- 01 Oct, 2026 25 commits
-
-
next-js-bot[bot] authored
-
Jiwon Choi authored
> [!TIP] > Recommended to review commit by commit with "Hide whitespace" enabled. ### Why? Measure how often humans and agents receive upgrade nudges, start AI upgrades, and finish them. ### How? Record the configured policy, shown nudges, human decisions, AI upgrade runs, CLI results, and agent results through existing Next.js telemetry. Existing telemetry opt-out settings apply. `next upgrade --ai` prepares an upgrade prompt and hands it to an agent. The prompt asks the agent to run `next internal report-ai-upgrade <run-id> success|failure` after completing the upgrade and verification. The run ID links this report to the original upgrade run. Completion reporting depends on the agent following that instruction. A missing report means unknown, not failure. The prompt does not instruct the agent to hide reporting or require it to mention the telemetry submission to the user. The reporting command suppresses the normal telemetry notice. Closes NAR-938
-
Luke Sandberg authored
## What decrease the size of the used key amqfs to 32 bits ## Why Save disk space, compaction will also be a bit faster due to smaller filters which are cheaper to query and merge. All filters keep the same fingerprint size, so their union still uses qfilter's sorted merge. ## Verification - Re-encoding the used keys AMQFs of the final cache of a 31-build vercel-site run (GC on): **61.8 MB → 17.8 MB** (7.96 → 2.29 bytes per key), about 1.5% of the 2.96 GB cache. GC off: 58.9 → 16.9 MB. - `cargo test -p turbo-persistence` and `cargo test -p turbo-tasks-backend --tests` pass. <!-- NEXT_JS_LLM -->
-
Jimmy Miller authored
## What Remove the ability to suspend in flight operations to allow for snapshotting and GC ## Why Exstensive testing with GC has revealed a number of bugs latent in this system * transient task ids are getting serialized as part of operations, this confuses our new stricter task access semantics in fresh sessions * child execution contexts don't properly participate, they both need to suspend when requested to avoid exceeding retries since the task they are waiting for has already suspended, and cannot suspend because that could deadlock the scope_bounded call Instead we just don't suspend. This simplifies a lot, should be a small perf win (purely from not touching the atomics as often) The major risk is that waiting for everything to settle adds latency. To mitigate that: * idle snapshots don't wait for operations to settle, they just bail out instead * other snapshots do wait but log warnings if it takes more than 10s, this should never happen but if it does we will have a smoking gun --------- Co-authored-by:Luke Sandberg <lukesandberg@users.noreply.github.com>
-
Josh Story authored
## Summary Migrate the remaining runtime, middleware, instrumentation, telemetry, logging, and local-process deployment exclusions to `@force-gate` and remove their `skipDeployment`/`skipped` control flow. This stack is rooted at canary and independent of #98430; it preserves deployment coverage already enabled on canary. The sub-shell-generation-middleware migration is deferred to #98155 while its adapter deployment cache-upgrade failure is investigated. Its test file stays byte-for-byte identical to this PR's base, preserving the existing adapter exception and assertions. Adds the static nodeMiddleware condition and its unit coverage. The custom-server and OpenTelemetry files covered by #98994 are omitted. ## Verification - Compared test registration against canary in development, production, and deploy modes with test bodies and hooks disabled (Webpack). Existing active deployment assertions are preserved; development/production test selection is unchanged apart from replacing empty skip placeholders with skipped real tests. - Gate unit tests: 79 passed. - TypeScript, Prettier, and ESLint checks passed. Registration probes use a default config for lazy predicates and do not exercise fixture setup or test bodies. - Verified the deferred suite is absent from this PR diff and the upper-layer patches are unchanged. Full bootstrap passed on the combined canary-based stack after the final restack; deployment execution remains for CI. <!-- NEXT_JS_LLM --> --------- Co-authored-by:Jamiboy Mohammad <jamiboym@gmail.com>
-
Josh Story authored
## Summary Migrate the remaining image, font, CSS, and other asset integration deployment exclusions to `@force-gate` and remove their `skipDeployment`/`skipped` control flow. This stack is rooted at canary and independent of #98430; it preserves deployment coverage already enabled on canary. Shared image-optimizer exclusions are placed on the calling test suites because gate pragmas are transformed only in test files. Existing bundler guards in the shared helper are preserved. ## Verification - Compared test registration against canary in development, production, and deploy modes with test bodies and hooks disabled. Existing active deployment assertions are preserved; development/production test selection is unchanged apart from replacing empty skip placeholders with skipped real tests. Checked both Webpack and Turbopack. - TypeScript, Prettier, and ESLint checks passed. Registration probes use a default config for lazy predicates and do not exercise fixture setup or test bodies. - Full bootstrap attempted, but the local native SWC build fails at `cargo metadata` with `EPERM`; full integration/deployment execution remains for CI. <!-- NEXT_JS_LLM --> --------- Co-authored-by:Jamiboy Mohammad <jamiboym@gmail.com>
-
Josh Story authored
## Summary Migrate the remaining Pages Router, API route, i18n, custom route, rewrite, redirect, and error-page deployment exclusions to `@force-gate` and remove their `skipDeployment`/`skipped` control flow. This stack is rooted at canary and independent of #98430; it preserves deployment coverage already enabled on canary. The custom-app-render and client-max-body-size files covered by #98994 and #98911 are omitted. ## Verification - Compared test registration against canary in development, production, and deploy modes with test bodies and hooks disabled (Webpack). Existing active deployment assertions are preserved; development/production test selection is unchanged apart from replacing empty skip placeholders with skipped real tests. - TypeScript, Prettier, and ESLint checks passed. Registration probes use a default config for lazy predicates and do not exercise fixture setup or test bodies. - Full bootstrap attempted, but the local native SWC build fails at `cargo metadata` with `EPERM`; full integration/deployment execution remains for CI. <!-- NEXT_JS_LLM --> --------- Co-authored-by:Jamiboy Mohammad <jamiboym@gmail.com>
-
Janka Uryga authored
#97393 was a stale merge. we've renamed `unstable_ensureStatic` to `ensureStatic` in the meantime.
