mirror of
https://github.com/cloudflare/vinext.git
synced 2026-09-14 19:04:59 +08:00
vinext@1.0.0-beta.2
5 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Jerry Zhao
|
1860632d87 |
refactor: migrate path handling from node:path to pathslash (#2502)
* refactor: migrate the CLI cluster path handling to pathslash * refactor(routing): migrate the routing cluster path handling to pathslash * refactor(plugins): migrate the plugins cluster path handling to pathslash * refactor(server): migrate the server cluster path handling to pathslash * refactor(build): migrate the build cluster path handling to pathslash * refactor(core): migrate entries, config, and shared utils to pathslash * refactor(plugin): migrate index.ts path handling to pathslash * refactor: finish the pathslash migration — test sweep, stragglers, lint guard * refactor: address final-review findings on the pathslash migration * docs(agents): document the pathslash path-handling convention * fix(build): emit inlined dependencies outside dist/node_modules * fix: normalize wire- and bundler-shaped ids unconditionally * docs(agents): document the unconditional-normalization exception * fix: gate Windows-shaped path fixtures to Windows * docs(agents): gate Windows-shaped fixtures with runIf * refactor: migrate upstream prerender path discovery to pathslash * fix(create-vinext-app): bundle pathslash into the scaffolder dist --------- Co-authored-by: James <james@eli.cx> |
||
|
Steve Faulkner
|
54497d42e7 |
fix: dev-server polish batch — HTML charset, client global polyfill, trailingSlash image endpoint (#2512)
* fix(pages): send charset=utf-8 on HTML Content-Type Next.js serves every HTML response (SSR and prerendered) with `Content-Type: text/html; charset=utf-8`, but vinext's Pages Router paths sent a bare `text/html`. Without the header charset — and without an early <meta charset> in the page — Chromium falls back to windows-1252, so non-ASCII content renders as mojibake (e.g. nbsp as 'Â ') and the resulting DOM diverges from the hydrated tree, triggering the full-screen hydration error overlay in dev. Reproduced against nextjs-notion-starter-kit; baseline `next dev`/`next start` both send the charset. Append `; charset=utf-8` on every Pages Router HTML producer (the App Router paths already send it): - pages-page-response.ts — page render response headers + gSSP header merge - dev-server.ts — streaming SSR, ISR HIT/STALE, and static-HTML dev responses - pages-page-data.ts — ISR cache HIT/STALE responses in prod - pages-request-pipeline.ts — the `defaultContentType` a buffering adapter applies when a render response carries no Content-Type - static-file-cache.ts — `.html` static files (prerendered pages), matching Next.js static serving Compression negotiation is unaffected: COMPRESSIBLE_TYPES matching splits the media type on ";" before lookup. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(client): polyfill `global` in browser bundles Next.js exposes the Node-style `global` alias to client code: webpack via its `node.global` runtime shim, Turbopack by compile-time rewriting the free `global` identifier to its globalThis shortcut and folding `typeof global` to "object" (turbopack-ecmascript references). vinext provided nothing, so any client dependency that reads `global` (e.g. use-dark-mode via nextjs-notion-starter-kit) threw `ReferenceError: global is not defined` after hydration. Add a `vinext:client-global-define` plugin that scopes `define: { global: "globalThis" }` to the client environment: - builds statically rewrite free `global` references (Turbopack-style) - dev injects `"global": globalThis` into the client runtime defines (/@vite/env), assigning `globalThis.global` before user code runs (webpack-style) - the same define is layered into the client dep optimizer (rolldownOptions.transform.define / esbuildOptions.define) because pre-bundled deps bypass the plugin transform pipeline `typeof global` evaluates to "object" in the browser either way, matching Next.js. Server environments are untouched — `global` remains the real Node global — and a user-configured `compiler.define.global` takes precedence, mirroring Turbopack's or_insert free-var semantics. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(images): accept /_next/image/ with trailingSlash: true With `trailingSlash: true`, the App Router dev handler 308-redirects `/_next/image?url=...` to `/_next/image/?url=...