Control: tags -1 moreinfo I see there are 3 threads in the full backtrace. In the version in bookworm there was still an async cleanup routine that used posix threads. This is known to mix really badly with glibc's handling of malloc internals, so I strongly suspect that's the issue. This was replaced with using subprocesses in trixie.
Are you able to reproduce the issue with stable-backports or testing/unstable?
