Merge pull request #205 from azooKey/fix/roman_triplet_input

fix: getWiseData関連の実装を修正し、bya→ビャなどの変換を適切に表示

Sources/CliTool/Subcommands/SessionCommand.swift

@@ -13,6 +13,8 @@ extension Subcommands {
         var displayTopN: Int = 1
         @Option(name: [.customLong("zenz")], help: "gguf format model weight for zenz.")
         var zenzWeightPath: String = ""
+        @Flag(name: [.customLong("mix_english_candidate")], help: "Enable mixing English Candidates.")
+        var mixEnglishCandidate = false
         @Flag(name: [.customLong("disable_prediction")], help: "Disable producing prediction candidates.")
         var disablePrediction = false
         @Flag(name: [.customLong("enable_memory")], help: "Enable memory.")
@@ -290,7 +292,7 @@ extension Subcommands {
                 keyboardLanguage: .ja_JP,
                 typographyLetterCandidate: false,
                 unicodeCandidate: true,
-                englishCandidateInRoman2KanaInput: true,
+                englishCandidateInRoman2KanaInput: self.mixEnglishCandidate,
                 fullWidthRomanCandidate: false,
                 halfWidthKanaCandidate: false,
                 learningType: learningType,

Sources/KanaKanjiConverterModule/ComposingText.swift

@@ -501,7 +501,7 @@ extension ComposingText {
     /// - Returns: 領域がvalidであれば`convertTarget`を返し、invalidなら`nil`を返す。
     /// - Note: `elements = [r(k, a, n, s, h, a)]`のとき、`k,a,n,s,h,a`や`k, a`は正当だが`a, n`や`s, h`は正当ではない。`k, a, n`は特に正当であるとみなす。
     static func getConvertTargetIfRightSideIsValid(lastElement: InputElement, of originalElements: [InputElement], to rightIndex: Int, convertTargetElements: [ConvertTargetElement]) -> [Character]? {
-        debug("getConvertTargetIfRightSideIsValid", lastElement, rightIndex)
+        debug(#function, lastElement, rightIndex)
         if originalElements.endIndex < rightIndex {
             return nil
         }

Sources/KanaKanjiConverterModule/DicdataStore/DicdataStore.swift

@@ -281,7 +281,7 @@ public final class DicdataStore {
         let toIndexLeft = toIndexRange?.startIndex ?? fromIndex
         let toIndexRight = min(toIndexRange?.endIndex ?? inputData.input.count, fromIndex + self.maxlength)
         if fromIndex > toIndexLeft || toIndexLeft >= toIndexRight {
-            debug("getLOUDSDataInRange: index is wrong")
+            debug(#function, "index is wrong")
             return []
         }
 
@@ -388,15 +388,12 @@ public final class DicdataStore {
     private func getFrozenLOUDSDataInRange(inputData: ComposingText, from fromIndex: Int, toIndexRange: Range<Int>? = nil) -> [LatticeNode] {
         let toIndexLeft = toIndexRange?.startIndex ?? fromIndex
         let toIndexRight = min(toIndexRange?.endIndex ?? inputData.input.count, fromIndex + self.maxlength)
-        debug("getLOUDSDataInRange", fromIndex, toIndexRange?.description ?? "nil", toIndexLeft, toIndexRight)
+        debug(#function, fromIndex, toIndexRange?.description ?? "nil", toIndexLeft, toIndexRight)
         if fromIndex > toIndexLeft || toIndexLeft >= toIndexRight {
-            debug("getLOUDSDataInRange: index is wrong")
+            debug(#function, "index is wrong")
             return []
         }
 
-        let segments = (fromIndex ..< toIndexRight).reduce(into: []) { (segments: inout [String], rightIndex: Int) in
-            segments.append((segments.last ?? "") + String(inputData.input[rightIndex].character.toKatakana()))
-        }
         let character = String(inputData.input[fromIndex].character.toKatakana())
         let characterNode = LatticeNode(data: DicdataElement(word: character, ruby: character, cid: CIDData.一般名詞.cid, mid: MIDData.一般.mid, value: -10), inputRange: fromIndex ..< fromIndex + 1)
         if fromIndex == .zero {
@@ -404,9 +401,10 @@ public final class DicdataStore {
         }
 