-
Josh Story authored
## Summary Let Cache Components applications configure parameter matching separately from build-time prerenders. `generateStaticParams` continues to supply concrete examples; the new `unstable_paramMatching` and `unstable_generateParamMatching` exports control whether a novel value returns 404, blocks on generation, receives a fallback shell, or stays dynamic. With `cacheComponents: true`: ```ts // app/[lang]/layout.tsx export const unstable_paramMatching = { lang: 'not-found' } // app/[lang]/catalog/[top]/items/[bottom]/page.tsx export const unstable_paramMatching = { top: 'blocking', bottom: 'fallback', } export function generateStaticParams() { return [{ lang: 'en', top: 't1', bottom: 'b1' }] } ``` This prerenders `/en/catalog/t1/items/b1`. A language other than `en` returns 404. On a cache miss, `/en/catalog/t2/items/b2` blocks on generation, while `/en/catalog/t1/items/b2` can receive fallback UI immediately. `dynamic` keeps a parameter out of prerenders. Alternatively, `export async function unstable_generateParamMatching()` returns the same partial object. It takes no arguments and runs independently of `generateStaticParams`, once per module within each route evaluation. Public `use cache` helpers are supported. No additional configuration flag is required, and `ParamMatching` from `next` provides optional strict typing. Policies merge from layouts toward pages, with descendants overriding only supplied keys: - A module can configure parameters at or above its segment. Explicit modes must follow URL order: `not-found → blocking → fallback → dynamic`. - A `not-found` parameter requires every preceding parameter to be explicitly `not-found`, too. Every route or parallel branch sharing that parameter definition must agree. - Unconfigured parameters retain inference. An unconfigured tail without examples stays dynamic; explicitly dynamic parameters cannot be included in prerenders. - Shell validation follows explicit fallback boundaries instead of silently replacing them with blocking behavior; `instant = false` opts out of that validation. With these APIs, `output: 'export'` requires all parameters to be explicitly `not-found`. - `unstable_ensureStatic = 'navigation'` allows `blocking` and `not-found`, but rejects effective `fallback` and `dynamic` policies in both builds and development. It still requires complete `generateStaticParams` examples and fully static server output, even with `instant = false`. Supporting fallbacks for fully static navigations is deferred to a follow-up. Runtime admission, cache regeneration, and client route discovery preserve those boundaries. Adapter output uses the existing routing contract, including prefixed query keys for root fallbacks. ## Verification - Unit and integration coverage for policy merging, validation, generated types, shell selection, blocking/fallback behavior, navigation, parameter encoding, revalidation, and root-cache isolation. - Real fixtures cover direct, inherited, generated, and parallel-slot combinations with `ensureStatic`, including valid overrides, missing examples, dynamic-data errors, and preserved closed prefixes. Both bundlers and both Partial Prefetching settings are covered in build/start and dev. - Existing `ensure-static-navigation` build/dev validation and `ensure-static-navigation-runtime` production suites pass without changing their fixtures or validation implementation. - Routing, generator, and root-parameter fixtures have passed real adapter deployment tests in both ordinary and Cache Components jobs. Local adapter-output assertions separately check the root-fallback query contract. - [ ] Confirm cold open-suffix requests and revalidation in a deployment known to use grouped functions, and verify rejected prefixes do not invoke a function or consult ISR. Passing HTTP assertions alone does not establish these properties. <!-- NEXT_JS_LLM --> -
Swarnava Sengupta authored
Updates the Security section in packages/next/README.md to send bug bounty reports to [Vercel’s HackerOne program](https://hackerone.com/vercel) instead of the deprecated email address. Expires: 2026-09-30
-
Aurora Scharff authored
## Summary - document the `ensureStatic` route segment config for Next.js 16.4, including the `shell`, `prefetch`, and `navigation` levels, their prerequisites, and how requirements compose across pages, layouts, and parallel routes - add the validation experience for `ensureStatic = 'navigation'`, including direct Learn more links, route, metadata, and viewport guidance, cause-oriented fix cards, and build diagnostics - add focused coverage for route content, metadata, viewport, cache lifetimes, server promises, and dynamic parameters This PR documents the final public `ensureStatic` name. The framework export change lands separately. Short-lived cache failures use the normal runtime-data or uncached-data message selected by the existing validator. They do not have a separate