X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/df7b81dc06bea97c926bf113d4dfe0f6e0669923..33f7bba69858405517c68a4aaf609ce69e00e277:/sql/functions.sql diff --git a/sql/functions.sql b/sql/functions.sql index 03104dd5..839d086c 100644 --- a/sql/functions.sql +++ b/sql/functions.sql @@ -5,7 +5,6 @@ -- score integer --); - CREATE OR REPLACE FUNCTION getclasstypekey(c text, t text) RETURNS TEXT AS $$ DECLARE @@ -96,7 +95,7 @@ BEGIN SELECT min(word_id), max(search_name_count) FROM word WHERE word_token = lookup_token and class is null and type is null into return_word_id, count; IF return_word_id IS NULL THEN return_word_id := nextval('seq_word'); - INSERT INTO word VALUES (return_word_id, lookup_token, regexp_replace(lookup_token,E'([^0-9])\\1+',E'\\1','g'), null, null, null, null, 0, null); + INSERT INTO word VALUES (return_word_id, lookup_token, null, null, null, null, 0); ELSE IF count > get_maxwordfreq() THEN return_word_id := NULL; @@ -118,7 +117,7 @@ BEGIN SELECT min(word_id) FROM word WHERE word_token = lookup_token and class='place' and type='house' into return_word_id; IF return_word_id IS NULL THEN return_word_id := nextval('seq_word'); - INSERT INTO word VALUES (return_word_id, lookup_token, null, null, 'place', 'house', null, 0, null); + INSERT INTO word VALUES (return_word_id, lookup_token, null, 'place', 'house', null, 0); END IF; RETURN return_word_id; END; @@ -136,7 +135,7 @@ BEGIN SELECT min(word_id) FROM word WHERE word_token = lookup_token and country_code=lookup_country_code into return_word_id; IF return_word_id IS NULL THEN return_word_id := nextval('seq_word'); - INSERT INTO word VALUES (return_word_id, lookup_token, null, null, null, null, lookup_country_code, 0, null); + INSERT INTO word VALUES (return_word_id, lookup_token, null, null, null, lookup_country_code, 0); END IF; RETURN return_word_id; END; @@ -154,7 +153,7 @@ BEGIN SELECT min(word_id) FROM word WHERE word_token = lookup_token and class=lookup_class and type = lookup_type into return_word_id; IF return_word_id IS NULL THEN return_word_id := nextval('seq_word'); - INSERT INTO word VALUES (return_word_id, lookup_token, null, null, lookup_class, lookup_type, null, 0, null); + INSERT INTO word VALUES (return_word_id, lookup_token, null, lookup_class, lookup_type, null, 0); END IF; RETURN return_word_id; END; @@ -172,7 +171,7 @@ BEGIN SELECT min(word_id) FROM word WHERE word_token = lookup_token into return_word_id; IF return_word_id IS NULL THEN return_word_id := nextval('seq_word'); - INSERT INTO word VALUES (return_word_id, lookup_token, null, null, null, null, null, 0, null); + INSERT INTO word VALUES (return_word_id, lookup_token, null, null, null, null, 0); END IF; RETURN return_word_id; END; @@ -204,7 +203,7 @@ BEGIN SELECT min(word_id) FROM word WHERE word_token = lookup_token and class=lookup_class and type = lookup_type and operator = op into return_word_id; IF return_word_id IS NULL THEN return_word_id := nextval('seq_word'); - INSERT INTO word VALUES (return_word_id, lookup_token, null, null, lookup_class, lookup_type, null, 0, op, null); + INSERT INTO word VALUES (return_word_id, lookup_token, null, lookup_class, lookup_type, null, 0, op); END IF; RETURN return_word_id; END; @@ -223,7 +222,7 @@ BEGIN SELECT min(word_id) FROM word WHERE word_token = lookup_token and class is null and type is null into return_word_id; IF return_word_id IS NULL THEN return_word_id := nextval('seq_word'); - INSERT INTO word VALUES (return_word_id, lookup_token, regexp_replace(lookup_token,E'([^0-9])\\1+',E'\\1','g'), src_word, null, null, null, 0, null); + INSERT INTO word VALUES (return_word_id, lookup_token, src_word, null, null, null, 0); -- nospace_lookup_token := replace(replace(lookup_token, '-',''), ' ',''); -- IF ' '||nospace_lookup_token != lookup_token THEN -- INSERT INTO word VALUES (return_word_id, '-'||nospace_lookup_token, null, src_word, null, null, null, 0, null); @@ -346,6 +345,12 @@ BEGIN result := result || w; END IF; + w := getorcreate_word_id(s); + + IF w IS NOT NULL AND NOT (ARRAY[w] <@ result) THEN + result := result || w; + END IF; + words := string_to_array(s, ' '); IF array_upper(words, 1) IS NOT NULL THEN FOR j IN 1..array_upper(words, 1) LOOP @@ -408,6 +413,12 @@ BEGIN result := result || w; END IF; + w := getorcreate_word_id(s); + + IF w IS NOT NULL AND NOT (ARRAY[w] <@ result) THEN + result := result || w; + END IF; + words := string_to_array(s, ' '); IF array_upper(words, 1) IS NOT NULL THEN FOR j IN 1..array_upper(words, 1) LOOP @@ -484,6 +495,12 @@ BEGIN --DEBUG: RAISE WARNING 'get_country_code, start: %', ST_AsText(place_centre); + -- Try for a OSM polygon + FOR nearcountry IN select country_code from location_area_country where country_code is not null and not isguess and st_covers(geometry, place_centre) limit 1 + LOOP + RETURN nearcountry.country_code; + END LOOP; + --DEBUG: RAISE WARNING 'osm fallback: %', ST_AsText(place_centre); -- Try for OSM fallback data @@ -493,12 +510,6 @@ BEGIN RETURN nearcountry.country_code; END LOOP; - -- Try for a OSM polygon - FOR nearcountry IN select country_code from location_area_country where country_code is not null and not