`useBeforeLeave` and scroll restoration break on Back after 50 history entries (`_depth` taken from the capped `history.length`)
维护者通常 1 天内回复
评估
- 难度
- 4/5
- 预计耗时
- 1-2 天
- 新手友好度
- 57/100
- Issue 类型
- 缺陷
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
- 技术栈
- typescript
- 领域
- frontend
调研方向
Start by reading saveCurrentDepth() and notifyIfNotBlocked in src/lifecycle.ts, then trace set() in src/routers/history.ts on next and the matching router code in src/routers/scrollRestoration.ts. Check how depth is assigned after the browser history reaches its entry cap and how Back and Forward use it. Done means navigation blocking and scroll restoration work across the cap, including after going Back and pushing a new entry.
由索引模型根据 Issue 内容生成。
描述
Describe the bug
saveCurrentDepth() (src/lifecycle.ts) stamps a new entry with _depth: window.history.length - 1:
export function saveCurrentDepth() {
if (!window.history.state || window.history.state._depth == null) {
window.history.replaceState({ ...window.history.state, _depth: window.history.length - 1 }, "");
}
depth = window.history.state._depth;
}
Chromium (kMaxSessionHistoryEntries) and Firefox (browser.sessionhistory.max_entries) cap the session history at 50 entries (WebKit at 100, not tested). Once it is full, pushState drops the oldest entry and history.length stays at 50, so every new entry gets the same _depth as the one below it.
Everything keyed on _depth then misbehaves between those entries:
notifyIfNotBlocked(src/lifecycle.ts): Back from the newest entry computesdelta = 0, soif (delta && block(delta))skipsblock:useBeforeLeavelisteners are never called and the navigation can't be prevented.createScrollRestoration(src/routers/scrollRestoration.ts, same path onnext): positions are stored per depth, so consecutive entries share one slot, andonPush()deletes every key>= depth. Back restores the offset of the entry just left instead of its own.
Present in 1.0.0 (latest on npm) / main (set() in Router and HashRouter), and on next (2.0.0-next.42) in src/routers/history.ts (saveCurrentDepth, set() in browserHistory() and hashHistory()), where scroll restoration is on by default with the default browser history, so it hits without opting in.
Your Example Website or App
Minimal code below; no hosted repro needed, it shows from the console in any Chromium or Firefox tab.
Steps to Reproduce the Bug or Issue
Fresh tab (so the history starts at length 1):
import { render } from "solid-js/web";
import { Router, Route, useNavigate, useBeforeLeave, useParams } from "@solidjs/router";
function Page() {
const params = useParams();
(window as any).nav = useNavigate();
useBeforeLeave((e) => {
if ((window as any).guard && !confirm("Leave?")) e.preventDefault();
});
return <p>page {params.n}</p>;
}
render(
() => (
<Router>
<Route path="/:n?" component={Page} />
</Router>
),
document.getElementById("root")!,
);
In the console:
for (let i = 1; i <= 52; i++) {
nav(`/${i}`);
await new Promise((r) => setTimeout(r));
console.log(i, history.length, history.state._depth);
}
Then run guard = true and press the browser Back button.
Expected behavior
_depthequalsion every line: the last three print 50, 51, 52.- Back triggers the
useBeforeLeaveconfirm.
Actual:
- From
i = 50on,history.lengthstays 50 and_depthstays 49 (Firefox: 50, as it applies its cap asynchronously). - Back navigates to
/51without the confirm (deltais 0). With<Router scrollRestoration>and a page taller than the window, scrolled before the last push, Back restores the offset of the entry just left.
Below the cap (e.g. Back from /10) the confirm shows and blocking works.
Platform
- OS: Linux
- Browser: Chromium 156.0.8078.4, Firefox 157
@solidjs/router: 1.0.0 (same code onnext2.0.0-next.42)
Additional context
Suggested fix: stamp each entry the router pushes from the depth of the entry it leaves, rather than from history.length. depth isn't exported from lifecycle.ts, so e.g. a small helper there:
export const nextDepth = () => depth + 1;
// in set() of Router / HashRouter:
window.history.pushState({ ...state, _depth: nextDepth() }, "", value);
and the fallback in saveCurrentDepth() becomes _depth: depth == null ? window.history.length - 1 : depth + 1, so entries pushed outside the router (e.g. location.hash) keep counting up past the cap too.
depth + 1 equals today's value below the cap, including after going Back (pushState drops the forward entries). A Math.max(history.length - 1, depth + 1) variant is wrong there: history.length is read before the forward entries are dropped, so after going Back 3 entries and navigating, the new entry is over-stamped, so a Back that useBeforeLeave prevents still goes through: the listener sees delta -3 instead of -1, and the router's go(3) undo has no forward entries to return to. With the change above, I checked in Chromium and Firefox that Back and Forward past the cap are blocked, scroll is restored, and a reload keeps the stamped depth. replace goes through keepDepth, which the change doesn't touch (from reading the code).
- 主要语言
- TypeScript
- 星标
- 1.3k
- 派生
- 185
- 平均合并
- 15 小时 47 分钟
- 30 天内合并 PR
- 47
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 没有 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-router 的其他 Issue
-
难度 5/5 一周以上 新手友好度 35/100
solidjs/solid-router#696 ·
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 48/100
solidjs/solid-router#691 ·
维护者通常 1 天内回复
-
Forward pressed while a Back navigation is still loading is dropped; the URL and the rendered route get out of sync可能已有人在做 @sean0407 于 10 天前认领。 未关闭1.x
难度 4/5 1-2 天 新手友好度 54/100
solidjs/solid-router#687 ·
维护者通常 1 天内回复
-
<A> costs ~6us of server CPU per instance during SSR (20x a plain <a>), mostly mergeProps/splitProps可能已有人在做 @birkskyum 于 81 天前认领。 未关闭1.x
难度 4/5 3-5 天 新手友好度 55/100
solidjs/solid-router#583 ·
维护者通常 1 天内回复
-
1.x enhancement
难度 3/5 1-2 天 新手友好度 38/100
solidjs/solid-router#518 · 2 条评论 ·
维护者通常 1 天内回复
查看 solidjs/solid-router 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 78/100
-
area:desktop area:ui bug platform:macos
难度 2/5 1-3 小时 新手友好度 67/100
anthropics/claude-code#106183 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 76/100
-
难度 1/5 1-3 小时 新手友好度 76/100
维护者通常 1 天内回复
-
core: OpenRouter BYOK sessions record cost = 0 since v2.2.0 (ignores cost_details.upstream_inference_cost)可能已有人在做 关联的 PR 仍在进行中或已合并。 未关闭reproduced
难度 2/5 1-3 小时 新手友好度 72/100
anomalyco/opencode#60558 · 1 条评论 ·
维护者通常 1 天内回复