cache-lifetime classifier. ## Design and capture review - Route: [runtime data](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/125-ensure-static-cookies), [uncached data](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/128-ensure-static-uncached), [connection](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/129-ensure-static-connection) - Metadata: [runtime data](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/130-ensure-static-metadata-runtime), [uncached data](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/131-ensure-static-metadata-dynamic) - Viewport: [runtime data](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/132-ensure-static-viewport-runtime), [uncached data](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/133-ensure-static-viewport-dynamic) - Cache lifetimes: [short expire](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/134-ensure-static-short-expire), [short stale](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/135-ensure-static-short-stale), [`revalidate: 0`](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/136-ensure-static-zero-revalidate) - Other validation: [`instant = false`](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/137-ensure-static-instant-false), [Suspense](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/138-ensure-static-suspense), [server promise](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/139-ensure-static-client-promise), [missing parameters](https://error-messages-overhaul-ibsl.labs.vercel.dev/scenario/144-ensure-static-missing-params) ## Verification - `pnpm --filter=next build` - Development E2E: 27 tests and 19 snapshots passed - Production E2E: 27 tests and 20 snapshots passed - Focused overlay message and fix-card unit suites: 137 tests passed - Error-overhaul production build: 124 routes generated - Captured build output for 22 Ensure Static scenarios and dev overlays for all applicable insight cases - Prettier, ESLint, and `git diff --check` passed for the changed framework files - Documentation link validation passed <!-- NEXT_JS_LLM --> --------- Co-authored-by:
Joseph Chamochumbi <joseph.chamochumbi@vercel.com>
-
Tobias Koppers authored
### What? `experimental.turbopackMangleExportNames` is now on by default for production builds (`next build`) in both stable and canary releases. It stays off by default in development (`next dev`), and an explicit `true`/`false` in `next.config` still wins in either mode. ### Why? Mangling export names shortens the keys modules use to link to each other, which reduces bundle size. Until now it was only on by default in canary builds, because stable releases pinned it to `false`. ### How? Turbopack already chooses the default from the build mode in `NextConfig::turbopack_mangle_export_names`: if the option is unset, it's enabled everywhere except development, and an explicit value overrides that. The only thing keeping it off on stable was the JS default config, which forced `false` when `isStableBuild()` was true. That pin is removed, so the default stays unset on every release channel and the Rust side decides from the build mode. There's no change to Rust or to the mangling itself. The JSDoc on the config type and the Turbopack config docs table now say the default is `false` in dev and `true` for builds. ### Verification - Prettier and ESLint pass on the changed files. - `pnpm --filter=next types`: only 3 `TS2307` errors, all for missing generated `dist` modules (`@next/font`, `@vercel/og`) in an unbootstrapped checkout. The same errors show up without this change, and none are in `config-shared.ts`. - Not run: a bootstrapped integration build. CI covers the build-mode behavior end to end. <!-- NEXT_JS_LLM --> <!-- fleet dfbd6860-64e0-4a98-ac6f-220e8131fee2 --> --------- Co-authored-by:
vercel-fleet-prod[bot] <318278635+vercel-fleet-prod[bot]@users.noreply.github.com> Co-authored-by:
Tobias Koppers <1365881+sokra@users.noreply.github.com>
-
Jiwon Choi authored
Stacked on #99471. ### Why? Upgrading React or ESLint can cause peer dependency conflicts. After restoring compatible versions, retrying the upgrade can bump them again. ### How? Add `--skip-react-upgrade` and `--skip-eslint-upgrade` to skip the corresponding dependency updates and codemods. The React flag also preserves types and overrides. Add troubleshooting checklists to the version guides and point upgrade agents to them. Required migrations still need to be completed before verification; if Next.js already matches the target, run the interrupted migrations individually.
-
Jiwon Choi authored
### Why? Running a codemod through `npx` can select npm even when the app declares Yarn, pnpm, or Bun, causing dependency installation to fail. ### How? Read `package.json#packageManager` before `npm_config_user_agent`. Keep the existing user-agent, lockfile, and npm fallbacks unchanged. Add shared checks for installation and recovery after an interrupted upgrade.
