[postgis-tickets] r16421 - Use more precise bounds when testing overlap of temporal ranges.

[Paul Ramsey]

Author: pramsey
Date: 2018-02-26 10:27:03 -0800 (Mon, 26 Feb 2018)
New Revision: 16421

Modified:
   branches/2.3/NEWS
   branches/2.3/liblwgeom/lwlinearreferencing.c
Log:
Use more precise bounds when testing overlap of temporal ranges.
For 2.3, references #4025


Modified: branches/2.3/NEWS
===================================================================
--- branches/2.3/NEWS	2018-02-26 18:25:45 UTC (rev 16420)
+++ branches/2.3/NEWS	2018-02-26 18:27:03 UTC (rev 16421)
@@ -7,6 +7,7 @@
   - #4003, lwpoly_construct_circle: Avoid division by zero (Raúl Marín Rodríguez)
   - #4020, Casting from box3d to geometry now returns correctly connected
            PolyhedralSurface (Matthias Bay)
+  - #4025, Incorrect answers for temporally "amost overlapping" ranges
 
 PostGIS 2.3.6
 2018/01/17

Modified: branches/2.3/liblwgeom/lwlinearreferencing.c
===================================================================
--- branches/2.3/liblwgeom/lwlinearreferencing.c	2018-02-26 18:25:45 UTC (rev 16420)
+++ branches/2.3/liblwgeom/lwlinearreferencing.c	2018-02-26 18:27:03 UTC (rev 16421)
@@ -1073,7 +1073,7 @@
 {
 	LWLINE *l1, *l2;
 	int i;
-	const GBOX *gbox1, *gbox2;
+	GBOX gbox1, gbox2;
 	double tmin, tmax;
 	double *mvals;
 	int nmvals = 0;
@@ -1101,20 +1101,19 @@
 		return -1;
 	}
 
-	/* WARNING: these ranges may be wider than real ones */
-	gbox1 = lwgeom_get_bbox(g1);
-	gbox2 = lwgeom_get_bbox(g2);
+	/* We use lwgeom_calculate_gbox() instead of lwgeom_get_gbox() */
+	/* because we cannot afford the float rounding inaccuracy when */
+	/* we compare the ranges for overlap below */
+	lwgeom_calculate_gbox(g1, &gbox1);
+	lwgeom_calculate_gbox(g2, &gbox2);
 
-	assert(gbox1); /* or the npoints check above would have failed */
-	assert(gbox2); /* or the npoints check above would have failed */
-
 	/*
 	 * Find overlapping M range
 	 * WARNING: may be larger than the real one
 	 */
 
-	tmin = FP_MAX(gbox1->mmin, gbox2->mmin);
-	tmax = FP_MIN(gbox1->mmax, gbox2->mmax);
+	tmin = FP_MAX(gbox1.mmin, gbox2.mmin);
+	tmax = FP_MIN(gbox1.mmax, gbox2.mmax);
 
 	if ( tmax < tmin )
 	{