Bun v1.4.1 (392624bed) still segfaults intermittently on Linux x64

Environment

• Claude Code version: 2.1.251
• Bun version: 1.4.1 (392624bed)
• OS: Linux x64, Kernel 6.8.12 (Proxmox LXC)
• glibc: 2.41
• CPU features: sse42 popcnt avx avx2
• RAM: 8GB container, RSS at crash time: 0.23–0.31GB (not OOM)
• Usage: Running as an always-on agent backend (OpenClaw) serving Discord

Problem

After upgrading from Claude Code 2.1.239 (Bun 1.4.0 fdb5e06cc) to 2.1.251 (Bun 1.4.1 392624bed), the segfault crashes reduced from ~10/day to ~3/day but did not stop.

The crash signature is a Bun panic/segfault that kills the CLI process before it produces any output, causing silent failures (no response to users).

Crash data

Aug 30 — Bun 1.4.0 (fdb5e06cc): ~10 crashes

panic(main thread): Segmentation fault at address 0x1C
panic(main thread): Segmentation fault at address 0x20

Bun crash reports:

• https://bun.report/1.4.0/l_1fdb5e06kggggIugogC+2tjwE_gp7z7Dms8r7Dg0w34Ew4j24E6v3r+Ew+oihF435whC_2/j07Dms8r7Dsv4wuEkrio1Eu0uk7BixlngC+k8o+Cs3/+3Csq9o+C2wslxCA2A4B
• https://bun.report/1.4.0/l_1fdb5e06kggggIggggCsu0r6D0kki6D+qy4sE0423sE48kjtE6j6ptEyoj6sE6662sE4+zlgF6xxlgF6h923B6r3j5E6/5n5Ews7n5Eyzyu+E_gp7z7D2/j07Dms8r7Dsv4wuEA2AgC

Aug 31 — Bun 1.4.1 (392624bed): 3 crashes

15:27 UTC+7 — Timeout (180s, no output, no panic line)

error=FailoverError durationMs=180005
reason: "timeout"
errorPreview: "CLI produced no output for 180s and was terminated."

15:45 UTC+7 — Bun crash (466ms, instant death)

Bun v1.4.1 (392624bed) Linux x64
Linux Kernel v6.8.12 | glibc v2.41
durationMs=466 error=FailoverError

18:23 UTC+7 — Bun crash (855ms / 833ms elapsed)

Bun v1.4.1 (392624bed) Linux x64
Linux Kernel v6.8.12 | glibc v2.41
Elapsed: 833ms | User: 806ms | Sys: ...
durationMs=855 error=FailoverError

All three show fallbackConfigured: false and fallbackStepFinalOutcome: "chain_exhausted" — meaning the crash produces complete silence to the end user.

Key observations

• Not OOM: RSS was 0.23–0.31GB at crash time in an 8GB container
• Crash addresses 0x1C and 0x20: Consistent with null-pointer dereference (low address near zero), likely a use-after-free in JSC/GC
• Happens during both fresh starts and resumed sessions (--resume flag present)
• Reduced but not fixed by upgrading from 1.4.0 → 1.4.1

Related issues

• #89437 — Bun fdb5e06cc crash rate 71%
• #88820 — Bun fdb5e06cc segfault
• #77579 — JSC GC use-after-free (SlotVisitor::drain)
• #87868 — Crash after long session + subagents
• #75538 — Concurrent subagents crash
• #76241 — glibc 2.41 crash
• oven-sh/bun#26843 — Upstream root cause (subprocess stdin use-after-free)

Expected behavior

Claude Code CLI should not segfault. If a crash happens, it should be caught and retried gracefully.

Workaround

Configuring a fallback model in the wrapper (OpenClaw) so crashes degrade to a secondary model instead of silence.