[2.0 rc.13] An async read that rejects under `<Errored>` still triggers `unhandledrejection` in the browser
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Newbie friendliness
- 58/100
- Issue type
- Bug
- Clarity
- Clearly specified
- Activity status
- Active
- Tech stack
- javascript, node.js, playwright, typescript
- Domain
- backend, frontend, testing-qa
Research direction
Start by running the self-contained repro.mjs with Playwright and compare the async-reject and streamed cases. Read packages/solid/src/server/signals.ts, packages/web/src/server.ts, client/hydration.ts, and the seroval PROMISE_FAILURE path, then inspect the hydration-records harness. Done means the Errored fallback still renders without browser unhandledrejection or pageerror reports, with regression coverage for pre-shell and streamed rejection.
Written by the indexing model from the issue text.
Description
Describe the bug
An async source that is serialized for hydration rejects, and an <Errored> above it renders its fallback. The HTML is correct, but the inline payload script still creates the source's promise and rejects it with no handler attached. The browser reports Uncaught (in promise) Error: Internal Server Error as an unhandledrejection. The page needs no client JS for this: the event fires from the inline script alone. Error monitoring that listens for it reports an error, and Playwright pageerror checks fail, for an error the boundary already handled.
It happens both when the rejection lands before the shell and when it lands after it under a <Loading>. A synchronous throw caught by the same <Errored> doesn't do this.
Your Example Website or App
Self-contained script below. No JSX, no router, no client bundle; it runs with plain node against the published packages.
Steps to Reproduce the Bug or Issue
// repro.mjs: plain Node ESM, no JSX, no router, no client bundle.
// npm i [email protected] @solidjs/[email protected] (optional: playwright)
// node repro.mjs -> writes out-<case>.html; with playwright, opens each in Chromium
import { writeFileSync } from "node:fs";
import { HydrationScript, createComponent, escape, renderToStream, ssr, ssrHydrationKey } from "@solidjs/web";
import { Errored, Loading, createMemo } from "solid-js";
const read = (ms, fail) => () => {
const value = createMemo(async () => {
await new Promise(r => setTimeout(r, ms));
if (fail) throw new Error("boom");
return "ok";
});
return ssr(["<p", ">", "</p>"], ssrHydrationKey(), escape(value()));
};
const syncThrow = () => {
throw new Error("boom");
};
const errored = child => () =>
createComponent(Errored, { fallback: () => ssr(["<p>fallback</p>"]), get children() { return child(); } });
const loading = child => () =>
createComponent(Loading, { fallback: ssr(["<p>loading</p>"]), get children() { return child(); } });
const page = tree => () =>
ssr(
["<!doctype html><html><head>", "</head><body><main", ">", "</main></body></html>"],
escape(createComponent(HydrationScript, {})),
ssrHydrationKey(),
escape(tree())
);
// Stand-in for a client that reads key "100" late: a module script runs after
// the inline payload, like a client entry would. This is not Solid's hydration.
const CONSUME = `<script type="module">_$HY.r["100"].then(undefined, () => {});</script>`;
const cases = {
"async-reject": { tree: errored(read(0, true)) },
"async-resolve": { tree: errored(read(0, false)) },
"sync-throw": { tree: errored(syncThrow) },
"async-reject-consumed": { tree: errored(read(0, true)), extra: CONSUME },
// shell flushes with the Loading fallback, the read rejects 100 ms later
"streamed": { tree: loading(errored(read(100, true))), pipe: true }
};
const files = [];
for (const [name, { tree, extra, pipe }] of Object.entries(cases)) {
let html;
if (pipe) {
const chunks = [];
const t0 = Date.now();
await new Promise(end =>
renderToStream(page(tree)).pipe({ write: c => chunks.push([Date.now() - t0, String(c)]), end })
);
console.log(`${name}: ${chunks.length} chunks at ${chunks.map(c => c[0]).join(", ")} ms`);
html = chunks.map(c => c[1]).join("");
} else html = String(await renderToStream(page(tree)));
if (extra) html = html.replace("</body>", `${extra}</body>`);
const file = new URL(`out-${name}.html`, import.meta.url);
writeFileSync(file, html);
files.push([name, file]);
}
const { chromium } = await import("playwright").catch(() => ({}));
if (!chromium) {
console.log("playwright not installed: open the out-*.html files and check the console");
process.exit(0);
}
const browser = await chromium.launch();
console.log(`Chromium ${browser.version()}`);
for (const [name, file] of files) {
const tab = await browser.newPage();
await tab.addInitScript(() => {
window.__events = [];
addEventListener("unhandledrejection", e => __events.push(`unhandledrejection: ${e.reason?.message}`));
addEventListener("rejectionhandled", () => __events.push("rejectionhandled"));
});
const pageErrors = [];
tab.on("pageerror", e => pageErrors.push(e.message));
const cdp = await tab.context().newCDPSession(tab);
const devtools = [];
cdp.on("Runtime.exceptionThrown", ({ exceptionDetails: d }) =>
devtools.push(`${d.text} ${d.exception?.description?.split("\n")[0]}`)
);
cdp.on("Runtime.exceptionRevoked", ({ reason }) => devtools.push(`revoked: ${reason}`));
await cdp.send("Runtime.enable");
await tab.goto(file.href);
await tab.waitForTimeout(1000);
console.log(`\n[${name}] <main>: ${JSON.stringify(await tab.locator("main").innerText())}`);
console.log(` pageerror: ${JSON.stringify(pageErrors)}`);
console.log(` window events: ${JSON.stringify(await tab.evaluate(() => __events))}`);
console.log(` devtools console: ${JSON.stringify(devtools)}`);
await tab.close();
}
await browser.close();
npm i [email protected] @solidjs/[email protected] playwright(Playwright is optional; it needsnpx playwright install chromiumonce).node repro.mjs- Without Playwright, open
out-async-reject.htmlfrom disk in a browser. The console showsUncaught (in promise) Error: Internal Server Error.