-
Will Binns-Smith authored
Use the App Router error boundary and `next-themes` provider instead of custom client-side equivalents. Persist environment, file type, and debounced file search filters in URL search parameters so analyzer views can be shared. Remove the incompatible `next start` script from the statically exported app. ## What? - Add an App Router `error.tsx` boundary with segment reset support - Replace the inline theme bootstrap script with `next-themes` - Store analyzer filters and file search in URL search parameters - Remove the static export's `next start` script ## Why? Use the framework's existing error, theme, and navigation integration while keeping the analyzer's static export deployment model. ## How was it tested? - `pnpm --filter=@next/bundle-analyzer-ui exec tsc --noEmit` - `pnpm --filter=@next/bundle-analyzer-ui lint` - `pnpm --filter=@next/bundle-analyzer-ui build` <!-- NEXT_JS_LLM --> Co-authored-by:
vercel-fleet-prod[bot] <318278635+vercel-fleet-prod[bot]@users.noreply.github.com> Co-authored-by:
Will Binns-Smith <755844+wbinnssmith@users.noreply.github.com>
-
Josh Story authored
## Summary Continue decoupling App Router rendering from incoming request objects by capturing HMR state at the handler boundary and passing it through DevRenderContext. The context contains the HMR refresh hash and shared server-component HMR cache, and is constructed only under `next dev`. Production omits this state, and prerendering does not consume it. Rendering no longer reads the hash from request metadata. Existing upstream metadata transport and cache invalidation behavior are preserved; status and debugging callbacks are left for separate migrations. ## Verification - `pnpm test-dev-turbo test/development/app-dir/use-cache-dev/use-cache-dev.test.ts` (7 tests) - `pnpm test-dev-webpack test/development/app-dir/use-cache-dev/use-cache-dev.test.ts` (7 tests) - `pnpm test-start-webpack test/e2e/app-dir/app/standalone.test.ts` (isolated production build and Edge coverage) <!-- NEXT_JS_LLM -->
-
Josh Story authored
## Summary Continue decoupling App Router rendering from incoming request objects by keeping the request inside handler-owned instrumentation callbacks. Renderers now report the error and its context without forwarding the request. The public onRequestError request descriptor and error-handling behavior are unchanged. Remove unused callback plumbing and extend the existing Node and Edge instrumentation coverage to check query strings and custom headers. ## Verification - Existing server-action instrumentation tests passed in development and production with both Turbopack and Webpack. - `pnpm test-start-turbo test/e2e/on-request-error/isr/isr.test.ts` - `pnpm test-start-webpack test/e2e/app-dir/app/standalone.test.ts` (isolated Edge build) <!-- NEXT_JS_LLM -->
-
Niklas Mischkulnig authored
<!-- Thanks for opening a PR! Your contribution is much appreciated. To make sure your PR is handled as smoothly as possible we request that you follow the checklist sections below. Choose the right checklist for the change(s) that you're making: ## For Contributors We encourage you to use AI to assist you in researching, creating, and reviewing changes. However, you must review and deeply understand the contributions you are making. For this reason, **pull request descriptions from external contributors must be written by a human**. ### Improving Documentation - Run `pnpm prettier-fix` to fix formatting issues before opening the PR. - Read the Docs Contribution Guide to ensure your contribution follows the docs guidelines: https://nextjs.org/docs/community/contribution-guide ### Fixing a bug - Related issues linked using `fixes #number` - Tests added. See: https://github.com/vercel/next.js/blob/canary/contributing/core/testing.md#writing-tests-for-nextjs - Errors have a helpful link attached, see https://github.com/vercel/next.js/blob/canary/contributing.md ### Adding a feature - Implements an existing feature request or RFC. Make sure the feature request has been accepted for implementation before opening a PR. (A discussion must be opened, see https://github.com/vercel/next.js/discussions/new?category=ideas) - Related issues/discussions are linked using `fixes #number` - e2e tests added (https://github.com/vercel/next.js/blob/canary/contributing/core/testing.md#writing-tests-for-nextjs) - Documentation added - Telemetry added. In case of a feature if it's used or not. - Errors have a helpful link attached, see https://github.com/vercel/next.js/blob/canary/contributing.md ### Signed commits - This repository requires verified commit signatures on protected branches. - If this pull request is blocked for unsigned commits, re-sign the commits and force-push the branch. - A `Signed-off-by` line in the commit message is not enough. ## For Maintainers - Minimal description (aim for explaining to someone not on the team to understand the PR) - When linking to a Slack thread, you might want to share details of the conclusion - Link both the Linear (Fixes NEXT-xxx) and the GitHub issues - Add review comments if necessary to explain to the reviewer the logic behind a change ### What? ### Why? ### How? Closes NEXT- Fixes # -->
