shithub: nanobsp

Info  •  Files  •  Log  •  Branches

ensure a final partition's delta is not too small.

Well this "bug" took all day to investigate, but the culprit turns out
to be the existing R_PointOnSide() code, which has very poor accuracy
when the delta is small, leading to bad results during BSP traversals.

I wasted the better part of two hours yesterday to discover that the
existing M_AddToBox() code in m_bbox.c has a serious bug.  Such are
the joys of programming I guess :)

--- a/nano_bsp.c

+++ b/nano_bsp.c

@@ -347,6 +347,11 @@

 	eval->right = 0;

 	eval->split = 0;

+	// do not create tiny partitions

+	if (abs (part->v2->x - part->v1->x) < 4*DIST_EPSILON &&

+		abs (part->v2->y - part->v1->y) < 4*DIST_EPSILON)

+		return false;

+

 	seg_t * S;

 	for (S = soup ; S != NULL ; S = S->next)

 	{

@@ -657,6 +662,18 @@

 	N->y  = part->v1->y;

 	N->dx = part->v2->x - N->x;

 	N->dy = part->v2->y - N->y;

+

+	// ensure partitions are a minimum length, since the engine's

+	// R_PointOnSide() functions have very poor accuracy when the

+	// delta is too small, and that WILL BREAK a map.

+

+	fixed_t min_size = 64 * FRACUNIT;

+

+	while (abs (N->dx) < min_size && abs (N->dy) < min_size)

+	{

+		N->dx *= 2;

+		N->dy *= 2;

+	}

 	// these are the new lists (after splitting)

 	seg_t * lefts  = NULL;

-- 

⑨