Merge pull request #189 from azooKey/codex/modify-learningmanager.init-and-add-error-handling

feat: `.pause`が残っている場合に学習データの復旧を試みるよう修正

Docs/conversion_algorithms.md

@@ -171,6 +171,8 @@ ComposingText(
 
 3のステップの実行中にエラーが生じた場合、`.pause`があるため、次回キーボードを開いた際は学習を停止状態にします。ついで適切なタイミングで再度ステップ3を実行することで、安全に全てのファイルを更新することができます。
 
+azooKeyKanaKanjiConverter では、変換器を開いた際に `.pause` ファイルが残っている場合、自動的に空の一時記憶とマージを試みて `.pause` を削除し、学習機能を復旧します。
+
 ## 変換候補の並び順
 
 変換候補の並び順の決定はとても難しい問題です。azooKeyではおおよそ以下のようになっています。`Converter.swift`が並び順を決めていますが、とても複雑な実装になっているため、改善したいと思っています。

README.md

@@ -76,6 +76,8 @@ let options = ConvertRequestOptions.withDefaultDictionary(
 )
 ```
 
+開く際に保存処理が中断された `.pause` ファイルが残っている場合は、変換器が自動的に復旧を試みてファイルを削除します。
+
 ### `ComposingText`
 `ComposingText`は入力管理を行いつつ変換をリクエストするためのAPIです。ローマ字入力などを適切にハンドルするために利用できます。詳しくは[ドキュメント](./Docs/composing_text.md)を参照してください。
 

Sources/KanaKanjiConverterModule/DicdataStore/DicdataStore.swift

@@ -139,7 +139,7 @@ public final class DicdataStore {
             } else {
                 self.requestOptions = value
             }
-            let shouldReset = self.learningManager.setRequestOptions(options: value)
+            let shouldReset = self.learningManager.setRequestOptions(value)
             if shouldReset {
                 self.reloadMemory()
             }
@@ -226,7 +226,7 @@ public final class DicdataStore {
         }
     }
 
-    private func perfectMatchLOUDS(query: String, charIDs: [UInt8]) -> [Int] {
+    func perfectMatchLOUDS(query: String, charIDs: [UInt8]) -> [Int] {
         guard let louds = self.loadLOUDS(query: query) else {
             return []
         }

Sources/KanaKanjiConverterModule/DicdataStore/LearningMemory.swift

@@ -626,7 +626,7 @@ final class LearningManager {
             debug("Error: louds/charID.chidが存在しません。このエラーは深刻ですが、テスト時には無視できる場合があります。Description: \(error)")
         }
     }
-    private var char2UInt8: [Character: UInt8] = [:]
+    var char2UInt8: [Character: UInt8] = [:]
 
     static var today: UInt16 {
         UInt16(Int(Date().timeIntervalSince1970) / 86400) - 19000
@@ -653,34 +653,50 @@ final class LearningManager {
         (!self.memoryCollapsed) && self.options.learningType.needUsingMemory
     }
 
-    init() {
-        self.memoryCollapsed = LongTermLearningMemory.memoryCollapsed(directoryURL: self.options.memoryDirectoryURL)
-        if memoryCollapsed {
-            // 学習データが壊れている状態であることを警告する
-            debug("LearningManager init: Memory Collapsed")
-        }
-        if !options.learningType.needUsingMemory {
-            return
-        }
-        Self.updateChar2Int8(bundleURL: options.dictionaryResourceURL, target: &char2UInt8)
-    }
+    init() {}
 
     /// - Returns: Whether cache should be reseted or not.
-    func setRequestOptions(options: ConvertRequestOptions) -> Bool {
-        // 変更があったら`char2Int8`を読み込み直す
-        if options.dictionaryResourceURL != self.options.dictionaryResourceURL {
-            Self.updateChar2Int8(bundleURL: options.dictionaryResourceURL, target: &char2UInt8)
+    func setRequestOptions(_ newOptions: ConvertRequestOptions) -> Bool {
+        // 更新の必要がなければ何もしない
+        if !newOptions.learningType.needUsingMemory {
+            self.options = newOptions
+            return false
         }
-        self.options = options
+        // 変更があったら`char2Int8`を読み込み直す