-
Jamiboy Mohammad authored
## Summary Enable deployment coverage for seven fixture suites while preserving dependency declarations, test bodies, assertions, and unrelated mode restrictions. - Set `ENABLE_EXPERIMENTAL_COREPACK: '1'` for `geist-font`, `gssp-redirect`, `rewrite-with-browser-history`, `router-hash-navigation`, and `scroll-back-restoration` so Vercel installs with the pnpm version declared by the generated `packageManager` field. These fixtures do not require committed pnpm lockfiles or `npm_config_frozen_lockfile` overrides. - Use fixture-relative tarball dependencies for `self-importing-package` and `typeof-window` so deployment installation does not refer to paths on the GitHub Actions runner. - Apply the Corepack setting to both suites using the shared source-map fixture. Retain their deployment exclusions because their assertions require runtime logs or a local Node.js inspector; they are not counted as deployment passes. ## Verification Ran focused `pnpm test-deploy-turbo --runTestsByPath` tests with `VERCEL_TURBOPACK_TEST_TEAM=vtest314-next-e2e-tests`, `NEXT_E2E_TEST_TIMEOUT=240000`, and `NEXT_TEST_VERSION=canary` (resolved to `16.4.0-canary.53`): - All five affected deploy-enabled suites passed in normal mode: **26/26 tests**. - With `__NEXT_CACHE_COMPONENTS=true` and `__NEXT_EXPERIMENTAL_CACHED_NAVIGATIONS=true`, the same five suites passed: **26/26 tests**. Remote build logs confirmed both features were enabled. - All ten deployments detected the Corepack flag and `pnpm@10.33.0`, installed successfully without source lockfiles, and executed the browser/HTTP assertions. - The unchanged `self-importing-package` and `typeof-window` tarball fixtures passed in both modes in the preceding validation (two tests per mode); this update does not touch them. - Prettier, ESLint, and `git diff --check` passed. The Next.js watch build completed with zero type errors. The full pre-push build completed 17/18 tasks; native SWC remains blocked because Cargo is unavailable locally. These focused runs validate the fixture changes against published canary, not all CI checks on the updated PR head. <!-- NEXT_JS_LLM -->
-
next-js-bot[bot] authored
-
Niklas Mischkulnig authored
https://github.com/vercel/next.js/pull/89325 changed the build id to be a hardcoded constant if `config.deploymentId` is set (= skew protection is enabled) This was done as an optimization (more deterministic build outputs) Instead, always use generateBuildId if set explicitly and only use the new behavior if `generateBuildId` isn't set. Closes https://github.com/vercel/next.js/issues/92154 --------- Co-authored-by:
vercel-fleet-prod[bot] <318278635+vercel-fleet-prod[bot]@users.noreply.github.com>
-
Luka Hartwig authored
### What? Sets the correct Next.js library version for Otel traces. ### Why? The tracer instance was configured with a hard coded "0.0.1" library version. This is confusing since it makes it look like trace spans were produced by `next@0.0.1`. Instead I'm setting the library version to the current Next.js version. ### How? I'm using the `__NEXT_VERSION` env var, which gets replaced by the package.json version during build.
-
Zakharii (Zakhar) Melnyk authored
## What Propagate a repeated revalidation of the same tag and profile from an `after()` callback when its timestamp advances during the request. Add an E2E test that refills the tagged cache entry between the two calls. ## Why `revalidateTag()` updates `revalidatedAt` on an existing pending item in place. `withExecuteRevalidates()` currently saves a shallow copy of that item, and `diffRevalidationState()` only compares tag and profile. The second call is therefore dropped, leaving an entry refilled between calls stale. The reproduction uses the built-in cache handler and returns `{ source: 2, cached: 1 }` where `{ source: 2, cached: 2 }` is expected. ## How Snapshot the pending tag items and compare `revalidatedAt` for matching tag/profile pairs. The regression test holds the `after()` callback until a request has refilled the cache, then checks that the final read is fresh. The production reproduction passed with the equivalent runtime change on `next@16.4.0-canary.53`. `git diff --check` passed. The full monorepo E2E suite has not been run locally. Fixes #99498 -
Janka Uryga authored
Closes NAR-933 Drop the `unstable_` prefix from `ensureStatic`, because we're reasonably confident that the API shape won't change.
-
Benjamin Woodruff authored
This has been on in canary for over a month with zero issues, [nobody within vercel has opted out of it](https://github.com/search?q=org%3Avercel+turbopackSharedRuntime&type=code), let's turn it on-by-default for 16.4! Leaving the flag around for now in case anybody needs to turn it off, but the current plan is to delete this flag (and the non-shared runtime codepath) in a future release.
