When checking visibility, we now use the Domains more extensively.
The base rules are:
* If an entity is a child of a domain-less entity, visiblity for the
child is the same as for the parent.
* If an entity is a child of a domain-holding entity, visibility is
determined by the Domain.
* Entities cannot observe things above a Domain.
* Entities can observe things below a Domain, if the Domain allows it.
These rules are not yet set in stone. We need some more time figuring out
exactly how we want visibility implemented. We're striving for a balance
between a powerful but not too complex system, with good performance.
It would be better to use a Broadcast of the Disappear op though, but
that would require a change so that the location is encoded in the op
which will be broadcasted. Otherwise the WorldRouter will just use the
location of the entity as it is when the op is processed, which will be
wrong for this case.
Collision detection should fully be handled by the domain, as should
sight checking.
The plan is to let the domain keep a separate structure of all collision
geometry, optimized in something like a quad or oct tree. Since the vast
majority of geometry is static, in the sense that it very seldom moves,
the Domain should keep a flat representation of all geometry, with
precomputer meshes for fast lookup.
Similarily it should keep a separate structure, or structures, for sight
checks.
And it should of course also use Bullet for collision detection.
As any world can contain multiple domains, we let the domain handle the
sights calculations.
This is a change from how this previously was done, as previously
sight checks were only done for the top level World entity. This has
changes however, so that sights checks now are done for all entities.
When determining whether an entity can be seen, the system first checks
the size of the entity against the distance from the observer. If this
check fails, the system makes exceptions if the entity in case is
outfitted or wielded.
