fix(wheel): re-render mirror after pin order commit

.loom/design/task-brief-wheel-drag-refresh.md

@@ -0,0 +1,118 @@
+# Devin Brief: 拖拽提交后 UI 不刷新（P0，Linux 真机实测抓到）
+
+## 项目
+`/home/haa/sites/promptkey`，当前 `master` HEAD `baa37fe`（2.0.4）。新建分支 `fix/wheel-drag-commit-no-refresh`。
+
+## 背景
+
+我在 **Linux 上用 Playwright 真实驱动前端**（mock `__TAURI__` IPC），跑完整个冒烟套件。16/17 通过，**抓到一个 P0 真 bug** —— 这正是你在 Windows 上装 2.0.4 后拖排序会遇到的"拖了没反应"。
+
+## 实测证据（完整链路，逐环都有）
+
+```
+MID-DRAG:   {target row gets `.dragging`, insertion slot appears}          ✅ 交互识别正常
+AFTER DROP: set_pin_order 收到 { ids: [2, 1] }                            ✅ 顺序计算完全正确
+FINAL DOM:  #wheelMirror 顺序仍是 ["1","2"]                               ❌ 提交后没有重新渲染
+```
+
+**即：拖拽逻辑对、提交参数对，唯一断点是"提交成功后 UI 没刷新"。**
+
+这解释了为什么 2.0.4 的 e2e 用例声称"断言 DOM reorder + `set_pin_order` 被调用"却全绿 —— 见下方"测试盲区"。
+
+## 疑似位置
+
+`src/js/views/prompts.js` 的拖拽提交函数（PR #14 的 pointer 实现，约 `commitPinOrder` / drop handler 一带）：
+
+```js
+try {
+  await ipc('set_pin_order', { ids: pins });
+  await loadPrompts(); rebuildIndex(); renderPrompts();   // ← 这条链看起来有，但实测没生效
+} catch { /* toasted */ }
+```
+
+请先**逐行读实际代码**，查清为什么刷新没发生。至少这几种可能，请逐一证伪/证实：
+
+1. **刷新被调用了但抛错被 catch 吞掉**（`loadPrompts` 内部依赖某个 mock 里没有的东西，或某个调用 throw）
+2. **`renderPrompts()` 只重渲染 `#grid`，根本不碰 `#wheelMirror`** —— 那顺序就永远不更新。**这个可能性最大，优先查**
+3. `wheelPrompts()` 的排序在 `wheelSort !== 'manual'` 时不认 `inject_order`，mock 切了 manual 但真实路径有问题
+4. 提交后 `state.prompts` 被替换但 mirror 的渲染走的是别的缓存
+5. drop 之后某个 return 提前退出，刷新语句根本没执行到
+
+## 修复要求
+
+1. **提交成功后 `#wheelMirror` 必须立即反映新顺序**（不需要刷新页面、不需要切页面再切回）
+2. **顺序要真的持久化** —— 重载页面后顺序不变（我的 mock 里已实现"真后端"语义，能验证这点）
+3. **失败时回滚视觉状态**：IPC 失败要恢复到拖拽前的顺序，不能停在中间态
+4. 我实测发现：**拖拽必须按在 `.drag` 句柄上才触发**（按行 body 无效）—— 这是 PR #14 的合理设计，**不要改掉**，但请在交付里确认这是有意为之，并考虑要不要给用户更明显的视觉提示（句柄 hover/光标变化），你论证
+
+## 附带：修测试盲区（必须）
+
+2.0.4 的 e2e 号称断言"DOM reorder"，却漏掉了这个 bug。原因是：
+
+- e2e 的 mock `set_pin_order` 返回 `null`，**不真的持久化顺序**
+- 且断言只看"调用发生"，**没断言松手后 DOM 顺序真的变了**
+
+请修 `tests/e2e/wheel_sort_drag_e2e.py`：
+
+1. mock 的 `set_pin_order` 必须**真的写回顺序**（真后端语义）
+2. 断言必须包含：**松手后 `#wheelMirror` 的 DOM 顺序确实变化**
+3. 再断言一次 `set_pin_order` 收到的 `ids` 顺序 = DOM 新顺序
+4. 加一个失败回滚用例：mock `set_pin_order` reject → UI 恢复原顺序
+5. 加一个"刷新后顺序仍在"用例（重新 load 页面）
+
+---
+
+# 硬约束
+
+- **我在 Linux 上给你复现步骤**（见下），你可以自己复现验证
+- 无 cargo，本机不编译，CI 全绿即可
+- 不得回归 Windows 行为
+- 一个 commit `fix(wheel): re-render mirror after pin order commit`，push，PR `--base master`，**不合并**
+- 遵守 LOOM 流程
+
+## 我在 Linux 上的复现方式（你可以照做）
+
+```bash
+cd /home/haa/sites/promptkey/src && python3 -m http.server 8913   # 后台起静态服务
+# 然后用 Playwright + mock __TAURI__（mock 必须实现 set_pin_order 真写回）
+# 我的冒烟脚本在 /tmp/pk_linux_smoke.py，可直接参考或复用
+```
+
+关键：**mock 一定要让 `set_pin_order` 真的更新数据**，否则这个 bug 测不出来（这正是原 e2e 漏掉它的原因）。
+
+---
+
+## 最终交付
+
+```
+## 根因
+（为什么提交后不刷新，逐行证据）
+
+## 修复
+（改了什么；为什么这样改）
+
+## 测试盲区修正
+（e2e 怎么改的；现在能抓住这个 bug 吗——**请真的跑一次证明**）
+
+## 复现验证
+（你在 Linux 上跑的证据：拖拽前/中/后 DOM + set_pin_order 参数 + 刷新后顺序）
+
+## 顺带观察
+（拖拽句柄的视觉提示、以及你发现的任何其他问题）
+```
+
+```json
+COMPLETION_NOTIFY
+source: assistant
+task: promptkey-wheel-drag-refresh
+deliverables:
+- root cause of "commits but UI never re-renders" proven with code lines
+- mirror re-renders immediately after successful commit and survives reload
+- failure rolls back visual order
+- e2e mock now persists order and asserts post-drop DOM order + received ids + reload persistence + rollback
+- e2e actually re-run to prove it now catches the bug (or would have)
+- CI green
+- branch fix/wheel-drag-commit-no-refresh + PR against master (not merged)
+status: success
+errors: none
+```