+        if newOptions.dictionaryResourceURL != self.options.dictionaryResourceURL {
+            Self.updateChar2Int8(bundleURL: newOptions.dictionaryResourceURL, target: &self.char2UInt8)
+        }
+        // ここで更新
+        self.options = newOptions
 
-        switch options.learningType {
+        // 学習の壊れ状態を確認
+        self.memoryCollapsed = LongTermLearningMemory.memoryCollapsed(directoryURL: newOptions.memoryDirectoryURL)
+        if self.memoryCollapsed && newOptions.learningType.needUsingMemory {
+            do {
+                try LongTermLearningMemory.merge(
+                    tempTrie: TemporalLearningMemoryTrie(),
+                    directoryURL: newOptions.memoryDirectoryURL,
+                    maxMemoryCount: newOptions.maxMemoryCount,
+                    char2UInt8: self.char2UInt8
+                )
+            } catch {
+                debug(#file, #function, "automatic merge failed", error)
+            }
+            self.memoryCollapsed = LongTermLearningMemory.memoryCollapsed(directoryURL: newOptions.memoryDirectoryURL)
+        }
+        if self.memoryCollapsed {
+            // 学習データが壊れている状態であることを警告する
+            debug(#file, #function, "LearningManager init: Memory Collapsed")
+        }
+
+        switch self.options.learningType {
         case .inputAndOutput, .onlyOutput: break
         case .nothing:
             self.temporaryMemory = TemporalLearningMemoryTrie()
         }
 
         // リセットチェックも実施
-        if options.shouldResetMemory {
+        if self.options.shouldResetMemory {
             self.reset()
             self.options.shouldResetMemory = false
             return true

Sources/KanaKanjiConverterModule/States.swift

@@ -20,8 +20,11 @@ public enum KeyboardLanguage: String, Codable, Equatable, Sendable {
 }
 
 public enum LearningType: Int, CaseIterable, Sendable {
+    /// 学習情報は変換結果(output)に反映され、学習情報は更新(input)されます
     case inputAndOutput
+    /// 学習情報は変換結果(output)に反映されるのみで、学習情報は更新されません
     case onlyOutput
+    /// 学習情報は一切用いません
     case nothing
 
     package var needUpdateMemory: Bool {

Tests/KanaKanjiConverterModuleTests/LearningMemoryTests.swift

@@ -0,0 +1,103 @@
+@testable import KanaKanjiConverterModule
+import XCTest
+
+final class LearningMemoryTests: XCTestCase {
+    static let resourceURL = Bundle.module.resourceURL!.appendingPathComponent("DictionaryMock", isDirectory: true)
+
+    private func getOptionsForMemoryTest(memoryDirectoryURL: URL) -> ConvertRequestOptions {
+        var options = ConvertRequestOptions.default
+        options.memoryDirectoryURL = memoryDirectoryURL
+        options.dictionaryResourceURL = Self.resourceURL
+        options.learningType = .inputAndOutput
+        options.maxMemoryCount = 32
+        return options
+    }
+
+    func testPauseFileIsClearedOnInit() throws {
+        let dir = FileManager.default.temporaryDirectory.appendingPathComponent("LearningMemoryTest-\(UUID().uuidString)", isDirectory: true)
+        try FileManager.default.createDirectory(at: dir, withIntermediateDirectories: true)
+        defer { try? FileManager.default.removeItem(at: dir) }
+
+        let options = self.getOptionsForMemoryTest(memoryDirectoryURL: dir)
+        let manager = LearningManager()
+        _ = manager.setRequestOptions(options)
+
+        let element = DicdataElement(word: "テスト", ruby: "テスト", cid: CIDData.一般名詞.cid, mid: MIDData.一般.mid, value: -10)
