Reproduction link or steps

Minimal repo: https://github.com/binoy14/rolldown-repro

git clone https://github.com/binoy14/rolldown-repro
cd rolldown-repro
npm install
node main.mjs; echo "exit=$?"   # → exit=139 (SIGSEGV); "PASS" never prints

Two files:

worker.mjs

import {parentPort} from 'node:worker_threads'
import {rolldown} from 'rolldown'
const VIRTUAL = '\0virtual-entry'
const bundle = await rolldown({
  input: VIRTUAL,
  cwd: process.cwd(),
  logLevel: 'silent',
  plugins: [
    {
      name: 'virtual',
      resolveId(id) {
        return id === VIRTUAL ? id : null
      },
      load(id) {
        return id === VIRTUAL ? `export const x = 1; console.log(x)` : null
      },
    },
  ],
})
await bundle.generate({format: 'esm'})
await bundle.close()
parentPort?.postMessage('done')

main.mjs

import {fileURLToPath} from 'node:url'
import {Worker} from 'node:worker_threads'
const WORKER = fileURLToPath(new URL('./worker.mjs', import.meta.url))
const runOnce = () => new Promise((resolve) => new Worker(WORKER).on('exit', resolve))
await runOnce() // first worker — fine
await runOnce() // second worker — segfaults here
console.log('PASS — did not crash') // never printed

What is expected?

Loading rolldown in a fresh worker_thread should succeed regardless of whether a previous worker in the same process already loaded rolldown and exited. main.mjs should print PASS — did not crash and exit 0.

What is actually happening?

The host process segfaults (SIGSEGV, exit code 139) when the second worker_thread loads rolldown. The first worker is fine; the crash happens on the second load, after the first worker has exited. PASS is never printed.

Scope of the crash (isolated):

• A single worker never crashes — only the 2nd rolldown load in the same process.
• Replacing the worker_thread with a child_process.fork (separate process) does not crash.
• The same harness with @swc/core (also Rust + napi + native threads) instead of rolldown does not crash — so it is not a generic "native addon in a worker_thread" issue; it is specific to rolldown.
• Reproduces on rolldown 1.0.3 and 1.1.0, and on Node 20, 22, and 24.
• Deterministic in this minimal form (no concurrency/scale needed).

Downstream impact: Vite 8 bundles rolldown, so tools that run Vite inside worker threads hit this — e.g. test runners (vitest's forks pool) report it as Worker forks emitted error / Worker exited unexpectedly when a test fork that spawned such a worker dies.

System Info

System:
    OS: macOS 26.5.1
    CPU: (14) arm64 Apple M4 Pro
    Memory: 753.22 MB / 48.00 GB
    Shell: 5.9 - /bin/zsh
  Binaries:
    Node: 24.16.0
    Yarn: 1.22.22
    npm: 11.13.0
    pnpm: 10.28.0
    bun: 1.3.5
  npmPackages:
    rolldown: 1.1.0 => 1.1.0


(Also reproduced with `rolldown: 1.0.3`, and on Node 20.20.0 / 22.22.0.)

Any additional comments?

No response