Gremlins happened whenever P_TryMove and P_SlideMove/P_BounceMove disagreed on what an object collided with. When TryMove said you collided with a line, but P_BounceMove said that you didn't, then you'd get gremlin'd.
To fix this, P_TryMove now can edit a struct to contain information on what it collides with. P_SlideMove and P_BounceMove no longer try to detect walls on their own and now requires this result from P_TryMove. If a slide/bounce is needed without moving the object, then you'd want to use P_CheckMove to get the result.
Nep Note: I tried implementing lua support, its probably shit lmao
This makes it significantly easier to save/restore the state of these variables, whenever we need to do so for calling movement functions in the middle of other movement functions. This will also make it easier to move it out of global variable hell if desired later.
Many improvements that were in the raycast-physics branch, backported since that branch is probably dead, and if I ever revive it then it will be with a new approach. Many are sourced from PRBoom +ZDoom + Eternity Engine.
- Some line / divline comparison functions use const.
- P_PointOnLineSide and P_BoxOnLineSize use PrBoom's version.
- P_InterceptVector is more resistant to overflow.
- PIT_AddThingIntercepts checks against actual bounding box, instead of a corner-to-corner cross-section.
- P_TraverseIntercepts doesn't run the callback function on NULL intercepts.
- P_PathTraverse handles traces that go through blockmap corners properly.
- Removed unused earlyout flag to simplify logic.
- Added g_ prefix to trace.
An optimized error check was put into P_BlockThingsIterator for tmthing, but this meant that anything that was not using or even setting this variable (the supermajority of them) would exit after finding 1 thing in each block.
To keep this error check & keep it optimized, the PIT_ functions were expanded to three options instead of two: BMIT_CONTINUE lets the search continue. If used as the last mobj, it will return true when the iterator finishes. BMIT_ABORT ends the search and forces the iterator to return false.
BMIT_STOP ends the search and forces the iterator to return true.
See notably the finish line. Basically respawning exactly on a line can let you
cross it twice, if you crossed it before respawning, or NOT cross it, depending
on which direction you drive after landing. So this just respawns very slightly
before the line so you can cross (or not cross) it normally.
Now FixedHypot uses the code from R_PointToDist2, and R_PointToDist2 just calls FixedHypot.
Ultimately, this branch was intended to get rid of a redundant way to retrieve distance and replace it with the one that was actually good at its job. So consolidating FixedHypot and R_PointToDist2 together is just an extension of that.
Not convinced that the small speed benefit from P_AproxDistance is worth the "aproximate"[sic] results it gives. Let's instead try a define to replace it with FixedHypot. In Lua, the function gives a deprecated warning.
Inspired by the hyperwall fix for vanilla, except for everything. From little testing, actively improves waypoint checks, bumping, speed checks, wall collisions, Jawz targetting, Lightning Shield attacks, so on.
The only way I see this as a potential downgrade is A_Look (and related functions) getting slower, which are barely used in Kart.
This should fix the janky falling off steep slopes and let you cling to them.
Because when falling off a sector the edge of the player must not intersect
the edge of the higher sector, it is trickier than step up. The height
difference at the line must be stored to be useful.
This is for slopes. When moving across a slope the height changes. Therefore
compare the height at the edge where slopes meet so only a wall will block the
player, and not a very steep slope.
. tmthing can be NULL if called from PTR_SlideTraverse, so we should use slidemo instead
This fixes a crash that occurs in yet ANOTHER SUGOI map, involving bouncy walls next to sloped floors/ceilings
