v.distance: 3d point inside area is classified as outside

[annakrat]

When a point is 3D and has different z coordinate than 0 and this point is inside 2D area, v.distance reports that this point is outside the area. The problem is in ​distance.c, which calls ​Vect_point_in_box - the top and bottom of the area's box are 0, the z coordinate (of the 3D point) is not.

One solution is to do something like this:

Index: distance.c
===================================================================
--- distance.c	(revision 66134)
+++ distance.c	(working copy)
@@ -291,7 +291,7 @@
 
     for (i = 0; i < Points->n_points; i++) {
 	if (Vect_point_in_box(Points->x[i], Points->y[i],
-			      Points->z[i], abox)) {
+			      with_z ? Points->z[i] : 0, abox)) {
 
 	    int poly;

but then the comparison of doubles (0 == 0) in Vect_point_in_box might not always work.

Or we could introduce a 2D version of the Vect_point_in_box function which checks x and y only.

[wenzeslaus]

On Wed, Oct 7, 2015 at 4:32 AM, Moritz Lennert wrote at ​grass-user:

I have the feeling that the second proposed solution, i.e. "introduce a 2D version of the Vect_point_in_box function which checks x and y only" is the better long-run solution.

Good idea. I could use this, e.g. in v.decimate​ where I have now point_in_region_2d() and point_in_region_3d. I'm using Cell_head as the region/bbox, but it doesn't contain any info whether it is 2D or 3D, if something like depths=0 (perhaps wrapped in G_is_cell_head_3d()) would tell if it is 3D, then I could have just point_in_region().

[wenzeslaus]

In r66459 I have added Vect_point_in_box_2d() function which is simply a 2D version of Vect_point_in_box() function which is 3D only. Alternative would be to add one function which would optionally consider z.

Using Vect_point_in_box_2d() and existing with_z logic I have tried to fix this ticket and ​reported issue in r66460 in v.decimate. The test written for v.decimate would suggest that the change helped.

Vect_point_in_box() function is used a lot in the modules and very often with hardcoded 0.0 as z parameter (and there might be also cases when z is optionally zero). These should be replaced by the new function or its universal alternative.

If Vect_point_in_box_2d() will be useful for v.decimate remains unclear. Usage of Vect_region_box() function would be necessary.

[neteler]

Ticket retargeted after milestone closed

[neteler]

Ticket retargeted after milestone closed

[neteler]

Ticket retargeted after 7.0.3 milestone closed

[wenzeslaus]

Changes to the library and v.distance backported to release branch 7.0 in r68143 and r68146. Test them using NC SPM full dataset:

v.distance -p from=precip_30ynormals_3d to=boundary_county dmax=0 upload=cat,dist

You should get:

from_cat|(null)|(null)
1|501|0
2|511|0
3|19|0
4|261|0
...

In previous versions you will get:

from_cat|(null)|(null)
1|null|null
2|null|null
3|null|null
4|null|null
...

The changes to rest of the modules are outside of scope of this ticket and are tracked in #2970.