@qnroa/qtype 0.1.0 → 0.1.1

package/CHANGELOG.md

@@ -13,6 +13,37 @@ can land in any minor bump (`0.x.0`).
 
 ## [Unreleased]
 
+## [0.1.1] — 2026-08-28
+
+Fixes typing comparison for equivalent-but-different characters that
+the OS/IME hands out (fullwidth space in Chinese input mode, smart
+quotes on macOS/iOS, CJK punctuation vs ASCII, etc.). Display side is
+unchanged; equivalence only runs when comparing a keystroke against
+the expected character.
+
+### Fixed
+- **Fullwidth / half-width space now match on input.** Typing `　`
+  (Chinese IME space) against a card whose target is ` ` no
+  longer misses. Also covers non-breaking space ` `. Mobile users
+  on Chinese keyboards used to have to switch to English just to hit
+  space; that friction is gone.
+- **Smart quotes match straight quotes on input.** iOS / macOS auto-
+  substituted `“`/`”`/`‘`/`’` now compare equal to
+  ASCII `"` and `'`. Cards typed on a phone stop failing on quotes.
+- **CJK punctuation matches ASCII punctuation on input.** `，。？！：
+  ；（）【】《》／＋＝～` and friends are compared equivalent to their
+  ASCII forms. Users don't need to swap IME state mid-card.
+- **En-dash / em-dash / fullwidth hyphen match ASCII `-` on input.**
+
+### Changed
+- **Display is honest about the source.** `normalizeText` no longer
+  rewrites user-visible characters at import time (previously
+  `　  ` were folded to ASCII space, `—` was expanded
+  to `--`, etc.). Cards render exactly what the author wrote; the
+  equivalence table lives in `typing/engine/normalize.ts` and only
+  runs at keystroke comparison time. Untypable characters (control,
+  zero-width, emoji) are still stripped.
+
 ## [0.1.0] — 2026-08-27
 
 Breaking dist format. Source `material/*.md` unchanged — the source

package/CHANGELOG.zh.md

@@ -12,6 +12,29 @@ English: [CHANGELOG.md](https://www.npmjs.com/package/@qnroa/qtype?activeTab=cod
 
 ## [Unreleased]
 
+## [0.1.1] — 2026-08-28
+
+修复输入法/终端产出的等价字符（中文输入法下的全角空格、macOS/iOS
+自动替换的智能引号、中文标点 vs 英文标点等）造成的误判。显示层
+不变，等价比较只在"按键 vs 期望字符"的那一步生效。
+
+### 修复
+- **全角/半角空格视为等价**。中文输入法下敲 `　`(U+3000) 匹配卡片
+  里的半角 ` ` 不再算错；` ` 不间断空格同样等价。手机上中文
+  输入法用户不再需要切英文只为了打个空格。
+- **智能引号 = 直引号**。iOS/macOS 自动替换的 `""''` 现在跟 ASCII
+  `"` `'` 比较相等。手机上打卡不会因为引号变形而卡住。
+- **中文标点 = 英文标点**。`，。？！：；（）【】《》／＋＝～` 等在
+  比较时等价于对应的 ASCII 标点，无需切换输入法状态。
+- **En-dash / em-dash / 全角减号 = ASCII `-`**。
+
+### 变更
+- **显示忠实于源码**。`normalizeText` 不再在 import 时改写用户可见
+  字符（之前 `　  ` 会被折叠成 ASCII 空格，`—` 被扩展成 `--`）。
+  卡片显示什么就是作者写的什么；等价表放在
+  `typing/engine/normalize.ts`，只在打字比较时生效。控制字符 / 零
+  宽字符 / emoji 依然会剔除。
+
 ## [0.1.0] — 2026-08-27
 
 破坏性 dist 格式变更。源 `material/*.md` 未变 —— 语法和语义端到端

package/dist/core/keystroke/textNormalize.js

@@ -1,15 +1,18 @@
 // Import-time text sanitization + validation.
-// Goal: guarantee that every character surviving normalization has a valid
-// keystroke plan, so runtime never has to skip or bail on unknown chars.
+//
+// This layer used to rewrite user-visible characters (smart quotes to
+// straight, fullwidth space to half, em-dash to double hyphen, etc.) at
+// import time — which meant the *displayed* text no longer matched what
+// the author wrote. That's the wrong layer: display should be honest
+// about the source; runtime equivalence for typing lives in
+// `typing/engine/normalize.ts` and only affects comparison.
+//
+// So the rewrite table is empty. We still strip characters that are
+// truly untypable (control chars, zero-width, emoji), because those
+// have no keyboard input at all.
 import { charToKeystrokes } from './index.js';
-// Character rewrites: smart quotes, unusual dashes, etc.
-const REWRITE = {
-    '—': '——', // em-dash → double hyphen (matches keystroke plan for 中文 —)
-    '–': '-', // en-dash → hyphen
-    '…': '……', // horizontal ellipsis → double per convention (chineseMap uses ……)
-    ' ': ' ', // non-breaking space → regular space
-    '　': ' ', // ideographic space → regular space
-};
+// Rewrites are intentionally empty. See file-level comment.
+const REWRITE = {};
 // Ranges of code points to strip entirely (they can't be typed).
 const STRIP_RANGES = [
     [0x0000, 0x0008], // C0 controls before \t

