[BUG] v2.1.243 native binary segfaults on startup — NULL deref in free() called from __newlocale

### Preflight Checklist

- [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet
- [x] This is a single bug report (please file separate reports for different bugs)
- [x] I am using the latest version of Claude Code

### What's Wrong?

Environment
- OS: CachyOS (Arch-based), kernel 7.2.0-1-cachyos
- Arch: x86_64
- glibc: 2.44 (ldd (GNU libc) 2.44)
- Install method: native/standalone installer (~/.local/share/claude/versions/, curl install script)
- Locale: en_AU.UTF-8 — ruled out as the cause, see below

Summary

Every launch of v2.1.243 (fresh downloads via both the background auto-updater and a manual curl -fsSL https://claude.ai/install.sh | bash re-run) segfaults immediately, including the installer's own
internal "$binary_path" install step, so a clean install is impossible. v2.1.241, v2.1.240, and v2.1.239 all launch fine on the same machine.

Crash details

dmesg shows the identical crash on every invocation, same instruction pointer, different PIDs/cores:

claude[10989]: segfault at 0 ip 0000000001d10458 sp 00007ffccf885140 error 4 in 2.1.243[1b0f458,1953000+3b33000] likely on CPU 4 (core 24, socket 0)

error 4 = user-mode read fault at address 0 — a NULL pointer dereference, not a write/exec fault. Fully deterministic: same ip across 5+ separate crashes on different CPU cores.

Backtrace (via coredumpctl gdb claude)

Stack trace of thread 11965:
#0 0x0000000001d10458 free (2.1.243 + 0x1b10458)
#1 0x00007fc5f603a1ab __newlocale (libc.so.6 + 0x3a1ab)
#2 0x0000000001abd282 n/a (2.1.243 + 0x18bd282)
#3 0x00007fc5f60b2584 n/a (libc.so.6 + 0xb2584)
#4 0x00007fc5f60b26a9 pthread_once (libc.so.6 + 0xb26a9)
#5 0x0000000001abc2d2 n/a (2.1.243 + 0x18bc2d2)
#6 0x0000000001ace6cf n/a (2.1.243 + 0x18ce6cf)
#7 0x0000000001abdc0a n/a (2.1.243 + 0x18bdc0a)
#8 0x0000000001abe9c4 n/a (2.1.243 + 0x18be9c4)
#9 0x00007fc5f60b2584 n/a (libc.so.6 + 0xb2584)
#10 0x00007fc5f60b26a9 pthread_once (libc.so.6 + 0xb26a9)
#11 0x0000000001abea15 n/a (2.1.243 + 0x18bea15)
#12 0x0000000001abead7 n/a (2.1.243 + 0x18bead7)
#13 0x0000000001abb922 n/a (2.1.243 + 0x18bb922)
#14 0x0000000001abb724 n/a (2.1.243 + 0x18bb724)
#15 0x0000000001aadadd n/a (2.1.243 + 0x18adadd)
#16 0x00007fc5f6027e23 __libc_start_main (libc.so.6 + 0x27e23)
#17 0x0000000001953a2e n/a (2.1.243 + 0x1753a2e)

Notably, frame #0's free resolves to an address inside the app binary (2.1.243 + 0x1b10458), not inside libc.so.6, even though it's called from glibc's own __newlocale (libc.so.6 + 0x3a1ab). This looks
like the binary statically links a custom allocator (mimalloc or similar — the pthread_once-driven lazy-init pattern in frames #4–#15 is consistent with a Bun-built single executable) that globally
overrides the free symbol. When glibc's internal __newlocale frees memory it allocated through its own internal path, that call gets intercepted by the app's custom allocator instead of glibc's real
one, which has no valid metadata for a block it never allocated — hence the NULL deref.

Ruling out other causes

- Not a corrupted download: reproduced identically across the original auto-update, a deleted-and-redownloaded copy, and a second full curl | bash reinstall.
- Not architecture/library mismatch: file confirms valid x86_64 ELF; ldd resolves every shared library with no missing symbols.
- Not disk/memory exhaustion: 20Gi free RAM, 46Gi swap, 188G free disk at time of crash.
- Not locale-specific: reproduced identically with LC_ALL=C forced for the entire install pipeline (LC_ALL=C bash -c 'curl -fsSL https://claude.ai/install.sh | bash'), ruling out en_AU.UTF-8 as the
trigger. Given the crash is in glibc's locale-init path, the likely trigger is this machine's newer glibc (2.44) exposing an allocator-interposition edge case that older/more common glibc versions
don't hit.

Workaround in use

Symlinked ~/.local/bin/claude back to the working 2.1.241 binary and set DISABLE_AUTOUPDATER=1 in ~/.claude/settings.json to stop the updater from overwriting it with the broken build again.

### What Should Happen?

claude 2.1.243 should not fail installing nor fail to launch

### Error Messages/Logs

```shell

```

### Steps to Reproduce

Let claude auto-update or manual update via `curl -fsSL https://claude.ai/install.sh | bash`

### Claude Model

None

### Is this a regression?

Yes, this worked in a previous version

### Last Working Version

2.1.241

### Claude Code Version

2.1.241 (Claude Code)

### Platform

Anthropic API

### Operating System

Other Linux

### Terminal/Shell

Other

### Additional Information

Cachyos, GNOME Console 50.0

Reproduce the failure using the native standalone installer (`curl ... | bash`) on CachyOS with glibc 2.44, then inspect the v2.1.243 startup path around `__newlocale` and the binary's overridden `free`; compare with v2.1.241. Done means the installer completes and v2.1.243 launches on the reported environment without the NULL dereference.

Rédigé par le modèle d'indexation à partir du texte de l'issue.