X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/09bde0baf96af07d5873e952f604fe4e2409e455..8564b76add152949d0493f8d15c5049f154118eb:/sql/functions.sql diff --git a/sql/functions.sql b/sql/functions.sql index 818f6ac1..4aef74d1 100644 --- a/sql/functions.sql +++ b/sql/functions.sql @@ -271,6 +271,20 @@ END; $$ LANGUAGE plpgsql IMMUTABLE; +CREATE OR REPLACE FUNCTION get_name_ids(lookup_word TEXT) + RETURNS INTEGER[] + AS $$ +DECLARE + lookup_token TEXT; + return_word_ids INTEGER[]; +BEGIN + lookup_token := ' '||trim(lookup_word); + SELECT array_agg(word_id) FROM word WHERE word_token = lookup_token and class is null and type is null into return_word_ids; + RETURN return_word_ids; +END; +$$ +LANGUAGE plpgsql IMMUTABLE; + CREATE OR REPLACE FUNCTION array_merge(a INTEGER[], b INTEGER[]) RETURNS INTEGER[] AS $$ @@ -347,7 +361,7 @@ BEGIN w := getorcreate_word_id(s); - IF NOT (ARRAY[w] <@ result) THEN + IF w IS NOT NULL AND NOT (ARRAY[w] <@ result) THEN result := result || w; END IF; @@ -415,7 +429,7 @@ BEGIN w := getorcreate_word_id(s); - IF NOT (ARRAY[w] <@ result) THEN + IF w IS NOT NULL AND NOT (ARRAY[w] <@ result) THEN result := result || w; END IF; @@ -554,20 +568,6 @@ END; $$ LANGUAGE plpgsql IMMUTABLE; -CREATE OR REPLACE FUNCTION get_country_code(place geometry, in_country_code VARCHAR(2)) RETURNS TEXT - AS $$ -DECLARE - nearcountry RECORD; -BEGIN - FOR nearcountry IN select country_code from country_name where country_code = lower(in_country_code) - LOOP - RETURN nearcountry.country_code; - END LOOP; - RETURN get_country_code(place); -END; -$$ -LANGUAGE plpgsql IMMUTABLE; - CREATE OR REPLACE FUNCTION get_country_language_code(search_country_code VARCHAR(2)) RETURNS TEXT AS $$ DECLARE @@ -596,10 +596,9 @@ END; $$ LANGUAGE plpgsql IMMUTABLE; -CREATE OR REPLACE FUNCTION get_partition(place geometry, in_country_code VARCHAR(10)) RETURNS INTEGER +CREATE OR REPLACE FUNCTION get_partition(in_country_code VARCHAR(10)) RETURNS INTEGER AS $$ DECLARE - place_centre GEOMETRY; nearcountry RECORD; BEGIN FOR nearcountry IN select partition from country_name where country_code = in_country_code @@ -792,7 +791,7 @@ LANGUAGE plpgsql; CREATE OR REPLACE FUNCTION create_interpolation(wayid BIGINT, interpolationtype TEXT) RETURNS INTEGER AS $$ DECLARE - + newpoints INTEGER; waynodes BIGINT[]; nodeid BIGINT; @@ -805,117 +804,103 @@ DECLARE originalnumberrange INTEGER; housenum INTEGER; linegeo GEOMETRY; + splitline GEOMETRY; + sectiongeo GEOMETRY; search_place_id BIGINT; defpostalcode TEXT; - havefirstpoint BOOLEAN; - linestr TEXT; BEGIN - newpoints := 0; - IF interpolationtype = 'odd' OR interpolationtype = 'even' OR interpolationtype = 'all' THEN - - select postcode from placex where osm_type = 'W' and osm_id = wayid INTO defpostalcode; - select nodes from planet_osm_ways where id = wayid INTO waynodes; ---RAISE WARNING 'interpolation % % %',wayid,interpolationtype,waynodes; - IF array_upper(waynodes, 1) IS NOT NULL THEN - - havefirstpoint := false; - - FOR nodeidpos in 1..array_upper(waynodes, 1) LOOP - - -- 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 - -- 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 - -- because an excerpt was updated and the node never imported - -- because the interpolation is outside the region of the excerpt. - -- Give up. - RETURN newpoints; - END IF; - ELSE - select * from placex where place_id = search_place_id INTO nextnode; - END IF; + IF interpolationtype = 'odd' OR interpolationtype = 'even' THEN + stepsize := 2; + ELSEIF interpolationtype = 'all' THEN + stepsize := 1; + ELSEIF interpolationtype ~ '^\d+$' THEN + stepsize := interpolationtype::INTEGER; + ELSE + RETURN 0; + END IF; ---RAISE WARNING 'interpolation node % % % ',nextnode.housenumber,ST_X(nextnode.geometry),ST_Y(nextnode.geometry); - - IF havefirstpoint THEN + select nodes from planet_osm_ways where id = wayid INTO waynodes; - -- add point to the line string - linestr := linestr||','||ST_X(nextnode.geometry)||' '||ST_Y(nextnode.geometry); - endnumber := ('0'||substring(nextnode.housenumber,'[0-9]+'))::integer; + IF array_upper(waynodes, 1) IS NULL THEN + RETURN 0; + END IF; - IF startnumber IS NOT NULL and startnumber > 0 AND endnumber IS NOT NULL and endnumber > 0 AND @(startnumber - endnumber) < 1000 THEN + select postcode, geometry from placex where osm_type = 'W' and osm_id = wayid + INTO defpostalcode, linegeo; ---RAISE WARNING 'interpolation end % % ',nextnode.place_id,endnumber; + IF ST_GeometryType(linegeo) != 'ST_LineString' THEN + RETURN 0; + END IF; - IF startnumber != endnumber THEN + startnumber := NULL; + newpoints := 0; - linestr := linestr || ')'; ---RAISE WARNING 'linestr %',linestr; - linegeo := ST_GeomFromText(linestr,4326); - linestr := 'LINESTRING('||ST_X(nextnode.geometry)||' '||ST_Y(nextnode.geometry); - IF (startnumber > endnumber) THEN - housenum := endnumber; - endnumber := startnumber; - startnumber := housenum; - linegeo := ST_Reverse(linegeo); - END IF; - orginalstartnumber := startnumber; - originalnumberrange := endnumber - startnumber; - --- Too much broken data worldwide for this test to be worth using --- IF originalnumberrange > 500 THEN --- RAISE WARNING 'Number block of % while processing % %', originalnumberrange, prevnode, nextnode; --- END IF; - - IF (interpolationtype = 'odd' AND startnumber%2 = 0) OR (interpolationtype = 'even' AND startnumber%2 = 1) THEN - startnumber := startnumber + 1; - stepsize := 2; - ELSE - IF (interpolationtype = 'odd' OR interpolationtype = 'even') THEN - startnumber := startnumber + 2; - stepsize := 2; - ELSE -- everything else assumed to be 'all' - startnumber := startnumber + 1; - stepsize := 1; - END IF; - END IF; - endnumber := endnumber - 1; - delete from placex where osm_type = 'N' and osm_id = prevnode.osm_id and type = 'house' and place_id != prevnode.place_id; - 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, addr_place, isin, postcode, - country_code, parent_place_id, rank_address, rank_search, indexed_status, geometry) - 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_Line_Interpolate_Point(linegeo, (housenum::float-orginalstartnumber::float)/originalnumberrange::float)); - newpoints := newpoints + 1; ---RAISE WARNING 'interpolation number % % ',prevnode.place_id,housenum; - END LOOP; - END IF; - havefirstpoint := false; + FOR nodeidpos in 1..array_upper(waynodes, 1) LOOP + + -- 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 NOT NULL THEN + select * from placex where place_id = search_place_id INTO nextnode; + + IF nodeidpos < array_upper(waynodes, 1) THEN + -- Make sure that the point is actually on the line. That might + -- be a bit paranoid but ensures that the algorithm still works + -- should osm2pgsql attempt to repair geometries. + splitline := split_line_on_node(linegeo, nextnode.geometry); + sectiongeo := ST_GeometryN(splitline, 1); + IF ST_GeometryType(ST_GeometryN(splitline, 2)) = 'ST_LineString' THEN + linegeo := ST_GeometryN(splitline, 2); END IF; + ELSE + sectiongeo = linegeo; END IF; + endnumber := substring(nextnode.housenumber,'[0-9]+')::integer; + + IF startnumber IS NOT NULL AND endnumber IS NOT NULL + AND @(startnumber - endnumber) < 1000 AND startnumber != endnumber + AND ST_GeometryType(linegeo) = 'ST_LineString' THEN + + IF (startnumber > endnumber) THEN + housenum := endnumber; + endnumber := startnumber; + startnumber := housenum; + sectiongeo := ST_Reverse(sectiongeo); + END IF; + orginalstartnumber := startnumber; + originalnumberrange := endnumber - startnumber; - IF NOT havefirstpoint THEN - startnumber := ('0'||substring(nextnode.housenumber,'[0-9]+'))::integer; - IF startnumber IS NOT NULL AND startnumber > 0 THEN - havefirstpoint := true; - linestr := 'LINESTRING('||ST_X(nextnode.geometry)||' '||ST_Y(nextnode.geometry); - prevnode := nextnode; + startnumber := startnumber + stepsize; + -- correct for odd/even + IF (interpolationtype = 'odd' AND startnumber%2 = 0) OR (interpolationtype = 'even' AND startnumber%2 = 1) THEN + startnumber := startnumber - 1; END IF; ---RAISE WARNING 'interpolation start % % ',nextnode.place_id,startnumber; + endnumber := endnumber - 1; + + delete from placex where osm_type = 'N' and osm_id = prevnode.osm_id and type = 'house' and place_id != prevnode.place_id; + FOR housenum IN startnumber..endnumber BY stepsize LOOP + -- this should really copy postcodes but it puts a huge burden 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, addr_place, isin, postcode, + country_code, parent_place_id, rank_address, rank_search, + indexed_status, geometry) + 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(sectiongeo, (housenum::float-orginalstartnumber::float)/originalnumberrange::float)); + newpoints := newpoints + 1; +--RAISE WARNING 'interpolation number % % ',prevnode.place_id,housenum; + END LOOP; END IF; - END LOOP; + + startnumber := substring(nextnode.housenumber,'[0-9]+')::integer; + prevnode := nextnode; END IF; - END IF; + END LOOP; --RAISE WARNING 'interpolation points % ',newpoints; @@ -937,6 +922,11 @@ DECLARE BEGIN --DEBUG: RAISE WARNING '% %',NEW.osm_type,NEW.osm_id; + -- 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; @@ -965,9 +955,9 @@ BEGIN NEW.place_id := nextval('seq_place'); NEW.indexed_status := 1; --STATUS_NEW - NEW.calculated_country_code := lower(get_country_code(NEW.geometry, NEW.country_code)); + NEW.calculated_country_code := lower(get_country_code(NEW.geometry)); - NEW.partition := get_partition(NEW.geometry, NEW.calculated_country_code); + NEW.partition := get_partition(NEW.calculated_country_code); NEW.geometry_sector := geometry_sector(NEW.partition, NEW.geometry); -- copy 'name' to or from the default language (if there is a default language) @@ -1019,6 +1009,13 @@ BEGIN NEW.rank_address := 5; END IF; + ELSEIF NEW.calculated_country_code = 'sg' THEN + + IF NEW.postcode ~ '^([0-9]{6})$' THEN + NEW.rank_search := 25; + NEW.rank_address := 11; + END IF; + ELSEIF NEW.calculated_country_code = 'de' THEN IF NEW.postcode ~ '^([0-9]{5})$' THEN @@ -1078,18 +1075,6 @@ BEGIN ELSEIF NEW.type in ('village','hamlet','municipality','district','unincorporated_area','borough') THEN NEW.rank_search := 19; NEW.rank_address := 16; - ELSEIF NEW.type in ('airport') AND ST_GeometryType(NEW.geometry) in ('ST_Polygon','ST_MultiPolygon') THEN - NEW.rank_search := 18; - NEW.rank_address := 17; - ELSEIF NEW.type in ('moor') AND ST_GeometryType(NEW.geometry) in ('ST_Polygon','ST_MultiPolygon') THEN - NEW.rank_search := 17; - NEW.rank_address := 18; - ELSEIF NEW.type in ('moor') THEN - NEW.rank_search := 17; - NEW.rank_address := 0; - ELSEIF NEW.type in ('national_park') THEN - NEW.rank_search := 18; - NEW.rank_address := 18; ELSEIF NEW.type in ('suburb','croft','subdivision','isolated_dwelling') THEN NEW.rank_search := 20; NEW.rank_address := NEW.rank_search; @@ -1100,12 +1085,9 @@ BEGIN IF (NEW.extratags -> 'locality') = 'townland' THEN NEW.rank_address := 20; END IF; - ELSEIF NEW.type in ('hall_of_residence','neighbourhood','housing_estate','nature_reserve') THEN + ELSEIF NEW.type in ('neighbourhood') THEN NEW.rank_search := 22; NEW.rank_address := 22; - ELSEIF NEW.type in ('airport','street') THEN - NEW.rank_search := 26; - NEW.rank_address := NEW.rank_search; ELSEIF NEW.type in ('house','building') THEN NEW.rank_search := 30; NEW.rank_address := NEW.rank_search; @@ -1123,13 +1105,26 @@ 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; + IF NEW.type in ('residential', 'farm', 'farmyard', 'industrial', 'commercial', 'allotments', 'retail') THEN + NEW.rank_address := NEW.rank_search; + ELSE + NEW.rank_address := 0; + END IF; ELSEIF NEW.class = 'natural' and NEW.type in ('peak','volcano','mountain_range') THEN NEW.rank_search := 18; NEW.rank_address := 0; + ELSEIF NEW.class = 'natural' and NEW.type = 'sea' THEN + NEW.rank_search := 4; + NEW.rank_address := NEW.rank_search; + ELSEIF NEW.class = 'natural' and NEW.type in ('coastline') THEN + RETURN NULL; -- any feature more than 5 square miles is probably worth indexing ELSEIF ST_GeometryType(NEW.geometry) in ('ST_Polygon','ST_MultiPolygon') AND ST_Area(NEW.geometry) > 0.1 THEN NEW.rank_search := 22; @@ -1143,18 +1138,18 @@ BEGIN ELSEIF NEW.class = 'waterway' AND NEW.name is NULL THEN RETURN NULL; ELSEIF NEW.class = 'waterway' THEN - NEW.rank_address := 17; + IF NEW.osm_type = 'R' THEN + NEW.rank_search := 16; + ELSE + NEW.rank_search := 17; + END IF; + NEW.rank_address := 0; ELSEIF NEW.class = 'highway' AND NEW.osm_type != 'N' AND NEW.type in ('service','cycleway','path','footway','steps','bridleway','motorway_link','primary_link','trunk_link','secondary_link','tertiary_link') THEN NEW.rank_search := 27; NEW.rank_address := NEW.rank_search; ELSEIF NEW.class = 'highway' AND NEW.osm_type != 'N' THEN NEW.rank_search := 26; NEW.rank_address := NEW.rank_search; - ELSEIF NEW.class = 'natural' and NEW.type = 'sea' THEN - NEW.rank_search := 4; - 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; @@ -1175,7 +1170,7 @@ BEGIN END IF; -- a country code make no sense below rank 4 (country) - IF NEW.rank_address < 4 THEN + IF NEW.rank_search < 4 THEN NEW.calculated_country_code := NULL; END IF; @@ -1188,60 +1183,61 @@ BEGIN RETURN NEW; -- @DIFFUPDATES@ The following is not needed until doing diff updates, and slows the main index process down - IF (ST_GeometryType(NEW.geometry) in ('ST_Polygon','ST_MultiPolygon') AND ST_IsValid(NEW.geometry)) THEN - -- Performance: We just can't handle re-indexing for country level changes - IF st_area(NEW.geometry) < 1 THEN - -- mark items within the geometry for re-indexing --- RAISE WARNING 'placex poly insert: % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type; - - -- 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 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 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 :( - diameter := 0; - -- 16 = city, anything higher than city is effectively ignored (polygon required!) - IF NEW.type='postcode' THEN - diameter := 0.05; - ELSEIF NEW.rank_search < 16 THEN + IF NEW.rank_address > 0 THEN + IF (ST_GeometryType(NEW.geometry) in ('ST_Polygon','ST_MultiPolygon') AND ST_IsValid(NEW.geometry)) THEN + -- Performance: We just can't handle re-indexing for country level changes + IF st_area(NEW.geometry) < 1 THEN + -- mark items within the geometry for re-indexing + -- RAISE WARNING 'placex poly insert: % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type; + + -- 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 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 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 :( diameter := 