Build hangs in @tailwindcss/oxide-darwin-x64 on Node 26 / Vite 8 (darwin-x64)

[HerbiusYang]

Build hangs in @tailwindcss/oxide-darwin-x64 on Node 26 / Vite 8 (darwin-x64)

Question

Is a build hang in @tailwindcss/oxide-darwin-x64 on Node 26 a known issue, and is this
the right repo to report it — or should it go to Vite / Rolldown?

I'm looking for confirmation before filing a bug, because the stack is bleeding-edge and
the root cause may not be Tailwind itself.

What's happening

After upgrading Tailwind CSS from 4.3.0 to 4.3.1, vite build production builds
started hanging indefinitely.

Sampling shows the process parked in tailwindcss-oxide.darwin-x64.node at 0% CPU.
That looks like blocking/waiting rather than busy source scanning.

I do not have a clean minimal reproduction yet. The failure appeared after several
pnpm install --force cycles during unstable network conditions, so I cannot fully rule
out a bad native dependency install or cache state. That's why I'm asking for routing /
known-issue confirmation before opening a formal bug.

Environment

• macOS Intel / darwin-x64 / x86_64
• Node.js v26.0.0
• pnpm 11.1.3
• pnpm workspace / monorepo
• Vite 8.0.14 (Rolldown-backed Vite 8)
• Tailwind CSS 4.3.0 -> 4.3.1
• @tailwindcss/postcss 4.3.0 -> 4.3.1
• .npmrc contains ignore-scripts=true

Why I upgraded

Frontend tests under Node 26 produced:

[DEP0205] DeprecationWarning: `module.register()` is deprecated. Use `module.registerHooks()` instead.

Running with NODE_OPTIONS=--trace-deprecation traced this to @tailwindcss/node@4.3.0.

Upgrading Tailwind / @tailwindcss/postcss to 4.3.1 removed the warning in tests, but
the production build then started hanging.

## The hang

Command:

pnpm --dir frontend build

Output stops at:

vite v8.0.14 building client environment for production...
transforming...

Observed behavior:

- vite build process stays at 0.0% CPU
- no error output
- no normal exit
- still hung after manual 180s / 300s timeouts
- tsc --noEmit passes
- vitest passes
- only production vite build hangs

Sampling observations:

- main thread / stack includes tailwindcss-oxide.darwin-x64.node
- rolldown-binding.darwin-x64.node worker threads are also present

## Things I tried

I first suspected source scanning, but that now looks unlikely.

Tried:

@import "tailwindcss" source("../");

Then tried:

@import "tailwindcss" source(none);
@source "../";

Neither changed the hang.

This is why I think source scanning is probably not the root cause:

- CPU stays at 0.0%, which does not look like broad filesystem scanning
- disabling automatic source detection did not resolve the hang
- the process appears blocked in native code rather than busy in JS or scanning

I also tried clearing Vite cache:

rm -rf frontend/node_modules/.vite

No change.

## Native binary state

The platform binary exists:

node_modules/.pnpm/@tailwindcss+oxide-darwin-x64@4.3.1/node_modules/@tailwindcss/oxide-darwin-x64/tailwindcss-
oxide.darwin-x64.node

It is approximately 2.5MB, and file reports:

Mach-O 64-bit dynamically linked shared library x86_64

One note: a root-level probe like this is not valid in this pnpm workspace:

node -e "require('@tailwindcss/oxide')"

It fails with MODULE_NOT_FOUND, because @tailwindcss/oxide is not a direct root dependency
under pnpm's isolated dependency layout. So I do not currently have a clean "native binary
loads / does not load" probe.

## Why I'm unsure where to report

The sample includes Tailwind's native oxide binary, but this is running inside Vite 8's
production build path, and rolldown-binding.darwin-x64.node worker threads are also present.

So I'm trying to determine whether this is likely:

1. a Tailwind oxide darwin-x64 / Node 26 issue,
2. a bad native dependency install/cache state,
3. an ignore-scripts=true / optional native dependency install issue,
4. or an interaction with Vite 8 / Rolldown.

## Current workaround

Revert Tailwind / @tailwindcss/postcss from 4.3.1 back to 4.3.0.

That brings back the Node 26 DEP0205 warning, but it restores a buildable path in this repo.
For now the warning is less risky than a production build hang.

## Questions

1. Is this a known issue or weak spot for @tailwindcss/oxide-darwin-x64 on Node 26?
2. Can ignore-scripts=true leave Tailwind's native packages in an incomplete or unverified state?
3. Is the Tailwind repo the right place to start, or should this be reported to Vite / Rolldown first?
4. Is there a recommended direct probe for verifying that the selected @tailwindcss/oxide-*
   native package is correctly installed and loadable under pnpm?

I can provide the full macOS sample output or test a clean reinstall if useful.

[wongjn]

Your problem might already be fixed by #20263. However, the fix hasn't been released in stable version yet, so try using insiders version:

$ npm install tailwindcss@insiders @tailwindcss/vite@insiders

Otherwise, try running with DEBUG=* to create log file of Tailwind scanning behavior. This might reveal some clues about the hang.