Support terminal Hangul graphemes

Author: SqlRush

docs/cc-100-roadmap.md

@@ -902,6 +902,8 @@ M7 补充：prompt history `LogEntry` 读取现在接受 `sessionID`/`session`/`
 
 本轮补充：terminal parser 的 text grapheme 分段继续补齐 emoji keycap sequence，`1️⃣`/`2⃣` 这类 keycap 在完整输入和跨 `Feed()` 边界切在 base 或 variation selector 后时都会保持单个宽 grapheme；完整 Unicode UAX #29 分段仍未宣称完成。
 
+本轮补充：terminal parser 的 text grapheme 分段继续补齐 Hangul L/V/T jamo 连接规则，decomposed `한` 这类音节在完整输入以及跨 `Feed()` 边界切在 leading/vowel jamo 后时都会保持单个宽 grapheme；完整 Unicode UAX #29 分段仍未宣称完成。
+
 本轮补充：terminal CSI parser 现在对 tokenizer flush 出来的非 final-byte incomplete CSI 返回 unknown action，而不是丢弃，贴近官方 `parseCSI` 对 flushed partial sequence 的 fallback 行为。
 
 本轮补充：terminal sequence dispatcher 对 tokenizer flush 出来的 OSC partial sequence 使用 `ParseOSCContent` fallback，允许无 BEL/ST terminator 的 title/link/tab-status content 按官方 parser 语义产出 action。

docs/claude-code-go-rewrite-plan.md

@@ -409,6 +409,7 @@ test/parity/                 # golden tests against TS/official behavior
 - 本轮补充：terminal CSI parser 按 ANSI 默认参数解析 scroll-region (`CSI r`/`CSI ;10r`)，缺失 top 默认为 1，缺失 bottom 保持 0 表示 reset/full-height，避免 reset scroll region 被误判为单行区域。
 - 本轮补充：terminal CSI parser 把 DECSTR soft reset (`CSI !p`) 归入 reset action，并在 terminal parser 中清理 SGR/link 状态。
 - 本轮补充：terminal parser 的 text grapheme 分段继续补齐 emoji keycap sequence，`1️⃣`/`2⃣` 这类 keycap 在完整输入和跨 `Feed()` 边界切在 base 或 variation selector 后时都会保持单个宽 grapheme；完整 Unicode UAX #29 分段仍未宣称完成。
+- 本轮补充：terminal parser 的 text grapheme 分段继续补齐 Hangul L/V/T jamo 连接规则，decomposed `한` 这类音节在完整输入以及跨 `Feed()` 边界切在 leading/vowel jamo 后时都会保持单个宽 grapheme；完整 Unicode UAX #29 分段仍未宣称完成。
 - 本轮补充：prompt history 写入按官方 `history.ts` 跳过 image pasted content，避免把 image base64/filename/mediaType 存入 `history.jsonl`，读取路径仍兼容旧 image metadata。
 - 本轮补充：paste-cache 增加按 cutoff mtime 清理旧 `.txt` paste 文件的 best-effort 入口，忽略缺失目录、非 `.txt` 文件和单文件错误，和官方 `cleanupOldPastes` 语义对齐。
 - 本轮补充：Buffered prompt history writer 支持撤销最近 pending entry，覆盖官方 `removeLastFromHistory` 在异步 flush 前直接从 pending buffer 移除的 fast path。

docs/first-second-parity-audit.md

@@ -348,6 +348,7 @@ M7 progress now includes:
 - `internal/tui`: terminal sequence dispatch now parses SS3 application cursor sequences such as `ESC OA`/`OB`/`OC`/`OD` as structured cursor move actions.
 - `internal/tui`: terminal grapheme width now keeps base+combining-mark clusters at the base glyph width while preserving wide treatment for emoji presentation, ZWJ, regional-indicator, and tag sequences.
 - `internal/tui`: terminal grapheme segmentation now treats emoji keycap sequences such as `1️⃣` and `2⃣` as single wide graphemes, including streaming chunks split after the keycap base or variation selector.
+- `internal/tui`: terminal grapheme segmentation now applies Hangul L/V/T jamo joining, keeping decomposed syllables such as `한` as one wide grapheme across full input and streaming chunk boundaries.
 - `internal/tui`: image hint parsing now accepts OSC ST terminators and base64 `name=` filenames while preserving prompt pasted-content metadata.
 - `internal/session`: prompt history writing now skips image pasted-content records like official `history.ts`, while the reader still accepts older image metadata entries.
 - `internal/session`: paste-cache now has a best-effort cutoff-mtime cleanup helper for old `.txt` paste files, matching official `cleanupOldPastes` behavior.

