What is the problem with visible mesh collision???

#1 Visible mesh collision can/may lead to performance problems if used exclusively. It is generally more optimal to use a collision mesh for an object that doesn't require absolute precision for collision. If a low detailed collision volume causes noticeable lack of collision precision then visible mesh collision may be more desirable.

#2 Is it safe to just use visible mesh collision until I run into some sort of issue with it, then when that happens work around it with an actual collision mesh, or will I run into traps building my set pieces that way?

#3 Not necessarily traps, but you can find yourself in place where you might have to make a hard decision of whether or not to redesign something later on. Something also to keep in mind is decals. If your building has a lot of contours and such then you'll want to use visible mesh for the decals, which unfortunately requires visible mesh collision to work.

#4 dup

#5 One more sorta question...

So after playing with this a bunch my workflow looks like...

1. Make game elements in max that have a simple(ish) visible mesh collision or box collision.

2. Arrange these items in the Torque3d World editor to make my level

3. After I am happy with the way the game level is working come back with detail meshes that I place directly over my collision meshes for added detail/visual interest, but assign no collision mesh to these objects.

There would be an extra step in there when I place objects with the forest brush... Probably when I am at the add detail stage...

Does it look like I am understanding the workflow?

#6 The visible mesh collision property is simply a toggle for TSStatics that makes the engine calculate collision at the mesh level, which is why it can become a performance problem - especially for highly detailed shapes. This is the easiest way for artists to drop shapes into Torque.

Actual collision meshes are created as part of a shapes hierarchy inside your modelling app, and not as separate shapes. Some bit of precision can be approximated by using multiple convex shapes instead of one general convex shape.