- 29 Jun, 2026 18 commits
-
-
next-js-bot[bot] authored
-
Sam Poder authored
This PR is the PR that "turns on" the service worker feature, this is because in the analyzer we are now looking for code like: ``` await navigator.serviceWorker.register(new URL('../lib/service-workers/1.js', import.meta.url), ...) ``` And transforms it into registering the URL that Next.js will now serve based on scope. URL scheme is: ``` scope: "/" -> `sw.js` scope: "/offline/mode" -> `sw-offline-mode.js` ``` etc. It also inserts the `ServiceWorkerEntryModule`s so they can be discovered by `next-api` (see https://github.com/vercel/next.js/pull/94922). -
Sam Poder authored
<!-- NEXT_JS_LLM_PR -->
-
Sam Poder authored
This PR changes `compute_chunk_group_info` so that chunks inherit heuristics from the entry chunk groups that reference them. For entry-points, if a chunk group is referenced by any entry point, it inherits a true value. For clusters, chunk groups inherit all the clusters that the groups that reference them are a part of.
-
Sam Poder authored
This takes `chunkingHeuristics` from `next.config.js` and puts it on `ChunkGroupEntry` so it can be on `ChunkGroupInfo` which will then make it available in `make_production_chunks`
-
Sam Poder authored
This PR adds new experimental options to `next.config.js`, they are: the request cost, “clusters” of routes where should chunk more (because the Z+Z case in `make_production_chunks` is more likely), “entry points” where we prioritize larger merged chunks for fast initial page loads (achieved by increasing P(N=1)), and a bounce rate to configure P(N=1) and P(N=2) based on site data.
-
Janka Uryga authored
Stopgap until we make this a proper part of the CI matrix
-
Sukka authored
- Implements `location` scope check (suggested by @bgw in https://github.com/vercel/next.js/pull/92900#discussion_r3107167621), ensuring `location` is a true global variable before reporting. - Implements basic variable tracking (suggested by @icyJoseph in https://github.com/vercel/next.js/pull/93057#issuecomment-4281583840) --------- Co-authored-by:
Benjamin Woodruff <benjamin.woodruff@vercel.com>
-
Will Binns-Smith authored
This adds documentation for this config flag. Test Plan: Tested each hyperlink within the repo. --------- Co-authored-by:vercel[bot] <35613825+vercel[bot]@users.noreply.github.com>
-
Andrew Clark authored
On a client-side navigation, when a proxy/middleware redirected the requested URL to a destination that was itself rewritten to a dynamically rendered route, the page content updated correctly but the address bar stayed on the original URL. Hard navigations (reload, typing the URL) were unaffected. This happened because the navigation optimistically committed the predicted URL up front, and when the redirect resolved to a route with the same shape as the prediction, the router treated the navigation as already correct and never reconciled the address bar. A redirect discovered during a navigation now invalidates the prediction and re-resolves the route, so the address bar reflects the redirect destination. Adds an end-to-end test covering the regression. Fixes #95195
-
Tobias Koppers authored
## Summary Reworks the cost model of the experimental graph CSS chunking algorithm (`experimental.cssChunking: 'graph'`, Turbopack only) and retunes its defaults. - **Replaces the `moduleFactorCost` option with `weightDistribution`.** The old `moduleFactorCost` term penalized a chunk purely by `chunk_size / group_total_size`, which charged a chunk group even for CSS it fully needs and never actually measured overshipping. - **New per-group cost** is `chunk_group_weight * (chunk_size + request_cost)`, summed over the chunk groups that load a chunk, where `chunk_group_weight = group_total_size ^ (-weightDistribution)` is precomputed once per group. `weightDistribution = 0` weights every chunk group equally; higher values give smaller chunk groups a larger weight, so the algorithm overships less to small pages at the cost of more requests. The size weighting subsumes the explicit overship penalty, so the metric stays chunk-local and the greedy merger is unchanged. - **Retunes defaults:** `requestCost` `20000` → `100000` (bytes) and `weightDistribution` default `0.1`. Config shape (object form): `{ type: 'graph', requestCost?, weightDistribution? }`. The change is wired end to end through the config schema/types, `StyleGroupsAlgorithm::Graph`, and the chunking algorithm. ## Verification - `cargo test -p turbopack-core --lib style_groups_graph` (54 passed, incl. a new test asserting `weightDistribution` keeps an unneeded module out of a small group's chunk) - `pnpm test-start-turbo test/e2e/app-dir/css-order/css-order.test.ts` (243 passed, 12 todo) — confirms chunk shapes/request counts at the new defaults across the `graph` and `graph-object` modes <!