-
- 30 Sep, 2026 15 commits
-
-
next-js-bot[bot] authored
-
Sebastian "Sebbie" Silbermann authored
https://github.com/vercel/next-js-mirror/pull/207 --------- Co-authored-by:
Josh Story <story@hey.com> Co-authored-by:
Josh Story <gnoff@storyposted.com>
-
Sebastian "Sebbie" Silbermann authored
https://github.com/vercel/next-js-mirror/pull/189Co-authored-by:
Josh Story <story@hey.com>
-
Sebastian "Sebbie" Silbermann authored
https://github.com/vercel/next-js-mirror/pull/201Co-authored-by:
Karim Rahal <karim@karimrahal.com>
-
Sebastian "Sebbie" Silbermann authored
https://github.com/vercel/next-js-mirror/pull/204Co-authored-by:
Hendrik Liebau <mail@hendrik-liebau.de>
-
Sebastian "Sebbie" Silbermann authored
https://github.com/vercel/next-js-mirror/pull/182Co-authored-by:
David Ilie <david@davidilie.com>
-
Sebastian "Sebbie" Silbermann authored
https://github.com/vercel/next-js-mirror/pull/158Co-authored-by:
Steven <steven@ceriously.com>
-
Luke Sandberg authored
## Summary Stacked on #99268. An intermediate merge merged every file above a shard's bottom run. Small commits, such as the snapshots of a development session, write small files: they push a shard over `max_files_above_bottom` quickly, and the merge then rewrote the large files of earlier build-sized commits along with them. Like the size ratio of RocksDB's universal compaction, an intermediate merge now takes the newest file, and then each next older one while it's at most `size_ratio_percent` (default 100%) larger than the files taken so far, with at least the two newest. Small files are merged with each other; a large file only joins once the newer files add up to about its size. Because the merge always takes the newest files, its output can still be placed after all existing files, and the older files it leaves out stay older. The trigger (`max_files_above_bottom`) and bottom merges are unchanged. ## Verification - A selector simulation where every build-like commit is followed by ten tiny commits: write amplification **6.43 → 1.98**, with the same maximum of 5 files per shard. The build-only simulation is unchanged (space amplification 1.34, write amplification 1.25). - Selector tests: small files after a large one merge without it; similar sizes all merge; steeply growing sizes merge at least the two newest; the ratio boundary. - `cargo test -p turbo-persistence`, `cargo test -p turbo-tasks-backend --tests`, and the Miri job's command for turbo-persistence pass. - End to end, 31 consecutive vercel-site builds with GC on (about half the TaskData commits are small), against #99268: | | #99268 | this PR | |---|---|---| | compaction bytes written per byte committed | 1.61× (12.52 GB) | **1.45×** (10.99 GB) | | …by intermediate merges | 3.69 GB | **2.34 GB (−37%)** | | …by bottom merges | 8.82 GB | 8.66 GB | | cache size, final / max | 2.94 / 3.24 GB | 2.92 / 3.13 GB | Intermediate merges run more often (126 vs 104 TaskData jobs) but each rewrites less. A development session, with many small snapshot commits, should benefit more. <!-- NEXT_JS_LLM -->
-
Luke Sandberg authored
## Summary Turbo-tasks GC can only shrink the persistent cache if compaction merges each tombstone with the data it deletes. Canary's compaction algorithm leaves many tombstones behind so with GC enabled the cache is only 5% smaller than without it. This PR replaces compaction with a space-amplification policy on hash shards, modeled on RocksDB universal compaction and Cassandra UCS. With GC enabled, the cache stays ~flat. ### Cache size over 31 consecutive vercel-site builds Same canary base and the same commits in every arm. The size is measured after each build, including its compaction on shutdown. | build | canary, GC off | canary, GC on | this PR, GC off | this PR, GC on | |---|---|---|---|---| | 1 | 2.59 GB | 2.60 GB | 2.63 GB | 2.64 GB | | 10 | 3.15 GB | 3.12 GB | 3.24 GB | 3.07 GB | | 20 | 3.42 GB | 2.92 GB | 3.20 GB | 3.06 GB | | 22 | 4.57 GB | 3.87 GB | 3.55 GB | 3.18 GB | | 31 | **4.19 GB** | **3.96 GB** | **3.75 GB** | **3.00 GB** | | growth, build 1 → 31 | +1.60 GB | +1.36 GB | +1.12 GB | **+0.36 GB** | - **GC saves 5% with canary's compaction and 20% with this one** (final size, GC on vs. off). A full compaction of the final cache puts the live data at about 2.8 GB, so this PR with GC is within about 8% of it. - **Build 22 rewrites a lot of task data.** Canary grows by about 1 GB at that build in both modes and never recovers the space. This PR absorbs it and merges the superseded versions away within a few builds. ### Why canary's compaction can't keep up with GC - **Old versions and tombstones don't reliably meet the data they supersede.** Canary wrote recently read keys and the rest into separate warm and cold files and never merged across them. - **Its merge triggers couldn't see the duplication.