.loom/state.json

@@ -4,7 +4,7 @@
     "name": "promptkey",
     "status": "complete",
     "created_at": "2026-10-06T08:16:56.246Z",
-    "updated_at": "2026-10-07T06:08:09.960Z"
+    "updated_at": "2026-10-07T07:31:11.309Z"
   },
   "understanding": {
     "confirmed": [

.loom/tasks.json

@@ -1034,6 +1034,109 @@
       "created_at": "2026-10-07T05:55:04.721Z",
       "updated_at": "2026-10-07T06:08:09.959Z",
       "completed_at": "2026-10-07T06:08:09.959Z"
+    },
+    {
+      "id": "TASK-011",
+      "title": "Fix wheel-mirror stale order after pin-order commit (pinOrder parse + e2e blind spots)",
+      "outcome": "After a pointer/keyboard reorder commit, #wheelMirror immediately renders the persisted order, the order survives reload, a rejected commit rolls back the visual order, and the e2e proves each of these against a durable mock backend.",
+      "acceptance": [
+        {
+          "criterion": "store.js pinOrder() honors any finite inject_order (including 0 / weight-0 pack data); only missing/unparseable values get the 9999 tail sentinel",
+          "verify_by": "read src/js/store.js pinOrder; e2e 'inject_order 0 ranks first' case",
+          "evidence": "src/js/store.js:93 `return Number.isFinite(n) ? n : 9999;` — 0-based and '0' weight writes now rank as explicit positions"
+        },
+        {
+          "criterion": "e2e mock set_pin_order is durable (localStorage-backed) and writes 1-based inject_order strings matching Rust (i+1).to_string()",
+          "verify_by": "read tests/e2e/wheel_sort_drag_e2e.py INIT block",
+          "evidence": "tests/e2e/wheel_sort_drag_e2e.py INIT: db in localStorage '__pk_e2e_prompts__', writes String(i+1), persist() after each mutation"
+        },
+        {
+          "criterion": "e2e asserts post-drop DOM order equals submitted set_pin_order ids for pointer AND keyboard commits",
+          "verify_by": "python3 tests/e2e/wheel_sort_drag_e2e.py — 'submitted ids == rendered DOM order' checks",
+          "evidence": "'submitted ids == rendered DOM order' + 'keyboard commit: submitted ids == rendered DOM order' checks PASS"
+        },
+        {
+          "criterion": "e2e covers reload persistence, reject-rollback (DOM + backend data unchanged, no mid-state), and inject_order=0 ordering",
+          "verify_by": "python3 tests/e2e/wheel_sort_drag_e2e.py ALL PASS (26 checks)",
+          "evidence": "'order persisted after reload', 'failed commit: DOM rolls back / no mid-state residue / backend data untouched', \"inject_order '0' ranks first, not last\" — all PASS (26/26)"
+        },
+        {
+          "criterion": "e2e provably catches the bug class: with the refresh chain removed it FAILs on named DOM-order checks; with pinOrder reverted it FAILs the '0' check",
+          "verify_by": "local runs recorded in PR body / session evidence",
+          "evidence": "Broken-prompts.js run: FAIL 'DOM re-rendered to committed order', 'submitted ids == rendered DOM order', 'sort toggle now shows manual', + 2 keyboard-path FAILs. Reverted pinOrder run: FAIL \"inject_order '0' ranks first\" (got ['1','3','4','2'])"
+        },
+        {
+          "criterion": "CI green; PR opened against master, not merged",
+          "verify_by": "gh pr checks on fix/wheel-drag-commit-no-refresh",
+          "evidence": "PR #16 https://github.com/Haaaiawd/PromptKey/pull/16 — cargo check/clippy/test pass, frontend static pass, playwright e2e pass; open, not merged"
+        }
+      ],
+      "done_when": [],
+      "boundaries": [
+        "No change to Rust backend or schema; inject_order stays TEXT 1-based writes",
+        "No change to drag arming rules: mouse may drag from anywhere on a row, touch/pen only from the .drag grip",
+        "Windows behavior must not regress"
+      ],
+      "depends_on": [],
+      "reads": [