-- NEXT_JS_LLM_PR --> -
Vercel Release Bot authored
Co-authored-by:next-js-bot[bot] <279046576+next-js-bot[bot]@users.noreply.github.com>
-
Aurora Scharff authored
### What? Tightens `skills/next-cache-components-adoption/SKILL.md` based on four agent friction logs. ### Why? - Non-interactive strategy default was ambiguous on small apps. - `experimental.dynamicIO` is fatal at config-schema time, not segment-config time — needs its own bullet. - The `cache-components-instant-false` codemod doesn't translate `revalidate`/`dynamic`/`fetchCache`; agents expected an immediate passing build and got a failed one. - "Build passes immediately" overpromised — only true once sync-IO + segment configs are translated. - "Green" → "passing" for clarity. ### How? 13 insertions / 7 deletions, single file. New `dynamicIO` bullet, restructured pre-codemod work into a numbered list (sync-IO + segment configs), softened the Incremental headline, simplified the non-interactive default. No docs material duplicated — recipes stay in the [migration guide](https://nextjs.org/docs/app/guides/migrating-to-cache-components). <!-- NEXT_JS_LLM_PR -->
-
Sebastian "Sebbie" Silbermann authored
There are more client-node modules being bundled into `client-browser` which I'll fix in a followup (with codegen). Our internal `unstable-rethrow` will now have a version without suffix (unbundled or client-node) and a client-browser only version (`.browser`) that bundlers alias to for client-browser which will always be bundled. For the specific modules being removed from client-browser, check the second commit with the actual change. The first commit fixes the `browser-chunk` test to include `next/src/server` in a snapshot. Closes #95189
-
Tariqul Islam Tuhin authored
### What Turbopack's `server-reference-manifest` emitter stores `filename` and `exportedName` **twice**: once hoisted to the action-entry level (correct), **and** redundantly inside *every* per-page `ActionManifestWorkerEntry`. This PR removes the per-worker copies so the worker map only carries `{ moduleId, async }` — matching the webpack backend, which already emits them at the entry level only. The emitter even comments the intent (`// Hoist the filename and exported_name to the entry level`) and does the hoist — the per-worker fields are vestigial. ```rust // crates/next-api/src/server_actions.rs entry.workers.insert(&key, ActionManifestWorkerEntry { module_id: loader_id.clone(), is_async: …, exported_name, // ← removed (already on entry) filename: filename.as_ref(), // ← removed (already on entry) }); entry.exported_name = exported_name; // hoist (kept) entry.filename = filename.as_ref(); // hoist (kept) ``` ### Why it matters `filename` and `exportedName` are invariant per action, but the worker map has **one entry per page that can reach the action**, so these strings get repeated `O(actions × pages)` times. On a large App Router app (`output: 'standalone'`, `cacheComponents: true`) this is the dominant cost of the manifest: - ~1,650 routes, **1,847 references → 589,349 `workers` entries** - `server-reference-manifest.json` = **170 MB** / `.js` = **182 MB** (plus the duplicate copy under `.next/standalone`, plus ~314 MB of per-route manifests) - A single `'use cache'`/action reachable from a shared client is registered on **1,638 pages**, each repeating the full `filename` string Dropping the two redundant per-worker fields removes the largest repeated strings from every worker entry, shrinking the manifest substantially with no information loss. ### Why it's safe (consumer-compatible) - The webpack backend