]> git.openstreetmap.org Git - nominatim.git/blobdiff - sql/functions/placex_triggers.sql
Merge pull request #2027 from lonvia/remove-duplicate-admin-boundaries
[nominatim.git] / sql / functions / placex_triggers.sql
index ad0f041de7a5b4c4f927f96e1eef616ed3fefa9c..1e2aac4c96fcc88a6e188eacf43a0888135458e6 100644 (file)
@@ -93,13 +93,15 @@ BEGIN
 
     IF fallback THEN
       IF addr_street is null and addr_place is not null THEN
-        -- The address is attached to a place we don't know. Find the
-        -- nearest place instead.
+        -- The address is attached to a place we don't know.
+        -- Instead simply use the containing area with the largest rank.
         FOR location IN
-          SELECT place_id FROM getNearFeatures(poi_partition, bbox, 26, '{}'::INTEGER[])
-            ORDER BY rank_address DESC, isguess asc, distance LIMIT 1
+          SELECT place_id FROM placex
+            WHERE bbox @ geometry AND _ST_Covers(geometry, ST_Centroid(bbox))
+                  AND rank_address between 5 and 25
+            ORDER BY rank_address desc
         LOOP
-          parent_place_id := location.place_id;
+            RETURN location.place_id;
         END LOOP;
       ELSEIF ST_Area(bbox) < 0.005 THEN
         -- for smaller features get the nearest road
@@ -209,7 +211,11 @@ BEGIN
     FOR linked_placex IN
       SELECT placex.* from placex
       WHERE make_standard_name(name->'name') = bnd_name
-        AND ((bnd.rank_address > 0 and placex.rank_address = bnd.rank_address)
+        AND ((bnd.rank_address > 0
+              and bnd.rank_address = (compute_place_rank(placex.country_code,
+                                                         'N', placex.class,
+                                                         placex.type, 15::SMALLINT,
+                                                         false, placex.postcode)).address_rank)
              OR (bnd.rank_address = 0 and placex.rank_search = bnd.rank_search))
         AND placex.osm_type = 'N'
         AND placex.rank_search < 26 -- needed to select the right index
@@ -253,21 +259,16 @@ CREATE OR REPLACE FUNCTION insert_addresslines(obj_place_id BIGINT,
                                                OUT nameaddress_vector INT[])
   AS $$
 DECLARE
-  current_rank_address INTEGER := 0;
-  location_distance FLOAT := 0;
-  location_parent GEOMETRY := NULL;
-  parent_place_id_rank SMALLINT := 0;
+  address_havelevel BOOLEAN[];
 
   location_isaddress BOOLEAN;
-
-  address_havelevel BOOLEAN[];
-  location_keywords INT[];
+  current_boundary GEOMETRY := NULL;
+  current_node_area GEOMETRY := NULL;
 
   location RECORD;
   addr_item RECORD;
 
   isin_tokens INT[];
-  isin TEXT[];
 BEGIN
   parent_place_id := 0;
   nameaddress_vector := '{}'::int[];
@@ -296,77 +297,71 @@ BEGIN
   END IF;
 
   ---- now compute the address terms
-  FOR i IN 1..28 LOOP
+  FOR i IN 1..maxrank LOOP
     address_havelevel[i] := false;
   END LOOP;
 
   FOR location IN
-    SELECT * FROM getNearFeatures(partition, geometry, maxrank, isin_tokens)
+    SELECT * FROM getNearFeatures(partition, geometry, maxrank)
+    ORDER BY rank_address, isin_tokens && keywords desc, isguess asc,
+             distance *
+               CASE WHEN rank_address = 16 AND rank_search = 15 THEN 0.2
+                    WHEN rank_address = 16 AND rank_search = 16 THEN 0.25
+                    WHEN rank_address = 16 AND rank_search = 18 THEN 0.5
+                    ELSE 1 END ASC
   LOOP
-    IF location.rank_address != current_rank_address THEN
-      current_rank_address := location.rank_address;
-      IF location.isguess THEN
-        location_distance := location.distance * 1.5;
-      ELSE
-        IF location.rank_address <= 12 THEN
-          -- for county and above, if we have an area consider that exact
-          -- (It would be nice to relax the constraint for places close to
-          --  the boundary but we'd need the exact geometry for that. Too
-          --  expensive.)
-          location_distance = 0;
-        ELSE
-          -- Below county level remain slightly fuzzy.
-          location_distance := location.distance * 0.5;
-        END IF;
+    -- Ignore all place nodes that do not fit in a lower level boundary.
+    CONTINUE WHEN location.isguess
+                  and current_boundary is not NULL
+                  and not ST_Contains(current_boundary, location.centroid);
+
+    -- If this is the first item in the rank, then assume it is the address.
+    location_isaddress := not address_havelevel[location.rank_address];
+
+    -- Further sanity checks to ensure that the address forms a sane hierarchy.
+    IF location_isaddress THEN
+      IF location.isguess and current_node_area is not NULL THEN
+        location_isaddress := ST_Contains(current_node_area, location.centroid);
+      END IF;
+      IF not location.isguess and current_boundary is not NULL
+         and location.rank_address != 11 AND location.rank_address != 5 THEN
+        location_isaddress := ST_Contains(current_boundary, location.centroid);
       END IF;
-    ELSE
-      CONTINUE WHEN location.keywords <@ location_keywords;
     END IF;
 