package/dist/core/typing/engine/TypingInput.js

@@ -19,6 +19,7 @@
  */
 import { CharAttr } from './chars.js';
 import { Feedback } from './feedback.js';
+import { normalizeForCompare } from './normalize.js';
 import { defaultTypingSettings, } from './settings.js';
 import { computeStats } from './stats.js';
 export class TypingInput {
@@ -73,9 +74,14 @@ export class TypingInput {
                 return Feedback.Correct;
             expected = this.#targetCodepoints[this.#position];
         }
+        // Normalize both sides through the same equivalence table so
+        // fullwidth/half-width / smart-quote / CJK-punctuation variants
+        // don't cause false misses. Display remains untouched.
+        const nCp = normalizeForCompare(codePoint);
+        const nExp = normalizeForCompare(expected);
         const isHit = this.settings.ignoreCase
-            ? sameLetter(codePoint, expected)
-            : codePoint === expected;
+            ? sameLetter(nCp, nExp)
+            : nCp === nExp;
         // Clone the array so subscribers see a new reference. Only the touched
         // Char is a new object — the rest are shared.
         const next = this.#chars.slice();

package/dist/core/typing/engine/normalize.js

@@ -0,0 +1,87 @@
+/**
+ * Runtime input normalization for typing comparison.
+ *
+ * The engine used to compare `codePoint === expected` strictly, which broke
+ * for equivalent-but-different characters that the OS/IME hands out
+ * depending on state:
+ *
+ *   - Fullwidth space U+3000 vs regular space U+0020 (Chinese IME)
+ *   - Non-breaking space U+00A0 vs U+0020
+ *   - Smart quotes U+201C/U+201D/U+2018/U+2019 vs straight quotes
+ *   - Fullwidth CJK punctuation (U+FF0C etc.) vs ASCII punctuation
+ *   - En-dash U+2013 vs hyphen U+002D
+ *   - Horizontal ellipsis U+2026 vs three dots
+ *
+ * These are all pairs where the two forms carry the same *intent* — the
+ * user is trying to type the character they see, and shouldn't be
+ * penalised because their input method emitted a different codepoint.
+ *
+ * The rule: normalize BOTH the user's keystroke and the expected
+ * character through the same table, then compare. Display is untouched.
+ *
+ * Character classes we do NOT normalize:
+ *   - Fullwidth Latin letters (U+FF21..U+FF3A / U+FF41..U+FF5A) — if a
+ *     card contains "Ａ" that's the author's explicit intent; ignoreCase
+ *     handles the ASCII case elsewhere.
+ *   - Fullwidth digits (U+FF10..U+FF19) — same reasoning.
+ *   - Curly braces / angles that carry semantic meaning in Chinese text.
+ */
+/** Char → equivalent char used for typing comparison. */
+const EQUIV = {
+    // ─── Spaces ───────────────────────────────────────────
+    0x3000: 0x0020, // U+3000 IDEOGRAPHIC SPACE
+    0x00A0: 0x0020, // U+00A0 NO-BREAK SPACE
+    // ─── Quotes ────────────────────────────────────────────
+    0x201C: 0x0022, // U+201C LEFT DOUBLE QUOTATION MARK
+    0x201D: 0x0022, // U+201D RIGHT DOUBLE QUOTATION MARK
+    0x201E: 0x0022, // U+201E DOUBLE LOW-9
+    0x2033: 0x0022, // U+2033 DOUBLE PRIME
+    0x2018: 0x0027, // U+2018 LEFT SINGLE QUOTATION MARK
+    0x2019: 0x0027, // U+2019 RIGHT SINGLE QUOTATION MARK
+    0x2032: 0x0027, // U+2032 PRIME
+    // ─── Dashes ────────────────────────────────────────────
+    0x2013: 0x002D, // U+2013 EN DASH
+    0x2014: 0x002D, // U+2014 EM DASH  (kept: single "-" match; author's "——"
+    //                             double-hyphen convention still works because
+    //                             it's already two ASCII "-" in storage)
+    0x2212: 0x002D, // U+2212 MINUS SIGN
+    0xFF0D: 0x002D, // U+FF0D FULLWIDTH HYPHEN-MINUS
+    // ─── CJK punctuation → ASCII ───────────────────────────
+    0xFF0C: 0x002C, // ，
+    0x3002: 0x002E, // 。 (map to `.`; author writing 句号 the user types `.`)
+    0xFF1F: 0x003F, // ？
+    0xFF01: 0x0021, // ！
+    0xFF1A: 0x003A, // ：
+    0xFF1B: 0x003B, // ；
+    0xFF08: 0x0028, // （
+    0xFF09: 0x0029, // ）
+    0x3010: 0x005B, // 【
+    0x3011: 0x005D, // 】
+    0x300A: 0x003C, // 《
+    0x300B: 0x003E, // 》
+    0xFF0F: 0x002F, // ／
+    0xFF0B: 0x002B, // ＋
+    0xFF1D: 0x003D, // ＝
+    0xFF5E: 0x007E, // ～
+    0xFF03: 0x0023, // ＃
+    0xFF04: 0x0024, // ＄
+    0xFF05: 0x0025, // ％
+    0xFF06: 0x0026, // ＆
+    0xFF0A: 0x002A, // ＊
+    0xFF20: 0x0040, // ＠
+    0xFF3B: 0x005B, // ［
+    0xFF3D: 0x005D, // ］
+    0xFF5B: 0x007B, // ｛
+    0xFF5D: 0x007D, // ｝
+    // ─── Middle dot / interpunct ───────────────────────────
+    0x00B7: 0x00B7, // U+00B7 keep (rendered as ·)
+    0x2027: 0x00B7, // U+2027 HYPHENATION POINT
+    0x30FB: 0x00B7, // U+30FB KATAKANA MIDDLE DOT
+    // ─── Ellipsis ──────────────────────────────────────────
+    0x2026: 0x002E, // U+2026 HORIZONTAL ELLIPSIS -> "."  (matched char-by-char
+    //                                     if the card stores "..." literally)
+};
+/** Return the canonical form of `codePoint` for typing comparison. */
+export function normalizeForCompare(codePoint) {
+    return EQUIV[codePoint] ?? codePoint;
+}