fix(web): send without a microtask when prepareRequest is sync

[brenelz]

Summary

Fixes #3959. With any prepareRequest hook configured, even init => init, a server-function read that hydration adopted was fetched again from the browser.

Hydration runs an adopted source's compute once inside subFetch, which swaps window.fetch for a never-settling mock only for that synchronous call. createRequest always awaited the hook, so even a synchronous hook pushed send(base, init) one microtask later, after the real fetch was restored, and the request reached the network.

createRequest now awaits the hook's result only when it is a thenable. A synchronous hook sends in the calling tick, like the transport does without a hook. An async hook is still awaited and keeps its current behavior; the prepareRequest docs now say that an async hook makes hydrated reads fetch again.

Left open: calls whose arguments need the rich-argument codec still await serializeArguments(args) before sending (the POST path in initializeResponse and the GET path in fetchServerFunction), so a hydrated read with such arguments refetches with or without a hook. Fixing that needs synchronous serialization, which is a larger change.

How did you test this change?

• test/server/server-functions-extensions.spec.tsx adds a case that checks a synchronous hook sends before the call returns, with and without arguments, and that an async hook still sends after the await. It fails on next (0 synchronous sends) and passes with the fix: pnpm build in packages/web, then npx vitest run --config vite.config.server.mjs test/server/server-functions-extensions.spec.tsx.
• test/hydration/prepare-request-hydration-3959.spec.tsx hydrates a createMemo(() => read()) over adopted data, once without a hook and once with init => init, and asserts no request is sent. On next the hook case sends /_server/data/prepare-request-3959; with the fix neither case sends anything: npx vitest run --config vite.config.hydrate.mjs test/hydration/prepare-request-hydration-3959.spec.tsx.
• I ran the @solidjs/web client, server and hydration suites before and after the change. The local compiler build is older than Client element claims fire before dynamic attributes are applied, and only href/action writes re-claim #3923 and Select loses its bound value when options load asynchronously #3928, so some element-claim and server specs fail on next as well; the failing test lists are identical before and after, apart from the new test.
• node size.mjs in scripts/size: the server-function scenarios pass. "app: CSR, observe tier" is over its cap by 5 B, with the same byte count on next.

🤖 Generated with Claude Code

[changeset-bot]

🦋 Changeset detected

Latest commit: 84ee3a5

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 12 packages

Name	Type
@solidjs/web	Patch
@solidjs/babel-plugin	Patch
@solidjs/diagnostics	Patch
@solidjs/element	Patch
@solidjs/h	Patch
@solidjs/html	Patch
test-integration	Patch
todos-server-example	Patch
@solidjs/compiler	Patch
@solidjs/signals	Patch
solid-js	Patch
@solidjs/universal	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[codspeed]

Merging this PR will not alter performance

⚠️ 2 benchmarks spent significant time in system calls

System calls cannot be consistently instrumented, so they are not included in the measure, which understates the real cost. Please switch to the Walltime instrument to accurately measure system calls.

Measurement and system calls

• readAllowed
• propagation:diamond

✅ 195 untouched benchmarks

---

_{Comparing brenelz:fix/prepare-request-sync-3959 (84ee3a5) with next (84dada7)}

[ryansolid]

Review: one type fix, then ready.

The job check fails in @solidjs/web#test-types:

test/hydration/prepare-request-hydration-3959.spec.tsx(9,58): error TS2305: Module '"solid-js"' has no exported member 'sharedConfig'.

sharedConfig comes from solid-js/internal, as in the other hydration specs:

import { createMemo, createRoot, enableHydration, flush } from "solid-js";
import { sharedConfig } from "solid-js/internal";

With that change test-types is clean and the spec still passes.

The fix itself is right. createRequest reaches the hook with no await in between, so a synchronous return now sends within the subFetch window, and a thenable is still awaited. I checked it locally on 84ee3a5, with web rebuilt:

Check	Result
Both new tests with server-functions/src/client.ts reverted to next	fail
Both new tests on the PR	pass (13/13 server, 2/2 hydration)

The gap you left open is worth its own issue: a hydrated read whose arguments go through the rich-argument codec still refetches, because of await serializeArguments(args), with or without a hook.

— Claude via Claude Code

---

Generated by Claude Code

[ryansolid]

Merged through #3990. This session could not push to your fork, so your commit went in unchanged and still authored by you (c2d4571 on next), followed by the solid-js/internal import fix for test-types (4101c67). Closing this one as superseded. Thanks!

— Claude via Claude Code

---

Generated by Claude Code