ST_CreateTopoGeo failed

[strk]

The attached SQL file imports a table of multipolygons which when collected and given in input to ST_CreateTopoGeo trigger an error about a mismatch between a node and a curve endpoint.

The cause of the bug is in issue #1273 (a point is not equal to the same point but with a bounding box added). I've filed this other bug because I belive another bug is hiding somewhere else in that nodes in the topology being constructed are being points with a bounding box cache attached, which isn't expected.

Also, I'll attach the input to this ticket.

[strk]

The fix for #1273 wasn't enough to fix this. Next stop: GetNodeByPoint returning 0 for a given point. Chances are is && or ST_Intersects or ST_DWithin failing.

[strk]

GetNodeByPoint returning 0 means no point found, which means a point added with ST_AddIsoNode isn't found later. Probably still suffering from float vs. double but I'll need closer inspection

[strk]

Oddly enough, interrupting the insertion of edges on error and querying the database I find 2 candidate points but both seem to be pretty far from each-other:

=# with p as ( select '0101000000B0B7BE9F8369294180870580ACEA2841'::geometry g ) select node_id, st_distance(geom, p.g) from small_sample_topo.node, p where geom && p.g;
 node_id |     st_distance
---------+----------------------
    2515 | 3.97360129963356e-08
    2516 | 0.000998956812181611
(2 rows)

Now to figure out how did that point move… (hoping ST_Distance is _not_ using bounding box)

[strk]

smaller sample

[strk]

Second sub-product of this dataset is in #1284

This dataset is triggering multiple sub-problems. Numbers can be pretty tricky!

[strk]

Third sub-product of this dataset was #1302, squashed this morning. Let's see who's next.

[strk]

Alright, it looks like there's nothing else. As of r8212 the small_sample.sql.gz dataset can be converted to a topology on both 32bit and 64bit system.

Summary:

 Topology small_sample_topo (2042), SRID 0, precision 0
 83 nodes, 156 edges, 74 faces, 0 topogeoms in 0 layers

It's interesting to note how once constructed the topology can be more easily cleaned by inspecting suspicious areas:

Edges which partecipate in the boundary of a faces whose bounding box has an area < 0.1:

with smallfaces as ( select face_id from small_sample_topo.face where st_area(mbr) < 0.1 ) select edge_id from small_sample_topo.edge_data where left_face in (select * from smallfaces ) or right_face in (select * from smallfaces) order by edge_id;

Automating further cleanup could mean getting the longest edge in each "small_face" and removing it, so to assign the small face area to an adiacent face.