isguess and st_covers(geometry, place_centre) limit 1 - LOOP - RETURN nearcountry.country_code; - END LOOP; - --DEBUG: RAISE WARNING 'natural earth: %', ST_AsText(place_centre); -- Natural earth data @@ -571,6 +582,20 @@ END; $$ LANGUAGE plpgsql IMMUTABLE; +CREATE OR REPLACE FUNCTION get_country_language_codes(search_country_code VARCHAR(2)) RETURNS TEXT[] + AS $$ +DECLARE + nearcountry RECORD; +BEGIN + FOR nearcountry IN select country_default_language_codes from country_name where country_code = search_country_code limit 1 + LOOP + RETURN lower(nearcountry.country_default_language_codes); + END LOOP; + RETURN NULL; +END; +$$ +LANGUAGE plpgsql IMMUTABLE; + CREATE OR REPLACE FUNCTION get_partition(place geometry, in_country_code VARCHAR(10)) RETURNS INTEGER AS $$ DECLARE @@ -624,7 +649,7 @@ BEGIN -- RAISE WARNING 'Adding location with rank > 25 (% rank %)', place_id, rank_search; - x := deleteLocationArea(partition, place_id); + x := deleteLocationArea(partition, place_id, rank_search); isarea := false; IF (ST_GeometryType(geometry) in ('ST_Polygon','ST_MultiPolygon') AND ST_IsValid(geometry)) THEN @@ -691,7 +716,7 @@ CREATE OR REPLACE FUNCTION update_location( DECLARE b BOOLEAN; BEGIN - b := deleteLocationArea(partition, place_id); + b := deleteLocationArea(partition, place_id, rank_search); -- result := add_location(NEW.place_id, NEW.country_code, NEW.partition, name_vector, NEW.rank_search, NEW.rank_address, NEW.geometry); RETURN add_location(place_id, place_country_code, name, rank_search, rank_address, geometry); END; @@ -769,8 +794,8 @@ CREATE OR REPLACE FUNCTION create_interpolation(wayid BIGINT, interpolationtype DECLARE newpoints INTEGER; - waynodes integer[]; - nodeid INTEGER; + waynodes BIGINT[]; + nodeid BIGINT; prevnode RECORD; nextnode RECORD; startnumber INTEGER; @@ -798,10 +823,14 @@ BEGIN FOR nodeidpos in 1..array_upper(waynodes, 1) LOOP - select min(place_id) from placex where osm_type = 'N' and osm_id = waynodes[nodeidpos]::INTEGER and type = 'house' INTO search_place_id; + -- If there is a place of a type other than place/house, use that because + -- it is guaranteed to be the original node. For place/house types use the + -- one with the smallest id because the original node was created first. + -- Ignore all nodes marked for deletion. (Might happen when the type changes.) + select place_id from placex where osm_type = 'N' and osm_id = waynodes[nodeidpos]::BIGINT and indexed_status < 100 order by (type = 'house'),place_id limit 1 INTO search_place_id; IF search_place_id IS NULL THEN - -- null record of right type - select * from placex where osm_type = 'N' and osm_id = waynodes[nodeidpos]::INTEGER and type = 'house' limit 1 INTO nextnode; + -- if no such node exists, create a record of the right type + select * from placex where osm_type = 'N' and osm_id = waynodes[nodeidpos]::BIGINT and type = 'house' limit 1 INTO nextnode; select ST_SetSRID(ST_Point(lon::float/10000000,lat::float/10000000),4326) from planet_osm_nodes where id = waynodes[nodeidpos] INTO nextnode.geometry; IF nextnode.geometry IS NULL THEN -- we don't have any information about this point, most likely @@ -863,10 +892,10 @@ BEGIN FOR housenum IN startnumber..endnumber BY stepsize LOOP -- this should really copy postcodes but it puts a huge burdon on the system for no big benefit -- ideally postcodes should move up to the way - insert into placex (osm_type, osm_id, class, type, admin_level, housenumber, street, isin, postcode, + insert into placex (osm_type, osm_id, class, type, admin_level, housenumber, street, addr_place, isin, postcode, country_code, parent_place_id, rank_address, rank_search, indexed_status, geometry) - values ('N',prevnode.osm_id, prevnode.class, prevnode.type, prevnode.admin_level, housenum, prevnode.street, prevnode.isin, coalesce(prevnode.postcode, defpostalcode), - prevnode.country_code, prevnode.parent_place_id, prevnode.rank_address, prevnode.rank_search, 1, ST_Line_Interpolate_Point(linegeo, (housenum::float-orginalstartnumber::float)/originalnumberrange::float)); + values ('N',prevnode.osm_id, 'place', 'house', prevnode.admin_level, housenum, prevnode.street, prevnode.addr_place, prevnode.isin, coalesce(prevnode.postcode, defpostalcode), + prevnode.country_code, prevnode.parent_place_id, prevnode.rank_address, prevnode.rank_search, 1, ST_LineInterpolatePoint(linegeo, (housenum::float-orginalstartnumber::float)/originalnumberrange::float)); newpoints := newpoints + 1; --RAISE WARNING 'interpolation number % % ',prevnode.place_id,housenum; END LOOP; @@ -908,11 +937,12 @@ DECLARE BEGIN --DEBUG: RAISE WARNING '% %',NEW.osm_type,NEW.osm_id; - -- just block these - IF NEW.class = 'highway' and NEW.type in ('turning_circle','traffic_signals','mini_roundabout','noexit','crossing') THEN --- RAISE WARNING 'bad highway %',NEW.osm_id; - RETURN null; + -- remove operator tag for most places, messes too much with search_name indexes + IF NEW.class not in ('amenity', 'shop') THEN + NEW.name := delete(NEW.name, 'operator'); END IF; + + -- just block these IF NEW.class in ('landuse','natural') and NEW.name is null THEN -- RAISE WARNING 'empty landuse %',NEW.osm_id; RETURN