Stabilize feature order in overzoom (#210)

* Stabilize feature order in overzoom

* I want my sorts to be stable, please

* Revert "[ci] test in debug mode (#202)"

This reverts commit 853ada8.

* No need to reinitialize here

.github/workflows/test.yml

@@ -5,10 +5,7 @@ on: [push]
 jobs:
   test:
     runs-on: ubuntu-latest
-    strategy:
-      matrix:
-        version: ['Release', 'Debug']
     steps:
       - uses: actions/checkout@v3
-      - run: uname -a; BUILDTYPE=${{ matrix.version }} make
-      - run: make test
+      - run: uname -a; make
+      - run: make test

CHANGELOG.md

@@ -1,3 +1,7 @@
+# 2.47.0
+
+* Stabilize feature order in tippecanoe-overzoom when --preserve-feature-order is specified but the sequence attribute is not present
+
 # 2.46.0
 
 * Polygon dust returns to having the attributes of the contributing feature nearest the placeholder instead of the contributing feature with the largest area.

Makefile

@@ -278,6 +278,12 @@ overzoom-test: tippecanoe-overzoom
 	./tippecanoe-decode tests/pbf/13-1310-3166.pbf 13 1310 3166 > tests/pbf/13-1310-3166.pbf.json.check
 	cmp tests/pbf/13-1310-3166.pbf.json.check tests/pbf/13-1310-3166.pbf.json
 	rm tests/pbf/13-1310-3166.pbf tests/pbf/13-1310-3166.pbf.json.check
+	# Make sure feature order is stable
+	./tippecanoe-overzoom --preserve-input-order -o tests/pbf/11-327-791-out.pbf tests/pbf/11-327-791.pbf 11/327/791 11/327/791
+	./tippecanoe-decode tests/pbf/11-327-791.pbf 11 327 791 > tests/pbf/11-327-791.json
+	./tippecanoe-decode tests/pbf/11-327-791-out.pbf 11 327 791 > tests/pbf/11-327-791-out.json
+	cmp tests/pbf/11-327-791.json tests/pbf/11-327-791-out.json
+	rm tests/pbf/11-327-791.json tests/pbf/11-327-791-out.json tests/pbf/11-327-791-out.pbf
 	# Different detail and buffer, and attribute stripping
 	./tippecanoe-overzoom -d8 -b30 -y NAME -y name -y scalerank -o tests/pbf/13-1310-3166-8-30.pbf tests/pbf/11-327-791.pbf 11/327/791 13/1310/3166
 	./tippecanoe-decode tests/pbf/13-1310-3166-8-30.pbf 13 1310 3166 > tests/pbf/13-1310-3166-8-30.pbf.json.check

clip.cpp

@@ -1026,7 +1026,7 @@ std::string overzoom(const mvt_tile &tile, int oz, int ox, int oy, int nz, int n
 		}
 
 		if (preserve_input_order) {
-			std::sort(outlayer.features.begin(), outlayer.features.end(), preservecmp);
+			std::stable_sort(outlayer.features.begin(), outlayer.features.end(), preservecmp);
 		}
 
 		if (outlayer.features.size() > 0) {

dirtiles.cpp

@@ -176,7 +176,7 @@ std::vector<zxy> enumerate_dirtiles(const char *fname, int minzoom, int maxzoom)
 		closedir(d1);
 	}
 
-	std::sort(tiles.begin(), tiles.end());
+	std::stable_sort(tiles.begin(), tiles.end());
 	return tiles;
 }
 

geometry.cpp

@@ -1189,7 +1189,7 @@ drawvec polygon_to_anchor(const drawvec &geom) {
 				std::vector<candidate> candidates;
 
 				for (size_t pass = 0; pass < 4; pass++) {
-					std::sort(points.begin(), points.end(), sorty_sorter(pass));
+					std::stable_sort(points.begin(), points.end(), sorty_sorter(pass));
 
 					for (size_t i = 1; i < points.size(); i++) {
 						double dx = points[i].x - points[i - 1].x;
@@ -1213,7 +1213,7 @@ drawvec polygon_to_anchor(const drawvec &geom) {
 				// segment, if we find one whose midpoint is inside the polygon and
 				// far enough from any edge to be good enough, stop looking.
 