already emits worker entries as `{ moduleId, async }` only, with `filename`/`exportedName` at the entry level — so this aligns Turbopack with webpack, not introduces a new shape. - The only first-party reader, `packages/next/src/server/mcp/tools/get-server-action-by-id.ts`, reads `entry.filename` / `entry.exportedName` (entry level) and types `workers` as `Record<string, any>` — it never reads the per-worker copies. ### Changes - `crates/next-core/src/next_manifests/mod.rs` — drop `exported_name` + `filename` from `ActionManifestWorkerEntry` (keep `module_id`, `is_async`). - `crates/next-api/src/server_actions.rs` — stop assigning them in the worker insert (entry-level hoist unchanged). ### Notes - Verified by reading the emitter, the manifest types, and the consumer; I don't have a local Turbopack build, so CI is the source of truth here — if any manifest **snapshot fixtures** include per-worker `filename`/`exportedName`, they'll need regenerating. - Context / measurements: vercel/next.js#95241. -
Sebastian "Sebbie" Silbermann authored
-
Joseph authored
Fixes: #94821, #92787, #93104 Using `Object.fromEntries` on URLSearchParams, drops repeated keys. ```js const s = '?foo=bar&foo=baz' Object.fromEntries(new URLSearchParams(s)) // { foo: 'baz' } ``` There's already a function, `urlSearchParamsToParsedUrlQuery`, which handles this case. https://github.com/vercel/next.js/blob/40296fe53f40ffe6a0e2184879bd5e294d271ce4/packages/next/src/client/route-params.ts#L235 ```js urlSearchParamsToParsedUrlQuery(new URLSearchParams(s)) // { foo: ['bar', 'baz'] } ``` - [x] The e2e off one of the issue reports, I'll try to simplify it. -
Niklas Mischkulnig authored
The types were unnecessarily specific (it doesn't always return a `T[]`)
-
- 28 Jun, 2026 2 commits
-
-
next-js-bot[bot] authored
-
Niklas Mischkulnig authored
We already switched to 128 bit hashes for many operations, these were still using 64 bit hashes. - make `hash_xxh3_hash128` return `u128`, just like `hash_xxh3_hash64` returns a `u64` - Make `Code::source_code_hash` return a u128
-
- 26 Jun, 2026 15 commits
-
-
next-js-bot[bot] authored
-
Luke Sandberg authored
## What Allow some modifications that turn out to be no-ops to be undone on TaskStorage. This saves us some redundant check-then-mutate cycles. This doesn't apply to every field type * `Option<T>` - still read-check-modify * `CounterMap<T>` - updates are unconditional tracks and removals use the undo approach * `AutoSet`/`AutoMap` - mostly uses the new 'undo' approach * `take` as a bulk operation does not, taking an empty set is rare and cheap to check for * `AutoMap::insert` doesn't use the undo approach or even check, we always track a modification since all callers are only inserting unique values anyway A few other incidental cleanups were landed as well. ## Why Avoid redundant hash lookups. ## How Have `track_modification` return a new TrackOutcome that records what modifications were made and then in the undo case we can pass it to a new `undo_track_modification` function which can reverse it. This is a little brittle since both operations need to occur within a single lock transition I considered using a 'lambda' oriented approach but decided against it since it would be awkward for some of the mutators and as all the callers are generated by a macro (well _almost_ all callers) it isn't too risky.
-
Luke Sandberg authored
Add data about persistence and compaction size to our spans This will show up in trace data collected by `NEXT_TURBOPACK_TRACING` and in the default trace spans collected by next.js
-
Luke Sandberg authored
Many fields in TaskStorage store a mix of transient and persistent data. When we serialize we leverage the `filter_transient` tag (and the `IsTransient` trait) to avoid persisting it, however, when mutations occur we are still marking the tasks as `modified` even when we are only modifying transient data. This leads to unnecessary persistence cycles. The solution is simply to not call `track_modification` when the modification is about transient data items. For most fields this is handled in the macro accessors, but for `cell_data` this is managed via some new 'handwritten' mutators implemented on `TaskGuard`. This is desirable to decrease the number of times we do registry lookups since most callers who modify the cell_data already have the ValueType in hand
