perf: exact rectangle clip for valid polygons (planet 25% faster, 31% less CPU)

Makefile

@@ -157,6 +157,7 @@ test: \
 	test_attribute_store \
 	test_config_validator \
 	test_deque_map \
+	test_geom \
 	test_helpers \
 	test_options_parser \
 	test_pbf_reader \
@@ -196,6 +197,11 @@ test_deque_map: \
 	test/deque_map.test.o
 	$(CXX) $(CXXFLAGS) -o test.deque_map $^ $(INC) $(LIB) $(LDFLAGS) && ./test.deque_map
 
+test_geom: \
+	src/geom.o \
+	test/geom.test.o
+	$(CXX) $(CXXFLAGS) -o test.geom $^ $(INC) $(LIB) $(LDFLAGS) && ./test.geom
+
 test_helpers: \
 	src/helpers.o \
 	src/external/libdeflate/lib/adler32.o \

include/geom.h

@@ -81,8 +81,10 @@ void make_valid(MultiPolygon &mp);
 // Attempt to repair an invalid areal geometry in place: dissolve-based
 // make_valid first (preserves area), then a zero-width buffer as a last
 // resort. Returns true if mp is valid afterwards; on failure mp is left as the
-// best-effort input so callers never regress.
-bool repair_multi_polygon(MultiPolygon &mp, bool knownInvalid = false);
+// best-effort input so callers never regress. If wholeValid is given, it is set
+// to true only if mp as a whole is known to pass geom::is_valid afterwards (a true
+// result alone can leave individually valid polygons that overlap each other).
+bool repair_multi_polygon(MultiPolygon &mp, bool knownInvalid = false, bool *wholeValid = nullptr);
 
 void union_many(std::vector<MultiPolygon> &mps);
 
@@ -91,6 +93,10 @@ char bit_code(Point const &p, Box const &bbox);
 // fast_clip returns true if the geometry was modified (false means identity).
 bool fast_clip(Ring &points, Box const &bbox);
 bool fast_clip(MultiPolygon &mp, Box const &bbox);
+// Exact clip of a valid multipolygon to bbox that keeps it valid (fast_clip can
+// leave zero-width bridges along the box edge). Returns false if it could not
+// clip robustly; `out` is then unspecified and the caller must use geom::intersection.
+bool rect_clip(const MultiPolygon &mp, Box const &bbox, MultiPolygon &out);
 
 // Cheap conservative validity tests, run before geom::is_valid. They return
 // true only when the geometry is certainly valid (is_valid would return true

include/osm_mem_tiles.h

@@ -61,7 +61,8 @@ class OsmMemTiles : public TileDataSource {
 	// CorrectGeometry found them Valid (or Reversed: valid after reversing the ring)
 	// on the same lon/latp coordinates.
 	bool polygonKnownValid(NodeID objectID) const override {
-		return objectID >= OSM_THRESHOLD && (IS_WAY(objectID) || IS_WAY_REV(objectID));
+		if (objectID < OSM_THRESHOLD) return TileDataSource::polygonKnownValid(objectID);
+		return IS_WAY(objectID) || IS_WAY_REV(objectID);
 	}
 
 	const NodeStore& nodeStore;

include/tile_data.h

@@ -298,6 +298,8 @@ class TileDataSource {
 	std::vector<linestring_store_t> linestringStores;
 	std::vector<multi_linestring_store_t> multilinestringStores;
 	std::vector<multi_polygon_store_t> multipolygonStores;
+	// Per multipolygonStores entry: stored geometry is known to pass geom::is_valid
+	std::vector<std::vector<bool>> multipolygonValid;
 
 	ClipCache<MultiPolygon> multiPolygonClipCache;
 	ClipCache<MultiLinestring> multiLinestringClipCache;
@@ -359,6 +361,7 @@ class TileDataSource {
 		pointStores[0].push_back(Point(0,0));
 		linestringStores[0].push_back(linestring_t());
 		multipolygonStores[0].push_back(multi_polygon_t());
+		multipolygonValid[0].push_back(false);
 		multilinestringStores[0].push_back(multi_linestring_t());
 	}
 	void reportSize() const;
@@ -380,7 +383,11 @@ class TileDataSource {
 	virtual void populateMultiPolygon(MultiPolygon& dst, NodeID objectID);
 	// True if the polygon populateMultiPolygon returns for objectID is known to
 	// have passed geom::is_valid on exactly the same coordinates.
-	virtual bool polygonKnownValid(NodeID objectID) const { return false; }
+	virtual bool polygonKnownValid(NodeID objectID) const {
+		const auto& valid = multipolygonValid[getShard(objectID)];
+		const auto offset = getId(objectID);
+		return offset < valid.size() && valid[offset];
+	}
 
 	unsigned int getIndexZoom() const { return indexZoom; }
 	inline size_t getId(NodeID id) const {
@@ -415,7 +422,8 @@ class TileDataSource {
 		return shard.at(offset);
 	}
 
-	NodeID storeMultiPolygon(const MultiPolygon& src);
+	// knownValid: src has passed geom::is_valid (see polygonKnownValid)
+	NodeID storeMultiPolygon(const MultiPolygon& src, bool knownValid = false);
 
 	multi_polygon_t const &retrieveMultiPolygon(NodeID id) const {
 		const auto& shardId = getShard(id);