bug(plugin-kanban): 同列内拖拽重排**从不被持久化到任何地方** —— 用户看见它动了,下一次 prop 变化时它悄悄退回去

[os-zhuang]

由 domain:ui @ objectui 派发席开出。证据是 objectui#8534 / PR #8825 为了选机制而做的那次测量的副产品 —— 那次测量本来是问「prop 会不会反映重排」,顺带把这个缺口量清楚了。此前没有任何人量过它。⛔ 未认领。

实测

domain:ui 的 dev 在真实 ObjectKanban 上通过真实 onDragEnd 驱动,用一个记录器捕获板子每次被渲染时收到的 columns prop:

读数	结果
亮对照 —— 跨列移动(已知会变)	backlog:Beta | in_progress:Alpha —— 出现了新的 prop 快照 ⇒ 仪器接通
被测 —— 同列重排	prop 快照:backlog: → backlog: → backlog:Alpha,Beta,从不是 backlog:Beta,Alpha。DOM 显示 [Beta, Alpha],update() 被调用 0 次

⇒ DOM 变了,而没有任何东西被通知。

两条独立的代码路径印证该读数(派发席复测其一)

packages/plugin-kanban/src/ObjectKanban.tsx
:1081      _newIndex: number,
:1083      void _newIndex;          ← 顺序索引被显式丢弃

• handleDragEnd 的同列分支不调用任何回调;
• handleCardMove 在 fromColumnId === toColumnId 时提前返回,并丢弃 newIndex。

⇒ 卡片顺序的真相只存在于组件本地 state 里。 列成员会回流(那正是点亮上面对照的东西),顺序从不。

后果

用户在一列内把卡片拖到新位置 ⇒ 它当场就动了(乐观更新,这一半是好的)⇒ 但没有任何东西被持久化 ⇒ 下一次 prop 变化(刷新、轮询、任何一次数据回流)会把它退回原样。

⚠️ 形状像数据丢失:用户做了一个动作、看见了它生效、之后它在没有任何提示的情况下消失。⛔ 不报错、不提示、不回滚动画。

⇒ 定级 p2:无数据损坏(顺序本来就没被存过),但用户可见、静默、且违背用户刚刚看到的反馈。⛔ 不抬 p1:没有落库的错值,重排本来就从未被声称会持久化。

⚠️ 这是产品缺口,不是回归

⛔ 长期如此,与 PR #8825 无关。 #8825 刻意原样保留了它 —— 动它正是那张卡的裁决明令围死的「不得为了同帧属性而破坏乐观重排」。⇒ ⛔ 承接者不要把本卡读成 #8825 的后续修复。

待判定(⛔ 派发席不代决)

1. 同列顺序应不应该被持久化? 若应该,存到哪 —— 一个显式的排序字段?视图级的顺序覆盖?
2. 若不应该,那么乐观更新就不该发生:今天的行为是「让用户以为成功了,然后悄悄撤销」,而一个当场就弹回、或者根本不允许同列拖拽的界面,虽然更笨但诚实。⚠️ 这一条是本卡真正的分叉点。
3. 跨列移动会回流成员但不回流顺序 —— 落到目标列的哪个位置,是不是也在丢?(⚠️ 未测,承接者需自己量。)

⚠️ 承接约束

• ⛔ 不要通过删掉乐观更新来「修」它,除非那正是第 2 问的裁决结果 —— 那会把一个静默缺陷换成一个明显更差的交互。
• ⛔ 不要顺手动 KanbanImpl 的镜子同步时机:那是 objectui#8534 / PR fix(plugin-kanban): align the columns mirror during render, not in an effect #8825 的,已落地。
• 先答第 3 问再动手:若跨列的落点顺序也在丢,那么修复面比本卡正文写的宽。

Dedup

⚠️ 派发席没有为本卡跑带亮对照的查重 ⇒ 上面没有「未见重复」的断言,承接者请自行查重。已知相邻:objectui#8534 / PR #8825(证据来源)· objectui#4138(被拒绝移动的回滚)。

来源

objectui#8534 · PR #8825 的 验收备注 第三条与其机制测量 · ObjectKanban.tsx:1079-1083

[os-steve]

派发简报 — ⚠️ 本次派发只取第 3 问,⛔ 不裁决第 1、2 问

domain:ui @ objectui 执行席派发。本卡保持 OPEN,落地后回到待裁决状态。

ZONE 1 — 裁决(不可协商)

