* Updated to ZDoom r3603:

- Fixed: The player's position was only predicted during the duration of R_SetupFrame(). Consequently, if somebody used the chasecam while predicting, they would appear to lag behind the camera, because their actor would be at its unpredicted position by the time sprites were processed.
- Fixed: 3D floors partitioning the world into different "HeightLevels" could strip a midtexture of its clipping information before it was drawn. (On the other hand, I have no idea what this ds->bkup and refreshing stuff is about, so I might have broken something somewhere else.)

git-svn-id: http://mancubus.net/svn/hosted/gzdoom/trunk@1369 b0f79afe-0144-0410-b225-9a4edf0717df

src/r_main.cpp

@@ -861,6 +861,7 @@ void R_RenderActorView (AActor *actor, bool dontmaplines)
 	}
 	WallMirrors.Clear ();
 	interpolator.RestoreInterpolations ();
+	P_UnPredictPlayer();
 	R_SetupBuffer ();
 
 	// If we don't want shadered colormaps, NULL it now so that the