+        manager.update(data: [element])
+        manager.save()
+
+        // ポーズファイルを設置
+        let pauseURL = dir.appendingPathComponent(".pause", isDirectory: false)
+        FileManager.default.createFile(atPath: pauseURL.path, contents: Data())
+        XCTAssertTrue(LongTermLearningMemory.memoryCollapsed(directoryURL: dir))
+
+        // ここで副作用が発生
+        _ = manager.setRequestOptions(options)
+
+        // 学習の破壊状態が回復されていることを確認
+        XCTAssertFalse(LongTermLearningMemory.memoryCollapsed(directoryURL: dir))
+        try? FileManager.default.removeItem(at: pauseURL)
+    }
+
+    func testMemoryFilesCreateAndRemove() throws {
+        let dir = FileManager.default.temporaryDirectory.appendingPathComponent("LearningMemoryTest-\(UUID().uuidString)", isDirectory: true)
+        try FileManager.default.createDirectory(at: dir, withIntermediateDirectories: true)
+        defer { try? FileManager.default.removeItem(at: dir) }
+
+        let options = self.getOptionsForMemoryTest(memoryDirectoryURL: dir)
+        let manager = LearningManager()
+        _ = manager.setRequestOptions(options)
+
+        let element = DicdataElement(word: "テスト", ruby: "テスト", cid: CIDData.一般名詞.cid, mid: MIDData.一般.mid, value: -10)
+        manager.update(data: [element])
+        manager.save()
+
+        let files = try FileManager.default.contentsOfDirectory(at: dir, includingPropertiesForKeys: nil)
+        XCTAssertTrue(files.contains { $0.lastPathComponent == "memory.louds" })
+        XCTAssertTrue(files.contains { $0.lastPathComponent == "memory.loudschars2" })
+        XCTAssertTrue(files.contains { $0.lastPathComponent == "memory.memorymetadata" })
+        XCTAssertTrue(files.contains { $0.lastPathComponent.hasSuffix(".loudstxt3") })
+
+        manager.reset()
+        let filesAfter = try FileManager.default.contentsOfDirectory(at: dir, includingPropertiesForKeys: nil)
+        XCTAssertTrue(filesAfter.isEmpty)
+    }
+
+    func testForgetMemory() throws {
+        let dir = FileManager.default.temporaryDirectory.appendingPathComponent("LearningManagerPersistence-\(UUID().uuidString)", isDirectory: true)
+        try FileManager.default.createDirectory(at: dir, withIntermediateDirectories: true)
+        defer { try? FileManager.default.removeItem(at: dir) }
+
+        let options = self.getOptionsForMemoryTest(memoryDirectoryURL: dir)
+        let manager = LearningManager()
+        _ = manager.setRequestOptions(options)
+        let element = DicdataElement(word: "テスト", ruby: "テスト", cid: CIDData.一般名詞.cid, mid: MIDData.一般.mid, value: -10)
+        manager.update(data: [element])
+        manager.save()
+
+        let dicdataStore = DicdataStore(requestOptions: options)
+        dicdataStore.sendToDicdataStore(.setRequestOptions(options))
+        let charIDs = "テスト".map { dicdataStore.character2charId($0) }
+        let indices = dicdataStore.perfectMatchLOUDS(query: "memory", charIDs: charIDs)
+        let dicdata = dicdataStore.getDicdataFromLoudstxt3(identifier: "memory", indices: indices)
+        XCTAssertFalse(dicdata.isEmpty)
+        XCTAssertTrue(dicdata.contains { $0.word == element.word && $0.ruby == element.ruby })
+
+        dicdataStore.sendToDicdataStore(
+            .forgetMemory(
+                Candidate(
+                    text: element.word,
