[2.0 rc.15 / next] GRAPH_GROWTH reports a keyed cache filling with new keys as a leak, and a cache has no way to say what it is
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 15/100
- Issue 类型
- 缺陷
- 描述清晰度
- 基本清楚
- 活跃度
- 活跃
- 技术栈
- typescript
调研方向
Start with walkGraph in packages/signals/src/core/attribution.ts, which counts nodes alive at both navigation start and settle as survived, and with createProjection next to createStore, which records owners in storeOwners but does not. Run the three-variant repro from the issue under --conditions=browser --conditions=development. Done means variant B goes quiet, variants A and C still report, and the three new graph-growth tests pass. The { cache: true } option is a design proposal, so confirm the API with maintainers before implementing.
由索引模型根据 Issue 内容生成。
描述
Summary
A cache that keeps entries by key reports GRAPH_GROWTH when visits fill it with new keys, and the message says a visit left a createRoot() behind with no dispose. This is the usual shape of a data layer's entity cache: one entry per item, each under its own root so it can be evicted, all under the cache's long-lived root. Since #4020 (fixing #4007), a series must also climb in what outlived each navigation, but a cache's entries are alive when the next navigation starts and survive it, exactly as a leaked root would. Nothing in the graph's shape tells a cache filling up from a leak, and there is no way for the cache to say what it is. OBSERVE.exclude is meant for an observer's own subtree and silences every diagnostic under it, not only this check.
Reproduction
solid-js 2.0.0-rc.15, no DOM. Each visit to /items/:page lists two items the cache hasn't seen; / sits between visits.
import {
OBSERVE,
createEffect,
createMemo,
createProjection,
createRenderEffect,
createRoot,
createSignal,
flush,
getOwner,
runWithOwner,
} from "solid-js";
import { attribution } from "solid-js/attribution";
const variant = process.argv[2] ?? "A";
attribution.enable({ log: false });
const findings: { route: string; grew: string[] }[] = [];
OBSERVE!.diagnostics.subscribe((event) => {
if (event.code === "GRAPH_GROWTH") findings.push(event.data as { route: string; grew: string[] });
});
// An entry per item (a signal and a projection of it), each under its own
// root so it can be evicted, all under the cache's root.
const cache = createRoot((dispose) => ({ owner: getOwner()!, dispose }), {
cache: variant !== "A",
});
const entries = new Map<number, { value: number }>();
const fill = (id: number) =>
runWithOwner(cache.owner, () =>
createRoot(() => {
const [value] = createSignal(id);
entries.set(
id,
createProjection<{ value: number }>(
(draft) => {
draft.value = value() * 2;
},
{ value: 0 },
),
);
}),
);
const leaked: (() => void)[] = [];
const app = createRoot((dispose) => {
const [location, setLocation] = createSignal("/");
createRenderEffect(
() => (location(), [...entries.values()].reduce((sum, entry) => sum + entry.value, 0)),
() => {},
);
return { setLocation, dispose };
});
flush();
const navigate = (name: string, to: string, before?: () => void) => {
OBSERVE!.attribution.withOrigin({ kind: "navigation", name, to }, () => {
before?.();
app.setLocation(to);
});
flush();
};
for (const page of [1, 2, 3]) {
navigate("/items/:page", `/items/${page}`, () => {
fill(page * 2 - 1);
fill(page * 2);
if (variant === "C")
runWithOwner(null, () =>
createRoot((dispose) => {
leaked.push(dispose);
const [n] = createSignal(0);
createEffect(n, () => {});
createMemo(() => n() + 1);
}),
);
});
navigate("/", "/");
}
console.log(variant, findings);
Run with node --conditions=browser --conditions=development repro.ts <A|B|C>.
| Variant | 2.0.0-rc.15 | 2.0.0-rc.15 with the fix below |
|---|---|---|
| A: the cache under a plain root | reports | reports |
B: the cache root created with { cache: true } |
reports | quiet |
| C: B, plus a detached root left behind on every visit | reports | reports |
Without the fix the option doesn't exist, so B is the same as A.
Cause
walkGraph (packages/signals/src/core/attribution.ts) counts every node alive at a navigation's start and still alive at its settle as survived, whoever owns it. Projections also record no owner: createStore puts its owner in storeOwners, and store property nodes are stamped from it, but createProjection doesn't, so a projection's property nodes have none.
Expected
A cache can declare itself, and what it keeps is left out of what survived, while a leak beside it still reports. Related: #4007, #4020.
Possible fix
Let a root say it is a cache: createRoot(fn, { cache: true }) (and createOwner). Observe tiers set a config bit for it, as they do for CONFIG_WIDE and CONFIG_SUPERSEDES, and prod ignores the option. walkGraph then leaves out of survived what such a root keeps: the owners and computations under it, and the signals created there (_owner). The whole-graph counts (graphSize()) are unchanged. createProjection records the owner in scope in storeOwners, as createStore does, so the stores and projections a cache keeps are left out too, and their attribution owner paths become complete as well.
Tested on the RC.15 release commit (ea4438b1; next at 9f4174b6 has no signals changes since): the signals suite passes with it (5,212 tests, 20 expected failures, 2 skipped; 5,209 without it). Three new graph-growth tests cover a declared cache staying quiet, the same cache undeclared still reporting, and a leak beside a declared cache still reporting. The two declared-cache tests fail without the change, and the first one also fails when only the projection's owner isn't recorded.
(submitted by Claude Opus 5.5 on behalf of rvlzzr)
- 主要语言
- TypeScript
- 星标
- 36.1k
- 派生
- 1.1k
- 平均合并
- 10 小时 41 分钟
- 30 天内合并 PR
- 363
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
solidjs/solid 的其他 Issue
-
难度 3/5 半天 新手友好度 55/100
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 30/100
维护者通常 1 天内回复
-
难度 3/5 半天 新手友好度 45/100
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 35/100
维护者通常 1 天内回复
-
signals: a landing reveals a held write its stale reader re-derives onto a new flight (U-5, D-9)未关闭
难度 5/5 一周以上 新手友好度 12/100
维护者通常 1 天内回复
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 74/100
QuantumNous/new-api#7839 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 70/100
维护者通常 1 天内回复
-
bug p2 triaged
难度 2/5 1-3 小时 新手友好度 68/100
维护者通常 1 天内回复
-
test: make the #3408 Claude OAuth token fixture obviously fake so secret scanners stop flagging it未关闭
难度 2/5 1-3 小时 新手友好度 76/100
code-yeongyu/senpi#3436 ·
维护者通常 1 天内回复
-
community first-timers-only good first issue hacktoberfest help wanted low hanging fruit up-for-grabs
难度 1/5 1 小时以内 新手友好度 78/100
lingdojo/kana-dojo#32505 · 1 条评论 · 5 个 reaction ·
维护者通常 1 天内回复