-				std::sort(candidates.begin(), candidates.end());
+				std::stable_sort(candidates.begin(), candidates.end());
 				// only check the top 50 stride midpoints, since this list can be quite large
 				for (size_t i = 0; i < candidates.size() && i < 50; i++) {
 					double maybe_goodness = label_goodness(geom, candidates[i].x, candidates[i].y);

main.cpp

@@ -2704,7 +2704,7 @@ std::pair<int, metadata> read_input(std::vector<source> &sources, char *fname, i
 				}
 			}
 
-			std::sort(ddv.begin(), ddv.end());
+			std::stable_sort(ddv.begin(), ddv.end());
 
 			size_t i = 0;
 			for (int z = 0; z <= basezoom; z++) {

mvt.cpp

@@ -377,7 +377,7 @@ std::string mvt_tile::encode() {
 			sorted_values.push_back(std::move(sv));
 		}
 
-		std::sort(sorted_values.begin(), sorted_values.end());
+		std::stable_sort(sorted_values.begin(), sorted_values.end());
 		std::vector<size_t> mapping;
 		mapping.resize(sorted_values.size());
 

pmtiles_file.cpp

@@ -174,7 +174,7 @@ void mbtiles_map_image_to_pmtiles(char *fname, metadata m, bool tile_compression
 		sqlite3_finalize(stmt);
 	}
 
-	std::sort(tile_ids.begin(), tile_ids.end());
+	std::stable_sort(tile_ids.begin(), tile_ids.end());
 
 	std::unordered_map<std::string, std::pair<unsigned long long, unsigned long>> hash_to_offset_len;
 	std::vector<pmtiles::entryv3> entries;
@@ -263,7 +263,7 @@ void mbtiles_map_image_to_pmtiles(char *fname, metadata m, bool tile_compression
 
 	// finalize PMTiles archive.
 	{
-		std::sort(entries.begin(), entries.end(), pmtiles::entryv3_cmp());
+		std::stable_sort(entries.begin(), entries.end(), pmtiles::entryv3_cmp());
 
 		std::string root_bytes;
 		std::string leaves_bytes;

serial.cpp

@@ -604,7 +604,7 @@ int serialize_feature(struct serialization_state *sst, serial_feature &sf, std::
 				locs.push_back(encode_index(SHIFT_LEFT(scaled_geometry[i].x), SHIFT_LEFT(scaled_geometry[i].y)));
 			}
 		}
-		std::sort(locs.begin(), locs.end());
+		std::stable_sort(locs.begin(), locs.end());
 		size_t n = 0;
 		double sum = 0;
 		for (size_t i = 1; i < locs.size(); i++) {

shared_borders.cpp

@@ -136,7 +136,7 @@ bool find_common_edges(std::vector<serial_feature> &features, int z, int line_de
 		}
 	}
 
-	std::sort(edges.begin(), edges.end(), edgecmp_ring);
+	std::stable_sort(edges.begin(), edges.end(), edgecmp_ring);
 	std::set<draw> necessaries;
 
 	// Now mark all the points where the set of rings using the edge on one side
@@ -397,7 +397,7 @@ bool find_common_edges(std::vector<serial_feature> &features, int z, int line_de
 			}
 		}
 	}
-	std::sort(order.begin(), order.end());
+	std::stable_sort(order.begin(), order.end());
 
 	size_t merged = 0;
 	for (size_t o = 0; o < order.size(); o++) {

tile-join.cpp

@@ -584,7 +584,7 @@ struct tileset_reader {
 			next_overzoomed_tiles.clear();
 		}
 
-		std::sort(overzoomed_tiles.begin(), overzoomed_tiles.end(), tilecmp);
+		std::stable_sort(overzoomed_tiles.begin(), overzoomed_tiles.end(), tilecmp);
 		overzoom_consumed_at_this_zoom = false;
 	}
 

tile.cpp

@@ -426,7 +426,7 @@ static std::vector<serial_feature> disassemble_multiplier_clusters(std::vector<s
 