null; @@ -1065,10 +1095,10 @@ BEGIN ELSEIF NEW.type in ('national_park') THEN NEW.rank_search := 18; NEW.rank_address := 18; - ELSEIF NEW.type in ('suburb','croft','subdivision') THEN + ELSEIF NEW.type in ('suburb','croft','subdivision','isolated_dwelling') THEN NEW.rank_search := 20; NEW.rank_address := NEW.rank_search; - ELSEIF NEW.type in ('farm','locality','islet','isolated_dwelling','mountain_pass') THEN + ELSEIF NEW.type in ('farm','locality','islet','mountain_pass') THEN NEW.rank_search := 20; NEW.rank_address := 0; -- Irish townlands, tagged as place=locality and locality=townland @@ -1098,7 +1128,11 @@ BEGIN return NULL; END IF; NEW.rank_search := NEW.admin_level * 2; - NEW.rank_address := NEW.rank_search; + IF NEW.type = 'administrative' THEN + NEW.rank_address := NEW.rank_search; + ELSE + NEW.rank_address := 0; + END IF; ELSEIF NEW.class = 'landuse' AND ST_GeometryType(NEW.geometry) in ('ST_Polygon','ST_MultiPolygon') THEN NEW.rank_search := 22; NEW.rank_address := NEW.rank_search; @@ -1130,6 +1164,9 @@ BEGIN NEW.rank_address := NEW.rank_search; ELSEIF NEW.class = 'natural' and NEW.type in ('coastline') THEN RETURN NULL; + ELSEIF NEW.class = 'mountain_pass' THEN + NEW.rank_search := 20; + NEW.rank_address := 0; END IF; END IF; @@ -1168,9 +1205,9 @@ BEGIN -- work around bug in postgis, this may have been fixed in 2.0.0 (see http://trac.osgeo.org/postgis/ticket/547) update placex set indexed_status = 2 where (st_covers(NEW.geometry, placex.geometry) OR ST_Intersects(NEW.geometry, placex.geometry)) - AND rank_search > NEW.rank_search and indexed_status = 0 and ST_geometrytype(placex.geometry) = 'ST_Point' and (rank_search < 28 or name is not null); + AND rank_search > NEW.rank_search and indexed_status = 0 and ST_geometrytype(placex.geometry) = 'ST_Point' and (rank_search < 28 or name is not null or (NEW.rank_search >= 16 and addr_place is not null)); update placex set indexed_status = 2 where (st_covers(NEW.geometry, placex.geometry) OR ST_Intersects(NEW.geometry, placex.geometry)) - AND rank_search > NEW.rank_search and indexed_status = 0 and ST_geometrytype(placex.geometry) != 'ST_Point' and (rank_search < 28 or name is not null); + AND rank_search > NEW.rank_search and indexed_status = 0 and ST_geometrytype(placex.geometry) != 'ST_Point' and (rank_search < 28 or name is not null or (NEW.rank_search >= 16 and addr_place is not null)); END IF; ELSE -- mark nearby items for re-indexing, where 'nearby' depends on the features rank_search and is a complete guess :( @@ -1195,7 +1232,16 @@ BEGIN END IF; IF diameter > 0 THEN -- RAISE WARNING 'placex point insert: % % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type,diameter; - update placex set indexed_status = 2 where indexed_status = 0 and rank_search > NEW.rank_search and ST_DWithin(placex.geometry, NEW.geometry, diameter) and (rank_search < 28 or name is not null); + IF NEW.rank_search >= 26 THEN + -- roads may cause reparenting for >27 rank places + update placex set indexed_status = 2 where indexed_status = 0 and rank_search > NEW.rank_search and ST_DWithin(placex.geometry, NEW.geometry, diameter); + ELSEIF NEW.rank_search >= 16 THEN + -- up to rank 16, street-less addresses may need reparenting + update placex set indexed_status = 2 where indexed_status = 0 and rank_search > NEW.rank_search and ST_DWithin(placex.geometry, NEW.geometry, diameter) and (rank_search < 28 or name is not null or addr_place is not null); + ELSE + -- for all other places the search terms may change as well + update placex set indexed_status = 2 where indexed_status = 0 and rank_search > NEW.rank_search and ST_DWithin(placex.geometry, NEW.geometry, diameter) and (rank_search < 28 or name is not null); + END IF; END IF; END IF; @@ -1204,7 +1250,7 @@ BEGIN -- Note: won't work on initial import because the classtype tables -- do not yet exist. It won't hurt either. classtable := 'place_classtype_' || NEW.class || '_' || NEW.type; - SELECT count(*)>0 FROM pg_tables WHERE tablename = classtable INTO result; + SELECT count(*)>0 FROM pg_tables WHERE tablename = classtable and schemaname = current_schema() INTO result; IF result THEN EXECUTE 'INSERT INTO ' || classtable::regclass || ' (place_id, centroid) VALUES ($1,$2)' USING NEW.place_id, ST_Centroid(NEW.geometry); @@ -1275,7 +1321,7 @@ BEGIN RETURN NULL; END IF; - IF NEW.indexed_status != 0 OR OLD.indexed_status = 0 OR NEW.linked_place_id is not null THEN + IF NEW.indexed_status != 0 OR OLD.indexed_status = 0 THEN RETURN NEW; END IF; @@ -1289,23 +1335,25 @@ BEGIN RETURN NEW; END IF; + -- TODO: this test is now redundant? IF OLD.indexed_status != 0 THEN - --DEBUG: RAISE WARNING 'placex_update_0 % %',NEW.osm_type,NEW.osm_id; NEW.indexed_date = now(); - IF NEW.class = 'place' AND NEW.type = 'houses' THEN - i := create_interpolation(NEW.osm_id, NEW.housenumber); + result := deleteSearchName(NEW.partition, NEW.place_id); + DELETE FROM place_addressline WHERE place_id = NEW.place_id; + DELETE FROM place_boundingbox where place_id = NEW.place_id; + result := deleteRoad(NEW.partition, NEW.place_id); + result := deleteLocationArea(NEW.partition, NEW.place_id, NEW.rank_search); + UPDATE placex