Stable updates. Refcount leak + SET change on alloc fail

[dormando]

Uh, I'm an idiot. that's the DPM repo.

git clone http://consoleninja.net/code/memcached/memcached.git - sigh.

dormando wrote:
> Yo,
> 
> git-send-email is being a punk, so please check the three latest commits
> here:
> 
> http://consoleninja.net/gitweb/gitweb.cgi?p=memcached.git;a=shortlog;h=stable
> 
> (can also clone the repo:
> git clone http://consoleninja.net/code/dpm/dpm.git and fetch my 'stable'
> branch).
> 
> This is first in a small series of patches going in before 1.2.6-rc1.
> I've audited all item_get* calls and ensured they were matched with
> item_remove() calls properly. I might've missed something, so feel free
> to check for yourselves :)
> 
> The other change is a little weird, but makes sense to me. If you 'set'
> an item and the set fails alloc (too large, OOM), unlink an item if it
> already exists. This would help in the case of someone getting OOM
> errors, or if they're sloppy and occasionally fail to update their
> cache. Giving into the side of additional cache misses for data correctness.
> 
> Please review, or if you hate gitweb whine and I'll make git-send-email
> cooperate again.
> 
> I'm going through the stack for anything else to apply to stable. If
> you've sent a relatively minor patch and would totally love to see it in
> 1.2.6, now's the time to speak up.
> 
> -Dormando