0; - ELSEIF NEW.rank_search < 18 THEN - diameter := 0.1; - ELSEIF NEW.rank_search < 20 THEN - diameter := 0.05; - ELSEIF NEW.rank_search = 21 THEN - diameter := 0.001; - ELSEIF NEW.rank_search < 24 THEN - diameter := 0.02; - ELSEIF NEW.rank_search < 26 THEN - diameter := 0.002; -- 100 to 200 meters - ELSEIF NEW.rank_search < 28 THEN - diameter := 0.001; -- 50 to 100 meters - END IF; - IF diameter > 0 THEN --- RAISE WARNING 'placex point insert: % % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type,diameter; - 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); + -- 16 = city, anything higher than city is effectively ignored (polygon required!) + IF NEW.type='postcode' THEN + diameter := 0.05; + ELSEIF NEW.rank_search < 16 THEN + diameter := 0; + ELSEIF NEW.rank_search < 18 THEN + diameter := 0.1; + ELSEIF NEW.rank_search < 20 THEN + diameter := 0.05; + ELSEIF NEW.rank_search = 21 THEN + diameter := 0.001; + ELSEIF NEW.rank_search < 24 THEN + diameter := 0.02; + ELSEIF NEW.rank_search < 26 THEN + diameter := 0.002; -- 100 to 200 meters + ELSEIF NEW.rank_search < 28 THEN + diameter := 0.001; -- 50 to 100 meters + END IF; + IF diameter > 0 THEN + -- RAISE WARNING 'placex point insert: % % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type,diameter; + 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; - END IF; -- add to tables for special search -- 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); @@ -1284,6 +1280,7 @@ DECLARE search_maxrank INTEGER; address_maxrank INTEGER; address_street_word_id INTEGER; + address_street_word_ids INTEGER[]; parent_place_id_rank BIGINT; isin TEXT[]; @@ -1357,10 +1354,10 @@ BEGIN -- 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)); + NEW.partition := get_partition(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); + NEW.partition := get_partition(NEW.calculated_country_code); ELSE NEW.calculated_country_code := lower(NEW.country_code); END IF; @@ -1371,10 +1368,29 @@ BEGIN ELSE NEW.calculated_country_code := NULL; END IF; - NEW.partition := get_partition(place_centroid, NEW.calculated_country_code); + NEW.partition := get_partition(NEW.calculated_country_code); END IF; NEW.geometry_sector := geometry_sector(NEW.partition, place_centroid); + -- waterway ways are linked when they are part of a relation and have the same class/type + IF NEW.osm_type = 'R' and NEW.class = 'waterway' THEN + FOR relation IN select * from planet_osm_rels r where r.id = NEW.osm_id and r.parts != array[]::bigint[] + LOOP + FOR i IN 1..array_upper(relation.members, 1) BY 2 LOOP + IF relation.members[i+1] in ('', 'main_stream', 'side_stream') AND substring(relation.members[i],1,1) = 'w' THEN + --DEBUG: RAISE WARNING 'waterway parent %, child %/%', NEW.osm_id, i, relation.parts[i]; + FOR location IN SELECT * FROM placex + WHERE osm_type = 'W' and osm_id = substring(relation.members[i],2,200)::bigint + and class = NEW.class and type = NEW.type + and ( relation.members[i+1] != 'side_stream' or NEW.name->'name' = name->'name') + LOOP + UPDATE placex SET linked_place_id = NEW.place_id WHERE place_id = location.place_id; + END LOOP; + END IF; + END LOOP; + END LOOP; + END IF; + -- Adding ourselves to the list simplifies address calculations later INSERT INTO place_addressline VALUES (NEW.place_id, NEW.place_id, true, true, 0, NEW.rank_address); @@ -1431,6 +1447,16 @@ BEGIN -- Note that addr:street links can only be indexed once the street itself is indexed IF NEW.parent_place_id IS NULL AND NEW.osm_type = 'N' THEN + -- if there is no address information, see if we can get it from a surrounding building + IF NEW.street IS NULL AND NEW.addr_place IS NULL