set linked_place_id = null where linked_place_id = NEW.place_id; + + IF NEW.linked_place_id is not null THEN RETURN NEW; END IF; - IF OLD.indexed_status > 0 THEN - result := deleteSearchName(NEW.partition, NEW.place_id); - DELETE FROM place_addressline WHERE place_id = NEW.place_id; - DELETE FROM place_boundingbox where place_id = NEW.place_id; - result := deleteRoad(NEW.partition, NEW.place_id); - result := deleteLocationArea(NEW.partition, NEW.place_id); - UPDATE placex set linked_place_id = null where linked_place_id = NEW.place_id; + IF NEW.class = 'place' AND NEW.type = 'houses' THEN + i := create_interpolation(NEW.osm_id, NEW.housenumber); + RETURN NEW; END IF; -- Speed up searches - just use the centroid of the feature @@ -1314,13 +1362,26 @@ BEGIN NEW.centroid := null; -- reclaculate country and partition - IF NEW.rank_search >= 4 THEN - --NEW.calculated_country_code := lower(get_country_code(NEW.geometry, NEW.country_code)); - NEW.calculated_country_code := lower(get_country_code(place_centroid)); + IF NEW.rank_search = 4 THEN + -- for countries, believe the mapped country code, + -- so that we remain in the right partition if the boundaries + -- suddenly expand. + NEW.partition := get_partition(place_centroid, lower(NEW.country_code)); + IF NEW.partition = 0 THEN + NEW.calculated_country_code := lower(get_country_code(place_centroid)); + NEW.partition := get_partition(place_centroid, NEW.calculated_country_code); + ELSE + NEW.calculated_country_code := lower(NEW.country_code); + END IF; ELSE - NEW.calculated_country_code := NULL; + IF NEW.rank_search > 4 THEN + --NEW.calculated_country_code := lower(get_country_code(NEW.geometry, NEW.country_code)); + NEW.calculated_country_code := lower(get_country_code(place_centroid)); + ELSE + NEW.calculated_country_code := NULL; + END IF; + NEW.partition := get_partition(place_centroid, NEW.calculated_country_code); END IF; - NEW.partition := get_partition(place_centroid, NEW.calculated_country_code); NEW.geometry_sector := geometry_sector(NEW.partition, place_centroid); -- Adding ourselves to the list simplifies address calculations later @@ -1369,6 +1430,9 @@ BEGIN --RAISE WARNING 'finding street for %', NEW; + -- We won't get a better centroid, besides these places are too small to care + NEW.centroid := place_centroid; + NEW.parent_place_id := null; -- to do that we have to find our parent road @@ -1384,7 +1448,7 @@ BEGIN FOR i IN 1..array_upper(relation.members, 1) BY 2 LOOP IF NEW.parent_place_id IS NULL AND relation.members[i+1] = 'street' THEN --RAISE WARNING 'node in relation %',relation; - SELECT place_id from placex where osm_type='W' and osm_id = substring(relation.members[i],2,200)::integer + SELECT place_id from placex where osm_type='W' and osm_id = substring(relation.members[i],2,200)::bigint and rank_search = 26 INTO NEW.parent_place_id; END IF; END LOOP; @@ -1413,7 +1477,7 @@ BEGIN FOR i IN 1..array_upper(relation.members, 1) BY 2 LOOP IF NEW.parent_place_id IS NULL AND relation.members[i+1] = 'street' THEN --RAISE WARNING 'node in way that is in a relation %',relation; - SELECT place_id from placex where osm_type='W' and osm_id = substring(relation.members[i],2,200)::integer + SELECT place_id from placex where osm_type='W' and osm_id = substring(relation.members[i],2,200)::bigint and rank_search = 26 INTO NEW.parent_place_id; END IF; END LOOP; @@ -1422,20 +1486,25 @@ BEGIN END IF; -- If the way contains an explicit name of a street copy it - IF NEW.street IS NULL AND location.street IS NOT NULL THEN + IF NEW.street IS NULL AND NEW.addr_place IS NULL AND location.street IS NOT NULL THEN --RAISE WARNING 'node in way that has a streetname %',location; NEW.street := location.street; END IF; + -- IF the way contains an explicit name of a place copy it + IF NEW.addr_place IS NULL AND NEW.street IS NULL AND location.addr_place IS NOT NULL THEN + NEW.addr_place := location.addr_place; + END IF; + -- If this way is a street interpolation line then it is probably as good as we are going to get - IF NEW.parent_place_id IS NULL AND NEW.street IS NULL AND location.class = 'place' and location.type='houses' THEN + IF NEW.parent_place_id IS NULL AND NEW.street IS NULL AND NEW.addr_place IS NULL AND location.class = 'place' and location.type='houses' THEN -- Try and find a way that is close roughly parellel to this line FOR relation IN SELECT place_id FROM placex WHERE ST_DWithin(location.geometry, placex.geometry, 0.001) and placex.rank_search = 26 and st_geometrytype(location.geometry) in ('ST_LineString') - ORDER BY (ST_distance(placex.geometry, ST_Line_Interpolate_Point(location.geometry,0))+ - ST_distance(placex.geometry, ST_Line_Interpolate_Point(location.geometry,0.5))+ - ST_distance(placex.geometry, ST_Line_Interpolate_Point(location.geometry,1))) ASC limit 1 + ORDER BY (ST_distance(placex.geometry, ST_LineInterpolatePoint(location.geometry,0))+ + ST_distance(placex.geometry, ST_LineInterpolatePoint(location.geometry,0.5))+ + ST_distance(placex.geometry, ST_LineInterpolatePoint(location.geometry,1))) ASC limit 1 LOOP --RAISE WARNING 'using nearest street to address interpolation line,0.001 %',relation; NEW.parent_place_id := relation.place_id; @@ -1458,7 +1527,7 @@ BEGIN FOR i IN 1..array_upper(relation.members, 1) BY 2 LOOP IF NEW.parent_place_id IS NULL AND relation.members[i+1] = 'street' THEN --RAISE WARNING 'way that is in a relation %',relation; - SELECT place_id from placex where osm_type='W' and osm_id = substring(relation.members[i],2,200)::integer + SELECT place_id from placex where osm_type='W' and osm_id = substring(relation.members[i],2,200)::bigint and rank_search = 26 INTO NEW.parent_place_id; END IF; END LOOP; @@ -1469,9 +1538,18 @@ BEGIN --RAISE WARNING 'x3 %',NEW.parent_place_id; IF NEW.parent_place_id IS NULL AND NEW.street IS NOT NULL THEN - address_street_word_id := get_name_id(make_standard_name(NEW.street)); + address_street_word_id := get_name_id(make_standard_name(NEW.street)); IF address_street_word_id IS NOT NULL THEN FOR location IN SELECT * from getNearestNamedRoadFeature(NEW.partition, place_centroid, address_street_word_id) LOOP + NEW.parent_place_id := location.place_id; + END LOOP; + END IF; + END IF; + + IF NEW.parent_place_id IS NULL AND NEW.addr_place IS NOT NULL THEN + address_street_word_id := get_name_id(make_standard_name(NEW.addr_place)); + IF address_street_word_id IS NOT NULL THEN + FOR location IN SELECT * from getNearestNamedPlaceFeature(NEW.partition, place_centroid, address_street_word_id) LOOP NEW.parent_place_id := location.place_id; END LOOP; END IF; @@ -1512,6 +1590,7 @@ BEGIN -- Merge address from parent nameaddress_vector := array_merge(nameaddress_vector, location.nameaddress_vector); + nameaddress_vector := array_merge(nameaddress_vector, location.name_vector); --return NEW; -- Performance, it would be more acurate to do all the rest of the import process but it takes too long -- Just be happy with inheriting from parent road only @@ -1520,7 +1599,7 @@ BEGIN result := add_location(NEW.place_id, NEW.calculated_country_code, NEW.partition, name_vector, NEW.rank_search, NEW.rank_address, NEW.geometry); END IF; - result := insertSearchName(NEW.partition, NEW.place_id, NEW.calculated_country_code, name_vector, nameaddress_vector, NEW.rank_search, NEW.rank_address, NEW.importance, place_centroid); + result := insertSearchName(NEW.partition, NEW.place_id, NEW.calculated_country_code, name_vector, nameaddress_vector, NEW.rank_search, NEW.rank_address, NEW.importance, place_centroid, NEW.geometry); return NEW; END IF; @@ -1550,7 +1629,7 @@ BEGIN -- merge in the label name, re-init word vector IF NOT linkedPlacex.name IS NULL THEN NEW.name := linkedPlacex.name || NEW.name; - name_vector := make_keywords(NEW.name); + name_vector := array_merge(name_vector, make_keywords(linkedPlacex.name)); END IF; -- merge in extra tags @@ -1558,6 +1637,10 @@ BEGIN NEW.extratags := linkedPlacex.extratags || NEW.extratags; END IF; + IF NOT NEW.extratags ? linkedPlacex.class THEN + NEW.extratags := NEW.extratags || hstore(linkedPlacex.class, linkedPlacex.type); + END IF; + -- mark the linked place (excludes from search results) UPDATE placex set linked_place_id = NEW.place_id where place_id = linkedPlacex.place_id; @@ -1577,7 +1660,6 @@ BEGIN IF make_standard_name(NEW.name->'name') = make_standard_name(linkedPlacex.name->'name') AND NEW.rank_address = linkedPlacex.rank_address THEN - -- If we don't already have one use this as the centre point of the geometry IF NEW.centroid IS NULL THEN NEW.centroid := coalesce(linkedPlacex.centroid,st_centroid(linkedPlacex.geometry)); @@ -1594,6 +1676,10 @@ BEGIN NEW.extratags := linkedPlacex.extratags || NEW.extratags; END IF; + IF NOT NEW.extratags ? linkedPlacex.class THEN + NEW.extratags := NEW.extratags || hstore(linkedPlacex.class, linkedPlacex.type); + END IF; + -- mark the linked place (excludes from search results) UPDATE placex set linked_place_id = NEW.place_id where place_id = linkedPlacex.place_id; @@ -1608,6 +1694,11 @@ BEGIN END IF; END IF; + END IF; + + -- Name searches can be done for ways as well as relations + IF NEW.osm_type in ('W','R') AND NEW.rank_search < 26 THEN + -- not found one yet? how about doing a name search IF NEW.centroid IS NULL AND (NEW.name->'name') is not null and make_standard_name(NEW.name->'name') != '' THEN @@ -1631,6 +1722,10 @@ BEGIN -- merge in extra tags NEW.extratags := linkedPlacex.extratags || NEW.extratags; + IF NOT NEW.extratags ? linkedPlacex.class THEN + NEW.extratags := NEW.extratags || hstore(linkedPlacex.class, linkedPlacex.type); + END IF; + -- mark the linked place (excludes from search results) UPDATE placex set linked_place_id = NEW.place_id where place_id = linkedPlacex.place_id; @@ -1641,6 +1736,19 @@ BEGIN IF NEW.centroid IS NOT NULL THEN place_centroid := NEW.centroid; + -- Place might have had only a name tag before but has now received translations + -- from the linked place. Make sure a name tag for the default language exists in + -- this case. + IF NEW.name is not null AND array_upper(akeys(NEW.name),1) > 1 THEN + default_language := get_country_language_code(NEW.calculated_country_code); + IF default_language IS NOT NULL THEN + IF NEW.name ? 'name' AND NOT NEW.name ? ('name:'||default_language) THEN + NEW.name := NEW.name || hstore(('name:'||default_language), (NEW.name -> 'name')); + ELSEIF NEW.name ? ('name:'||default_language) AND NOT NEW.name ? 