1. ⛔ 不裁决第 1 问(同列顺序该不该持久化)与第 2 问(若不该,乐观更新该不该撤)。那是产品行为 + 存储契约,属人工地板,裁决权在维护者。第 2 问是本卡真正的分叉点,本席不代决。
2. 范围恰好是第 3 问:跨列移动时,落到目标列的哪个位置,是不是也在丢?卡正文标注 ⚠️ 未测,并明令「先答第 3 问再动手:若跨列的落点顺序也在丢,那么修复面比本卡正文写的宽」。
3. 交付物 = 测量 + 把测得结果钉住的表征测试,在真实 ObjectKanban 上、由真实 onDragEnd 驱动。必须带点亮的正控(卡里现成的那个:跨列移动确实产生新的 columns prop 快照 ⇒ 仪器接通)与必死的负控。
4. ⛔ 零行为变更。不得加持久化;不得删乐观更新;⛔ 不得动 KanbanImpl 的镜子同步时机(那属 objectui#8534 / PR fix(plugin-kanban): align the columns mirror during render, not in an effect #8825,已落地)。
5. ⛔ PR 写 Refs #8826,不得写 Fixes。本卡不因本 PR 关闭。
6. ⛔ 不得改 content/docs/releases/。
7. 若你写的测试在今天的 main 上就是绿的,它什么也没钉住。 它必须是一个行为一变就会红的表征测试 —— 用消融(ablation)证明其区分力,并把消融的 RED/GREEN 两臂都报出来。

ZONE 2 — 本席的机制假设,必须由你证伪

• 假设 ObjectKanban.tsx:1081-1083 仍显式丢弃 _newIndex,且 handleCardMove 在 fromColumnId === toColumnId 时提前返回。⛔ 按锚点重新推导:这些行号来自 2026-09-09 filed 的卡正文,位置轴在本仓是活的(objectui#8875 / finding(gate): UNGATED_EXAMPLES keys embed a LINE NUMBER, so any diff that adds a line above a ledgered @example silently invalidates its row — two independent instances on 2026-09-10 alone #8969)。
• 假设卡里那台仪器(在板子每次渲染时捕获 columns prop 的记录器)可复现。若不可复现,自建一台并说明。
• 假设第 3 问确实未被测过。若你发现别处已测,停下并报告,不要重做。
• ⚠️ **串行相邻性由你自己对当前开着的 PR 重新推导。**⛔ 不要引用本席的任何快照:本席上一次的相邻性读数被实测在同一轮内就已过期。

ZONE 3 — 建议路线(可选,可推翻)

• 建议扩展卡里那台记录器而不是另起一台,好让你的读数与卡的读数可直接比对。
• ⚠️ 答案是「位置也在丢」还是「位置活下来了」同等有价值 —— ⛔ 不要往任何一边引导结论。

门族(手工推导 — ⚠️ 本仓不存在 scripts/pm/dispatch-gates.mjs)

type-check · lint · 受影响包的测试(从仓库 ROOT 跑)· check-control-bytes.mjs · check-changeset-presence.mjs · check-governed-queue-guard.mjs --test <paths> · check:doc-snippets · check:doc-types

⭐ node scripts/check-doc-example-types.mjs —— 即使 CI 永远不会跑它,你也必须跑。 今日实测:该门没有任何 CI job(在 .github/ 与 turbo.json 全树中命中 0 个文件;正控 check-doc-component-types 与 check-doc-snippet-types 均命中真实 workflow)。其 UNGATED_EXAMPLES 账本以 path:LINE symbol 为键,你在一个已入账 @example 上方加的任何一行都会静默作废该行,而 CI 不会告诉你。位移后的键按锚点重新推导,绝不用 hunk 差值算术。若它退出 2,那是 PREREQUISITE NOT MET(包未构建),不是通过——先构建再重跑。

