Merge pull request #694 from Maxxen/v1.4-andium-dev

Setup CI for v1.4 branch,  fix bug in st_distance when processing zero-length segments in polygon rings

Author: Maxxen

.github/workflows/MainDistributionPipeline.yml

@@ -5,13 +5,13 @@ name: Main Extension Distribution Pipeline
 on:
   pull_request:
     branches:
-      - main
+      - v1.4-andium
     paths-ignore:
       - '**/README.md'
       - 'doc/**'
   push:
     branches:
-      - main
+      - v1.4-andium
     paths-ignore:
       - '**/README.md'
       - 'doc/**'
@@ -27,7 +27,7 @@ jobs:
     name: Build extension binaries
     uses: duckdb/extension-ci-tools/.github/workflows/_extension_distribution.yml@main
     with:
-      duckdb_version: main
+      duckdb_version: v1.4.1
       extension_name: spatial
       ci_tools_version: main
       vcpkg_commit: ce613c41372b23b1f51333815feb3edd87ef8a8b
@@ -38,7 +38,7 @@ jobs:
     uses: duckdb/extension-ci-tools/.github/workflows/_extension_deploy.yml@main
     secrets: inherit
     with:
-      duckdb_version: main
+      duckdb_version: v1.4.1
       ci_tools_version: main
       extension_name: spatial
       deploy_latest: ${{ startsWith(github.ref, 'refs/heads/v') || github.ref == 'refs/heads/main' }}

duckdb

@@ -3130,13 +3130,30 @@ static double point_segment_dist_sq(const vertex_xy &p, const vertex_xy &a, cons
 	return diff.norm_sq();
 }
 
+// Check if point P is on segment QR
 static bool point_on_segment(const vertex_xy &p, const vertex_xy &q, const vertex_xy &r) {
 	return q.x >= std::min(p.x, r.x) && q.x <= std::max(p.x, r.x) && q.y >= std::min(p.y, r.y) &&
 	       q.y <= std::max(p.y, r.y);
 }
 
 static bool segment_intersects(const vertex_xy &a1, const vertex_xy &a2, const vertex_xy &b1, const vertex_xy &b2) {
 	// Check if two segments intersect using the orientation method
+	// Handle degenerate cases where a segment is actually a single point
+	const bool a_is_point = (a1.x == a2.x && a1.y == a2.y);
+	const bool b_is_point = (b1.x == b2.x && b1.y == b2.y);
+
+	if (a_is_point && b_is_point) {
+		// Both are points: intersect only if identical
+		return (a1.x == b1.x && a1.y == b1.y);
+	}
+	if (a_is_point) {
+		// A is a point: check if A lies on segment B
+		return point_on_segment(a1, b1, b2);
+	}
+	if (b_is_point) {
+		// B is a point: check if B lies on segment A
+		return point_on_segment(b1, a1, a2);
+	}
 
 	const auto o1 = orient2d_fast(a1, a2, b1);
 	const auto o2 = orient2d_fast(a1, a2, b2);
@@ -3250,6 +3267,15 @@ static bool try_get_prepared_distance_lines(const prepared_geometry &lhs, const
 			for (uint32_t i = lhs_beg_idx + 1; i < lhs_end_idx; i++) {
 				memcpy(&lhs_next, lhs_vertex_array + i * lhs_vertex_width, sizeof(vertex_xy));
 
+				// If this is a zero-length segment, skip it
+				// LINESTRINGs must have at least two distinct vertices to be valid, so this is safe. Even if we skip
+				// this vertex now, we must eventually reach a non-zero-length segment that includes this vertex as
+				// its start point. It will therefore still contribute to the distance calculation once we process that
+				// segment.
+				if (lhs_prev.x == lhs_next.x && lhs_prev.y == lhs_next.y) {
+					continue;
+				}
+
 				// Quick check: If the distance between the segment and the box (all the segments)
 				// is greater than min_dist, we can skip the exact distance check
 
@@ -3268,6 +3294,13 @@ static bool try_get_prepared_distance_lines(const prepared_geometry &lhs, const
 				for (uint32_t j = rhs_beg_idx + 1; j < rhs_end_idx; j++) {
 					memcpy(&rhs_next, rhs_vertex_array + j * rhs_vertex_width, sizeof(vertex_xy));
 
+					// If this is a zero-length segment, skip it
+					// LINESTRINGs must have at least two distinct points to be valid, so this is safe.
+					// (see comment above)
+					if (rhs_prev.x == rhs_next.x && rhs_prev.y == rhs_next.y) {
+						continue;
+					}
+
 					// Quick check: If the distance between the segment bounds are greater than min_dist,
 					// we can skip the exact distance check
 					extent_xy rhs_seg;

src/sgl/sgl.cpp

@@ -26,3 +26,15 @@ SELECT ST_Distance(
 ----
 5
 
+
+# This test also comes from a bug report, where ST_Distance would incorrectly handle line segments with zero length.
+# The issue is that we were not properly considering segments where both endpoints are the same point, and conclude
+# that a zero length segment always intersects with any other segment, which is not correct. We now skip zero-length
+# segments when performing intersection tests between indexed polygon rings, and also handle zero-length segments as
+# points when performing regular segment-segment intersection tests.
+
+query I
+select st_distance('POLYGON((0 0, 1 1, 1 1, 10 10, 0 0))'::GEOMETRY, 'POLYGON((4 0, 4 1, 5 1, 5 0, 4 0))'::GEOMETRY) != 0;
+----
+true
+