 		// sort the other features by their drop sequence, for consistency across zoom levels
 		if (cluster.size() > 1) {
-			std::sort(cluster.begin() + 1, cluster.end(), drop_sequence_cmp());
+			std::stable_sort(cluster.begin() + 1, cluster.end(), drop_sequence_cmp());
 		}
 
 		for (auto const &feature : cluster) {
@@ -814,7 +814,7 @@ static unsigned long long choose_mingap(std::vector<unsigned long long> const &i
 // If there are no higher extents available, the tile has already been reduced as much as possible
 // and tippecanoe will exit with an error.
 static long long choose_minextent(std::vector<long long> &extents, double f, long long existing_extent) {
-	std::sort(extents.begin(), extents.end());
+	std::stable_sort(extents.begin(), extents.end());
 
 	size_t ix = (extents.size() - 1) * (1 - f);
 	while (ix + 1 < extents.size() && extents[ix] == existing_extent) {
@@ -829,7 +829,7 @@ static unsigned long long choose_mindrop_sequence(std::vector<unsigned long long
 		return ULLONG_MAX;
 	}
 
-	std::sort(drop_sequences.begin(), drop_sequences.end());
+	std::stable_sort(drop_sequences.begin(), drop_sequences.end());
 
 	size_t ix = (drop_sequences.size() - 1) * (1 - f);
 	while (ix + 1 < drop_sequences.size() && drop_sequences[ix] == existing_drop_sequence) {
@@ -2009,7 +2009,7 @@ long long write_tile(decompressor *geoms, std::atomic<long long> *geompos_in, ch
 		// Reorder and coalesce.
 		// Sort back into input order or by attribute value
 
-		std::sort(shared_nodes.begin(), shared_nodes.end());
+		std::stable_sort(shared_nodes.begin(), shared_nodes.end());
 
 		for (auto &kv : layers) {
 			std::string const &layername = kv.first;
@@ -2031,7 +2031,7 @@ long long write_tile(decompressor *geoms, std::atomic<long long> *geompos_in, ch
 				// these will be smaller numbers, and avoid the problem of the
 				// original sequence number varying based on how many reader threads
 				// there were reading the input
-				std::sort(feature_sequences.begin(), feature_sequences.end());
+				std::stable_sort(feature_sequences.begin(), feature_sequences.end());
 				for (size_t i = 0; i < feature_sequences.size(); i++) {
 					size_t j = feature_sequences[i].second;
 					serial_val sv(mvt_double, std::to_string(i));
@@ -2151,7 +2151,7 @@ long long write_tile(decompressor *geoms, std::atomic<long long> *geompos_in, ch
 			std::vector<serial_feature> &layer_features = features;
 
 			if (additional[A_REORDER]) {
-				std::sort(layer_features.begin(), layer_features.end(), coalindexcmp_comparator());
+				std::stable_sort(layer_features.begin(), layer_features.end(), coalindexcmp_comparator());
 			}
 
 			if (additional[A_COALESCE]) {
@@ -2215,13 +2215,13 @@ long long write_tile(decompressor *geoms, std::atomic<long long> *geompos_in, ch
 
 			if (prevent[P_INPUT_ORDER]) {
 				auto clustered = assemble_multiplier_clusters(layer_features);
-				std::sort(clustered.begin(), clustered.end(), preservecmp);
+				std::stable_sort(clustered.begin(), clustered.end(), preservecmp);
 				layer_features = disassemble_multiplier_clusters(clustered);
 			}
 
 			if (order_by.size() != 0) {
 				auto clustered = assemble_multiplier_clusters(layer_features);
-				std::sort(clustered.begin(), clustered.end(), ordercmp());
+				std::stable_sort(clustered.begin(), clustered.end(), ordercmp());
 				layer_features = disassemble_multiplier_clusters(clustered);
 			}
 

version.hpp

@@ -1,6 +1,6 @@
 #ifndef VERSION_HPP
 #define VERSION_HPP
 
-#define VERSION "v2.46.0"
+#define VERSION "v2.47.0"
 
 #endif