-    IF location.distance < location_distance OR NOT location.isguess THEN
-      location_keywords := location.keywords;
+    IF location_isaddress THEN
+      address_havelevel[location.rank_address] := true;
+      parent_place_id := location.place_id;
 
-      location_isaddress := NOT address_havelevel[location.rank_address];
-      --DEBUG: RAISE WARNING 'should be address: %, is guess: %, rank: %', location_isaddress, location.isguess, location.rank_address;
-      IF location_isaddress AND location.isguess AND location_parent IS NOT NULL THEN
-          location_isaddress := ST_Contains(location_parent, location.centroid);
+      -- Set postcode if we have one.
+      -- (Returned will be the highest ranking one.)
+      IF location.postcode is not NULL THEN
+        postcode = location.postcode;
       END IF;
 
-      --DEBUG: RAISE WARNING '% isaddress: %', location.place_id, location_isaddress;
-      -- Add it to the list of search terms
-      IF NOT %REVERSE-ONLY% THEN
-          nameaddress_vector := array_merge(nameaddress_vector,
-                                            location.keywords::integer[]);
-      END IF;
-
-      INSERT INTO place_addressline (place_id, address_place_id, fromarea,
-                                     isaddress, distance, cached_rank_address)
-        VALUES (obj_place_id, location.place_id, true,
-                location_isaddress, location.distance, location.rank_address);
-
-      IF location_isaddress THEN
-        -- add postcode if we have one
-        -- (If multiple postcodes are available, we end up with the highest ranking one.)
-        IF location.postcode is not null THEN
-            postcode = location.postcode;
-        END IF;
-
-        address_havelevel[location.rank_address] := true;
-        -- add a hack against postcode ranks
-        IF NOT location.isguess
-           AND location.rank_address != 11 AND location.rank_address != 5
-        THEN
+      -- Recompute the areas we need for hierarchy sanity checks.
+      IF location.rank_address != 11 AND location.rank_address != 5 THEN
+        IF location.isguess THEN
+          current_node_area := place_node_fuzzy_area(location.centroid,
+                                                     location.rank_search);
+        ELSE
+          current_node_area := NULL;
           SELECT p.geometry FROM placex p
-            WHERE p.place_id = location.place_id INTO location_parent;
-        END IF;
-
-        IF location.rank_address > parent_place_id_rank THEN
-          parent_place_id = location.place_id;
-          parent_place_id_rank = location.rank_address;
+              WHERE p.place_id = location.place_id INTO current_boundary;
         END IF;
       END IF;
     END IF;
 
+    -- Add it to the list of search terms
+    IF NOT %REVERSE-ONLY% THEN
+      nameaddress_vector := array_merge(nameaddress_vector,
+                                        location.keywords::integer[]);
+    END IF;
+
+    INSERT INTO place_addressline (place_id, address_place_id, fromarea,
+                                     isaddress, distance, cached_rank_address)
+        VALUES (obj_place_id, location.place_id, not location.isguess,
+                location_isaddress, location.distance, location.rank_address);
   END LOOP;
 END;
 $$
@@ -499,33 +494,6 @@ END;
 $$
 LANGUAGE plpgsql;
 
