At least in pike7.6, ./configure detects pthread_atfork() and then the
resulting program uses it, and fails because for now our implementation
If we weren't providing the function, pike would manage to work without
it. So until we have a proper implementation maybe we should add a
debian patch that drops the function, so as to have circumvents such as
(people wouldn't expect such function to fail, so having a stub that
returns ENOSYS may be asking for troubles).