AND NEW.housenumber IS NULL THEN + FOR location IN select * from placex where ST_Covers(geometry, place_centroid) and rank_search > 28 and (housenumber is not null or street is not null or addr_place is not null) AND ST_GeometryType(geometry) in ('ST_Polygon','ST_MultiPolygon') + LOOP + NEW.housenumber := location.housenumber; + NEW.street := location.street; + NEW.addr_place := location.addr_place; + END LOOP; + END IF; + -- Is this node part of a relation? FOR relation IN select * from planet_osm_rels where parts @> ARRAY[NEW.osm_id] and members @> ARRAY['n'||NEW.osm_id] LOOP @@ -1440,7 +1466,7 @@ BEGIN 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)::bigint - and rank_search = 26 INTO NEW.parent_place_id; + and rank_search = 26 and name is not null INTO NEW.parent_place_id; END IF; END LOOP; END IF; @@ -1469,7 +1495,7 @@ BEGIN 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)::bigint - and rank_search = 26 INTO NEW.parent_place_id; + and rank_search = 26 and name is not null INTO NEW.parent_place_id; END IF; END LOOP; END IF; @@ -1477,13 +1503,10 @@ BEGIN END IF; -- If the way contains an explicit name of a street copy it - IF NEW.street IS NULL AND NEW.addr_place IS NULL AND location.street IS NOT NULL THEN + -- Slightly less strict then above because data is copied from any object. + IF NEW.street IS NULL AND NEW.addr_place IS 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; @@ -1493,9 +1516,9 @@ BEGIN 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; @@ -1519,7 +1542,7 @@ BEGIN 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)::bigint - and rank_search = 26 INTO NEW.parent_place_id; + and rank_search = 26 and name is not null INTO NEW.parent_place_id; END IF; END LOOP; END IF; @@ -1529,18 +1552,18 @@ 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)); - IF address_street_word_id IS NOT NULL THEN - FOR location IN SELECT * from getNearestNamedRoadFeature(NEW.partition, place_centroid, address_street_word_id) LOOP + address_street_word_ids := get_name_ids(make_standard_name(NEW.street)); + IF address_street_word_ids IS NOT NULL THEN + FOR location IN SELECT * from getNearestNamedRoadFeature(NEW.partition, place_centroid, address_street_word_ids) 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 + address_street_word_ids := get_name_ids(make_standard_name(NEW.addr_place)); + IF address_street_word_ids IS NOT NULL THEN + FOR location IN SELECT * from getNearestNamedPlaceFeature(NEW.partition, place_centroid, address_street_word_ids) LOOP NEW.parent_place_id := location.place_id; END LOOP; END IF; @@ -1624,17 +1647,13 @@ BEGIN END IF; -- merge in extra tags - IF NOT linkedPlacex.extratags IS NULL THEN - 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; + NEW.extratags := hstore(linkedPlacex.class, linkedPlacex.type) || coalesce(linkedPlacex.extratags, ''::hstore) || coalesce(NEW.extratags, ''::hstore); -- mark the linked place (excludes from search results) UPDATE placex set linked_place_id = NEW.place_id where place_id = linkedPlacex.place_id; + -- keep a note of the node id in case we need it for wikipedia in a bit + linked_node_id := linkedPlacex.osm_id; END LOOP; END LOOP; @@ -1663,13 +1682,7 @@ BEGIN END IF; -- merge in extra tags - IF NOT linkedPlacex.extratags IS NULL THEN - 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; + NEW.extratags := hstore(linkedPlacex.class, linkedPlacex.type) || coalesce(linkedPlacex.extratags, ''::hstore) || coalesce(NEW.extratags, ''::hstore); -- mark the linked place (excludes from search results) UPDATE placex set linked_place_id = NEW.place_id where place_id = linkedPlacex.place_id; @@ -1711,11 +1724,7 @@ BEGIN name_vector := make_keywords(NEW.name); -- 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; + NEW.extratags := hstore(linkedPlacex.class, linkedPlacex.type) || coalesce(linkedPlacex.extratags, ''::hstore) || coalesce(NEW.extratags, ''::hstore); -- mark the linked place (excludes from search results) UPDATE placex set linked_place_id = NEW.place_id where place_id = linkedPlacex.place_id; @@ -1962,7 +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; @@ -1996,7 +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; @@ -2012,15 +2021,18 @@ LANGUAGE plpgsql; CREATE OR REPLACE FUNCTION place_delete() RETURNS TRIGGER AS $$ DECLARE - placeid BIGINT; + has_rank BOOLEAN; BEGIN --DEBUG: RAISE WARNING 'delete: % % % %',OLD.osm_type,OLD.osm_id,OLD.class,OLD.type; -- deleting large polygons can have a massive effect on the system - require manual intervention to let them through IF st_area(OLD.geometry) > 2 and st_isvalid(OLD.geometry) THEN - insert into import_polygon_delete values (OLD.osm_type,OLD.osm_id,OLD.class,OLD.type); - RETURN NULL; + SELECT bool_or(not (rank_address = 0 or rank_address > 26)) as ranked FROM placex WHERE osm_type = OLD.osm_type and osm_id = OLD.osm_id and class = OLD.class and type = OLD.type INTO has_rank; + IF has_rank THEN + insert into import_polygon_delete values (OLD.osm_type,OLD.osm_id,OLD.class,OLD.type); + RETURN NULL; + END IF; END IF; -- mark for delete @@ -2052,8 +2064,15 @@ BEGIN --DEBUG: RAISE WARNING '%', existingplacex; END IF; + -- 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; @@ -2078,8 +2097,8 @@ BEGIN -- Handle a place changing type by removing the old data -- My generated 'place' types are causing havok because they overlap with real keys -- TODO: move them to their own special purpose key/class to avoid collisions - IF existing.osm_type IS NULL AND (NEW.type not in ('postcode','house','houses')) THEN - DELETE FROM place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type not in ('postcode','house','houses'); + IF existing.osm_type IS NULL THEN + DELETE FROM place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class; END IF; --DEBUG: RAISE WARNING 'Existing: %',existing.osm_id; @@ -2099,8 +2118,19 @@ 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 @@ -2229,6 +2259,12 @@ 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, '') @@ -2254,6 +2290,16 @@ BEGIN geometry = NEW.geometry where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type; + IF NEW.class in ('place','boundary') AND NEW.type in ('postcode','postal_code') THEN + IF NEW.postcode IS NULL THEN + -- postcode was deleted, no longer retain in placex + DELETE FROM placex where place_id = existingplacex.place_id; + RETURN NULL; + END IF; + + NEW.name := hstore('ref', NEW.postcode); + END IF; + update placex set name = NEW.name, housenumber = NEW.housenumber, @@ -2303,29 +2349,29 @@ END; $$ LANGUAGE plpgsql IMMUTABLE; -CREATE OR REPLACE FUNCTION get_connected_ways(way_ids INTEGER[]) RETURNS SETOF planet_osm_ways - AS $$ -DECLARE - searchnodes INTEGER[]; - location RECORD; - j INTEGER; -BEGIN - - searchnodes := '{}'; - FOR j IN 1..array_upper(way_ids, 1) LOOP - FOR location IN - select nodes from planet_osm_ways where id = way_ids[j] LIMIT 1 - LOOP - IF not (ARRAY[location.nodes] <@ searchnodes) THEN - searchnodes := searchnodes || location.nodes; - END IF; - END LOOP; - END LOOP; - - RETURN QUERY select * from planet_osm_ways where nodes && searchnodes and NOT ARRAY[id] <@ way_ids; -END; -$$ -LANGUAGE plpgsql IMMUTABLE; +--CREATE OR REPLACE FUNCTION get_connected_ways(way_ids INTEGER[]) RETURNS SETOF planet_osm_ways +-- AS $$ +--DECLARE +-- searchnodes