'name' THEN + NEW.name := NEW.name || hstore('name', (NEW.name -> ('name:'||default_language))); + END IF; + END IF; + END IF; END IF; -- Did we gain a wikipedia tag in the process? then we need to recalculate our importance @@ -1670,8 +1778,15 @@ BEGIN FOR i IN 1..array_upper(isin, 1) LOOP address_street_word_id := get_name_id(make_standard_name(isin[i])); IF address_street_word_id IS NOT NULL AND NOT(ARRAY[address_street_word_id] <@ isin_tokens) THEN + nameaddress_vector := array_merge(nameaddress_vector, ARRAY[address_street_word_id]); isin_tokens := isin_tokens || address_street_word_id; END IF; + + -- merge word into address vector + address_street_word_id := get_word_id(make_standard_name(isin[i])); + IF address_street_word_id IS NOT NULL THEN + nameaddress_vector := array_merge(nameaddress_vector, ARRAY[address_street_word_id]); + END IF; END LOOP; END IF; END IF; @@ -1681,11 +1796,32 @@ BEGIN FOR i IN 1..array_upper(isin, 1) LOOP address_street_word_id := get_name_id(make_standard_name(isin[i])); IF address_street_word_id IS NOT NULL AND NOT(ARRAY[address_street_word_id] <@ isin_tokens) THEN + nameaddress_vector := array_merge(nameaddress_vector, ARRAY[address_street_word_id]); isin_tokens := isin_tokens || address_street_word_id; END IF; + + -- merge into address vector + address_street_word_id := get_word_id(make_standard_name(isin[i])); + IF address_street_word_id IS NOT NULL THEN + nameaddress_vector := array_merge(nameaddress_vector, ARRAY[address_street_word_id]); + END IF; END LOOP; END IF; END IF; + + -- for the USA we have an additional address table. Merge in zip codes from there too + IF NEW.rank_search = 26 AND NEW.calculated_country_code = 'us' THEN + FOR location IN SELECT distinct postcode from location_property_tiger where parent_place_id = NEW.place_id LOOP + address_street_word_id := get_name_id(make_standard_name(location.postcode)); + nameaddress_vector := array_merge(nameaddress_vector, ARRAY[address_street_word_id]); + isin_tokens := isin_tokens || address_street_word_id; + + -- also merge in the single word version + address_street_word_id := get_word_id(make_standard_name(location.postcode)); + nameaddress_vector := array_merge(nameaddress_vector, ARRAY[address_street_word_id]); + END LOOP; + END IF; + -- RAISE WARNING 'ISIN: %', isin_tokens; -- Process area matches @@ -1719,19 +1855,21 @@ BEGIN INSERT INTO place_addressline VALUES (NEW.place_id, location.place_id, true, location_isaddress, location.distance, location.rank_address); IF location_isaddress THEN - address_havelevel[location.rank_address] := true; - IF NOT location.isguess THEN - SELECT geometry FROM placex WHERE place_id = location.place_id INTO location_parent; - END IF; - END IF; ---RAISE WARNING ' Terms: (%) %',location, nameaddress_vector; + address_havelevel[location.rank_address] := true; + IF NOT location.isguess THEN + SELECT geometry FROM placex WHERE place_id = location.place_id INTO location_parent; + END IF; + + IF location.rank_address > parent_place_id_rank THEN + NEW.parent_place_id = location.place_id; + parent_place_id_rank = location.rank_address; + END IF; - IF location.rank_address > parent_place_id_rank THEN - NEW.parent_place_id = location.place_id; - parent_place_id_rank = location.rank_address; END IF; +--RAISE WARNING ' Terms: (%) %',location, nameaddress_vector; + END IF; END LOOP; @@ -1799,9 +1937,8 @@ BEGIN result := insertLocationRoad(NEW.partition, NEW.place_id, NEW.calculated_country_code, NEW.geometry); END IF; - result := insertSearchName(NEW.partition, NEW.place_id, NEW.calculated_country_code, name_vector, nameaddress_vector, NEW.rank_search, NEW.rank_address, NEW.importance, place_centroid); + result := insertSearchName(NEW.partition, NEW.place_id, NEW.calculated_country_code, name_vector, nameaddress_vector, NEW.rank_search, NEW.rank_address, NEW.importance, place_centroid, NEW.geometry); --- INSERT INTO search_name values (NEW.place_id, NEW.rank_search, NEW.rank_search, 0, NEW.calculated_country_code, name_vector, nameaddress_vector, place_centroid); END IF; -- If we've not managed to pick up a better one - default centroid @@ -1834,7 +1971,7 @@ BEGIN -- mark everything linked to this place for re-indexing --DEBUG: RAISE WARNING 'placex_delete:03 % %',OLD.osm_type,OLD.osm_id; UPDATE placex set indexed_status = 2 from place_addressline where address_place_id = OLD.place_id - and placex.place_id = place_addressline.place_id and indexed_status = 0; + and placex.place_id = place_addressline.place_id and indexed_status = 0 and place_addressline.isaddress; --DEBUG: RAISE WARNING 'placex_delete:04 % %',OLD.osm_type,OLD.osm_id; DELETE FROM place_addressline where address_place_id = OLD.place_id; @@ -1851,7 +1988,7 @@ BEGIN --DEBUG: RAISE WARNING 'placex_delete:08 % %',OLD.osm_type,OLD.osm_id; IF OLD.rank_address < 26 THEN - b := deleteLocationArea(OLD.partition, OLD.place_id); + b := deleteLocationArea(OLD.partition, OLD.place_id, OLD.rank_search); END IF; --DEBUG: RAISE WARNING 'placex_delete:09 % %',OLD.osm_type,OLD.osm_id; @@ -1868,7 +2005,7 @@ BEGIN -- remove from tables for special search classtable := 'place_classtype_' || OLD.class || '_' || OLD.type; - SELECT count(*)>0 FROM pg_tables WHERE tablename = classtable INTO b; + SELECT count(*)>0 FROM pg_tables WHERE tablename = classtable and schemaname = current_schema() INTO b; IF b THEN EXECUTE 'DELETE FROM ' || classtable::regclass || ' WHERE place_id = $1' USING OLD.place_id; END IF; @@ -1924,11 +2061,15 @@ BEGIN --DEBUG: RAISE WARNING '%', existingplacex; END IF; - -- Just block these - lots and pointless - IF NEW.class = 'highway' and NEW.type in ('turning_circle','traffic_signals','mini_roundabout','noexit','crossing') THEN - RETURN null; + -- remove operator tag for most places, messes too much with search_name indexes + IF NEW.class not in ('amenity', 'shop') THEN + NEW.name := delete(NEW.name, 'operator'); END IF; + + -- Just block these - lots and pointless IF NEW.class in ('landuse','natural') and NEW.name is null THEN + -- if the name tag was removed, older versions might still be lurking in the place table + DELETE FROM place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type; RETURN null; END IF; @@ -1974,16 +2115,30 @@ BEGIN DELETE from import_polygon_error where osm_type = NEW.osm_type and osm_id = NEW.osm_id; DELETE from import_polygon_delete where osm_type = NEW.osm_type and osm_id = NEW.osm_id; - -- To paraphrase, if there isn't an existing item, OR if the admin level has changed, OR if it is a major change in geometry - IF existingplacex.osm_type IS NULL THEN + -- To paraphrase, if there isn't an existing item, OR if the admin level has changed + IF existingplacex.osm_type IS NULL OR + (coalesce(existingplacex.admin_level, 15) != coalesce(NEW.admin_level, 15) AND existingplacex.class = 'boundary' AND existingplacex.type = 'administrative') + THEN + + IF existingplacex.osm_type IS NOT NULL THEN + -- sanity check: ignore admin_level changes on places with too many active children + -- or we end up reindexing entire countries because somebody accidentally deleted admin_level + --LIMIT INDEXING: SELECT count(*) FROM (SELECT 'a' FROM placex , place_addressline where address_place_id = existingplacex.place_id and placex.place_id = place_addressline.place_id and indexed_status = 0 and place_addressline.isaddress LIMIT 100001) sub INTO i; + --LIMIT INDEXING: IF i > 100000 THEN + --LIMIT INDEXING: RETURN null; + --LIMIT INDEXING: END IF; + END IF; IF existing.osm_type IS NOT NULL THEN + -- pathological case caused by the triggerless copy into place during initial import + -- force delete even for large areas, it will be reinserted later + UPDATE place set geometry = ST_SetSRID(ST_Point(0,0), 4326) where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type; DELETE from place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type; END IF; -- No - process it as a new insertion (hopefully of low rank or it will be slow) insert into placex (osm_type, osm_id, class, type, name, admin_level, housenumber, - street, isin, postcode, country_code, extratags, geometry) + street, addr_place, isin, postcode, country_code, extratags, geometry) values (NEW.osm_type ,NEW.osm_id ,NEW.class @@ -1992,6 +2147,7 @@ BEGIN ,NEW.admin_level ,NEW.housenumber ,NEW.street + ,NEW.addr_place ,NEW.isin ,NEW.postcode ,NEW.country_code @@ -2017,6 +2173,9 @@ BEGIN IF coalesce(existing.street, '') != coalesce(NEW.street, '') THEN RAISE WARNING 'update details, street: % % % %',NEW.osm_type,NEW.osm_id,existing.street,NEW.street; END IF; + IF coalesce(existing.addr_place, '') != coalesce(NEW.addr_place, '') THEN + RAISE WARNING 'update details, street: % % % %',NEW.osm_type,NEW.osm_id,existing.addr_place,NEW.addr_place; + END IF; IF coalesce(existing.isin, '') != coalesce(NEW.isin, '') THEN RAISE WARNING 'update details, isin: % % % %',NEW.osm_type,NEW.osm_id,existing.isin,NEW.isin; END IF; @@ -2059,6 +2218,7 @@ BEGIN IF FALSE AND existingplacex.rank_search < 26 AND coalesce(existing.housenumber, '') = coalesce(NEW.housenumber, '') AND coalesce(existing.street, '') = coalesce(NEW.street, '') + AND coalesce(existing.addr_place, '') = coalesce(NEW.addr_place, '') AND coalesce(existing.isin, '') = coalesce(NEW.isin, '') AND coalesce(existing.postcode, '') = coalesce(NEW.postcode, '') AND coalesce(existing.country_code, '') = coalesce(NEW.country_code, '') @@ -2081,6 +2241,7 @@ BEGIN IF coalesce(existing.name::text, '') != coalesce(NEW.name::text, '') OR coalesce(existing.housenumber, '') != coalesce(NEW.housenumber, '') OR coalesce(existing.street, '') != coalesce(NEW.street, '') + OR coalesce(existing.addr_place, '') != coalesce(NEW.addr_place, '') OR coalesce(existing.isin, '') != coalesce(NEW.isin, '') OR coalesce(existing.postcode, '') != coalesce(NEW.postcode, '') OR coalesce(existing.country_code, '') != coalesce(NEW.country_code, '') THEN @@ -2095,10 +2256,17 @@ BEGIN END IF; + -- refuse to update multiplpoygons with too many objects, too much of a performance hit + IF ST_NumGeometries(NEW.geometry) > 2000 THEN + RAISE WARNING 'Dropping update of % % because of geometry complexity.', NEW.osm_type, NEW.osm_id; + RETURN NULL; + END IF; + IF