+                    value: element.value(),
+                    correspondingCount: 3,
+                    lastMid: element.mid,
+                    data: [element]
+                )
+            )
+        )
+
+        let indices2 = dicdataStore.perfectMatchLOUDS(query: "memory", charIDs: charIDs)
+        let dicdata2 = dicdataStore.getDicdataFromLoudstxt3(identifier: "memory", indices: indices2)
+        XCTAssertFalse(dicdata2.contains { $0.word == element.word && $0.ruby == element.ruby })
+    }
+}
+

Tests/KanaKanjiConverterModuleTests/TemporalLearningMemoryTrieTests.swift

@@ -0,0 +1,56 @@
+@testable import KanaKanjiConverterModule
+import XCTest
+
+final class TemporalLearningMemoryTrieTests: XCTestCase {
+    static let resourceURL = Bundle.module.resourceURL!.appendingPathComponent("DictionaryMock", isDirectory: true)
+
+    static func loadCharMap() -> [Character: UInt8] {
+        let chidURL = resourceURL.appendingPathComponent("louds/charID.chid", isDirectory: false)
+        let string = try! String(contentsOf: chidURL, encoding: .utf8)
+        return Dictionary(uniqueKeysWithValues: string.enumerated().map { ($0.element, UInt8($0.offset)) })
+    }
+
+    func chars(for string: String) -> [UInt8] {
+        LearningManager.keyToChars(string, char2UInt8: Self.loadCharMap())!
+    }
+
+    func testMemorizeAndMatch() throws {
+        var trie = TemporalLearningMemoryTrie()
+        let element1 = DicdataElement(word: "テスト", ruby: "テスト", cid: CIDData.一般名詞.cid, mid: MIDData.一般.mid, value: -10)
+        let element2 = DicdataElement(word: "テスター", ruby: "テスター", cid: CIDData.一般名詞.cid, mid: MIDData.一般.mid, value: -12)
+
+        trie.memorize(dicdataElement: element1, chars: chars(for: element1.ruby))
+        trie.memorize(dicdataElement: element2, chars: chars(for: element2.ruby))
+
+        let result1 = trie.perfectMatch(chars: chars(for: element1.ruby))
+        XCTAssertEqual(result1.count, 1)
+        XCTAssertEqual(result1.first?.word, element1.word)
+        XCTAssertTrue(result1.first?.metadata.contains(.isLearned) ?? false)
+
+        let result2 = trie.throughMatch(chars: chars(for: element2.ruby), depth: (element2.ruby.count - 1)..<element2.ruby.count)
+        XCTAssertEqual(result2.map { $0.word }, [element2.word])
+
+        let prefixResult = trie.prefixMatch(chars: chars(for: "テス"))
+        XCTAssertEqual(Set(prefixResult.map { $0.word }), Set([element1.word, element2.word]))
+    }
+
+    func testMemorizeUpdateCountAndForget() throws {
+        var trie = TemporalLearningMemoryTrie()
+        let element = DicdataElement(word: "テスター", ruby: "テスター", cid: CIDData.一般名詞.cid, mid: MIDData.一般.mid, value: -10)
+        let charIDs = chars(for: element.ruby)
+
+        trie.memorize(dicdataElement: element, chars: charIDs)
+        var stored = trie.perfectMatch(chars: charIDs).first!
+        let adjust1 = stored.adjust
+
+        trie.memorize(dicdataElement: element, chars: charIDs)
+        stored = trie.perfectMatch(chars: charIDs).first!
+        let adjust2 = stored.adjust
+
+        XCTAssertGreaterThan(adjust2, adjust1)
+        XCTAssertEqual(trie.perfectMatch(chars: charIDs).count, 1)
+
+        XCTAssertTrue(trie.forget(dicdataElement: stored, chars: charIDs))
+        XCTAssertTrue(trie.perfectMatch(chars: charIDs).isEmpty)
+    }
+}