[sldev] Region crossing problem + patch

[Rob Lanphier]

Based on this feedback and some private feedback, let's push this to
1.2.  The convincing piece of feedback I've gotten is that this is the
type of fix that should go in early in a cycle rather than late in the
cycle, since the repro is so gnarly.

Thanks for taking a look.

Rob

On 07/29/2009 11:54 AM, Aleric Inglewood wrote:
> Hi,
>
> I have a bad feeling about this patch. It doesn't make
> things monolithic correct; it just patches one erroneous
> state that was detected under one particular (debug)
> condition.
>
> That state should not happen to begin with imho.
>
> Normally every object has a localID and an UUID,
> and the map should always map that localID to that
> UUID. If that ever gets out of sync then THAT is
> the bug.
>
> What happens in the patch is that it is detected
> that went wrong, therefore something is not done
> that normally should be done.
>
> Doesn't that sound wrong to you to?
> Why is this objects localID not mapped to its UUID?
> What happens because of this (wrong) state?
>
> If the localID --> UUID map contains wrong data
> (which apparently it still does with this patch) then
> calls to getUUIDFromLocal can return incorrect UUIDs
> for that very same object. getUUIDFromLocal is called
> while processing ObjectUpdate messages... I lose any
> overview of robustness here. I am not saying that this
> HAS to go wrong, but who is guaranteeing that we never
> receive an ObjectUpdate for this object while
> it is in this wrong state?
>