** There were duplication-byte thresholds and `max_merge_bytes`. Most of the garbage is superseded versions of a relatively small set of large, frequently rewritten `TaskData` values, so those triggers missed it and produced inefficient merge groups. ### New Compaction policy - **Shards:** each family's key space is split into power-of-two hash-prefix shards. The count starts at `FamilyConfig::initial_shard_bits` (TaskData 8 shards, TaskMeta 4, TaskCache and Infra 1) and then follows the family's size, targeting about 256 MB per shard. - It doubles once shards exceed 1.5× the target and halves below 0.5×, so a size near a boundary doesn't flip it. - Commits and merges split SST files at shard boundaries, so each shard is compacted on its own. Files written with an old count are re-split by their shard's next merge, not by a rewrite. - Write batches still buffer a family in at most 4 collectors (64 MB each) as on canary, split by a prefix of the shard bits, so their memory doesn't grow with the shard count. Families with more than 4 shards write proportionally smaller commit files. - **Bottom merge:** when the files above a shard's bottom run exceed 50% of it, the whole shard is merged into a new bottom run. Tombstones count as the entries they delete. This is the merge that reliably drops superseded entries and tombstones. - This is analogous to a 'full compaction' but it targets a single shard and is targeting files covering typically <0.5G of data. - **Intermediate merge:** more than 4 files above the bottom run are merged together, which bounds the number of files a lookup consults. - **Budget:** bottom merges per family stop after rewriting 3× the family's fresh (not yet compacted) bytes, so cost follows new writes and shards spread over multiple compactions. 3× is how often a written byte is copied in steady state: a bottom merge at the 50% trigger rewrites 1.5× the bytes above the bottom run for every 0.5× of them. - **Hard cap:** a shard over twice the trigger (100% space amplification), or with more than twice `max_files_above_bottom` files above its bottom run, is bottom-merged even when the budget is spent. This bounds how far a shard can fall behind, e.g. after a GC purge: its tombstones count as the entries they delete, but they're small and give little budget. - **Tiny families:** a 1 MB floor keeps families like Infra, which rewrite ~everything on every commit, out of bottom merges. - **Ranking:** jobs are ranked across families by how far over their trigger they are, so intermediate merges aren't starved. **Hot/cold split, only in bottom merges.** A bottom merge writes recently read keys and the rest into separate hot and cold files. It always includes every file of its shard, hot and cold, so this does not interfere with deduplication. Used-key tracking is unchanged from canary: each commit records the keys it read in its meta file, and compaction uses the union of all live meta files. The marks are never copied into meta files written by compaction, so they expire with the commit that recorded them. - This Hot/cold split is less aggressive than canary. See below for Read locality impacts. **Reads:** a lookup only consults the SST files of its key's shard. A per-family index keeps each shard's hash ranges densely, newest first, so lookups scan them sequentially. ## Verification - The cache size benchmark above: vercel-site, 31 commits, GC with `rootTtlMs=0`. Earlier runs on the same base reproduce the GC-on results (canary 3.93–3.99 GB, this PR 3.00–3.05 GB). - **Read locality**, measured with temporary instrumentation that isn't part of this PR. It counts the distinct SST blocks each build reads, and the distinct 128 KiB file regions they touch as an estimate of squashfs decompression. Same base, same builds, builds 2–31, GC on: | | canary | no split | this PR (bottom split) | |---|---|---|---| | distinct blocks read per build, median | 1.35 GB | 1.80 GB | 1.55 GB | | 128 KiB regions touched per build, median / total | 2.42 / 64.2 GB | — | 2.63 / 68.8 GB | The bottom split recovers most of the locality that dropping the split cost. Light builds still read about a third more than on canary, because a shard's hot set only refreshes at its bottom merge. Splitting intermediate merges as well gained about 1%, so it isn't included. - **Lookup microbenchmark**, 16 shards × 20 uncompacted commits (320 SSTs), alternating before and after: `get` hit −14%, miss −28%, `batch_get` −5%. - `cargo test -p turbo-persistence` and `cargo test -p turbo-tasks-backend --tests` pass. The selector has unit tests, and a simulation whose sweep supports the 50% threshold. <!-- NEXT_JS_LLM -->
-
Jamiboy Mohammad authored
## Summary Re-enable `gssp-redirect-base-path` deployment coverage with Corepack so installation uses the fixture's declared pnpm version. A redirect with `basePath: false` lands at `/404`, outside the application's `/docs` base path. Vercel may serve its own 404 page there. Check the destination URL and the browser's actual HTTP 404 navigation response in every mode, while retaining the custom-page text assertion for self-hosted modes. This follows the existing base-path error-page tests; the other redirect assertions remain unchanged. Follow-up to #99192. ## Verification - All 22 tests passed locally in production Turbopack mode, both normally and with `__NEXT_CACHE_COMPONENTS=true` and `__NEXT_EXPERIMENTAL_CACHED_NAVIGATIONS=true`. - Normal and cache-components deployment manifests select the suite without excluding individual cases. Deployment execution awaits CI. - The repository TypeScript check is blocked by the worktree's shared `node_modules` resolving an older Next.js build in another checkout (missing upgrade and static-config exports); it reported no errors in the changed test. <!-- NEXT_JS_LLM -->
-
Orkhan Rzazade authored