INTEGER[]; +-- location RECORD; +-- j INTEGER; +--BEGIN +-- +-- searchnodes := '{}'; +-- FOR j IN 1..array_upper(way_ids, 1) LOOP +-- FOR location IN +-- select nodes from planet_osm_ways where id = way_ids[j] LIMIT 1 +-- LOOP +-- IF not (ARRAY[location.nodes] <@ searchnodes) THEN +-- searchnodes := searchnodes || location.nodes; +-- END IF; +-- END LOOP; +-- END LOOP; +-- +-- RETURN QUERY select * from planet_osm_ways where nodes && searchnodes and NOT ARRAY[id] <@ way_ids; +--END; +--$$ +--LANGUAGE plpgsql IMMUTABLE; CREATE OR REPLACE FUNCTION get_address_postcode(for_place_id BIGINT) RETURNS TEXT AS $$ @@ -2457,7 +2503,7 @@ BEGIN CASE WHEN class = 'place' and type = 'postcode' THEN hstore('name', postcode) ELSE name END as name, class, type, admin_level, true as fromarea, true as isaddress, CASE WHEN rank_address = 0 THEN 100 WHEN rank_address = 11 THEN 5 ELSE rank_address END as rank_address, - 0 as distance, calculated_country_code + 0 as distance, calculated_country_code, postcode from placex where place_id = for_place_id LOOP @@ -2468,6 +2514,9 @@ BEGIN IF searchpostcode IS NOT NULL and location.type = 'postcode' THEN location.isaddress := FALSE; END IF; + IF searchpostcode IS NULL 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; @@ -2488,9 +2537,11 @@ BEGIN FOR location IN select placex.place_id, osm_type, osm_id, CASE WHEN class = 'place' and type = 'postcode' THEN hstore('name', postcode) ELSE name END as name, - class, type, admin_level, fromarea, isaddress, + CASE WHEN extratags ? 'place' THEN 'place' ELSE class END as class, + CASE WHEN extratags ? 'place' THEN extratags->'place' ELSE type END as 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) @@ -2505,6 +2556,9 @@ 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; @@ -2801,7 +2855,7 @@ DECLARE BEGIN place_centroid := ST_Centroid(pointgeo); - out_partition := get_partition(place_centroid, in_countrycode); + out_partition := get_partition(in_countrycode); out_parent_place_id := null; address_street_word_id := get_name_id(make_standard_name(in_street)); @@ -2916,14 +2970,14 @@ BEGIN wiki_article := regexp_replace(wiki_article,E'^(.*?)([a-z]{2,3}).wikipedia.org/wiki/',E'\\2:'); wiki_article := regexp_replace(wiki_article,E'^(.*?)([a-z]{2,3}).wikipedia.org/w/index.php\\?title=',E'\\2:'); wiki_article := regexp_replace(wiki_article,E'^(.*?)/([a-z]{2,3})/wiki/',E'\\2:'); - wiki_article := regexp_replace(wiki_article,E'^(.*?)([a-z]{2,3})[=:]',E'\\2:'); + --wiki_article := regexp_replace(wiki_article,E'^(.*?)([a-z]{2,3})[=:]',E'\\2:'); wiki_article := replace(wiki_article,' ','_'); - wiki_article_title := trim(split_part(wiki_article, ':', 2)); - IF wiki_article_title IS NULL OR wiki_article_title = '' THEN + IF strpos(wiki_article, ':') IN (3,4) THEN + wiki_article_language := lower(trim(split_part(wiki_article, ':', 1))); + wiki_article_title := trim(substr(wiki_article, strpos(wiki_article, ':')+1)); + ELSE wiki_article_title := trim(wiki_article); wiki_article_language := CASE WHEN langs[i] = 'english' THEN 'en' WHEN langs[i] = 'country' THEN get_country_language_code(country_code) ELSE langs[i] END; - ELSE - wiki_article_language := lower(trim(split_part(wiki_article, ':', 1))); END IF; select wikipedia_article.language,wikipedia_article.title,wikipedia_article.importance @@ -3075,6 +3129,7 @@ DECLARE diameter FLOAT; rank INTEGER; BEGIN + UPDATE placex SET indexed_status = 2 WHERE place_id = placeid; SELECT geometry, rank_search FROM placex WHERE place_id = placeid INTO placegeom, rank; IF placegeom IS NOT NULL AND ST_IsValid(placegeom) THEN IF ST_GeometryType(placegeom) in ('ST_Polygon','ST_MultiPolygon') THEN