Note
AI disclosure: The vast majority of this repro was created using Claude (Sonnet & Opus) 4.6. Everything below this section and anything not in the from_real_world_repo/ dir has been touched on, or created by AI in some way.
While the segfault does appear to be happening within Node.js, it's a bit beyond my skill level to confidently claim the bug is in V8's Sparkplug. Thus root cause identified below is what Claude has determined.
I supplied all version numbers used for testing, these have all been manually verified.
Claude also seemed to want to keep the patch-jest-resolve.js script, although this not necessary to trigger a segfault, so I've removed it from the npm postinstall script, but still remains as an optional npm run script.
Reliably reproduces a SIGSEGV crash in V8 Sparkplug triggered by the introduction of unrs-resolver in Jest ≥ 30.0.0-beta.6 via jestjs/jest#15619.
Claude's Root cause verdict: V8 Sparkplug (Baseline JIT) leaves a stale near-null pointer in an
InternalFrameslot. When GC fires andClearStaleLeftTrimmedPointerVisitorscans stack roots, it dereferences the stale slot and segfaults.unrs-resolver's NAPI object churn creates the heap pressure that triggers this. Running with--no-sparkplugprevents the crash entirely. This is a V8 bug, not a bug innapi-rsorunrs-resolver.
Most testing has been done on the M3 Pro with Node.js v24.14.1.
| Node.js Version | Status | Run count | Device |
|---|---|---|---|
| 22.12.0 | pass | 5 | M3 Pro |
| 22.13.0 | pass | 5 | M3 Pro |
| 22.17.0 | pass | 5 | M3 Pro |
| 22.22.2 | pass | 5 | M3 Pro |
| 23.0.0 | pass | 5 | M3 Pro |
| 24.0.0 | crash | 5 | M3 Pro |
| 24.8.0 | crash | 5 | M3 Pro |
| 24.8.0 | pass | 1 | M4 Pro |
| 24.11.0 | crash | 5 | M3 Pro |
| 24.14.1 | crash | many | M3 Pro |
| 24.14.1 | crash | 1 | M4 Pro |
| 24.15.0 | crash | many | M3 Pro |
| 25.0.0 | pass | 5 | M3 Pro |
| 25.15.0 | pass | many | M3 Pro |
| Device | OS | Arch | System Memory |
|---|---|---|---|
| Apple M3 Pro | Tahoe 26.4.1 | arm64 | 18GB |
| Apple M4 Pro | Tahoe 26.4.1 | arm64 | 48GB |
| Package | Version | Info | Status |
|---|---|---|---|
| Jest | <= 30.0.0-beta.5 |
pure-JS resolver - does not utilise unrs-resolver package |
pass |
| Jest | >= 30.0.0-beta.6 |
unrs-resolver package introduced |
crash |
| Jest | 30.3.0 (latest) |
unrs-resolver still present |
crash |
| unrs-resolver | v1.7.11 |
First version Jest supported | crash |
| unrs-resolver | v1.11.1 |
Latest version | crash |
npm install # patches jest-resolve, generates tests, warms transform cache
npm run test:crash # single run — crashes within seconds
npm run test:crash:loop # loop 50 attempts
npm run test:safe:nosparkplug # --no-sparkplug → no crash (confirms root cause)Or simply: npx jest - on Node 24.14.1 (extensively tested) this alone is sufficient to trigger a segfault, without any special flags.
jest-resolve creates a new ResolverFactory NAPI instance (wrapping Rust Arc<Resolver>) on every require(). The postinstall script patches jest-resolve to disable its resolver cache, so every module resolution creates a fresh instance. This produces sustained heap churn — many promoted-then-discarded objects — that drives V8's left-trimming into the buggy code path.
transformIgnorePatterns: []— transforms ALLnode_modulesthrough ts-jest, creating massive resolver cascades- Angular/rxjs re-imports every cycle — each
jest.resetModules()+ re-require of@angular/core,rxjs,zone.jsetc. triggers hundreds of nested resolver calls - 200 test files × 15 reset cycles each — sustained module churn across 4 workers
- 500
.tssource modules — cross-dependent (non-circular) to maximize resolver activity --no-maglev --max-old-space-size=256— keeps Sparkplug hot and increases GC pressure- Warm transform cache —
postinstallprimes the ts-jest cache so tests run fast, concentrating NAPI churn
| File | Purpose |
|---|---|
patch-jest-resolve.js |
Disables resolver caching (getResolver → undefined) |
generate-tests.js |
Generates 500 source modules + 200 test files |
warm-cache.js |
Primes ts-jest transform cache during postinstall |
setup.ts |
Jest setup — setupZoneTestEnv() from jest-preset-angular |
segfault-bail-reporter.js |
Jest reporter that exits immediately on worker SIGSEGV |
src/mod*.ts |
Generated source modules with cross-dependencies |
__tests__/test*.spec.ts |
Generated test files with module churn + NAPI churn |
| Configuration | Result | Conclusion |
|---|---|---|
| Default run (4 workers) | SIGSEGV | Baseline crash |
MallocScribble=1 MallocGuardEdges=1 |
SIGSEGV at same address | Rules out userspace heap corruption |
--no-maglev |
Still crashes | Rules out Maglev JIT |
--no-sparkplug |
No crash | Confirms Sparkplug as the root cause |
--detectOpenHandles (1 worker) |
SIGSEGV | Not a concurrency bug — single process crashes too |
Jest 30.0.0-beta.5 |
No crash | Uses pure-JS resolve — no NAPI churn |
v8::ClearStaleLeftTrimmedPointerVisitor::VisitRootPointers ← SIGSEGV at 0x6
v8::InternalFrame::Iterate
v8::Heap::IterateRoots
v8::MarkCompactCollector::MarkRoots
v8::MarkCompactCollector::CollectGarbage
v8::StackGuard::HandleInterrupts
Builtins_BaselineOutOfLinePrologue ← Sparkplug frame holding stale pointer
Builtins_ArrayMap ← Array.prototype.map() hot path
The faulting address is always near-null (0x6, 0xe) — a stale tagged pointer inside V8's own frame, not in addon memory.
Update Node to v25.9.0 and the segfault is no longer reproducible.
This tested as working for Node v24.14.1 & v24.15.0.
node --no-sparkplug ./node_modules/.bin/jestJest v30.0.0-beta.5 and earlier v29 releases uses the pure-JS resolve package instead of unrs-resolver which also doesn't reproduce this segfault issue.