` (the trailing-slash normalizer runs before the image-endpoint check), but the image endpoint only matched the exact `/_next/image` pathname — so the redirected request 404'd and every dev-mode next/image request broke. Reproduced on tailwind-nextjs-starter-blog, which ships trailingSlash: true. Baseline Next 16.2.10 behaves the same way up to the redirect (dev with trailingSlash: true also 308s `/_next/image` to `/_next/image/`) but then SERVES the slashed form: its route matching strips a trailing slash before matching internal paths (getItem in packages/next/src/server/lib/router-utils/filesystem.ts), so image requests never fail. Match that: isImageOptimizationPath() now strips a single trailing slash before comparing, which covers every caller — App Router dev/prod (app-rsc-handler), Pages Router dev middleware, the Node prod server, and the Cloudflare worker entry. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * test: align existing assertions with charset and trailing-slash fixes Three assertions added on main after these fixes were authored still asserted the old behavior: bare text/html Content-Type in static-file-cache and the pages pipeline defaultContentType, and isImageOptimizationPath rejecting the trailing-slash form. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com> |
||
|
James Anderson
|
76de660c20 |
fix(image): emit static imports as managed assets (#2040)
* fix(image): emit static imports as managed assets * fix(image): handle static asset rebuilds and caching * fix(image): scope managed media hash etags * test(image): escape all URL ampersands |
||
|
James Anderson
|
d504f022b0 |
fix(assets): default assetsDir to _next/static (Next.js parity) (#1411)
* fix(assets): default assetsDir to _next/static (Next.js parity) Closes #1337. Supersedes #1383. vinext's default `assetsDir` was Vite's historical `assets/`, so URL emission (`/_next/static/...` via `resolveAssetUrlPrefix("")`) and on-disk layout (`dist/client/assets/...`) disagreed in the empty-`assetPrefix` case. The `build.assetsDir` and `experimental.renderBuiltUrl` Vite overrides only applied when `assetPrefix` was configured, leaving the no-prefix branch serving from a different path than the URL contract Next.js's client runtime and test harness assert against. This commit flips the default to Next.js's canonical layout: - `resolveAssetsDir("")` returns `_next/static` (was `assets`) - `build.assetsDir` is now set unconditionally from `resolveAssetsDir` - Vite's default `base + assetsDir` composition produces correct URLs in the no-prefix case; `renderBuiltUrl` stays gated on `assetPrefix` because it's only needed for the configured cases - Drops the legacy `/assets/` branches in prod-server (3 sites) and static-file-cache (hard cutover — see PR description for rationale) - Updates `_headers` generation, precompress default, fonts plugin `DEFAULT_ASSETS_DIR` to match the new default With the layout aligned, invalid `_next/static/*` requests naturally return plain-text `404 + "Not Found"` from the static-file layer instead of falling through to the page renderer (which would produce an HTML 404 with bootstrap scripts and CSS). This replaces #1383's parity short-circuit with the natural code path: - Node prod-server (App + Pages branches): missing asset under `resolveAppRouterAssetPath` returns `text/plain; charset=utf-8` 404 - Cloudflare worker entry: `isNextStaticPath` recognises asset-shape after ASSETS-binding misses; returns `notFoundStaticAssetResponse` Mirrors Next.js: packages/next/src/server/lib/router-server.ts. Adds `tests/invalid-static-asset-404.test.ts` (3 App Router + 3 Pages Router cases — no prefix / basePath / assetPrefix) ported from Next.js e2e suites. Updates 11 source files + 12 test files; 914+ tests pass in the touched suites (asset-prefix, app-router, pages-router, deploy, font-google, static-file-cache, serve-static, precompress, features, routing, isr-cache, build-optimization, app-rsc/ssr, standalone, middleware, shims). Ported from Next.js: - test/e2e/invalid-static-asset-404-app/*.test.ts - test/e2e/invalid-static-asset-404-pages/*.test.ts * fix(e2e): update cloudflare-pages-router hydration spec for _next/static * PR #1411 approved. Clean fix, 2 nits. Co-authored-by: james-elicx <james-elicx@users.noreply.github.com> * refactor(assets): derive isHashed checks from ASSET_PREFIX_URL_DIR Replace hardcoded '_next/static' string literals in runtime code paths with the ASSET_PREFIX_URL_DIR constant so the asset URL convention has a single source of truth. - server/prod-server.ts isHashed: pathname.includes(`/${ASSET_PREFIX_URL_DIR}/`) - server/prod-server.ts