Bug Report - Segmentation fault on MacOS with workerd and wrangler and vitest

[paralin]

I am seeing a segfault crash in the Cloudflare Vitest worker pool on macOS arm64.

• macOS 15.7.3 (24G419)
• Apple Silicon
• bun 1.3.12
• node v25.9.0
• vitest 4.1.4
• @cloudflare/vitest-pool-workers 0.14.6
• wrangler 4.82.2

Some tests run, but tests that import certain Worker/Durable Object modules crash before any test body executes.

*** Received signal #11: Segmentation fault: 11

Error: [vitest-pool]: Worker cloudflare-pool emitted error.
Caused by: Error: Worker exited unexpectedly

What works

A simple non-DO test file passes under the same runner config.

What fails

Files that import more substantial Worker/Durable Object code paths consistently crash the Cloudflare pool before tests run.

Reproduction commands

These both fail the same way:

bunx vitest run --max-workers=1 --no-isolate path/to/crashing.test.ts

node ./node_modules/vitest/vitest.mjs run --max-workers=1 --no-isolate path/to/crashing.test.ts

The repo script also fails:

bun run test

Important notes

• This is not Bun-specific. Reproduces with direct node invocation too.
• --max-workers=1 --no-isolate does not avoid the crash.
• The crash happens before test execution for the affected files.

Tried

I also tried forcing Miniflare to use a newer external workerd binary from the latest GitHub release:

MINIFLARE_WORKERD_PATH=/absolute/path/to/workerd \
  bunx vitest run --max-workers=1 --no-isolate path/to/crashing.test.ts

The result was unchanged: same cloudflare-pool worker crash and segfault.

I tested with:

• external workerd binary version: workerd 2026-04-14

[paralin]

Repro repo: https://github.com/paralin/workerd-issue-6583

Commands

Passing control test:

bunx vitest run --max-workers=1 --no-isolate src/control.test.ts

Crashing repro:

bun run test:sqlite

Equivalent direct command:

bunx vitest run --max-workers=1 --no-isolate src/node-sqlite.test.ts

Expected

The node:sqlite test should pass and report one successful test.

Actual

The Cloudflare pool worker exits before any test executes:

*** Received signal #11: Segmentation fault: 11
Error: [vitest-pool]: Worker cloudflare-pool emitted error.
Caused by: Error: Worker exited unexpectedly

Test Files   (1)
     Tests  no tests
    Errors  1 error

Minimal crashing test

import { describe, expect, it } from 'vitest'
import { DatabaseSync } from 'node:sqlite'

describe('node:sqlite under cloudflare vitest pool', () => {
  it('opens an in-memory database', () => {
    const db = new DatabaseSync(':memory:')
    db.exec('CREATE TABLE t (id INTEGER PRIMARY KEY, name TEXT NOT NULL)')
    db.exec("INSERT INTO t (name) VALUES ('ok')")
    const row = db.prepare('SELECT name FROM t').get() as { name: string }
    expect(row.name).toBe('ok')
  })
})

Additional checks

• This is not Bun-specific. The broader crash also reproduces when invoking
  Vitest via node ./node_modules/vitest/vitest.mjs ....
• Pointing Miniflare at a locally built workerd binary did not change the
  behavior.

Example:

MINIFLARE_WORKERD_PATH=/absolute/path/to/workerd \
  bun run test:sqlite

still crashes with the same worker-pool error and segfault.

Native trace from source-built workerd

I attached LLDB to the spawned workerd process and reproduced the crash using
a local source build. The process stopped with:

EXC_BAD_ACCESS (code=2, address=0x16a6f3fe0)

Top of stack:

libc++abi.dylib`operator new
workerd`kj::_::HeapArrayDisposer::allocateImpl
workerd`kj::heapString(char const*, unsigned long)
workerd`kj::heapString(kj::ArrayPtr<char const>)
workerd`kj::Path::evalPart
workerd`kj::Path::evalImpl
workerd`kj::PathPtr::eval
workerd`workerd::jsg::ModuleRegistryImpl<...>::resolve
workerd`workerd::jsg::ModuleRegistryImpl<...>::resolve
workerd`workerd::jsg::ModuleRegistryImpl<...>::resolve
    ...

The resolve(...) frame repeats deeply, which makes this look like a module
resolution/path evaluation problem inside workerd, triggered by the
node:sqlite import path under the Vitest pool.