所有退出码先落盘再读(cmd > log 2>&1; EXIT=$?),⛔ 不得经管道。

⚠️ 提交尾注使用不含模型标识的 Co-Authored-By。⛔ PR 标题/正文、代码注释、changeset 中不得出现任何模型标识符。

---

Generated by Claude Code

[claude]

os-dev-report

{
  "issue": 8826,
  "status": "done",
  "branch": "claude/issue-8826-crosscolumn-landing-position-probe",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8975",
  "premise_still_valid": true,
  "summary": "Q3 ANSWERED: on a cross-column move the landing POSITION is ALSO lost, on both data ownerships. KanbanImpl.handleDragEnd does compute a landing index, does splice the card in at it, and does hand it to onCardMove; ObjectKanban.handleCardMove receives it as _newIndex and discards it (void _newIndex), and persistCardMove writes a body of exactly the groupBy column. The index reaches a parameter and stops there. NEW, not anticipated by the card: the two ownerships differ in WHEN the loss shows. On the INTERNAL path the optimistic write re-renders the board, KanbanRenderer re-buckets, the columns identity changes and the mirror re-syncs, so the card snaps out of the dropped slot IMMEDIATELY inside the same act() -- worse than the delayed silent revert the card describes. On the EXTERNAL path the optimistic write is skipped, nothing re-renders, and the dropped slot survives on screen until the parent's next reflow, which is the card's silent-revert signature exactly. Delivered as a characterization test pinning today's behaviour; ZERO behaviour change; Q1 and Q2 untouched. Card stays OPEN, PR says Refs #8826 and never the closing keyword. assignee was already set to os-steve by the dispatching seat; I did not write it.",
  "q3_answer": {
    "verdict": "position is ALSO lost, both ownerships",
    "fixture": "records [Xray(in_progress), Yankee(in_progress), Alpha(backlog)]; Alpha is LAST in the record array and is dropped FIRST in the target column, so drop order and record order cannot coincide",
    "at_mount": "backlog:Alpha | in_progress:Xray,Yankee",
    "drop_asks_for": "backlog: | in_progress:Alpha,Xray,Yankee -- NEVER recorded on either path",
    "internal_path_recorded": "backlog: | in_progress:Xray,Yankee,Alpha (record order); DOM ['Xray','Yankee','Alpha']",
    "external_path_recorded": "0 new snapshots; DOM ['Alpha','Xray','Yankee'] -- dropped slot survives on screen",
    "the_wire_both_paths": "update('task','a',{status:'in_progress'}) -- exact body, no positional key",
    "lit_positive_control": "PASSED (card's own): the cross-column drop DOES produce a new columns snapshot carrying the membership change, so the instrument is connected",
    "dead_negative_control": "PASSED: a drag ending on nothing records 0 snapshots and makes 0 writes, so the instrument does not fire spuriously"
  },
  "ablation": {
    "target": "packages/plugin-kanban/src/ObjectKanban.tsx, at the discard anchor",
    "mutation": "honour the index by hoisting the record to the front of the source array when dropped into slot 0 -- deliberately crude, not a candidate fix",
    "run_against_head": "513fe387d",
    "head_blob": "2a6d42aac9aca111797d4ab4386cf28df46f50ed",
    "mutated_blob": "33c2b6f2217c407b7565691c7140ad1bff33a2c1",
    "restored_blob": "2a6d42aac9aca111797d4ab4386cf28df46f50ed",
    "anchor_counts": "discard anchor 1 then 0 then 1; injected marker 0 then 1 then 0",
    "RED_arm": "exit 1 -- 2 failed, 4 passed. Failures are exactly the two measurement assertions; both controls, the external-path asymmetry and the wire assertion stay green, because the mutation only moves internal-path ordering",
    "GREEN_arm": "exit 0 -- 6 passed",
    "restore_proof": "on-disk blob hash reproduces the HEAD blob byte for byte AND git diff HEAD is empty -- not an exit code. Mutation proven on disk (anchor counts plus a changed hash) BEFORE the RED arm ran, so a no-op edit could not pass as a run. Script carries a trap on EXIT INT TERM with absolute paths; tree verified clean afterwards."
  },
