Would it make sense to make xbps_pkgdb_init and xbps_rpool_init thread-safe with pthread_mutex_t? This would eliminate the need for workarounds and make lazy initialization work safely whether called from single-threaded or multithreaded code. The performance overhead should be negligible since the mutex is only contended during the initial load.

This would also help prevent similar issues.

Happy to take a stab at implementing this if you think it's worthwhile.

My plan was to remove the lazy initialization instead.

I thought about that too, although it will be a breaking API change. How is something like that normally coordinated?