-
Tim Neutkens authored
### What? - Preserve the resolved routing `query` and dynamic `params` when `NextNodeServer.runApi` invokes a bundled Pages API route. - Keep `req.url` restored to the original browser-visible request URL for compatibility. - Add an end-to-end regression test for a middleware rewrite from `/foo/bar?key=value` to `/api/proxy/bar?added=1&extra=2` using an optional catch-all Pages API route. ### Why? Middleware routing correctly resolves the destination query and catch-all parameters, but the Node Pages API path restores `req.url` before invoking the bundled handler and previously forwarded only `waitUntil`. The handler then reparsed the original URL and produced only `{ key: "value" }`, dropping the rewrite-added `added` and `extra` values and the destination `slug` parameter. This restores the behavior from Next.js 14 that regressed during the Pages API handler-interface migration. The issue affects both webpack and Turbopack because the data is lost in shared server routing after bundling. The public documentation describes the pieces separately—rewrites preserve the original URL, dynamic API parameters appear in `req.query`, and resolved routing queries include middleware changes—so this is a bug fix rather than a new API contract. Fixes #94647 ### How? `runApi` now forwards the existing request metadata together with the resolved `query` and `match.params` through the bundled handler context. The Pages API entrypoint already installs supplied request metadata before `routeModule.prepare()`, which then uses these routed values while retaining the original `req.url`. The regression test asserts that: - `req.url` remains `/foo/bar?key=value`. - `req.query` contains the original `key`, rewrite-added `added` and `extra`, and catch-all `slug` values. ### Verification - `pnpm --filter=next build` - `NEXT_TEST_PREFER_OFFLINE=1 pnpm test-dev-turbo test/e2e/middleware-rewrites/test/index.test.ts -t "should preserve rewrite query and dynamic params in Pages API routes"` - `NEXT_TEST_PREFER_OFFLINE=1 pnpm test-dev-webpack test/e2e/middleware-rewrites/test/index.test.ts -t "should preserve rewrite query and dynamic params in Pages API routes"` - `NEXT_TEST_PREFER_OFFLINE=1 pnpm test-start-turbo test/e2e/middleware-rewrites/test/index.test.ts -t "should preserve rewrite query and dynamic params in Pages API routes"` - `NEXT_TEST_PREFER_OFFLINE=1 pnpm test-start-webpack test/e2e/middleware-rewrites/test/index.test.ts -t "should preserve rewrite query and dynamic params in Pages API routes"` - `NEXT_TEST_PREFER_OFFLINE=1 pnpm test-dev-turbo test/e2e/api-support/api-support.test.ts -t "query|dynamic"` - `NEXT_TEST_PREFER_OFFLINE=1 pnpm test-start-webpack test/e2e/api-support/api-support.test.ts -t "query|dynamic"` - `NEXT_TEST_PREFER_OFFLINE=1 pnpm test-dev-turbo test/e2e/custom-routes/custom-routes.test.ts -t "api rewrite"` - `NEXT_TEST_PREFER_OFFLINE=1 pnpm test-start-webpack test/e2e/custom-routes/custom-routes.test.ts -t "api rewrite"` - Pre-commit lint-staged checks (Prettier and ESLint) - Not run continuously: `pnpm --filter=next dev` (the local watcher hits the environment open-file limit with `EMFILE`; the focused package build passed) <!-- NEXT_JS_LLM_PR --> <!-- CURSOR_SUMMARY --> --- > [!NOTE] > **Medium Risk** > Changes shared Pages API request routing in the Node server and adapter i18n route generation; behavior is restored to pre-regression semantics but affects all bundled Pages API invocations. > > **Overview** > Fixes a regression where **Pages API** handlers saw only the original URL query after a middleware rewrite, because `runApi` restored `req.url` but did not pass the router’s resolved `query` and `params` into the bundled handler context. > > `NextNodeServer.runApi` now supplies `requestMeta` (including merged `query` and `match.params`) alongside `waitUntil`, so `req.query` keeps rewrite-added parameters and catch-all segments while `req.url` stays the browser-visible path. > > Adapter build output no longer applies i18n locale prefixing to dynamic **Pages API** routes (`shouldLocalize` excludes API pages); regular pages still get `nextLocale` routing. > > Regression coverage: middleware rewrite e2e (`/foo/bar` → catch-all API) and a production adapter test with `i18n` enabled. > > <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit ad415cfe574f67dd7385662a01c2c9f3c89af521. Bugbot is set up for automated code reviews on this repo. Configure [here](https://www.cursor.com/dashboard/bugbot).</sup> <!-- /CURSOR_SUMMARY --> -
Jiwon Choi authored