coalesce(existing.name::text, '') != coalesce(NEW.name::text, '') OR coalesce(existing.extratags::text, '') != coalesce(NEW.extratags::text, '') OR coalesce(existing.housenumber, '') != coalesce(NEW.housenumber, '') OR coalesce(existing.street, '') != coalesce(NEW.street, '') + OR coalesce(existing.addr_place, '') != coalesce(NEW.addr_place, '') OR coalesce(existing.isin, '') != coalesce(NEW.isin, '') OR coalesce(existing.postcode, '') != coalesce(NEW.postcode, '') OR coalesce(existing.country_code, '') != coalesce(NEW.country_code, '') @@ -2110,10 +2278,12 @@ BEGIN name = NEW.name, housenumber = NEW.housenumber, street = NEW.street, + addr_place = NEW.addr_place, isin = NEW.isin, postcode = NEW.postcode, country_code = NEW.country_code, extratags = NEW.extratags, + admin_level = NEW.admin_level, geometry = NEW.geometry where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type; @@ -2121,11 +2291,13 @@ BEGIN name = NEW.name, housenumber = NEW.housenumber, street = NEW.street, + addr_place = NEW.addr_place, isin = NEW.isin, postcode = NEW.postcode, country_code = NEW.country_code, parent_place_id = null, extratags = NEW.extratags, + admin_level = CASE WHEN NEW.admin_level > 15 THEN 15 ELSE NEW.admin_level END, indexed_status = 2, geometry = NEW.geometry where place_id = existingplacex.place_id; @@ -2157,7 +2329,9 @@ BEGIN END IF; END LOOP; - RETURN null; + -- anything will do as a fallback - just take the first name type thing there is + search := avals(name); + RETURN search[1]; END; $$ LANGUAGE plpgsql IMMUTABLE; @@ -2250,11 +2424,11 @@ END; $$ LANGUAGE plpgsql; -DROP TYPE addressline CASCADE; +DROP TYPE IF EXISTS addressline CASCADE; create type addressline as ( place_id BIGINT, osm_type CHAR(1), - osm_id INTEGER, + osm_id BIGINT, name HSTORE, class TEXT, type TEXT, @@ -2349,7 +2523,7 @@ BEGIN CASE WHEN class = 'place' and type = 'postcode' THEN hstore('name', postcode) ELSE name END as name, class, type, admin_level, fromarea, isaddress, CASE WHEN address_place_id = for_place_id AND rank_address = 0 THEN 100 WHEN rank_address = 11 THEN 5 ELSE rank_address END as rank_address, - distance,calculated_country_code + distance,calculated_country_code,postcode from place_addressline join placex on (address_place_id = placex.place_id) where place_addressline.place_id = for_place_id and (cached_rank_address > 0 AND cached_rank_address < searchrankaddress) @@ -2364,6 +2538,9 @@ BEGIN IF searchpostcode IS NOT NULL and location.type = 'postcode' THEN location.isaddress := FALSE; END IF; + IF searchpostcode IS NULL and location.isaddress and location.type != 'postcode' and location.postcode IS NOT NULL THEN + searchpostcode := location.postcode; + END IF; IF location.rank_address = 4 AND location.isaddress THEN hadcountry := true; END IF; @@ -2502,6 +2679,7 @@ BEGIN name = place.name, housenumber = place.housenumber, street = place.street, + addr_place = place.addr_place, isin = place.isin, postcode = place.postcode, country_code = place.country_code, @@ -2938,9 +3116,9 @@ BEGIN IF ST_GeometryType(placegeom) in ('ST_Polygon','ST_MultiPolygon') THEN FOR geom IN select split_geometry(placegeom) FROM placex WHERE place_id = placeid LOOP update placex set indexed_status = 2 where (st_covers(geom, placex.geometry) OR ST_Intersects(geom, placex.geometry)) - AND rank_search > rank and indexed_status = 0 and ST_geometrytype(placex.geometry) = 'ST_Point' and (rank_search < 28 or name is not null); + AND rank_search > rank and indexed_status = 0 and ST_geometrytype(placex.geometry) = 'ST_Point' and (rank_search < 28 or name is not null or (rank >= 16 and addr_place is not null)); update placex set indexed_status = 2 where (st_covers(geom, placex.geometry) OR ST_Intersects(geom, placex.geometry)) - AND rank_search > rank and indexed_status = 0 and ST_geometrytype(placex.geometry) != 'ST_Point' and (rank_search < 28 or name is not null); + AND rank_search > rank and indexed_status = 0 and ST_geometrytype(placex.geometry) != 'ST_Point' and (rank_search < 28 or name is not null or (rank >= 16 and addr_place is not null)); END LOOP; ELSE diameter := 0; @@ -2960,7 +3138,16 @@ BEGIN diameter := 0.001; -- 50 to 100 meters END IF; IF diameter > 0 THEN - update placex set indexed_status = 2 where indexed_status = 0 and rank_search > rank and ST_DWithin(placex.geometry, placegeom, diameter) and (rank_search < 28 or name is not null); + IF rank >= 26 THEN + -- roads may cause reparenting for >27 rank places + update placex set indexed_status = 2 where indexed_status = 0 and rank_search > rank and ST_DWithin(placex.geometry, placegeom, diameter); + ELSEIF rank >= 16 THEN + -- up to rank 16, street-less addresses may need reparenting + update placex set indexed_status = 2 where indexed_status = 0 and rank_search > rank and ST_DWithin(placex.geometry, placegeom, diameter) and (rank_search < 28 or name is not null or addr_place is not null); + ELSE + -- for all other places the search terms may change as well + update placex set indexed_status = 2 where indexed_status = 0 and rank_search > rank and ST_DWithin(placex.geometry, placegeom, diameter) and (rank_search < 28 or name is not null); + END IF; END IF; END IF; RETURN TRUE;