[BUG] Bundled Bun runtime Bus error on long-lived macOS Silicon sessions (2.1.245/2.1.246)

Status Open

Reported on v2.1.245

Maintainer reply None cached

Activity 0 comments · opened Aug 26, 2026

Summary

Multiple long-lived Claude Code sessions on macOS Silicon are hard-crashing with a Bun main-thread Bus error after many hours of uptime. This is not the known clipboard-paste 0xBAD4007 crash — fault addresses look like real heap pointers (several recent dumps end in …E30). System Bun is unrelated; crashes are inside the bundled standalone binary.

Until recently this machine ran many long sessions without Bun panics. After landing on 2.1.245 → 2.1.246 (native updater), several yesterday/today sessions died the same way.

Environment

Claude Code: 2.1.246 (also reproduced on 2.1.245)
Bundled Bun: v1.4.1 (aadadbbbe) on 2.1.246; v1.4.0 (fdb5e06cc) on 2.1.245
OS: macOS 26.6.2 (25G83), Apple Silicon (arm64)
Install: native binary at ~/.local/share/claude/versions/2.1.246
Updater: version_from: 2.1.245 → version_to: 2.1.246 at 2026-08-25T22:56:49Z
Typical launch args: --effort low --dangerously-skip-permissions --verbose

Crash dumps (same class)

1) 2.1.246 / Bun 1.4.1 — ~11.3 h uptime

Bun v1.4.1 (aadadbbbe) macOS Silicon
macOS v26.6.2
Args: ".../claude/versions/2.1.246" "--effort" "low" "--dangerously-skip-permissions" "--verbose"
Elapsed: 40837198ms | RSS: 0.43 GB | Peak: 0.67 GB | Faults: 1595
panic(main thread): Bus error at address 0x11D763E30
oh no: Bun has crashed. This indicates a bug in Bun, not your code.

(bun.report link was truncated in the terminal scrollback)

2) 2.1.246 / Bun 1.4.1 — ~7.9 h uptime

panic(main thread): Bus error at address 0x11E40FE30
Elapsed ≈ 28551s | RSS ≈ 0.53 GB

3) 2.1.245 / Bun 1.4.0 — ~25.9 h uptime

Bun v1.4.0 (fdb5e06cc) macOS Silicon
macOS v26.6.2
Args: ".../claude/versions/2.1.245" "--effort" "low" "--dangerously-skip-permissions" "--verbose"
Features: ... fetch(321) jsc spawn(1152) standalone_executable process_dlopen yaml_parse(131) claude_code
Elapsed: 93128579ms | User: 506624ms | Sys: 67914ms
RSS: 0.38 GB | Peak: 0.85 GB | Commit: 0.75 GB | Faults: 344 | Machine: 0.1 TB
panic(main thread): Bus error at address 0x116BD3E30
oh no: Bun has crashed. This indicates a bug in Bun, not your code.

Why this is not the known image-paste bug

| | This report | Known paste crash (#41603 / #66116 / etc.) |
|---|---|---|
| Address | 0x11……E30 (heap-like) | 0xBAD4007 (poison/sentinel) |
| Trigger | long-lived interactive session (hours) | Cmd+V clipboard image |
| Timing | mid-session after many hours | immediate on paste |

Memory at crash was modest (0.4–0.5 GB RSS, peak < 1 GB) — does not look like OOM.

Impact

Several concurrent long sessions (yesterday + today) terminated with the same Bun panic banner. Work is recoverable via --resume / session JSONL, but the interactive TUI dies hard with no graceful recovery.

Expected

Long-running sessions should not SIGBUS inside the bundled Bun runtime. If a native/JSC fault is unavoidable, Claude Code should catch/report and exit cleanly without the Bun panic dump, and preferably auto-resume.

Notes / possible clues

Fault addresses from three independent dumps all end in E30 (0x11D763E30, 0x11E40FE30, 0x116BD3E30) — may indicate one code path / object shape rather than random corruption.
Spans both Bun 1.4.0 (CC 2.1.245) and 1.4.1 (CC 2.1.246), so not unique to the 246 JS payload alone.
Happy to attach a full bun.report URL on the next occurrence if the link survives scrollback.

View original on GitHub ↗