 |
Added code to idBounds.Intersect() and idBounds.IntersectSelf() to handle null intersections. Without it, it was possible to generate a search area that couldn’t be handled by the hiding spot search, causing a hang. AAS data was not the problem. rev. 4806: bounds.h - the fix DarkModGlobals.cpp - added AAS logging DarkModGlobals.h - added AAS logging rev. 11844: tdm_defs.script - added AAS logging |
 |
Just bumped into this change of idBounds intersection. Well, I must say the new way is not how box intersection is expected to work. When boxes don't intersection, their "intersection" must be a box inverted across at least one dimension. Luckily, there are only a few places which call this method, so I can revert it by moving the check for intersection into call sites. What is important is that the IntersectSelf is used in interaction culling, where this old change efficiently disabled culling in most cases. I wonder how this did not cause noticeable performance regression. UPDATE: In fact, this change could not be noticed until Doom 3 source code was integrated into TDM codebase, since at the time of 1.04 the game engine was a separate .exe with its own unmodifiable instance of idlib. |
|
|
I don't understand what "their 'intersection' must be a box inverted across at least one dimension" means. That makes no sense to me. |
|
|
Committed the revert in rev 8468. |
- Anonymous