Results on Chromium 153, the same on rc.13 and on next at e44b2e4:
| Case | Tree | <main> |
pageerror |
Window events |
|---|---|---|---|---|
async-reject |
Errored > read, read rejects, await renderToStream |
fallback | 1 | unhandledrejection |
async-resolve |
same, read resolves | ok | 0 | none |
sync-throw |
Errored > child, child throws synchronously |
fallback | 0 | none |
async-reject-consumed |
async-reject plus the CONSUME module script |
fallback | 1 | unhandledrejection, then rejectionhandled |
streamed |
Loading > Errored > read, read rejects 100 ms after the shell, pipe() |
fallback | 1 | unhandledrejection |
streamed writes 2 chunks (about 1 ms and 100 ms). Served over a real http response, the result is the same.
Payload of async-reject, byte-identical on rc.13 and next:
<script>(self.$R=self.$R||{})[""]=[];_$HY.r["100"]=$R[0]=($R[1]=($R[2]=() => {
const resolver = {
p: 0,
s: 0,
f: 0
};
resolver.p = new Promise((resolve, reject) => {
resolver.s = resolve;
resolver.f = reject;
});
return resolver;
})()).p;($R[4]=(resolver, data) => {
resolver.f(data);
resolver.p.s = 2;
resolver.p.v = data;
})($R[1],$R[3]=new Error("Internal Server Error"));_$HY.r["1"]=$R[3];</script>
Key 100 is the memo's promise, rejected on the spot. Key 1 is the <Errored> record. In streamed, the shell declares _$HY.r["100000"] as a pending promise next to 1_fr. The second chunk rejects the 100000 promise, writes the boundary record at _$HY.r["1000"], and resolves 1_fr with true.
Expected behavior
When <Errored> contains the rejection, the page shows the fallback and the browser reports nothing, as in sync-throw. Instead there is one unhandledrejection for the memo's serialized promise.
Analysis
Links are to next at e44b2e4.
- The server memo hands its deferred promise to the serializer when the compute returns it (signals.ts#L1673-L1677). When the read rejects,
<Errored>renders its fallback and serializes the error at its own id (signals.ts#L3565-L3570). The promise already handed over stays in the payload.<Loading>buffers its subtree's serializations (ssrLoadingBoundary, server/hydration.ts#L265-L290); I didn't find anything similar in the servercreateErrorBoundary. - seroval writes the rejection as
PROMISE_FAILURE, which calls the payload's ownreject(constructors.ts#L34-L41 on seroval main; the same code is in the 1.6.8 dist that rc.13 resolves to today). No handler is attached at that point. - The client attaches a handler only when it consumes the key:
readHydratedValue(client/hydration.ts#L486-L491; the same code is in the rc.13 dist atsolid-js/dist/solid.js:170-171).takeHydrationValue, which the router reads through, does the same. - A hydrating
<Errored>that finds its serialized error throws it on the first run without calling its children (client/hydration.ts#L1500-L1522). So in the first hydration pass the child that would read key100(or100000) doesn't run. The sweeps after hydration skip it too:watchTruncationonly looks at_frkeys (#L2784) andrejectTruncatedRefsskips entries that already settled (#L2843). I'm reading this from the code; the repro has no hydrating client.
async-reject-consumed shows the timing. In this Chromium run, a handler attached from a module script, after the inline payload, only revokes the report: the page gets rejectionhandled and DevTools logs "Handler added to rejected promise", but the pageerror and unhandledrejection listeners have already fired.
next already covers a neighboring case. The comment above abandonSubtree (server.ts#L2543-L2550) says abandoned data ids resolve undefined because nothing consumes them and "a rejection would only raise unhandled-rejection noise". That runs only when a fragment settles with an error (server.ts#L2835). Here the <Errored> contains the error, so no fragment fails: async-reject has no fragment, and in streamed 1_fr resolves clean. The Node side of the same promise is already handled: since #3642, trackSerialized observes it so a pre-shell rejection can't exit the process (server.ts#L2534). The test harness also catches every _$HY.r record, because "without a consumer that leaks an unhandledRejection into the test process" (test/harness/hydration-records.ts#L27-L32), so the suite doesn't see this.
