Hello,

As seen in ccb80964cd7cd112e300c34d32f67125a6d6da9a, there’s a lock
ordering mismatch between ‘do_thread exit’ and ‘fat_mutex_lock’
wrt. ‘t->admin_mutex’ and ‘m->lock’.

I thought this commit solved the problem, but now I think it doesn’t
because it leaves a small window during which a mutex could be held by a
thread without being part of its `mutexes' list, thereby violating the
invariant tested at line 667:

        /* Since MUTEX is in `t->mutexes', T must be its owner.  */
        assert (scm_is_eq (m->owner, t->handle));

So I reverted the patch.

(The situation isn’t better without the patch since
“while ./check-guile srfi-18.test threads.test ; do : ; done”
eventually hits the assertion failure.)

I tried the attached patch, which is inelegant and leads to deadlocks
with canceled threads (namely the “cancel succeeds” test in
threads.test.)

IOW I would welcome fresh ideas to approach the problem.  :-)

Thanks,
Ludo’.