Repository navigation
Description
Link to the code that reproduces this issue
https://github.com/terrymun/Next.js-fails-with-BOM-marker-in-CSS
To Reproduce
- Check out the repo at . The trigger is to have one or more CSS files that contain the BOM marker, which is allowed based on CSS specifications.
- Build the Next.js app with
pnpm build - Notice you get the following error message:
> Build error occurred
Error: Turbopack build failed with 1 error:
./app/bom.css:1:2
Error: Parsing CSS source code failed
> 1 | @layer bom-layer {
| ^
2 | .bom {
3 | color: blue;
4 | }
Unexpected token AtKeyword("layer")
Generated code of PostCSS transform of file content of app/bom.css:
./app/bom.css:1:2
> 1 | @layer bom-layer {
| ^
2 | .bom {
3 | color: blue;
4 | }
Import trace:
Client Component Browser:
./app/bom.css [Client Component Browser]
./app/layout.tsx [Server Component]
at <unknown> (./app/bom.css:1:2)
at <unknown> (./app/bom.css:1:2)
Current vs. Expected behavior
When a BOM marker is introduced into a CSS file, any attempts to serve or build the Next.js app will fail.
Provide environment information
Operating System:
Platform: darwin
Arch: arm64
Version: Darwin Kernel Version 25.5.0: Tue Jun 9 22:18:58 PDT 2026; root:xnu-12377.121.10~1/RELEASE_ARM64_T6000
Available memory (MB): 32768
Available CPU cores: 10
Binaries:
Node: 26.3.1
npm: 11.16.0
Yarn: 1.22.19
pnpm: 10.21.0
Relevant Packages:
next: 16.3.0-canary.103 // Latest available version is detected (16.3.0-canary.103).
eslint-config-next: N/A
react: 19.2.8
react-dom: 19.2.8
typescript: 5.9.3
Next.js Config:
output: N/AWhich area(s) are affected? (Select all that apply)
CSS
Which stage(s) are affected? (Select all that apply)
next build (local), next dev (local)
Additional context
Original context
The issue originally surfaced in a monorepo where I'm working on that uses Next.js 16.2.12 and it has postcss bumped to 8.5.24, which caused the build to start failing. Somewhere in the transformation a BOM character was introduced, which lightningcss will throw an error on.
I have initially reported this on postcss, and with the help of others it is discovered that postcss 8.5.24 helped to surface the issue: postcss/postcss#2122
Attempting to debug the issue specific to our repo with Claude Opus 5 yielded the following findings:
Step 1 — dart-sass emits a BOM in compressed mode
For non-ASCII output, dart-sass signals the encoding. In expanded mode it emits @charset "UTF-8"; in
compressed mode it emits a UTF-8 BOM instead, because a BOM is fewer bytes.
Our input is pure ASCII: content: ':\a0'. The escape \a0 decodes to a literal NBSP (C2 A0) in the
output, which is enough to trigger the BOM.
const sass = require('sass')
const src = '@layer a{.x::after{content:":\\a0"}}'
sass.compileString(src, { style: 'expanded' }).css
// '@charset "UTF-8";\n@layer a {\n .x::after {\n content: ": ";\n }\n}'
sass.compileString(src, { style: 'compressed' }).css.charCodeAt(0) === 0xfeff
// true <-- BOM, not @charsetThis is long-standing, documented dart-sass behaviour and is suppressible with charset: false.
Step 2 — PostCSS 8.5.24 now re-emits that BOM
This is the only functional change between the two versions. lib/stringifier.js:
root(node) {
+ if (node.source && node.source.input.hasBOM) {
+ this.builder('', node, 'start')
+ }
this.body(node)(The only other diff between 8.5.23 and 8.5.24 is the version string in lib/processor.js.)
PostCSS has always detected a BOM (Input#hasBOM, lib/input.js); 8.5.24 started round-tripping it on
stringify. 8.5.23 silently discarded it.
const plugin = { postcssPlugin: 'noop', Once() {} }
const css = sass.compileString(src, { style: 'compressed' }).css // has BOM
postcss([plugin]).process(css, { from: undefined }).css.charCodeAt(0) === 0xfeff
// 8.5.23 -> false (BOM dropped)
// 8.5.24 -> true (BOM preserved)Note this requires at least one plugin with a visitor. With an empty plugin list PostCSS short-circuits and
returns the input string without stringifying, so the change is not observable — worth knowing when
reproducing.
Step 3 — Lightning CSS rejects the leading BOM
const lightningcss = require('lightningcss')
lightningcss.transform({ filename: 't.css', code: Buffer.from('@layer a{.x{color:red}}') })
// OK
lightningcss.transform({ filename: 't.css', code: Buffer.from('@layer a{.x{color:red}}') })
// ParserError: Unexpected token AtKeyword("layer")The BOM is tokenized as content rather than skipped, so the parser is misaligned when it reaches @layer.
The error text visible in the Next.js output comes from Lightning CSS via Turbopack, not from PostCSS.
Symbols in next-swc.darwin-arm64.node locate it at turbopack_css::process
(turbopack/crates/turbopack-css/src/process.rs), and Unexpected token AtKeyword(...) is Lightning CSS's
ParserError::UnexpectedToken. A useful tell in the raw Next.js error is the column number — it reports
1:2 on the compressed line while @layer visibly begins at column 1, because the invisible BOM occupies
column 1.
Reproduction
I tried to reproduce this error with Next.js canary version, and it can be reliably reproduced without upgrading to postcss 8.5.23, as long as one or more CSS files contains a BOM marker, which is allowed baesd on CSS specification.
Activity
test(css): add regression test for CSS files starting with a UTF-8 BOM
[turbopack] Strip leading BOM before parsing CSS (#96678)
[turbopack] Strip leading BOM before parsing CSS (vercel#96678)
github-actions commentedon Sep 6, 2026
This closed issue has been automatically locked because it had no new activity for 2 weeks. If you are running into a similar issue, please create a new issue with the steps to reproduce. Thank you.