public-dir guard: startsWith(`/${ASSET_PREFIX_URL_DIR}/`) - server/static-file-cache.ts isHashed: both startsWith + includes derived - plugins/fonts.ts DEFAULT_ASSETS_DIR = ASSET_PREFIX_URL_DIR Behavioural no-op — the constant is '_next/static' so the resolved strings are identical. Centralises the URL contract so future changes to ASSET_PREFIX_URL_DIR (or a refactor that derives it from build config) flow through to every consumer. Doc-comment references to '_next/static' left in place — they describe the canonical value for human readers and are not code paths. --------- Co-authored-by: ask-bonk[bot] <ask-bonk[bot]@users.noreply.github.com> Co-authored-by: james-elicx <james-elicx@users.noreply.github.com> |
||
|
Nathan Nguyen
|
6da45a680a |
perf: build-time precompression + startup metadata cache for static serving (#641)
* perf: add build-time precompression for hashed static assets
Generate .br (brotli q11), .gz (gzip l9), and .zst (zstandard l19)
files alongside compressible assets in dist/client/assets/ during
vinext build. These are served directly by the production server,
eliminating per-request compression overhead for immutable build output.
Only targets assets/ (hashed, immutable). Public directory files still
use on-the-fly compression since they may change between deploys.
* perf: add startup metadata cache for zero-syscall static serving
StaticFileCache walks dist/client/ once at server boot and caches:
- File metadata (path, size, content-type, cache-control, etag)
- Pre-computed response headers per variant (original, br, gz, zst)
- In-memory buffers for small files (< 64KB) for res.end(buffer)
- Precompressed variant paths and sizes
Per-request serving is Map.get() + res.end(buffer) with zero filesystem
calls, zero object allocation, and zero header construction. Modeled
after sirv's production mode but with in-memory buffering for small
files which eliminates createReadStream fd overhead.
* perf: refactor tryServeStatic to async + cache + precompressed serving
Refactor tryServeStatic to use StaticFileCache for the hot path:
- Pre-computed response headers (zero object allocation per request)
- In-memory buffer serving for small precompressed files
- 304 Not Modified via ETag + If-None-Match
- HEAD request optimization (headers only, no body)
- Zstandard serving (zstd > br > gzip > original fallback chain)
- Async filesystem fallback for non-cached files (replaces blocking
existsSync + statSync)
- Skip decodeURIComponent for clean URLs (no % in path)
Wire StaticFileCache.create() into both startAppRouterServer and
startPagesRouterServer at startup. Integrate precompressAssets() into
the vinext build pipeline.
CONTENT_TYPES is now a single source of truth exported from
static-file-cache.ts (was duplicated in prod-server.ts).
* docs: fix stale comments in precompress (mention .zst)
* fix: deduplicate buffer reads for HTML alias entries, fix stale JSDoc
* docs: fix stale comment in cli.ts (mention zstd)
* fix: address code review findings
- Add Vary: Accept-Encoding to original variant when compressed siblings exist
- Use pipeline() instead of .pipe() for stream error safety
- Remove legacy accessor fields from StaticFileEntry (all new code)
- Exclude Content-Type from 304 responses per RFC 9110
- Fix misleading Accept-Encoding comment in prod-server
- Batch stat() calls in walkFilesWithStats (chunks of 64)
- Bound precompression concurrency to availableParallelism (max 16)
- Gracefully skip zstd when zlib.zstdCompress unavailable (Node <22.15)
- Add --no-precompress CLI flag to skip build-time compression
- Drop brotli to q5 (fast, nearly same ratio as q11 at build time)
- Max out zstd at level 22 (build time, no reason to hold back)
* fix: address remaining review findings in prod-server
- Gate negotiateEncoding zstd on runtime availability (HAS_ZSTD)
- Add ZSTD_e_flush for streaming mode (progressive SSR decodability)
- Serve HEAD without body on slow path (compressed + uncompressed)
- Handle pipeline errors instead of silently swallowing them
- Lowercase Accept-Encoding in fast path per RFC 9110
- Fix stale comments: add zstd to module header and variant detection
- Add tests for slow path, HEAD, and malformed percent-encoded pathnames
* perf: move precompression to Vite plugin with edge auto-detection
Move build-time precompression from CLI post-step into a writeBundle
plugin hook (vinext:precompress). The plugin already knows the
deployment target — auto-skips for Cloudflare Workers and Nitro where
the CDN handles compression at the edge.
