claude-code - 💡(How to fix) Fix [BUG] Bun 1.3.14 JSC GC segfault (SlotVisitor::drain) on AMD Ryzen 5 5500 / Linux x86_64_baseline — all platform binaries identical [2 comments, 2 participants]

Preflight Checklist

• I have searched existing issues and this hasn't been reported yet
• This is a single bug report (please file separate reports for different bugs)
• I am using the latest version of Claude Code

What's Wrong?

Environment:

CC versions affected: 2.1.133, 2.1.136, 2.1.137 (all tested) OS: Ubuntu 24.04.4 LTS Kernel: 6.17.0-23-generic #23~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC CPU: AMD Ryzen 5 5500 (x86_64, AVX2 capable) Bun: 1.3.14, commit 050219c1 Install method: npm global install RAM: 33.5GB, healthy throughout crashes

Bug: Bun 1.3.14 crashes with a segfault in JSC's parallel GC during normal CC sessions. Reproducible at low context (under 30K tokens, under 10 minutes elapsed). Not a memory pressure issue. Crash fingerprint: panic(main thread): Segmentation fault at address 0xD0 SlotVisitorInlines.h:193: WTF::IterationStatus JSC::SlotVisitor::forEachMarkStack SlotVisitor.cpp:499: JSC::SlotVisitor::drain SlotVisitor.cpp:699: JSC::SlotVisitor::drainFromShared Heap.cpp:1535: JSC::Heap::runBeginPhase ParallelHelperPool.cpp:113: WTF::ParallelHelperPool::Thread::work Illegal instruction (core dumped) Full Bun crash report: https://bun.report/1.3.14/B_1050219cmgggkEuhogC4koo4E+ypRqxqyqF4ryyqF+j+jqF89kuqFmrpuqFkiwo4EmqlnB2muqCA2AgN Critical finding: all platform binaries are identical Every platform-specific package ships the same compiled binary:

@anthropic-ai/claude-code (linux-x64) @anthropic-ai/claude-code-linux-x64 @anthropic-ai/claude-code-linux-x64-musl

All three have BuildID 1c6448151b04aed46579bf7429d8704f6f2be83d. There is no non-baseline build available regardless of which package is installed. Version pinning and platform package selection provide no escape. This is a regression from prior stable behavior. CC was stable on this machine before May 8 2026. The auto-updater pulled 2.1.136/2.1.137 and introduced instability. Rolling back to 2.1.133 does not help because the Bun runtime is identical across all three versions. Reference: This appears related to closed issue #27497 (closed as stale, not fixed). CoreyCole in that thread reproduced on kernel 6.17.0 with identical CPU flags (sse42 popcnt avx avx2) in February 2026. The issue was never resolved, just went stale. Impact: CC is completely unusable on this hardware. Sessions crash reliably within minutes regardless of context size.

What Should Happen?

Claude Code should run stable sessions on AMD Ryzen 5 5500 / Ubuntu 24.04 / kernel 6.17.0 without crashing. The Bun 1.3.14 JSC parallel GC should not segfault during normal operation at low context. Platform-specific packages should ship distinct binaries so users on x86_64_baseline hardware have a non-baseline build path available.

Error Messages/Logs

Bun v1.3.14 (050219c1) on linux x86_64_baseline [StandaloneExecutable]

Segmentation fault at address 0x000000D0

- SlotVisitorInlines.h:193: WTF::IterationStatus JSC::SlotVisitor::forEachMarkStack<JSC::SlotVisitor::drain(...)::$_0>
- SlotVisitor.cpp:499: JSC::SlotVisitor::drain
- SlotVisitor.cpp:699: JSC::SlotVisitor::drainFromShared
- Heap.cpp:1535: JSC::Heap::runBeginPhase(...)::$_1::operator()(...) const
- SharedTask.h:91: WTF::SharedTaskFunctor<void (...)::$_1>::run
- ParallelHelperPool.cpp:113: WTF::ParallelHelperPool::Thread::work
- ParallelHelperPool.cpp:211: WTF::ParallelHelperPool::Thread::work
- AutomaticThread.cpp:241: WTF::AutomaticThread::start(...)::$_0::operator()(...) const
- Function.h:59: WTF::Detail::CallableWrapper<WTF::AutomaticThread::start(...)::$_0, void>::call
- Function.h:103: WTF::Function<void (...)(...) const
- Threading.cpp:303: WTF::Thread::entryPoint
- ThreadingPOSIX.cpp:254: WTF::wtfThreadEntryPoint

Full Bun crash report URL:
https://bun.report/1.3.14/B_1050219cmgggkEuhogC4koo4E+ypRqxqyqF4ryyqF+j+jqF89kuqFmrpuqFkiwo4EmqlnB2muqCA2AgN

Steps to Reproduce

1. Install Claude Code 2.1.133 via npm on Ubuntu 24.04.4 LTS, kernel 6.17.0-23-generic, AMD Ryzen 5 5500

2. Run: claude --dangerously-skip-permissions

3. Execute any moderately complex task involving multiple bash tool calls (database queries, file operations, service status checks)

4. Crash occurs within 2-10 minutes regardless of context size. Reproduced at under 30K tokens and under 10 minutes elapsed. Not context-dependent.

5. Rolling back to any prior version does not help — all versions from 2.1.133 through 2.1.137 ship identical Bun binary (BuildID: 1c6448151b04aed46579bf7429d8704f6f2be83d)

6. Installing platform-specific package @anthropic-ai/claude-code-linux-x64 does not help — same BuildID, same baseline binary

Note: Auto-updater pulled 2.1.136/2.1.137 mid-session on May 8 2026, causing initial crashes. Rolling back to 2.1.133 and disabling auto-updater did not resolve the issue as all versions share the same runtime.

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

2.1.133 before May 8 2026 (pre-auto-update).

Claude Code Version

2.1.133

Platform

Other

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Other

Additional Information

This bug has been occurring across multiple CC versions and is completely blocking productive use of Claude Code on this hardware. Key additional findings:

1. ALL platform-specific packages ship identical binaries:

   • @anthropic-ai/claude-code
   • @anthropic-ai/claude-code-linux-x64
   • @anthropic-ai/claude-code-linux-x64-musl All have BuildID: 1c6448151b04aed46579bf7429d8704f6f2be83d There is no escape path via platform package selection.

2. The auto-updater pulled 2.1.136/2.1.137 mid-session on May 8 2026, clobbering a pinned 2.1.133 install. The updater swaps the symlink while a session is actively running, destabilizing live sessions. This behavior should require at minimum a session boundary before taking effect.

3. This appears to be the same unresolved bug as issue #27497 (closed as stale February 2026, never fixed). CoreyCole in that thread reproduced identically on kernel 6.17.0 with the same CPU flags.

4. The crash is deterministic and reproducible. It is not random memory corruption. The same GC code path (SlotVisitor::drain in the parallel mark phase) crashes consistently on this CPU/kernel combination.

5. This is a Claude Max subscription ($100/month). CC is a core part of the value proposition and is currently completely unusable.