db4.5 rel 5 in th-test

[Jeff Johnson]

On Jun 29, 2007, at 5:38 AM, Arkadiusz Miskiewicz wrote:

>
> What I'm more interested in is impact of that robustness patch on  
> software
> other than rpm itself. Will it cause problems and so on...
>

When one inherits a lock, then one also inherits the responsibility
for insuring that whatever the lock was protecting is sane.

For rpm, the Berkeley DB access patterns are simple and predictable.
Header data used for secondary lookup is verified with digests.

The benefit of avoiding a deadlock (that can only be cleared by  
rebooting)
outweighs the risk of ignoring side effects and clearing the lock.  
Even if that
assumption is wrong, the header sha1 will detect damaged headers and
--rebuilddb will regenerate indices.

I don't think that the robust mutex patch can simply be pushed into  
system db.
One could/should add a flag to explicitly enable the functionality  
and the default
behavior of system db should not use robust mutexes.

73 de Jeff