internal/tui/terminal_parser.go

@@ -265,7 +265,7 @@ func terminalGraphemeMayContinueAtChunkBoundary(value string) bool {
 		return false
 	}
 	last, _ := utf8.DecodeLastRuneInString(value)
-	if last == 0x200d || isTerminalEmojiModifier(last) || terminalGraphemeCanStartKeycapSequence(value) {
+	if last == 0x200d || isTerminalEmojiModifier(last) || terminalGraphemeCanStartKeycapSequence(value) || terminalGraphemeMayContinueHangul(value) {
 		return true
 	}
 	regionalCount := 0
@@ -388,6 +388,7 @@ func nextTerminalGrapheme(text string) (string, int) {
 	if isTerminalRegionalIndicator(first) {
 		regionalCount = 1
 	}
+	hangulClass := terminalHangulClassOf(first)
 	for end < len(text) {
 		r, nextSize := utf8.DecodeRuneInString(text[end:])
 		if r == utf8.RuneError && nextSize == 0 {
@@ -413,6 +414,12 @@ func nextTerminalGrapheme(text string) (string, int) {
 			regionalCount++
 			continue
 		}
+		if nextHangulClass := terminalHangulClassOf(r); terminalHangulCanJoin(hangulClass, nextHangulClass) {
+			end += nextSize
+			hangulClass = terminalHangulJoinedClass(hangulClass, nextHangulClass)
+			previousWasZWJ = false
+			continue
+		}
 		break
 	}
 	return text[:end], end
@@ -425,7 +432,7 @@ func terminalGraphemeStringWidth(grapheme string) int {
 	for _, r := range grapheme {
 		if !hasBase {
 			hasBase = true
-			if isTerminalEmoji(r) || isTerminalEastAsianWide(r) {
+			if isTerminalEmoji(r) || isTerminalEastAsianWide(r) || isTerminalHangulWideBase(r) {
 				baseWidth = 2
 			}
 			continue
@@ -468,6 +475,86 @@ func terminalGraphemeCanStartKeycapSequence(value string) bool {
 	return len(runes) == 2 && isTerminalEmojiKeycapBase(runes[0]) && runes[1] == 0xfe0f
 }
 
+func terminalGraphemeMayContinueHangul(value string) bool {
+	class := terminalHangulNone
+	for _, r := range value {
+		if next := terminalHangulClassOf(r); next != terminalHangulNone {
+			class = next
+		}
+	}
+	return class != terminalHangulNone
+}
+
+type terminalHangulClass uint8
+
+const (
+	terminalHangulNone terminalHangulClass = iota
+	terminalHangulL
+	terminalHangulV
+	terminalHangulT
+	terminalHangulLV
+	terminalHangulLVT
+)
+
+func terminalHangulClassOf(r rune) terminalHangulClass {
+	switch {
+	case (r >= 0x1100 && r <= 0x115f) || (r >= 0xa960 && r <= 0xa97c):
+		return terminalHangulL
+	case (r >= 0x1160 && r <= 0x11a7) || (r >= 0xd7b0 && r <= 0xd7c6):
+		return terminalHangulV
+	case (r >= 0x11a8 && r <= 0x11ff) || (r >= 0xd7cb && r <= 0xd7fb):
+		return terminalHangulT
+	case r >= 0xac00 && r <= 0xd7a3:
+		if (r-0xac00)%28 == 0 {
+			return terminalHangulLV
+		}
+		return terminalHangulLVT
+	default:
+		return terminalHangulNone
+	}
+}
+
+func terminalHangulCanJoin(left, right terminalHangulClass) bool {
+	switch left {
+	case terminalHangulL:
+		return right == terminalHangulL || right == terminalHangulV || right == terminalHangulLV || right == terminalHangulLVT
+	case terminalHangulV, terminalHangulLV:
+		return right == terminalHangulV || right == terminalHangulT
+	case terminalHangulT, terminalHangulLVT:
+		return right == terminalHangulT
+	default:
+		return false
+	}
+}
+
+func terminalHangulJoinedClass(left, right terminalHangulClass) terminalHangulClass {
+	switch {
+	case left == terminalHangulL && right == terminalHangulL:
+		return terminalHangulL
+	case left == terminalHangulL && right == terminalHangulV:
+		return terminalHangulV
+	case left == terminalHangulL && right == terminalHangulLV:
+		return terminalHangulLV
+	case left == terminalHangulL && right == terminalHangulLVT:
+		return terminalHangulLVT
+	case (left == terminalHangulV || left == terminalHangulLV) && right == terminalHangulV:
+		return terminalHangulV
+	case (left == terminalHangulV || left == terminalHangulLV) && right == terminalHangulT:
+		return terminalHangulT
+	case (left == terminalHangulT || left == terminalHangulLVT) && right == terminalHangulT:
+		return terminalHangulT
+	case right != terminalHangulNone:
+		return right
+	default:
+		return left
+	}
+}
+
+func isTerminalHangulWideBase(r rune) bool {
+	class := terminalHangulClassOf(r)
+	return class == terminalHangulL || class == terminalHangulLV || class == terminalHangulLVT
+}
+
 func isTerminalEmoji(r rune) bool {
 	return (r >= 0x2600 && r <= 0x26ff) ||
 		(r >= 0x2700 && r <= 0x27bf) ||

