Repository navigation
[Decision] Should the eager-closure budget be tightened below today's payload — absolute ceiling, true ratchet, or per-chunk budgets? #5468
Description
Follow-up owed by the seat that accepted #5324 / PR #5463…#5466. ⛔ Filed as the fork only — not a grading and not a recommendation adopted. Dedupe done before filing: repo-scoped listing of all open issues, three pages, regex over titles and bodies for ratchet.*budget / per-chunk budget / vendor-objectstack / tighten.*ceiling / performance budget — the only hits are #5324 itself, #5359 (a specific regression, pm:blocked) and #5382 (unrelated). No twin.
Context
#5324 landed the eager-closure gauge (PR #5466). Per its dispatch, the dev shipped a truthful current-state ceiling and did not lower it unilaterally: 3,960,000 gzipped bytes = today's 3,881,609 plus 2.02% headroom, deliberately narrower than the 89 KiB #5266 regression it exists to catch (a repeat fails by 12.4 KB). Both constraints are expect()s in scripts/__tests__/check-eager-closure-budget.test.ts, so widening the ceiling past the regression size fails a test.
The gap that ceiling records is the point of this card. 3.79 MB gzipped before first render is a bad payload. Where the mass actually is, measured:
| eager chunk | gzipped | share of closure |
|---|---|---|
vendor-objectstack |
1,493 KB | 38% |
framework |
484 KB | 12% |
ui-components |
380 KB | 10% |
The fork
- A — leave the absolute ceiling, revisit after the closure is actually cleaved (e.g. after a finding(fields): every lazy widget import in
packages/fields/src/index.tsxis defeated by a static import of the same module — 13+INEFFECTIVE_DYNAMIC_IMPORTwarnings per console build #5325 route, or finding(app-shell/console): the DraftChangesPanel/metaspelling fold puts@objectstack/spec/sharedon the console's EAGER graph — +60.1 KB gzipped that no budget weighs #5359). Simplest. The gate already catches a apps/console: thevendor-objectstackadvancedChunks group folds the lazily-imported@objectstack/lintinto an eagerly-loaded chunk — 89 KiB gzipped on every page load #5266-class regression today. - B — convert to a true ratchet against
main(the direction triage originally converged on). Sharper.⚠️ Costs amainbuild per PR in CI, and carries a failure mode worth stating plainly: the merge queue rebuilds every PR on currentmain, so a strict ratchet turns any two concurrent size-adding PRs into a queue ejection neither author can see coming. - C — add per-chunk budgets on top of the aggregate, so
vendor-objectstackcannot grow unnoticed inside an aggregate that still has headroom.
Dev's recommendation: A now, C next, B only if C proves insufficient — the shipped headroom (78,391 bytes) is already narrower than the regression it targets, so B's marginal value over A today is small relative to its queue cost, while C is the cheapest real tightening and points at where 38% of the mass sits.
Why this is a decision and not a seat's call
It sets how much friction CI imposes on every future PR, and B's failure mode lands on authors who did nothing wrong. Gate-strength policy is the maintainer's. ⛔ No seat should pick this.
Triage may re-grade the label; filing it in the decision inbox because the fork is already fully measured and there is nothing left to investigate before answering it.
Generated by Claude Code
Activity
os-zhuang commentedon Aug 21, 2026
分诊决策箱勤务:本卡入箱时缺标准四棱卡面块,补齐如下(卡内测量与 dev 建议为输入;维护者回一个字母即可)。
<!-- os-decision-facets -->
一句话问题
首屏渲染前要下载 3.79MB(gzip)——门禁已钉住"不再变得更糟",但要不要收得更紧、用哪种闸,决定了以后每个 PR 在 CI 里付多少摩擦。
选项 × 真实代价
vendor-objectstack(38% 质量所在)等单块设上限四轴(业务立场)
packages/fields/src/index.tsxis defeated by a static import of the same module — 13+INEFFECTIVE_DYNAMIC_IMPORTwarnings per console build #5325/finding(app-shell/console): the DraftChangesPanel/metaspelling fold puts@objectstack/spec/sharedon the console's EAGER graph — +60.1 KB gzipped that no budget weighs #5359)互补;B 的队列弹出税随并行度线性放大,伤多 agent 工作流。推荐 A 现在 + C 下一步(即 dev 建议);回退 B 仅当 C 实测不够。
置信缺口:本分析看不见
vendor-objectstack块内部的可拆性(未测量其构成),也看不见 #5325/#5359 路由切分的落地时间表 —— 若那两张卡近期落地,C 的分块线应在其后设定而非之前。裁后执行(你不用管)
回一个字母。A → 本卡关 not planned(现状即裁定),C 候补开工卡入队;C → 立
pm:queue工卡(分块预算,S 级);B → 立卡并注明队列弹出税已被知情接受。Generated by Claude Code
os-zhuang commentedon Aug 21, 2026
Maintainer ruling (2026-08-21, live PM chat — recorded by the triage seat)
Verbatim (untranslated): 「其他接受」 — accepting the triage recommendation for this card: A now + C next; B rejected (the merge-queue double-ejection failure mode and per-PR main-build cost land on authors who did nothing wrong, for marginal value over the shipped ceiling).
Execution per the card's pre-written clause: the absolute ceiling stands as shipped (nothing to change = A); the C work card is filed as #5490 (
pm:queue, S-size: per-chunk budgets,vendor-objectstackfirst). This decision card closes as completed — the ruling is recorded here and on #5490.Generated by Claude Code
os-zhuang commentedon Aug 21, 2026
Closing as completed — the decision this card exists for is made and executed. Provenance: maintainer ruled A + C in the 2026-08-21 chat adjudication session (recorded on this card and in the triage seat post objectstack-ai/objectstack#6015 appendix, 03:54Z): keep the absolute ceiling now (A), and per-chunk budgets as the next tightening (C), with B only if C proves insufficient. The C work card exists as #5490. The close itself was owed from that session (blocked then by rate limiting) and is executed now by the triage seat, session
session_01YCm6utUKLsQ7B6VStm3XUM.Generated by Claude Code
chore(deps): pin @objectstack/* to 17.1.0 and adapt the surfaces it m…
Pin @objectstack/* to 17.1.0 (objectstack-ai#5529)