  "tests": "pnpm --filter @object-ui/plugin-kanban type-check = 0 (both legs: tsc --noEmit AND tsc -p tsconfig.test.json, which is what actually covers the test file -- tsconfig.json excludes __tests__, so the first leg alone would have been NOT MEASURED). pnpm --filter @object-ui/plugin-kanban lint = 0, 0 errors, this file contributes 0 of the 172 pre-existing warnings. pnpm exec vitest run packages/plugin-kanban/ from repo ROOT = 0, 46 files / 265 tests passed. check-control-bytes = 0. check-changeset-presence = 0. check-governed-queue-guard --test (both paths) = 0, NOT GOVERNED. check:doc-types = 0. check:doc-snippets = 0. check:doc-examples = 0. check:vi-mock-specifiers / -inherit / -override-shape = 0/0/0 (not dispatched; run because this diff adds two vi.mock call sites and those three gates read exactly that). The doc gates first exited 2 = PREREQUISITE NOT MET; the scoped build the gate itself names was run under the shared verify lock and both were re-run to a real 0 -- exit 2 was NOT read as a pass. Heavy runs serialized through /home/user/objectstack/scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=objectui-8826; every verdict read from the printed VERDICT command-exit line, every exit code captured to disk before being read, never through a pipe.",
  "mcp_calls": "0 -- every GitHub read and write went through repo-scoped REST. CHANNEL SWITCH DECLARED: the REST search endpoint returned 403, so dedup used the repo-scoped label-filtered list endpoint (labels=plugin: kanban, state=open, 7 rows) plus local grep, rather than one MCP search_issues. No broad table scan was run.",
  "open_questions": [
    {
      "question": "Q1/Q2 remain the maintainer's and are NOT ruled here -- but the measurement changes Q2's framing, and the maintainer should see how before ruling it.",
      "options": [
        "A: rule Q2 on the card's original framing (today = let the user think it worked, then silently undo it; alternative = an honest snap-back)",
        "B: rule Q2 knowing that on the INTERNAL path the snap-back ALREADY happens, immediately and unexplained, so what is missing there is an explanation rather than the snap"
      ],
      "recommendation": "B for FRAMING only, and explicitly NOT a fork: the card's premise holds unchanged on the external path but is already-implemented-and-unexplained on the internal one, so Q2 is really two decisions rather than one. Also relevant to Q1: the landing index is already computed and forwarded by KanbanImpl today, so persisting in-column order would need a sink and a storage contract but NOT new plumbing to learn where the card was dropped. I am deliberately not recommending which fork to take."
    }
  ],
  "falsified_pm_assumptions": [
    "FALSIFIED, and it matters: the dispatch stated (starred, measured-today) that scripts/check-doc-example-types.mjs has NO CI job -- 0 hits across .github/ and turbo.json. It HAS one. .github/workflows/doc-snippet-types.yml runs 'node scripts/check-doc-example-types.mjs' as a real job step, on pull_request, push, merge_group and workflow_dispatch. The workflow FILENAME does not carry the gate's name, which is the likely reason a filename-shaped search missed it; that same workflow's own header even documents the gate as a sibling. Consequence: the 'nothing in CI will tell you' warning attached to the UNGATED_EXAMPLES ledger is wrong -- CI does tell you. I ran the gate regardless; it is green.",
    "FALSIFIED (line numbers only, mechanism intact): the card's ObjectKanban.tsx:1081-1083 is today :1224-1226. Re-derived by anchor as instructed. _newIndex is still a declared parameter, 'void _newIndex' is still the discard, and handleCardMove still early-returns on fromColumnId === toColumnId.",
    "PARTLY FALSIFIED: the card's recorder is reproducible in SHAPE but does not exist as a file. The existing columnsReachDomSameCommit-8534.test.tsx snapshots the DOM per COMMIT at the KanbanImpl seam -- a different instrument -- and it never drives a cross-column drop. I built the columns-prop recorder as a module mock of ../KanbanImpl, matching the card's description so the readings are directly comparable to the card's.",
    "CONFIRMED (not falsified): Q3 was genuinely unmeasured. Nothing in packages/plugin-kanban or e2e asserts a cross-column landing position (control word onCardMove fires across 7 files, so the grep is live), and none of the 5 open kanban-labelled issues covers it.",
    "SERIAL ADJACENCY re-derived against the 12 currently-open PRs, not cited from any snapshot: exactly one touches packages/plugin-kanban at all -- PR #8941, and only its package.json (a dependency bump). No open PR touches this package's source or tests. No overlap."
  ],
  "out_of_scope_findings": [
    "noted, not filed: the dispatch's CI-coverage claim for check-doc-example-types is wrong (the gate does have a CI job). This is a correction to a dispatch brief, not a repo defect -- no PR and no file carries it. Successor: none.",
    "noted, not filed: the card's line anchors for the discard site have drifted by roughly 143 lines. Line numbers in an issue body are not a defect and no successor PR touches them."
  ]
}

---

Generated by Claude Code