### Why? Current behavior: https://github.com/vercel/next.js/actions/runs/28225010081/job/83615571987?pr=95184#step:31:8732 For a minimal-mode resume request carrying the Instant Navigation testing cookie, Next.js skips the postponed render so the test receives only the prefetched static shell. The resulting empty continuation was represented as `PAGES`, but the App Page handler requires `APP_PAGE`, causing error: ``` Error: Invariant app-page handler received invalid cache entry PAGES ``` ### How? Return an empty `APP_PAGE` response with the correct HTML or RSC content type and postponed metadata. If the document continuation has no `<head>` chunk, insert the Instant Navigation bootstrap while flushing the stream. Add production regression coverage for document and RSC-prefetch resume requests using build-generated postponed state. <!-- NEXT_JS_LLM_PR -->
-
Sebastian "Sebbie" Silbermann authored
Pass `--skip-git-metadata-upload` to all `datadog-ci junit upload` invocations. This stops the in-CI git metadata sync, which forces an expensive `git fetch` unshallow on every shallow CI checkout before each test-report upload. We don't need to upload git metadata manually since our GH actions are directly integrated with DataDog via app. > Git metadata upload is only required if you can’t integrate your CI provider directly with Datadog. If you are using a [source code provider integration](https://docs.datadoghq.com/code_coverage/setup/?tab=github#integrate-with-source-code-provider), such as Datadog GitHub app or Gitlab integration, you can disable the git metadata upload by passing the --skip-git-metadata-upload=1 flag to the datadog-ci coverage upload command, like this: --https://docs.datadoghq.com/code_coverage/setup/?tab=github#coverage-upload-outputs-could-not-sync-git-metadata-error ## test plan - [x] no more unshallow when uploading junit reports to DataDog: [upload from this branch shows no "syncing git metadata"](https://github.com/vercel/next.js/actions/runs/28252881331/job/83708778650?pr=95210#step:38:113) that [you'd used to see on prior uploads](https://github.com/vercel/next.js/actions/runs/28252765955/job/83708485242?pr=95207#step:38:112)
-
Jude Gao authored
## Summary Bump the `agent-browser` floor in the `next-dev-loop` skill from `>= 0.31.0` to `>= 0.31.1`. One-line change to the `requires` section; the listed capabilities (React introspection, worktree-scoped `session id`, idempotent `--restore`, launch-flag reconciliation) are unchanged. ## Verification - `prettier --check skills/next-dev-loop/SKILL.md` → clean. <!-- NEXT_JS_LLM_PR -->
-
Luke Sandberg authored
Bump rust nightly version Critically this fixes a bug in Rust Analyzer where it didn't understand 'pattern_types' a new rust feature that is used by `NonZero<T>` types for example)  So now we can get feedback on the size of types containing vc! From scanning release notes there is nothing major that i can see. Clippy did flag one new warning that is being addressed as well.
-
Aurora Scharff authored
The runtime error message for `blocking-prerender-dynamic` lists fixes in the order **[stream] → [cache] → [block]** (see `packages/next/src/server/app-render/blocking-route-messages.ts`). Every other surface (dev overlay fix cards, the blog post on `nextjs.org/blog/next-16-3-instant-navigations`, the in-post bullets, the `## Stream, Cache, or Block` section heading) uses the same order. Only this docs page was out of sync, listing Cache first. This PR: - Swaps the order of the `FixOption` cards under `## Ways to fix this` to `stream → cache → block`. - Swaps the body sections `## Cache the component or data` and `## Wrap in or move into Suspense` to match. No content changes inside either section — pure reorder.