internal/tui/terminal_parser_test.go

@@ -131,7 +131,7 @@ func TestTerminalParserDispatchesStringControlActions(t *testing.T) {
 
 func TestTerminalParserSegmentsCommonGraphemeClusters(t *testing.T) {
 	parser := NewTerminalParser()
-	actions := parser.Feed("e\u0301 \u2764\ufe0f 1\ufe0f\u20e3 2\u20e3 \U0001f44b\U0001f3fd \U0001f469\u200d\U0001f4bb \U0001f1fa\U0001f1f8 \U0001f3f4\U000e0067\U000e0062\U000e0073\U000e0063\U000e0074\U000e007f")
+	actions := parser.Feed("e\u0301 \u2764\ufe0f 1\ufe0f\u20e3 2\u20e3 \u1112\u1161\u11ab \U0001f44b\U0001f3fd \U0001f469\u200d\U0001f4bb \U0001f1fa\U0001f1f8 \U0001f3f4\U000e0067\U000e0062\U000e0073\U000e0063\U000e0074\U000e007f")
 	if len(actions) != 1 || actions[0].Type != TerminalActionText {
 		t.Fatalf("actions = %#v", actions)
 	}
@@ -144,6 +144,8 @@ func TestTerminalParserSegmentsCommonGraphemeClusters(t *testing.T) {
 		{Value: " ", Width: 1},
 		{Value: "2\u20e3", Width: 2},
 		{Value: " ", Width: 1},
+		{Value: "\u1112\u1161\u11ab", Width: 2},
+		{Value: " ", Width: 1},
 		{Value: "\U0001f44b\U0001f3fd", Width: 2},
 		{Value: " ", Width: 1},
 		{Value: "\U0001f469\u200d\U0001f4bb", Width: 2},
@@ -241,6 +243,33 @@ func TestTerminalParserKeepsChunkedGraphemeClustersTogether(t *testing.T) {
 		t.Fatalf("flush keycap base actions = %#v", actions)
 	}
 
+	parser = NewTerminalParser()
+	if actions := parser.Feed("\u1112"); len(actions) != 0 {
+		t.Fatalf("partial hangul leading actions = %#v", actions)
+	}
+	actions = parser.Feed("\u1161\u11ab!")
+	if len(actions) != 1 || len(actions[0].Graphemes) != 2 {
+		t.Fatalf("hangul actions = %#v", actions)
+	}
+	if got := actions[0].Graphemes[0]; got.Value != "\u1112\u1161\u11ab" || got.Width != 2 {
+		t.Fatalf("hangul grapheme = %#v", got)
+	}
+	if width := TerminalActionsVisibleWidth(actions); width != 3 {
+		t.Fatalf("hangul width = %d", width)
+	}
+
+	parser = NewTerminalParser()
+	if actions := parser.Feed("\u1112\u1161"); len(actions) != 0 {
+		t.Fatalf("partial hangul vowel actions = %#v", actions)
+	}
+	actions = parser.Feed("\u11ab!")
+	if len(actions) != 1 || len(actions[0].Graphemes) != 2 {
+		t.Fatalf("hangul trailing actions = %#v", actions)
+	}
+	if got := actions[0].Graphemes[0]; got.Value != "\u1112\u1161\u11ab" || got.Width != 2 {
+		t.Fatalf("hangul trailing grapheme = %#v", got)
+	}
+
 	parser = NewTerminalParser()
 	actions = parser.Feed("\U0001f1fa" + CSISequence(31, "m") + "red")
 	if len(actions) != 2 {