Skip to content

[pull] canary from vercel:canary - #2100

Merged
pull[bot] merged 8 commits into
MLH-Fellowship:canaryfrom
vercel:canary
Sep 30, 2026
Merged

pull[bot] merged 8 commits into
MLH-Fellowship:canaryfrom
vercel:canary

Conversation

@pull

@pull pull Bot commented Sep 30, 2026 •

Copy link
Copy Markdown

See Commits and Changes for more details.


Created by pull[bot] (v2.0.0-alpha.4)

Can you help keep this open source service alive? 💖 Please sponsor : )

aurorascharff and others added 8 commits September 30, 2026 14:19
## Summary

Metadata files such as `app/icon.tsx` export a default handler, rather
than HTTP methods. Validating them as ordinary route handlers rejects
the documented `size` and `contentType` exports with TS2559 during `next
typegen` + `tsc`.

Exclude metadata files from ordinary route-handler validation in both
validator modes by matching the `route.<extension>` file convention,
including custom page extensions. Add an icon fixture with the
documented exports and a regression test that checks generated
validation, retains an ordinary route-handler control, and runs `tsc`.

Correct the existing non-strict invalid-page-props assertion as well.
The same failure reproduces on the unchanged base commit: non-strict
`AppPageConfig` accepts arbitrary props, so the per-page check emits the
invalid-props diagnostic instead. The test still requires a failed build
and the specific invalid-props error; strict mode retains its
`AppPageConfig` assertion.

## Verification

- Reproduced the stale assertion on unchanged base commit `3932a6bd516`:
12/13 tests passed, with only invalid-page-props failing.
- Full Webpack validator suite: **14/14 passed** in normal mode and
**14/14 passed** with `__NEXT_EXPERIMENTAL_STRICT_ROUTE_TYPES=true`.
Both runs include the icon regression and invalid ordinary-handler
checks.
- Core Next.js build (including types), formatting, and ESLint passed.
- Direct before/after checks verified that both generators exclude
metadata files while retaining ordinary and custom-extension route
handlers.
- Tests use `NEXT_SKIP_ISOLATE=1` because the isolated install cannot
resolve the configured package registry. Full bootstrap in the temporary
worktree hit environment restrictions; the core package build passed.

<!-- NEXT_JS_LLM -->
Static metadata files nested under dynamic routes use placeholder paths
with empty params. The build treats those routes as missing
`generateStaticParams()`, which breaks adapter deployments.

This is a temporary hotfix, will follow up with better refactor of the
handling
We weren't running the use-cache/resume-data-cache test suites in PPF.
some of the tests fail due to the bug that is fixed in #98005 (although
interestingly the fetch-cache tests only fail in deploy mode, because in
`next start` we heal a bad RDC using outputs from
`.next/cache/fetch-cache`). a bunch of refactoring was needed (some
tests were producing false negatives, flakily), so i split this into a
separate PR to confirm behavior before/after the fix.

### `use-cache.test.ts`

- I've refactored the tests to use timestamps and be a bit more
debuggable. previously, they'd fail with a hydration error, and it was
hard to pinpoint the cause. including a name lets us know which cache
was mismatched, and assert that the values of the cache are consistent
across the HTML shell / resume.
- the tests for exluded caches now include more complete assertions on
what content is and isn't included in the HTML shell

### `resume-data-cache.test.ts`

This one needed quite a bit of work:

- The `use cache` and `fetch cache` tests were not independent. they
both used the same page and revalidated the same tag, which means that a
revalidation from the first test affected the outcome of the second
test. i've split them into two separate pages with two separate tags to
make the behavior consistent.
- Multiple tests fetched the same URL (with `fetch(..., { cache:
'force-cache' }`), which made them interfere with each other's fetch
caches. the URLs are now different so the tests are actually independent
- Some cache values are now tagged with a name and timestamp. The name
is used for asserting consistency, the timestamp is for improved
debuggability (provides ordering)
- The action revalidation tests weren't checking what the value from the
build-time prerender was actually supposed to be, so they were passing
even if the build-time RDC was not used at all
- The tests that use `browser()` now assert that there should be no
hydration warning, which can happen due to the PPF bug (due to
inconsistent cache values across HTML shell/resume)
- **the revalidation assertions are very flaky in deploy**, and i've had
to disable them to avoid introducing a test that fails often. i'm not
sure how they were passing before.
Closes NAR-954

In Partial Prefetching, the embedded runtime stream codepath awas
accidentally dropping the RDC. This PR fixes this by prefilling the
mutable cache from the RDC. we ungate the failing tests introduced in
#99417 and add some more e2e ones.

