These are all not thread safe and thus must lock the mutex or
risk undefined behavior.

Igor Kushnir writes:
> ItemRepository::close() was called from:
>
> 1. ~ItemRepository() and ItemRepository::open() - no need for a lock;
> 2. ItemRepositoryRegistry::unRegisterRepository() - no need for a lock because unRegisterRepository() is called only from within ~ItemRepository(). But there seems to be an inefficiency here: close() is called from unRegisterRepository() and then immediately from ~ItemRepository() again. Should perhaps this call be removed from unRegisterRepository()?
> 3. from ItemRepositoryRegistryPrivate::close(), which in turn is called from ~ItemRepositoryRegistry(). Here the missing mutex lock might have been a bug.

Then there was `ItemRepositoryRegistry::printAllStatistics` which
didn't lock the repository mutex when printing statistics.