-
Hendrik Liebau authored
When an App Router route begins with a dynamic segment (for example `app/[locale]/about`), a client-side navigation from a Pages Router page could render the wrong route. The Pages Router resolves the destination against its own routes only, so `/en/about` would match a less specific Pages Router dynamic route such as `pages/[locale]/[category]` instead of the App Router page. The server pools app and pages routes and ranks them by specificity, so a hard reload always rendered the correct App Router page; only the client-side soft navigation diverged. The client router filter is responsible for detecting destinations owned by the other router and forcing a hard navigation, but `createClientRouterFilter` only recorded the static prefix of a dynamic route. App routes whose first segment is dynamic have no static prefix, so they contributed nothing to the filter and the Pages Router never learned to hand them off. The filter now also stores a normalized pattern for those routes, with dynamic segments replaced by a placeholder token (`/[locale]/about` becomes `/[]/about`). After the Pages Router resolves a navigation to a dynamic route, `hasDynamicFilterCandidate` reconstructs the candidate app-route patterns from that route and the concrete path and triggers a hard navigation when any candidate is present in the dynamic filter. This also covers catch-all and optional catch-all pages routes, whose final parameter absorbs a variable number of segments. The shared placeholder token and the client-side check live in the new `dynamic-filter-pattern.ts`, while the build-only encoder stays in `create-client-router-filter.ts`. The change only ever turns a soft navigation into a hard navigation, and only when an app route is genuinely more specific than the resolved pages route, so a structural match always agrees with the server's resolution. Apps without dynamic app routes are unaffected because the dynamic filter stays empty. The `pages-to-app-routing` end-to-end suite is restructured around a `fixtures/` directory and gains coverage for the dynamic-segment shadowing, the catch-all and optional catch-all variants, `basePath`, middleware, and a guard that legitimate Pages Router dynamic routes still navigate client-side. The first request to a not-yet-compiled dynamic route can return a transient 404 in development, a pre-existing source of CI flakiness, so the tests now warm the route up with a direct request before driving the browser. fixes #74696
-
Aurora Scharff authored
## What Restructures the dev overlay's **Copy as prompt** body from three loosely-grouped instructions into a seven-step ordered checklist. ## Why Agents tend to do one thing per step. The old step 4 bundled "reload in browser," "set up browser tooling," "check the shell," "re-check siblings," and a before/after capture into a single paragraph, so most of it got skipped. Two other gotchas got their own steps: the empty-shell case (`<Suspense fallback={null}>` around the page body passes the build but renders nothing) and the sibling re-check on shared-code changes. ## How Same checklist shape as the `next-cache-components-adoption` skill's per-route loop, so agents following either workflow apply the same sequence. Used for the in-overlay copy button and the prompt mirrored in `next dev` / `next build` terminal output. -
Aurora Scharff authored
### What? Restructure the 15 Insight error pages (`errors/blocking-prerender-*.mdx`, `errors/instant-*.mdx`) so they orient the reader before diving into fixes, and point at the preview docs for the new 16.3 Instant Navigations guide and APIs. ### Why? These pages are the first thing devs see when they click the doc link from the dev overlay or the build output for a blocking-route / instant-navigation error. Previously they opened straight into the technical explanation with no signal that this error is part of the new 16.3 Instant Navigations feature, and several callouts mixed unrelated concerns (debug flags inside a top-of-page "Good to know", duplicated "Useful links" footers). ### How? - **Intro callout box** added to every page, pointing at the 16.3 launch blog post and the migration guide so newcomers have an anchor before reading the fix. - **`## Verifying the fix` section** added near the bottom, gathering the debug tips (empty-shell warning, `--debug-prerender`, `--debug-build-paths`) that previously sat in a top-of-page "Good to know" callout where they were misplaced. - **Removed** duplicated "Useful links" footers (already cross-linked in body), removed the misplaced top-of-page debug callout. - **Switched all 473 doc links** to `https://preview.nextjs.org/docs/...` so they resolve against the unpublished 16.3 docs (the live nextjs.org docs don't have the Instant Navigations guide or the route-segment-config/instant API reference yet). Split out of #95186 (Copy-as-prompt checklist) which was the same branch but a separate feature. Closes NEXT- <!-- NEXT_JS_LLM_PR -->
-
niketchandivade authored
## What? Fixes a typo in the preview message. - This is page is a preview. + This page is a preview. ## Why? The current text contains a grammatical error that is shown when preview mode is enabled. ## Testing * Verified the updated text renders correctly. * No functional changes.
-
Will Binns-Smith authored
-
- 25 Jun, 2026 5 commits
-
-
next-js-bot[bot] authored
-
Joseph authored
Narrowing down a few confusion items, and adding screenshots for the nav inspector.
-
dan authored
The underlying `agent-browser` was flaky about sessions, which is resolved in https://github.com/vercel-labs/agent-browser/pull/1486. This bumps the skill to use that version. ## Test Plan Verified manually that the dev loop skill now doesn't log me out all the time. I haven't finished the eval run but so far it also looks much more positive: <img width="518" height="99" alt="Screenshot 2026-06-25 at 20 59 46" src="https://github.com/user-attachments/assets/b8f47750-33a1-433c-a086-6f5e8cf4ca00" />
-
next-js-bot[bot] authored
-
next-js-bot[bot] authored
-