video = videoKey(videoUrl);
+    const QByteArray hash = contentHash(data);
+    PublishResult result{normalized(cacheFile), false};
+
+    if (videoUrl.isLocalFile()) {
+        QStringList candidates = sidecarPaths(videoUrl.toLocalFile(), suffix);
+        // Keep using the file written last time, so a translation does not move to
+        // ".vi.srt" and leave its ".qtpotato.vi.srt" copy behind.
+        if (const auto previous = registry.find(video, suffix);
+            previous && candidates.contains(previous->path) && isOwned(registry, previous->path)) {
+            candidates.removeAll(previous->path);
+            candidates.prepend(previous->path);
+        }
+        for (const QString &path : std::as_const(candidates)) {
+            if (QFileInfo::exists(path) && !isOwned(registry, path)) {
+                continue;
+            }
+            const auto current = readFile(path);
+            if ((current && contentHash(*current) == hash) || writeFile(path, data)) {
+                result = {path, true};
+            }
+            // Both names are in the same folder; when one can't be written, neither can.
+            break;
+        }
+    }
+
+    registry.put({video, suffix, result.path, configKey, QString::fromLatin1(hash)});
+    return result;
+}
+
+void recordRewrite(Registry &registry, const QString &path, const QByteArray &data)
+{
+    if (auto record = registry.findPath(path)) {
+        record->contentHash = QString::fromLatin1(contentHash(data));
+        registry.put(*record);
+    }
+}
+
+QString savedTranslation(const Registry &registry, const QUrl &videoUrl, const QString &suffix)
+{
+    const auto record = registry.find(videoKey(videoUrl), suffix);
+    return record && QFileInfo::exists(record->path) ? record->path : QString();
+}
+}

src/translate/translationstore.h

@@ -0,0 +1,83 @@
+#pragma once
+
+#include <QByteArray>
+#include <QList>
+#include <QString>
+#include <QStringList>
+#include <QUrl>
+
+#include <optional>
+
+// Where translated subtitles live. A translation is written next to the video as
+// "<video>.<lang>.srt" so mpv and other players load it, and recorded in a registry so
+// QtPotato knows which files it wrote, with which settings, and can find the copy it
+// had to keep in the cache when the video's folder was read-only or the video remote.
+namespace TranslationStore
+{
+// "Vietnamese" -> "vi", so the file ends in ".vi.srt" like other players expect.
+QString languageSuffix(const QString &language);
+
+// The registry's name for a video: its absolute path, or the URL of a stream.
+QString videoKey(const QUrl &videoUrl);
+
+// "<cacheDir>/<video basename>.<configKey>.<suffix>.srt". Every settings combination
+// gets its own file, so going back to earlier settings needs no new translation.
+QString cachePath(const QString &cacheDir, const QUrl &videoUrl, const QString &configKey, const QString &suffix);
+
+// Files next to \a videoPath to try, in order: "<base>.<suffix>.srt", then
+// "<base>.qtpotato.<suffix>.srt" for when the user has a subtitle of their own there.
+QStringList sidecarPaths(const QString &videoPath, const QString &suffix);
+
+QByteArray contentHash(const QByteArray &data);
+
+struct Record {
+    QString video;
+    QString language;
+    QString path;
+    // Hash of the source subtitle and the settings it was translated with.
+    QString configKey;
+    // Hash of the file as QtPotato last wrote it. A file that no longer matches was
+    // changed by someone else and is not overwritten.
+    QString contentHash;
+};
+
+// One record per video and language.
+struct Registry {
+    QList<Record> records;
+
+    std::optional<Record> find(const QString &video, const QString &language) const;
+    std::optional<Record> findPath(const QString &path) const;
+    void put(const Record &record);
+};
+
+Registry load(const QString &file);
+// Records whose file is gone are dropped.
+bool save(const QString &file, const Registry &registry, QString *error = nullptr);
+// ~/.local/share/qtpotato/translations.json
+QString registryFile();
+
+// True when \a path holds a translation QtPotato wrote and nobody changed since.
+bool isOwned(const Registry &registry, const QString &path);
+
+struct PublishResult {
+    QString path;
+    bool nextToVideo = false;
+};
+
+// Puts \a data, the translation of \a videoUrl already saved at \a cacheFile, where
+// players find it: next to the video when the folder takes it, without touching a
+// subtitle QtPotato did not write, otherwise \a cacheFile. Records the result.
+PublishResult publish(Registry &registry,
+                      const QUrl &videoUrl,
+                      const QString &suffix,
+                      const QString &configKey,
+                      const QByteArray &data,
+                      const QString &cacheFile);
+
+// After QtPotato edited one of its files (glossary changes), keeps it marked as its own.
+void recordRewrite(Registry &registry, const QString &path, const QByteArray &data);
+
+// The saved translation of \a videoUrl into \a suffix that still exists, or an empty
+// string. Also the check for "has this video been translated yet".
+QString savedTranslation(const Registry &registry, const QUrl &videoUrl, const QString &suffix);
+}