-CREATE OR REPLACE FUNCTION get_parent_address_level(geom GEOMETRY, in_level SMALLINT)
-  RETURNS SMALLINT
-  AS $$
-DECLARE
-  address_rank SMALLINT;
-BEGIN
-  IF in_level <= 3 or in_level > 15 THEN
-    address_rank := 3;
-  ELSE
-    SELECT rank_address INTO address_rank
-      FROM placex
-      WHERE osm_type = 'R' and class = 'boundary' and type = 'administrative'
-            and admin_level < in_level
-            and geometry && geom and ST_Covers(geometry, geom)
-      ORDER BY admin_level desc LIMIT 1;
-  END IF;
-
-  IF address_rank is NULL or address_rank <= 3 THEN
-    RETURN 3;
-  END IF;
-
-  RETURN address_rank;
-END;
-$$
-LANGUAGE plpgsql;
-
-
 CREATE OR REPLACE FUNCTION placex_update()
   RETURNS TRIGGER
   AS $$
@@ -614,14 +582,67 @@ BEGIN
   IF NEW.class = 'boundary' and NEW.type = 'administrative'
      and NEW.osm_type = 'R' and NEW.rank_address > 0
   THEN
-    parent_address_level := get_parent_address_level(NEW.centroid, NEW.admin_level);
-    IF parent_address_level >= NEW.rank_address THEN
-      IF parent_address_level >= 24 THEN
-        NEW.rank_address := 25;
+    -- First, check that admin boundaries do not overtake each other rank-wise.
+    parent_address_level := 3;
+    FOR location IN
+      SELECT rank_address,
+             (CASE WHEN extratags ? 'wikidata' and NEW.extratags ? 'wikidata'
+                        and extratags->'wikidata' = NEW.extratags->'wikidata'
+                   THEN ST_Equals(geometry, NEW.geometry)
+                   ELSE false END) as is_same
+      FROM placex
+      WHERE osm_type = 'R' and class = 'boundary' and type = 'administrative'
+            and admin_level < NEW.admin_level and admin_level > 3
+            and rank_address > 0
+            and geometry && NEW.centroid and _ST_Covers(geometry, NEW.centroid)
+      ORDER BY admin_level desc LIMIT 1
+    LOOP
+      IF location.is_same THEN
+        -- Looks like the same boundary is replicated on multiple admin_levels.
+        -- Usual tagging in Poland. Remove our boundary from addresses.
+        NEW.rank_address := 0;
       ELSE
-        NEW.rank_address := parent_address_level + 2;
+        parent_address_level := location.rank_address;
+        IF location.rank_address >= NEW.rank_address THEN
+          IF location.rank_address >= 24 THEN
+            NEW.rank_address := 25;
+          ELSE
+            NEW.rank_address := location.rank_address + 2;
+          END IF;
+        END IF;
       END IF;
+    END LOOP;
+
+    IF NEW.rank_address > 9 THEN
+        -- Second check that the boundary is not completely contained in a
+        -- place area with a higher address rank
+        FOR location IN
+          SELECT rank_address FROM placex
+          WHERE class = 'place' and rank_address < 24
+                and rank_address > NEW.rank_address
+                and geometry && NEW.geometry
+                and geometry ~ NEW.geometry -- needed because ST_Relate does not do bbox cover test
+                and ST_Relate(geometry, NEW.geometry, 'T*T***FF*') -- contains but not equal
+          ORDER BY rank_address desc LIMIT 1
+        LOOP
+          NEW.rank_address := location.rank_address + 2;
+        END LOOP;
     END IF;
+  ELSEIF NEW.class = 'place' and NEW.osm_type = 'N'
+     and NEW.rank_address between 16 and 23
+  THEN
+    -- If a place node is contained in a admin boundary with the same address level
+    -- and has not been linked, then make the node a subpart by increasing the
+    -- address rank (city level and above).
+    FOR location IN
+        SELECT rank_address FROM placex
+        WHERE osm_type = 'R' and class = 'boundary' and type = 'administrative'
+              and rank_address = NEW.rank_address
+              and geometry && NEW.centroid and _ST_Covers(geometry, NEW.centroid)
+        LIMIT 1
+    LOOP
+      NEW.rank_address = NEW.rank_address + 2;
+    END LOOP;
   ELSE
     parent_address_level := 3;
   END IF;
@@ -778,7 +799,8 @@ BEGIN
           IF NEW.rank_search <= 25 and NEW.rank_address > 0 THEN
             result := add_location(NEW.place_id, NEW.country_code, NEW.partition,
                                    name_vector, NEW.rank_search, NEW.rank_address,
-                                   upper(trim(NEW.address->'postcode')), NEW.geometry);
+                                   upper(trim(NEW.address->'postcode')), NEW.geometry,
+                                   NEW.centroid);
             --DEBUG: RAISE WARNING 'Place added to location table';
           END IF;
 
@@ -885,8 +907,9 @@ BEGIN
   END IF;
 
   SELECT * FROM insert_addresslines(NEW.place_id, NEW.partition,
-                                    CASE WHEN NEW.rank_address = 0
-                                      THEN NEW.rank_search ELSE NEW.rank_address END,
+                                    CASE WHEN NEW.rank_address = 0 THEN NEW.rank_search
+                                         WHEN NEW.rank_address > 25 THEN 25::smallint
+                                         ELSE NEW.rank_address END,
                                     NEW.address,
                                     CASE WHEN NEW.rank_search >= 26
                                              AND NEW.rank_search < 30
@@ -908,7 +931,7 @@ BEGIN
   IF NEW.name IS NOT NULL THEN
 
     IF NEW.rank_search <= 25 and NEW.rank_address > 0 THEN
-      result := add_location(NEW.place_id, NEW.country_code, NEW.partition, name_vector, NEW.rank_search, NEW.rank_address, upper(trim(NEW.address->'postcode')), NEW.geometry);
+      result := add_location(NEW.place_id, NEW.country_code, NEW.partition, name_vector, NEW.rank_search, NEW.rank_address, upper(trim(NEW.address->'postcode')), NEW.geometry, NEW.centroid);
       --DEBUG: RAISE WARNING 'added to location (full)';
     END IF;