+        ".loom/design/task-brief-wheel-drag-refresh.md",
+        "src/js/views/prompts.js",
+        "src/js/store.js",
+        "src/main.rs",
+        "tests/e2e/wheel_sort_drag_e2e.py"
+      ],
+      "touches": [
+        "src/js/store.js",
+        "tests/e2e/wheel_sort_drag_e2e.py"
+      ],
+      "implements": ".loom/design/task-brief-wheel-drag-refresh.md",
+      "design_exemption": "",
+      "capability_hooks": [],
+      "capability_exemption": "Bugfix within existing design-system/interaction scope — sort-parse defect + test strengthening; no new professional capability surface.",
+      "integrity_version": 1,
+      "covers": [],
+      "status": "done",
+      "progress": {
+        "completed": [],
+        "current": "complete",
+        "next": ""
+      },
+      "evidence": [
+        "Local: python3 tests/e2e/wheel_sort_drag_e2e.py 26/26 ALL PASS; hotkey_status 10/10; hotkey_recorder 22/22; no_tauri 13/13; user's Linux smoke script 16/17 -> 17/17 unchanged. Negative proofs: refresh chain removed -> 5 named FAILs; pinOrder reverted -> inject_order-0 FAIL. CI: PR #16 all checks green.",
+        {
+          "type": "acceptance_results",
+          "results": [
+            {
+              "criterion": "store.js pinOrder() honors any finite inject_order (including 0 / weight-0 pack data); only missing/unparseable values get the 9999 tail sentinel",
+              "evidence": "src/js/store.js:93 `return Number.isFinite(n) ? n : 9999;` — 0-based and '0' weight writes now rank as explicit positions"
+            },
+            {
+              "criterion": "e2e mock set_pin_order is durable (localStorage-backed) and writes 1-based inject_order strings matching Rust (i+1).to_string()",
+              "evidence": "tests/e2e/wheel_sort_drag_e2e.py INIT: db in localStorage '__pk_e2e_prompts__', writes String(i+1), persist() after each mutation"
+            },
+            {
+              "criterion": "e2e asserts post-drop DOM order equals submitted set_pin_order ids for pointer AND keyboard commits",
+              "evidence": "'submitted ids == rendered DOM order' + 'keyboard commit: submitted ids == rendered DOM order' checks PASS"
+            },
+            {
+              "criterion": "e2e covers reload persistence, reject-rollback (DOM + backend data unchanged, no mid-state), and inject_order=0 ordering",
+              "evidence": "'order persisted after reload', 'failed commit: DOM rolls back / no mid-state residue / backend data untouched', \"inject_order '0' ranks first, not last\" — all PASS (26/26)"
+            },
+            {
+              "criterion": "e2e provably catches the bug class: with the refresh chain removed it FAILs on named DOM-order checks; with pinOrder reverted it FAILs the '0' check",
+              "evidence": "Broken-prompts.js run: FAIL 'DOM re-rendered to committed order', 'submitted ids == rendered DOM order', 'sort toggle now shows manual', + 2 keyboard-path FAILs. Reverted pinOrder run: FAIL \"inject_order '0' ranks first\" (got ['1','3','4','2'])"
+            },
+            {
+              "criterion": "CI green; PR opened against master, not merged",
+              "evidence": "PR #16 https://github.com/Haaaiawd/PromptKey/pull/16 — cargo check/clippy/test pass, frontend static pass, playwright e2e pass; open, not merged"
+            }
+          ],
+          "at": "2026-10-07T07:31:11.294Z"
+        }
+      ],
+      "created_at": "2026-10-07T07:25:40.131Z",
+      "updated_at": "2026-10-07T07:31:11.305Z",
+      "completed_at": "2026-10-07T07:31:11.305Z"
     }
   ]
 }

src/js/store.js

@@ -90,7 +90,7 @@ export function wheelPrompts() {
 }
 function pinOrder(p) {
   const n = parseInt(p.inject_order, 10);
-  return Number.isFinite(n) && n > 0 ? n : 9999;
+  return Number.isFinite(n) ? n : 9999;
 }
 
 const VAR_RE = /\{\{\s*([a-zA-Z_][\w.-]*)\s*\}\}/g;