Where to fix it is your call. The options I can see:
- Server: when an
<Errored>switches to its fallback, keep the values serialized under it from reaching the payload as rejections (write them asundefinedlikeabandonSubtreedoes, or drop them if they haven't been flushed).abandonSubtreeitself wouldn't cover it as is: it only settles entries still inpendingSerialized, and a source that already rejected has left it (server.ts#L2539-L2540). - Payload: attach a no-op rejection handler to the promise the payload creates, in seroval's
PROMISE_FAILUREor in Solid's serialization layer, the wayrejectTruncatedRefsdoes for its own rejections (client/hydration.ts#L2863).
A client-side sweep after hydration would be too late: per async-reject-consumed, a handler added after the inline script only revokes the report.
Screenshots or Videos
N/A.
Platform
- OS: macOS 27.0
- Browser: Chromium 153.0.8010.12 through Playwright 1.63.0. Firefox and WebKit not tested.
- Node.js: v24.21.0
- Version:
solid-jsand@solidjs/web2.0.0-rc.13 from npm (seroval 1.6.8 through~1.6.7), andnextat e44b2e4 built locally (seroval 1.6.7). - Production conditions. With
node --conditions=developmentthe same cases reportboominstead ofInternal Server Error.
Additional context
- #2997 is the closest precedent. Its fix (62e78832) "plugged the unhandled-rejection leaks in this flow", and
readHydratedValuecites #2997 for observing a rejection when the client consumes it. This case has no consumer. - #3414 (closed) had a different symptom (a dead counter in the fallback, fixed in 5426ffb4 as an id mismatch), but its results table already listed a
pageerrorfor both async rejection cases. - If the fix keys off the
<Errored>id:query()in@solidjs/routerwrites under the query cache key (src/data/query.ts#L72-L77), not under the boundary's id, so a rejected query under that boundary might need a change in the router too.
- Dominant language
- TypeScript
- Stars
- 36.1k
- Forks
- 1.1k
- Avg merge
- 9h 33m
- Merged PRs (30d)
- 262
Getting set up
- No Dockerfile or Docker Compose file
- Has a pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from solidjs/solid
-
Difficulty 3/5 1-2 days Newbie friendliness 45/100
Maintainers usually reply within 1 day
-
Difficulty 4/5 3-5 days Newbie friendliness 48/100
Maintainers usually reply within 1 day
-
Difficulty 4/5 3-5 days Newbie friendliness 45/100
solidjs/solid#3766 · 1 comment ·
Maintainers usually reply within 1 day
-
Difficulty 5/5 Over a week Newbie friendliness 35/100
Maintainers usually reply within 1 day
-
Difficulty 4/5 3-5 days Newbie friendliness 48/100
Maintainers usually reply within 1 day
Similar issues
-
area/dashboard kind/bug QA/dev-automation
Difficulty 2/5 1-3 hours Newbie friendliness 65/100
rancher/dashboard#19379 · 2 comments ·
Maintainers usually reply within 5 days
-
perf(core): getComments() runs the approved count and the comment list as two sequential queriesOpenarea/core bot:bug bot:working
Difficulty 2/5 1-3 hours Newbie friendliness 76/100
emdash-cms/emdash#3905 · 2 comments ·
Maintainers usually reply within 1 day
-
community first-timers-only good first issue hacktoberfest help wanted low hanging fruit up-for-grabs
Difficulty 1/5 Under an hour Newbie friendliness 90/100
lingdojo/kana-dojo#31728 · 1 comment · 5 reactions ·
Maintainers usually reply within 1 day
-
selective-claw: freshTailTurns=0 keeps ALL turns verbatim and summarizes none (slice(-0) === slice(0))Possibly taken @zjncs claimed this today. Opencomponent:tokenless
Difficulty 2/5 1-3 hours Newbie friendliness 80/100
agentic-os-org/ANOLISA#6112 · 1 comment ·
Maintainers usually reply within 1 day
-
bug needs triage
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
rjsf-team/react-jsonschema-form#5439 ·
Maintainers usually reply within 1 day