bug: cancelling a published document like is not persisted

[PeterGuy326]

Problem

On a published document, cancelling a like reports success and reduces the displayed count optimistically, but does not persist the decrement. Reloading restores the higher count.

Reproduction

Use a disposable local database and a synthetic published document:

1. Record the initial count N.
2. Like the document and wait for the increase request to succeed; the count is N + 1.
3. Click again to cancel and wait for the decrease request to succeed; the interface shows N.
4. Reload the page: the count is N + 1.

This was observed during local appearance acceptance, with all writes confined to an isolated test environment.

Source evidence

• src/app/api/pub/thumb-up-decrease/[publishId]/route.ts:6 returns success unconditionally; the database decrement beginning at line 15 is commented out.
• src/components/thumb-up-button.tsx:28 reduces the displayed count optimistically and calls the decrease endpoint at line 31.

These API lines already exist on main at 0d88ec4b94f6fdb669fad1db858058650eb8c76e. The appearance change tracked in #57 does not modify this endpoint.

Expected result

A successful cancellation should be reflected after reload, with an appropriate nonnegative count and consistent error handling. A future fix should include a persisted readback regression and explicitly define any identity or duplicate-cancellation behavior. No API change is included in the appearance PR.

[waterbro-8]

Implementation candidate: open PR #70 (fix/persist-published-like-cancel). The decrease route currently succeeds without writing. I am not opening a second PR.

[waterbro-8]

平台组分工补派（冯浩然 / waterbro-8）

• 跟进人：李圆阳（GitHub sun-970，仓库暂不能指派，以钉钉+本评论为准）
• 仓库：bytefolk/doc bug: cancelling a published document like is not persisted #59
• 动作：现在提 PR
• 说明：取消已发布文档 like 未持久化，直接修并提 PR
• 流程：fork 后提 PR，标题带 bug: cancelling a published document like is not persisted #59，然后 @waterbro-8 审核。我不能合自己的 CODEOWNER PR。

[waterbro-8]

平台组群成员派活（冯浩然 / waterbro-8）

• 跟进人：施卓凡（钉钉平台组；本仓库可能无法 GitHub 指派，以钉钉+本评论为准）
• 仓库：bytefolk/doc bug: cancelling a published document like is not persisted #59
• 动作：现在提 PR
• 说明：取消已发布文档 like 未持久化，直接提 PR
• 流程：有 GitHub 权限就直接开 PR；没有就 fork 后提。标题带 bug: cancelling a published document like is not persisted #59。提完 @waterbro-8 审核。
• 李彬彬这批不用做。

[sun-970]

#59 on current main — already persisted, not opening a duplicate PR

Checked after syncing sun-970/doc to upstream main (gh api -X POST repos/sun-970/doc/merge-upstream -f branch=main). Head: b103638.

The original no-op is gone. src/app/api/pub/thumb-up-decrease/[publishId]/route.ts no longer returns success with the decrement commented out. Current decrease:

• 404 when the publication is missing (genErrorData('not found'))
• deleteMany of this viewer’s PubDocLike row and updateMany decrement share one $transaction
• decrement is clamped with thumbUpCount: { gt: 0 } (nonnegative)
• if this viewer holds no like, decrement is skipped and the current count is returned (duplicate cancel is a no-op, still errno: 0)
• response is a readback { liked: false, count } from findUnique after the write

That path landed with merged #79 (c3e89445, like identity / #72), which CODEOWNER noted is a strict superset of closed unmerged #70.

Identity is the #79 model, not #70 localStorage: PubDocLike unique on (viewerId, pubDocId); signed-in user:{id}, anonymous viewer_id httpOnly cookie. Increase is create-then-increment; P2002 does not increment. The published button loads GET state on mount and applies the PATCH readback (or rolls back on failure).

Regression coverage already on main in src/__tests__/api/pub-like-viewer.test.ts:

• decrease deletes + decrements in one transaction and returns the decremented count
• decrease does not decrement when this viewer has no like
• increase create-then-increment; P2002 is a no-op without increment
• GET reports whether this viewer currently holds a like

I am not opening another implementation PR. Re-applying #70’s updateMany-only decrease / localStorage identity would regress #79. No remaining production no-op on this endpoint at b103638.

@waterbro-8 — please treat this as the #59 follow-up: evidence only, no second persist PR.

[waterbro-8]

验收结论：已由 PR #79 的严格超集完成并合入 main（merge commit c3e8944）。最终 head da64ff2 已获两份正式 Approve，hosted test 和 docker-build 全绿；取消点赞按 viewer deleteMany，事务内仅在存在记录时非负递减，无记录幂等 no-op，并有 route 回归测试。该 issue 已满足，可关闭。