### What? An on-demand ISR page (dynamic route with `revalidate`, not prerendered at build) loses its `Cache-Control` header when the process serving it from the cache is not the one that rendered it. That happens with several instances sharing a cache handler, and with a single instance after a restart, including the default file-system cache. After a restart it is also served `STALE` on every request, because `calculateRevalidate` falls back to its 1-second default. A running instance can also retain an outdated lifetime after another instance replaces the shared entry with a shorter-lived result. ### Why? `IncrementalCache.get()` takes an entry's lifetime only from `SharedCacheControls`: an in-memory map that `IncrementalCache.set()` fills in the same process, with the prerender manifest as fallback. An on-demand page of a dynamic route is not in the manifest. So any other process has no lifetime for the entry, sends no header, and treats the entry as stale after one second. ### How? - `CacheHandlerValue` gets an optional `cacheControl`. Handlers already receive it as `ctx.cacheControl` in `set()`; now they can return it from `get()`. - In `IncrementalCache.get()`, the lifetime returned with the entry takes precedence over the in-memory map and is remembered for the route, so both the response header and revalidation use the entry's current lifetime. - The default `FileSystemCache` stores `cacheControl` in the `.meta` file of App Router pages and routes, and returns it. Pages Router ISR is not covered: the file-system cache writes no `.meta` for it. I kept that out of this PR. ### Tests `test/production/app-dir/isr-cache-control-restart` contains three separate on-disk fixture apps: - `default`: the built-in file-system cache, with `cacheComponents: false`. - `custom`: a persistent custom cache handler that stores `ctx.cacheControl`, with `cacheComponents: false`. - `cache-components`: `cacheLife` with a persistent custom route-cache handler, covering the same on-demand route-cache path with Cache Components enabled. The three restart cases render an on-demand page, age the entry beyond two seconds, and restart the server without rebuilding. The first response after restart must retain the original `Cache-Control` and report `HIT`; a repeat request must also report `HIT`. Aging past the broken one-second fallback makes the freshness assertion meaningful. A fourth test keeps servers A and B running from the same build with one shared cache handler and a controllable service. A first caches a success with a one-hour lifetime. After direct eviction from the shared store, B caches an error result with a two-second lifetime. A must use the replacement entry's shorter `Cache-Control` and report `STALE` after two seconds. Direct eviction avoids relying on the fixture handler's no-op tag invalidation. All four tests pass locally with Webpack and with Turbopack under `__NEXT_CACHE_COMPONENTS=true`. The new two-server test catches the previous implementation retaining A's one-hour lifetime. The three restart cases were also verified to fail on the original PR base (`1cc0f661`) with `STALE` instead of `HIT`. Prettier, ESLint, the root and Next.js type checks, and the full repository build pass. The local validation above ran on `7adcd57b`, before the subsequent merge of `canary` (`d093eb45`). That merge preserved the changed incremental-cache files and fixtures. Fixes #99288 --------- Co-authored-by:
orzazade <orzazade@users.noreply.github.com>
-
Josh Story authored
## Summary Allow the internal blur quality directly during development image request validation instead of appending it to `images.qualities`. This preserves quality validation while removing another runtime config mutation as part of supporting immutable configuration in Vercel adapter deployments. ## Verification Existing HTTP tests in `test/e2e/image-optimizer/sharp.test.ts` cover blur SVG responses, allowed image requests, and rejection of unconfigured qualities. The four selected cases pass in Turbopack development, Webpack development, and Webpack production modes. <!-- NEXT_JS_LLM -->
-
Josh Story authored
## Summary Remove `NEXT_PRIVATE_EXPERIMENTAL_CACHE_COMPONENTS` to simplify how deployment tests enable Cache Components. Capture `__NEXT_CACHE_COMPONENTS` in the generated test config instead, preserving deployment coverage and explicit config overrides without a separate environment-variable alias. ## Verification - `pnpm test-unit test/unit/next-deploy-failure/next-deploy-failure.test.ts --runTestsByPath --runInBand` — 19 tests passed. - Loaded nine generated configs in fresh processes: enabled, disabled, and unset test flags crossed with enabled, disabled, and unset `cacheComponents` config. Flag propagation and explicit config overrides were preserved. - Live Vercel deployment tests passed against the preview package for `654cad9519ce5af015d3cdbe23dd8ed1538662b8`, using Turbopack and `__NEXT_CACHE_COMPONENTS=true`: `not-found-non-document-dynamic` (4 tests, inherited flag) and `prefetch-inlining-no-cache-components` (3 tests, explicit opt-out). The inherited-flag build enabled Cache Components and produced partial prerenders. <!-- NEXT_JS_LLM -->
-
Josh Story authored
## 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 -->
-
Hendrik Liebau authored
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.
-