- Add precompress option to VinextOptions: 'auto' | boolean
- CLI --no-precompress forwards via VINEXT_NO_PRECOMPRESS env var
- Works for any build invocation (CLI, API, CI), not just vinext build
* docs: fix inaccurate comments in precompress and static-file-cache
- precompress: remove false idempotency claim (overwrites, not skips)
- static-file-cache: "frozen objects" → "reusable objects" (no freeze)
- static-file-cache: fix unrealistic compression ratio example
* fix: address code review findings
- Remove dead code: redundant .br/.gz/.zst extension checks already
caught by COMPRESSIBLE_EXTENSIONS filter
- Fix unreachable pipeline error handlers: res.writeHead() was called
before pipeline(), so res.headersSent was always true in the callback.
Use res.destroy(err) to cleanly tear down the socket on stream errors.
- Read file content lazily inside compression chunks instead of loading
all files upfront, keeping peak memory at O(CONCURRENCY * max_file_size)
* perf: tune compression levels and restore auto mode
- Drop zstd from level 22 to 8 (250x faster build, 7% larger output)
- Drop gzip from level 9 to 8 (marginal size difference, faster)
- Restore precompress: 'auto' | boolean config with auto as default
(auto-enables for Node.js targets, skips for Cloudflare/Nitro)
- Add progress bar for precompression step (matches prerender UX)
- Clarify compression ratio output as brotli-specific
* fix: use filename hash for ETag on hashed assets
Hashed assets (assets/app-DqZc3R4n.js) now derive their ETag from the
content hash in the filename instead of mtime. More robust across
Docker builds, CI artifacts, and cp -r where mtime changes between
deploys even when content is identical.
- Use weak validator (W/) since the tag is shared across content-encoded
variants (.br, .gz, .zst) which are byte-different (RFC 9110 §8.8.1)
- Fall back to mtime-based ETag when filename has no recognizable hash
(custom assetFileNames, plugin-emitted assets, public/ files)
* fix: make precompress opt-in per maintainer feedback
Real-world benchmark shows 13.5% build-time regression on fumadocs
with auto mode (2.4s → 2.7s). Switch to explicit opt-in via
`precompress: true` or `--precompress` CLI flag until the feature
stabilises.
