>>>>> "Guilhem" == Guilhem Bichot <guilhem@stripped> writes:
Guilhem> sanja@stripped a écrit, Le 12/05/2008 11:11 AM:
>> #At bzr+ssh://bk-internal.mysql.com/bzrroot/server/mysql-maria/
>> 2709 sanja@stripped 2008-12-05
>> Commit for review (debugging my_sleep()s are not removed)
Guilhem> Don't forget to remove them :)
Guilhem> I realize: by unlocking intern_lock in the middle of maria_close(), we
Guilhem> are making a half-destroyed share visible to checkpoint (for example
Guilhem> this share has some destroyed mutexes, freed components). I think that
Guilhem> the pieces of share which checkpoint may access are not destroyed yet,
Guilhem> so this should be safe, but could you please put a comment, where you
Guilhem> release the mutex, to warn against this? Something like "by unlocking
Guilhem> the mutex we make the half-destroyed share visible to checkpoint so we
Guilhem> have to be careful to what checkpoint reads of it".
Guilhem> Ok to push when you have Monty's approval. Note that after your push,
Guilhem> the warnings in master.err for test maria-lock.test in 6.0-maria should
Guilhem> go away (I mentioned this in the bug report).
Checkpoint shouldn't see the half-destroyed share as we remove the
share from the open list *before* we free intern_lock.