         // MARK: 誤り訂正なし
-        let stringToEndIndex = inputData.getRanges(fromIndex, rightIndexRange: toIndexLeft ..< toIndexRight)
+        let stringToEndIndex = inputData.getRangesWithoutTypos(fromIndex, rightIndexRange: toIndexLeft ..< toIndexRight)
         // MARK: 検索対象を列挙していく。
         guard let (minString, maxString) = stringToEndIndex.keys.minAndMax(by: {$0.count < $1.count}) else {
+            debug(#function, "minString/maxString is nil", stringToEndIndex)
             return [characterNode]
         }
         let maxIDs = maxString.map(self.character2charId)
@@ -424,11 +422,10 @@ public final class DicdataStore {
             // temporalな学習結果にpenaltyを加えて追加する
             dicdata.append(contentsOf: self.learningManager.temporaryThroughMatch(charIDs: consume maxIDs, depth: depth))
         }
-        for i in toIndexLeft ..< toIndexRight {
-            dicdata.append(contentsOf: self.getWiseDicdata(convertTarget: segments[i - fromIndex], inputData: inputData, inputRange: fromIndex ..< i + 1))
-        }
-        for item in stringToEndIndex {
-            dicdata.append(contentsOf: self.getMatchDynamicUserDict(String(item.key)))
+        for (key, value) in stringToEndIndex {
+            let convertTarget = String(key)
+            dicdata.append(contentsOf: self.getWiseDicdata(convertTarget: convertTarget, inputData: inputData, inputRange: fromIndex ..< value + 1))
+            dicdata.append(contentsOf: self.getMatchDynamicUserDict(convertTarget))
         }
         if fromIndex == .zero {
             return dicdata.compactMap {
@@ -629,7 +626,8 @@ public final class DicdataStore {
 
         // ローマ字入力の場合、単体でひらがな・カタカナ化した候補も追加
         if requestOptions.keyboardLanguage != .en_US && inputData.input[inputRange].allSatisfy({$0.inputStyle == .roman2kana}) {
-            if let katakana = Roman2Kana.katakanaChanges[convertTarget], let hiragana = Roman2Kana.hiraganaChanges[Array(convertTarget)] {
+            let roman = String(inputData.input[inputRange].map(\.character))
+            if let katakana = Roman2Kana.katakanaChanges[roman], let hiragana = Roman2Kana.hiraganaChanges[Array(roman)] {
                 result.append(DicdataElement(word: String(hiragana), ruby: katakana, cid: CIDData.固有名詞.cid, mid: MIDData.一般.mid, value: -13))
                 result.append(DicdataElement(ruby: katakana, cid: CIDData.固有名詞.cid, mid: MIDData.一般.mid, value: -14))
             }

Sources/KanaKanjiConverterModule/DicdataStore/TypoCorrection.swift

@@ -23,7 +23,7 @@ extension ComposingText {
     /// `left <= rightIndexRange.startIndex`が常に成り立つ
     func getRangesWithTypos(_ left: Int, rightIndexRange: Range<Int>) -> [[Character]: (endIndex: Int, penalty: PValue)] {
         let count = rightIndexRange.endIndex - left
-        debug("getRangesWithTypos", left, rightIndexRange, count)
+        debug(#function, left, rightIndexRange, count)
         let nodes = (0..<count).map {(i: Int) in
             Self.lengths.flatMap {(k: Int) -> [TypoCandidate] in
                 let j = i + k
@@ -100,15 +100,16 @@ extension ComposingText {
     /// closedRangeでもらう
     /// 例えば`left=4, rightIndexRange=6..<10`の場合、`4...6, 4...7, 4...8, 4...9`の範囲で計算する
     /// `left <= rightIndexRange.startIndex`が常に成り立つ
-    func getRanges(_ left: Int, rightIndexRange: Range<Int>) -> [[Character]: Int] {
+    func getRangesWithoutTypos(_ left: Int, rightIndexRange: Range<Int>) -> [[Character]: Int] {
         let count = rightIndexRange.endIndex - left
-        debug("getRangesWithTypos", left, rightIndexRange, count)
+        debug(#function, left, rightIndexRange, count)
         let nodes = (0..<count).map {(i: Int) in
             Self.lengths.flatMap {(k: Int) -> [TypoCandidate] in
                 let j = i + k
                 if count <= j {
                     return []
                 }
+                // frozen: trueとしているため、typo候補は含まれない
                 return Self.getTypo(self.input[left + i ... left + j], frozen: true)
             }
         }

Sources/KanaKanjiConverterModule/Roman2Kana.swift

@@ -8,7 +8,7 @@
 
 import Foundation
 enum Roman2Kana {
-    static let katakanaChanges: [String: String] = Dictionary(uniqueKeysWithValues: hiraganaChanges.map { (String($0.key), String($0.value)) })
+    static let katakanaChanges: [String: String] = Dictionary(uniqueKeysWithValues: hiraganaChanges.map { (String($0.key), String($0.value).toKatakana()) })
     static let hiraganaChanges: [[Character]: [Character]] = Dictionary(uniqueKeysWithValues: [
         "a": "あ",
         "xa": "ぁ",