* fix: address code review findings
- ETag regex: use lastIndexOf('-') to extract only the final hash
segment, avoiding capture of filename fragments in multi-hyphenated
names (e.g. `before-pop-state-D3R3Ien6` now extracts `D3R3Ien6`)
- Progress bar: track processed count (including skipped files) so
the bar always reaches 100%
- Tests: update ETag test for filename-hash behavior, add coverage
for mtime fallback on non-hashed and unhashed-in-assets files
* chore: use type instead of interface per lint rule
* fix final precompression review comments
* perf: optimize precompress build path
- Overlap precompression with SSR build phase (async writeBundle + closeBundle await)
- Lower concurrency cap from 16 to 8 (reduces zlib queue contention)
- Fix progress counter race in chunked processing
* fix: address final review feedback
- Fix progress counter race in precompress (single onProgress call site
after readFile, monotonic by construction)
- Add console.error before res.destroy in all pipeline error handlers
for operational visibility
- Chunk buffer reads at 64 concurrent to bound FD usage on large projects
- Defer createCompressor until after HEAD check on slow path
- Use Math.floor for mtime ETag (matches send/Express convention)
- Document Rolldown hash length assumption
* fix: address final bonk review nits
- Move progress reporting to after chunk completion to avoid
non-deterministic ordering within Promise.all
- Use Math.trunc instead of Math.floor for mtime-based ETags
(semantically correct for truncating to whole seconds)
* address bonk review comments: counters post-write, Math.floor, Content-Length comment, warn log level
- precompress.ts: move filesCompressed++ and totalOriginalBytes+= to after
writes complete so partial failures (e.g. ENOSPC) don't inflate totals
- precompress.ts: add comment clarifying that progress tracks all files
including skipped ones below MIN_SIZE, unlike filesCompressed
- static-file-cache.ts: Math.trunc -> Math.floor for mtime-based ETags
(matches send/Express convention; correct for pre-epoch mtimes)
- prod-server.ts: add comment explaining Content-Length is intentionally
omitted on slow-path compressed responses (chunked transfer encoding)
- prod-server.ts: downgrade all three pipeline error logs from console.error
to console.warn and add comments explaining the destroy-only pattern
* address bonk: slow-path ETag+304 parity, visible precompress error log
- prod-server.ts slow path: add mtime-based ETag (W/"size-mtime") and
304 Not Modified support, matching fast-path cache behaviour; also
propagate mtimeMs through resolveStaticFile/statIfFile
- index.ts: add console.error in precompress .catch() so build errors are
visible even if the SSR closeBundle hook never fires
* address bonk: slow-path 304 Vary header, HAS_ZSTD comment
- prod-server.ts slow path: include Vary: Accept-Encoding in 304
responses when the content type is compressible, matching fast-path
notModifiedHeaders behaviour and preventing shared caches from
collapsing compressed/uncompressed representations
- prod-server.ts fast path: add comment explaining HAS_ZSTD is not
checked during precompressed variant selection — we're serving a
pre-existing .zst file from disk, not invoking zstdCompress() at
runtime; HAS_ZSTD only guards the slow-path on-the-fly compressor
* address bonk: clarify Vary/compress=false intent in slow-path 304 comment
Expand the 304 comment to explain that Vary: Accept-Encoding is
intentionally omitted when compress=false (matching fast-path behaviour
where compress=false skips all compressed variants), and note that
spreading undefined is a deliberate no-op per ES2018+.
* address bonk: readFile-before-stat comment, document VINEXT_PRECOMPRESS env var
- precompress.ts: add comment on MIN_SIZE early-return explaining the
read-first vs stat-first tradeoff (one wasted read vs extra syscall
per tiny file; read-first cheaper for typical hashed asset sets)
- index.ts: document VINEXT_PRECOMPRESS=1 env var in VinextOptions.precompress
JSDoc alongside the existing --precompress CLI flag mention
* address bonk: filename-hash ETag on slow path, entry-level dedup in buffer pass
- static-file-cache.ts: export etagFromFilenameHash so prod-server can
reuse the same filename-hash extraction logic
- prod-server.ts: slow path now uses etagFromFilenameHash for /assets/*
files, falling back to mtime-based ETag for non-hashed files; this
matches fast-path cache behaviour and makes ETags stable across deploys
- static-file-cache.ts buffer pass: deduplicate at the StaticFileEntry
level (using a seenEntries Set) rather than at the variant path level;
HTML aliases share the same entry by reference so the old approach
iterated variants multiple times for sites with many HTML pages
* address bonk: fix etagFromFilenameHash arg, explicit re-run safety comment
- prod-server.ts: use resolved.path (not pathname) for etagFromFilenameHash
so that ext and the hash extraction always come from the same file;
pathname and resolved.path diverge after HTML fallback (e.g. a request
for /assets/widget-abc123 resolving to widget-abc123.html would have
extracted a hash from the wrong name)
- precompress.ts: add inline comment on the COMPRESSIBLE_EXTENSIONS filter
making explicit that .br/.gz/.zst are intentionally absent, so
precompressed variants from a previous run are never re-compressed
* address bonk: add slow-path 304 test coverage
---------
Co-authored-by: James <james@eli.cx>
|