As a drive-by improvement, i've also moved the `cacheReady()` call to
happen after the main render runs through all the stages. It's pretty
strange to call `cacheReady()` before we've even started the render that
might fill any caches. I'm guessing it works because the code in-between
runs quickly enough, but it's still a potential race and we shouldn't do
it.

### How this bug got introduced

In #94194 we accidentally started clobbering the resume data cache in
the `cachedNavigations` codepath (i.e. when rendering the embedded
runtime prefetch stream).
This assignment used to be non-destructive because
`prerenderResumeDataCache` was separate:

https://github.com/vercel/next.js/blob/b0420899b68a4f7506fdbc93ff5acf6edcfa3d3c/packages/next/src/server/app-render/app-render.tsx#L969-L972
But after unifying `prerenderResumeDataCache` and
`renderResumeDataCache`, we started essentially dropping the RDC and
using an empty one

https://github.com/vercel/next.js/blob/8c661813a1e66cd85b795e3aeae0f04ae3e3201e/packages/next/src/server/app-render/app-render.tsx#L970-L973
This wasn't caught because the tests that checked RDC behavior weren't
exercising this codepath (at the time: no `cachedNavigations` + runtime
prefetch opt in, and later on no `partialPrefetching` opt in)
Runtime shell prefetch requests were ignoring the provided RDC. I'm not
sure what the history was, i.e. whether this was introduced when
unifying RDCs like #98005 or if it was like that from the start, but
this PR fixes it.

(the added tests were failing before the fix)
## Summary

Image helpers and components sort arrays from their configuration, which
fails when that configuration comes from a frozen deployment manifest.
Prepare owned image options once and reuse them across rendering and
`getImageProps`. This supports immutable Next.js configuration in Vercel
adapter deployments.

Modern and legacy Image components share a `useImageConfig` hook that
preserves existing configuration selection and SSR `localPatterns`
validation. `getImageProps` prepares its fixed options once at module
initialization without context.

## Verification

Existing local-pattern, query-string validation, and custom-loader
production fixtures pass with Webpack and Turbopack. Development checks
cover local patterns, legacy remote-image rejection, and the
static-export diagnostic with both bundlers.

<!-- NEXT_JS_LLM -->
The existing `image-from-node-modules` fixture renders `next/image` when
Next.js prerenders its static page. It does not render the external
image component during an SSR request, when an adapter deployment loads
image configuration from the frozen `required-server-files.json`
manifest. This change adds a `getServerSideProps` route, keeps
`my-cool-image` external, and configures two `images.deviceSizes`.
Before #99394, the adapter deployment fails when `next/image` sorts the
frozen manifest configuration.
## Summary

Continue removing request-event dependencies from App Router rendering.
The request body's `end` event is not a render-completion signal and may
already have fired before a custom server invokes Next.

Keep the dev indicator Pending until HTML or Flight output completes,
then classify from render-phase dynamic API usage and route
configuration, including usage discovered during Client Component SSR.
Completion handling stays local and dev-only, without shared lifecycle
callbacks or stream helper APIs.

Actions and prefetches do not reclassify the page. HEAD requests
preserve the last completed classification because their output is not
consumed. Pages Router and Cache Components behavior is unchanged.

## Verification

- Gated browser regressions fail before the change and pass afterward,
including a Client Component that calls `unstable_noStore()` after
suspending during SSR.
- All 21 dev-indicator tests pass in each combination of
Turbopack/Webpack and Cache Components on/off.
- Isolated `pnpm test-start-webpack
test/e2e/app-dir/app/standalone.test.ts` passes (2/2).

<!-- NEXT_JS_LLM -->
@pull pull Bot locked and limited conversation to collaborators Sep 30, 2026
@pull pull Bot added the ⤵️ pull label Sep 30, 2026
@pull
pull Bot merged commit b61c8e2 into MLH-Fellowship:canary Sep 30, 2026
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants