X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/62b1fba68aa91a3e9fb65c85141d8ba55ce8f59e..3bb6ecdc3bcdfa9e8ae4e53eb8f7a10bced3b909:/sql/functions.sql?ds=inline diff --git a/sql/functions.sql b/sql/functions.sql index 8593a3b1..07fadba7 100644 --- a/sql/functions.sql +++ b/sql/functions.sql @@ -1,49 +1,14 @@ ---DROP TRIGGER IF EXISTS place_before_insert on placex; ---DROP TRIGGER IF EXISTS place_before_update on placex; ---CREATE TYPE addresscalculationtype AS ( --- word text, --- score integer ---); - -CREATE OR REPLACE FUNCTION getclasstypekey(c text, t text) RETURNS TEXT +-- Splits the line at the given point and returns the two parts +-- in a multilinestring. +CREATE OR REPLACE FUNCTION split_line_on_node(line GEOMETRY, point GEOMETRY) +RETURNS GEOMETRY AS $$ -DECLARE -BEGIN - RETURN c||'|'||t; -END; -$$ -LANGUAGE plpgsql IMMUTABLE; - -CREATE OR REPLACE FUNCTION isbrokengeometry(place geometry) RETURNS BOOLEAN - AS $$ -DECLARE - NEWgeometry geometry; BEGIN - NEWgeometry := place; - IF ST_IsEmpty(NEWgeometry) OR NOT ST_IsValid(NEWgeometry) OR ST_X(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') THEN - RETURN true; - END IF; - RETURN false; + RETURN ST_Split(ST_Snap(line, point, 0.0005), point); END; $$ -LANGUAGE plpgsql IMMUTABLE; +LANGUAGE plpgsql; -CREATE OR REPLACE FUNCTION clean_geometry(place geometry) RETURNS geometry - AS $$ -DECLARE - NEWgeometry geometry; -BEGIN - NEWgeometry := place; - IF ST_X(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') THEN - NEWgeometry := ST_buffer(NEWgeometry,0); - IF ST_X(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') THEN - RETURN ST_SetSRID(ST_Point(0,0),4326); - END IF; - END IF; - RETURN NEWgeometry; -END; -$$ -LANGUAGE plpgsql IMMUTABLE; CREATE OR REPLACE FUNCTION geometry_sector(partition INTEGER, place geometry) RETURNS INTEGER AS $$ @@ -52,12 +17,6 @@ DECLARE BEGIN -- RAISE WARNING '%',place; NEWgeometry := ST_PointOnSurface(place); --- IF ST_IsEmpty(NEWgeometry) OR NOT ST_IsValid(NEWgeometry) OR ST_X(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') THEN --- NEWgeometry := ST_buffer(NEWgeometry,0); --- IF ST_IsEmpty(NEWgeometry) OR NOT ST_IsValid(NEWgeometry) OR ST_X(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') THEN --- RETURN 0; --- END IF; --- END IF; RETURN (partition*1000000) + (500-ST_X(NEWgeometry)::integer)*1000 + (500-ST_Y(NEWgeometry)::integer); END; $$ @@ -124,25 +83,27 @@ END; $$ LANGUAGE plpgsql; -CREATE OR REPLACE FUNCTION getorcreate_country(lookup_word TEXT, lookup_country_code varchar(2)) +CREATE OR REPLACE FUNCTION getorcreate_postcode_id(postcode TEXT) RETURNS INTEGER AS $$ DECLARE lookup_token TEXT; + lookup_word TEXT; return_word_id INTEGER; BEGIN - lookup_token := ' '||trim(lookup_word); - SELECT min(word_id) FROM word WHERE word_token = lookup_token and country_code=lookup_country_code into return_word_id; + lookup_word := upper(trim(postcode)); + lookup_token := ' ' || make_standard_name(lookup_word); + SELECT min(word_id) FROM word WHERE word_token = lookup_token and class='place' and type='postcode' 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, lookup_country_code, 0); + INSERT INTO word VALUES (return_word_id, lookup_token, lookup_word, 'place', 'postcode', null, 0); END IF; RETURN return_word_id; END; $$ LANGUAGE plpgsql; -CREATE OR REPLACE FUNCTION getorcreate_amenity(lookup_word TEXT, lookup_class text, lookup_type text) +CREATE OR REPLACE FUNCTION getorcreate_country(lookup_word TEXT, lookup_country_code varchar(2)) RETURNS INTEGER AS $$ DECLARE @@ -150,49 +111,35 @@ DECLARE return_word_id INTEGER; BEGIN lookup_token := ' '||trim(lookup_word); - SELECT min(word_id) FROM word WHERE word_token = lookup_token and class=lookup_class and type = lookup_type into return_word_id; + 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, lookup_class, lookup_type, null, 0); + INSERT INTO word VALUES (return_word_id, lookup_token, null, null, null, lookup_country_code, 0); END IF; RETURN return_word_id; END; $$ LANGUAGE plpgsql; -CREATE OR REPLACE FUNCTION getorcreate_tagpair(lookup_class text, lookup_type text) +CREATE OR REPLACE FUNCTION getorcreate_amenity(lookup_word TEXT, normalized_word TEXT, lookup_class text, lookup_type text) RETURNS INTEGER AS $$ DECLARE lookup_token TEXT; return_word_id INTEGER; BEGIN - lookup_token := lookup_class||'='||lookup_type; - SELECT min(word_id) FROM word WHERE word_token = lookup_token into return_word_id; + lookup_token := ' '||trim(lookup_word); + SELECT min(word_id) FROM word WHERE word_token = lookup_token and word=normalized_word 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, null, null, 0); + INSERT INTO word VALUES (return_word_id, lookup_token, normalized_word, lookup_class, lookup_type, null, 0); END IF; RETURN return_word_id; END; $$ LANGUAGE plpgsql; -CREATE OR REPLACE FUNCTION get_tagpair(lookup_class text, lookup_type text) - RETURNS INTEGER - AS $$ -DECLARE - lookup_token TEXT; - return_word_id INTEGER; -BEGIN - lookup_token := lookup_class||'='||lookup_type; - SELECT min(word_id) FROM word WHERE word_token = lookup_token into return_word_id; - RETURN return_word_id; -END; -$$ -LANGUAGE plpgsql; - -CREATE OR REPLACE FUNCTION getorcreate_amenityoperator(lookup_word TEXT, lookup_class text, lookup_type text, op text) +CREATE OR REPLACE FUNCTION getorcreate_amenityoperator(lookup_word TEXT, normalized_word TEXT, lookup_class text, lookup_type text, op text) RETURNS INTEGER AS $$ DECLARE @@ -200,10 +147,10 @@ DECLARE return_word_id INTEGER; BEGIN lookup_token := ' '||trim(lookup_word); - 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; + SELECT min(word_id) FROM word WHERE word_token = lookup_token and word=normalized_word 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, lookup_class, lookup_type, null, 0, op); + INSERT INTO word VALUES (return_word_id, lookup_token, normalized_word, lookup_class, lookup_type, null, 0, op); END IF; RETURN return_word_id; END; @@ -271,6 +218,20 @@ 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; +$$ +LANGUAGE plpgsql IMMUTABLE; + CREATE OR REPLACE FUNCTION array_merge(a INTEGER[], b INTEGER[]) RETURNS INTEGER[] AS $$ @@ -295,6 +256,99 @@ END; $$ LANGUAGE plpgsql IMMUTABLE; + +CREATE OR REPLACE FUNCTION get_postcode_rank(country_code VARCHAR(2), postcode TEXT, + OUT rank_search SMALLINT, OUT rank_address SMALLINT) +AS $$ +DECLARE + part TEXT; +BEGIN + rank_search := 30; + rank_address := 30; + postcode := upper(postcode); + + IF country_code = 'gb' THEN + IF postcode ~ '^([A-Z][A-Z]?[0-9][0-9A-Z]? [0-9][A-Z][A-Z])$' THEN + rank_search := 25; + rank_address := 5; + ELSEIF postcode ~ '^([A-Z][A-Z]?[0-9][0-9A-Z]? [0-9])$' THEN + rank_search := 23; + rank_address := 5; + ELSEIF postcode ~ '^([A-Z][A-Z]?[0-9][0-9A-Z])$' THEN + rank_search := 21; + rank_address := 5; + END IF; + + ELSEIF country_code = 'sg' THEN + IF postcode ~ '^([0-9]{6})$' THEN + rank_search := 25; + rank_address := 11; + END IF; + + ELSEIF country_code = 'de' THEN + IF postcode ~ '^([0-9]{5})$' THEN + rank_search := 21; + rank_address := 11; + END IF; + + ELSE + -- Guess at the postcode format and coverage (!) + IF postcode ~ '^[A-Z0-9]{1,5}$' THEN -- Probably too short to be very local + rank_search := 21; + rank_address := 11; + ELSE + -- Does it look splitable into and area and local code? + part := substring(postcode from '^([- :A-Z0-9]+)([- :][A-Z0-9]+)$'); + + IF part IS NOT NULL THEN + rank_search := 25; + rank_address := 11; + ELSEIF postcode ~ '^[- :A-Z0-9]{6,}$' THEN + rank_search := 21; + rank_address := 11; + END IF; + END IF; + END IF; + +END; +$$ +LANGUAGE plpgsql IMMUTABLE; + +-- Find the nearest artificial postcode for the given geometry. +-- TODO For areas there should not be more than two inside the geometry. +CREATE OR REPLACE FUNCTION get_nearest_postcode(country VARCHAR(2), geom GEOMETRY) RETURNS TEXT + AS $$ +DECLARE + outcode TEXT; + cnt INTEGER; +BEGIN + -- If the geometry is an area then only one postcode must be within + -- that area, otherwise consider the area as not having a postcode. + IF ST_GeometryType(geom) in ('ST_Polygon','ST_MultiPolygon') THEN + SELECT min(postcode), count(*) FROM + (SELECT postcode FROM location_postcode + WHERE ST_Contains(geom, location_postcode.geometry) LIMIT 2) sub + INTO outcode, cnt; + + IF cnt = 1 THEN + RETURN outcode; + ELSE + RETURN null; + END IF; + END IF; + + SELECT postcode FROM location_postcode + WHERE ST_DWithin(geom, location_postcode.geometry, 0.05) + AND location_postcode.country_code = country + ORDER BY ST_Distance(geom, location_postcode.geometry) LIMIT 1 + INTO outcode; + + RETURN outcode; +END; +$$ +LANGUAGE plpgsql; + + CREATE OR REPLACE FUNCTION create_country(src HSTORE, lookup_country_code varchar(2)) RETURNS VOID AS $$ DECLARE @@ -347,7 +401,7 @@ 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 +469,7 @@ 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; @@ -460,31 +514,6 @@ END; $$ LANGUAGE plpgsql IMMUTABLE; -CREATE OR REPLACE FUNCTION get_word_score(wordscores wordscore[], words text[]) RETURNS integer - AS $$ -DECLARE - idxword integer; - idxscores integer; - result integer; -BEGIN - IF (wordscores is null OR words is null) THEN - RETURN 0; - END IF; - - result := 0; - FOR idxword in 1 .. array_upper(words, 1) LOOP - FOR idxscores in 1 .. array_upper(wordscores, 1) LOOP - IF wordscores[idxscores].word = words[idxword] THEN - result := result + wordscores[idxscores].score; - END IF; - END LOOP; - END LOOP; - - RETURN result; -END; -$$ -LANGUAGE plpgsql IMMUTABLE; - CREATE OR REPLACE FUNCTION get_country_code(place geometry) RETURNS TEXT AS $$ DECLARE @@ -493,7 +522,7 @@ DECLARE BEGIN place_centre := ST_PointOnSurface(place); ---DEBUG: RAISE WARNING 'get_country_code, start: %', ST_AsText(place_centre); +-- 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 @@ -501,7 +530,7 @@ BEGIN RETURN nearcountry.country_code; END LOOP; ---DEBUG: RAISE WARNING 'osm fallback: %', ST_AsText(place_centre); +-- RAISE WARNING 'osm fallback: %', ST_AsText(place_centre); -- Try for OSM fallback data -- The order is to deal with places like HongKong that are 'states' within another polygon @@ -510,7 +539,7 @@ BEGIN RETURN nearcountry.country_code; END LOOP; ---DEBUG: RAISE WARNING 'natural earth: %', ST_AsText(place_centre); +-- RAISE WARNING 'natural earth: %', ST_AsText(place_centre); -- Natural earth data FOR nearcountry IN select country_code from country_naturalearthdata where st_covers(geometry, place_centre) limit 1 @@ -518,7 +547,7 @@ BEGIN RETURN nearcountry.country_code; END LOOP; ---DEBUG: RAISE WARNING 'near osm fallback: %', ST_AsText(place_centre); +-- RAISE WARNING 'near osm fallback: %', ST_AsText(place_centre); -- FOR nearcountry IN select country_code from country_osm_grid where st_dwithin(geometry, place_centre, 0.5) order by st_distance(geometry, place_centre) asc, area asc limit 1 @@ -526,7 +555,7 @@ BEGIN RETURN nearcountry.country_code; END LOOP; ---DEBUG: RAISE WARNING 'near natural earth: %', ST_AsText(place_centre); +-- RAISE WARNING 'near natural earth: %', ST_AsText(place_centre); -- Natural earth data FOR nearcountry IN select country_code from country_naturalearthdata where st_dwithin(geometry, place_centre, 0.5) limit 1 @@ -534,40 +563,11 @@ BEGIN RETURN nearcountry.country_code; END LOOP; - -- WorldBoundaries data (second fallback - think there might be something broken in this data) --- FOR nearcountry IN select country_code from country where st_covers(geometry, place_centre) limit 1 --- LOOP --- RETURN nearcountry.country_code; --- END LOOP; - ---RAISE WARNING 'near country: %', ST_AsText(place_centre); - - -- Still not in a country - try nearest within ~12 miles of a country --- FOR nearcountry IN select country_code from country where st_distance(geometry, place_centre) < 0.5 --- order by st_distance(geometry, place) limit 1 --- LOOP --- RETURN nearcountry.country_code; --- END LOOP; - RETURN NULL; 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 @@ -629,39 +628,41 @@ CREATE OR REPLACE FUNCTION add_location( keywords INTEGER[], rank_search INTEGER, rank_address INTEGER, + in_postcode TEXT, geometry GEOMETRY ) RETURNS BOOLEAN AS $$ DECLARE locationid INTEGER; - isarea BOOLEAN; centroid GEOMETRY; diameter FLOAT; x BOOLEAN; splitGeom RECORD; secgeo GEOMETRY; + postcode TEXT; BEGIN IF rank_search > 25 THEN RAISE EXCEPTION 'Adding location with rank > 25 (% rank %)', place_id, rank_search; END IF; --- RAISE WARNING 'Adding location with rank > 25 (% rank %)', place_id, rank_search; - x := deleteLocationArea(partition, place_id, rank_search); - isarea := false; - IF (ST_GeometryType(geometry) in ('ST_Polygon','ST_MultiPolygon') AND ST_IsValid(geometry)) THEN + -- add postcode only if it contains a single entry, i.e. ignore postcode lists + postcode := NULL; + IF in_postcode is not null AND in_postcode not similar to '%(,|;)%' THEN + postcode := upper(trim (in_postcode)); + END IF; - isArea := true; + IF ST_GeometryType(geometry) in ('ST_Polygon','ST_MultiPolygon') THEN centroid := ST_Centroid(geometry); FOR secgeo IN select split_geometry(geometry) AS geom LOOP - x := insertLocationAreaLarge(partition, place_id, country_code, keywords, rank_search, rank_address, false, centroid, secgeo); + x := insertLocationAreaLarge(partition, place_id, country_code, keywords, rank_search, rank_address, false, postcode, centroid, secgeo); END LOOP; - ELSEIF rank_search < 26 THEN + ELSE diameter := 0.02; IF rank_address = 0 THEN @@ -683,17 +684,7 @@ BEGIN -- RAISE WARNING 'adding % diameter %', place_id, diameter; secgeo := ST_Buffer(geometry, diameter); - x := insertLocationAreaLarge(partition, place_id, country_code, keywords, rank_search, rank_address, true, ST_Centroid(geometry), secgeo); - - ELSE - - -- ~ 20meters - secgeo := ST_Buffer(geometry, 0.0002); - x := insertLocationAreaRoadNear(partition, place_id, country_code, keywords, rank_search, rank_address, true, ST_Centroid(geometry), secgeo); - - -- ~ 100meters - secgeo := ST_Buffer(geometry, 0.001); - x := insertLocationAreaRoadFar(partition, place_id, country_code, keywords, rank_search, rank_address, true, ST_Centroid(geometry), secgeo); + x := insertLocationAreaLarge(partition, place_id, country_code, keywords, rank_search, rank_address, true, postcode, ST_Centroid(geometry), secgeo); END IF; @@ -702,228 +693,102 @@ END; $$ LANGUAGE plpgsql; -CREATE OR REPLACE FUNCTION update_location( - partition INTEGER, - place_id BIGINT, - place_country_code varchar(2), - name hstore, - rank_search INTEGER, - rank_address INTEGER, - geometry GEOMETRY - ) - RETURNS BOOLEAN - AS $$ -DECLARE - b BOOLEAN; -BEGIN - b := deleteLocationArea(partition, place_id, rank_search); --- result := add_location(NEW.place_id, NEW.country_code, NEW.partition, name_vector, NEW.rank_search, NEW.rank_address, NEW.geometry); - RETURN add_location(place_id, place_country_code, name, rank_search, rank_address, geometry); -END; -$$ -LANGUAGE plpgsql; -CREATE OR REPLACE FUNCTION search_name_add_words(parent_place_id BIGINT, to_add INTEGER[]) - RETURNS BOOLEAN - AS $$ +-- find the parent road of the cut road parts +CREATE OR REPLACE FUNCTION get_interpolation_parent(wayid BIGINT, street TEXT, place TEXT, + partition INTEGER, centroid GEOMETRY, geom GEOMETRY) +RETURNS BIGINT AS $$ DECLARE - childplace RECORD; + addr_street TEXT; + addr_place TEXT; + parent_place_id BIGINT; + address_street_word_ids INTEGER[]; + + waynodes BIGINT[]; + + location RECORD; BEGIN + addr_street = street; + addr_place = place; - IF #to_add = 0 THEN - RETURN true; + IF addr_street is null and addr_place is null THEN + select nodes from planet_osm_ways where id = wayid INTO waynodes; + FOR location IN SELECT placex.address from placex + where osm_type = 'N' and osm_id = ANY(waynodes) + and placex.address is not null + and (placex.address ? 'street' or placex.address ? 'place') + and indexed_status < 100 + limit 1 LOOP + addr_street = location.address->'street'; + addr_place = location.address->'place'; + END LOOP; END IF; - -- this should just be an update, but it seems to do insane things to the index size (delete and insert doesn't) - FOR childplace IN select * from search_name,place_addressline - where address_place_id = parent_place_id - and search_name.place_id = place_addressline.place_id - LOOP - delete from search_name where place_id = childplace.place_id; - IF not (ARRAY[to_add] <@ childplace.nameaddress_vector) THEN - childplace.nameaddress_vector := childplace.nameaddress_vector || to_add; - END IF; - IF childplace.place_id = parent_place_id and not (ARRAY[to_add] <@ childplace.name_vector) THEN - childplace.name_vector := childplace.name_vector || to_add; + IF addr_street IS NOT NULL THEN + address_street_word_ids := get_name_ids(make_standard_name(addr_street)); + IF address_street_word_ids IS NOT NULL THEN + FOR location IN SELECT place_id from getNearestNamedRoadFeature(partition, centroid, address_street_word_ids) LOOP + parent_place_id := location.place_id; + END LOOP; END IF; - insert into search_name (place_id, search_rank, address_rank, country_code, name_vector, nameaddress_vector, centroid) - values (childplace.place_id, childplace.search_rank, childplace.address_rank, childplace.country_code, - childplace.name_vector, childplace.nameaddress_vector, childplace.centroid); - END LOOP; - - RETURN true; -END; -$$ -LANGUAGE plpgsql; - -CREATE OR REPLACE FUNCTION update_location_nameonly(partition INTEGER, OLD_place_id BIGINT, name hstore) RETURNS BOOLEAN - AS $$ -DECLARE - newkeywords INTEGER[]; - addedkeywords INTEGER[]; - removedkeywords INTEGER[]; -BEGIN - - -- what has changed? - newkeywords := make_keywords(name); - select coalesce(newkeywords,'{}'::INTEGER[]) - coalesce(location_point.keywords,'{}'::INTEGER[]), - coalesce(location_point.keywords,'{}'::INTEGER[]) - coalesce(newkeywords,'{}'::INTEGER[]) from location_point - where place_id = OLD_place_id into addedkeywords, removedkeywords; + END IF; --- RAISE WARNING 'update_location_nameonly for %: new:% added:% removed:%', OLD_place_id, newkeywords, addedkeywords, removedkeywords; + IF parent_place_id IS NULL AND addr_place IS NOT NULL THEN + address_street_word_ids := get_name_ids(make_standard_name(addr_place)); + IF address_street_word_ids IS NOT NULL THEN + FOR location IN SELECT place_id from getNearestNamedPlaceFeature(partition, centroid, address_street_word_ids) LOOP + parent_place_id := location.place_id; + END LOOP; + END IF; + END IF; - IF #removedkeywords > 0 THEN - -- abort due to tokens removed - RETURN false; + IF parent_place_id is null THEN + FOR location IN SELECT place_id FROM placex + WHERE ST_DWithin(geom, placex.geometry, 0.001) and placex.rank_search = 26 + ORDER BY (ST_distance(placex.geometry, ST_LineInterpolatePoint(geom,0))+ + ST_distance(placex.geometry, ST_LineInterpolatePoint(geom,0.5))+ + ST_distance(placex.geometry, ST_LineInterpolatePoint(geom,1))) ASC limit 1 + LOOP + parent_place_id := location.place_id; + END LOOP; END IF; - - IF #addedkeywords > 0 THEN - -- short circuit - no changes - RETURN true; + + IF parent_place_id is null THEN + RETURN 0; END IF; - UPDATE location_area set keywords = newkeywords where place_id = OLD_place_id; - RETURN search_name_add_words(OLD_place_id, addedkeywords); + RETURN parent_place_id; END; $$ LANGUAGE plpgsql; -CREATE OR REPLACE FUNCTION create_interpolation(wayid BIGINT, interpolationtype TEXT) RETURNS INTEGER +CREATE OR REPLACE FUNCTION osmline_insert() RETURNS TRIGGER AS $$ -DECLARE - - newpoints INTEGER; - waynodes BIGINT[]; - nodeid BIGINT; - prevnode RECORD; - nextnode RECORD; - startnumber INTEGER; - endnumber INTEGER; - stepsize INTEGER; - orginalstartnumber INTEGER; - originalnumberrange INTEGER; - housenum INTEGER; - linegeo 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; - ---RAISE WARNING 'interpolation node % % % ',nextnode.housenumber,ST_X(nextnode.geometry),ST_Y(nextnode.geometry); - - IF havefirstpoint THEN - - -- 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 startnumber IS NOT NULL and startnumber > 0 AND endnumber IS NOT NULL and endnumber > 0 AND @(startnumber - endnumber) < 1000 THEN - ---RAISE WARNING 'interpolation end % % ',nextnode.place_id,endnumber; + NEW.place_id := nextval('seq_place'); + NEW.indexed_date := now(); - IF startnumber != endnumber THEN + IF NEW.indexed_status IS NULL THEN + IF NEW.address is NULL OR NOT NEW.address ? 'interpolation' + OR NEW.address->'interpolation' NOT IN ('odd', 'even', 'all') THEN + -- other interpolation types than odd/even/all (e.g. numeric ones) are not supported + RETURN NULL; + END IF; - 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; - END IF; - END IF; + NEW.indexed_status := 1; --STATUS_NEW + NEW.country_code := lower(get_country_code(NEW.linegeo)); - 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; - END IF; ---RAISE WARNING 'interpolation start % % ',nextnode.place_id,startnumber; - END IF; - END LOOP; - END IF; + NEW.partition := get_partition(NEW.country_code); + NEW.geometry_sector := geometry_sector(NEW.partition, NEW.linegeo); END IF; ---RAISE WARNING 'interpolation points % ',newpoints; - - RETURN newpoints; + RETURN NEW; END; $$ LANGUAGE plpgsql; + CREATE OR REPLACE FUNCTION placex_insert() RETURNS TRIGGER AS $$ DECLARE @@ -934,49 +799,21 @@ DECLARE default_language VARCHAR(10); diameter FLOAT; classtable TEXT; + line RECORD; 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; - END IF; - IF NEW.class in ('landuse','natural') and NEW.name is null THEN --- RAISE WARNING 'empty landuse %',NEW.osm_id; - RETURN null; - END IF; - - IF ST_IsEmpty(NEW.geometry) OR NOT ST_IsValid(NEW.geometry) OR ST_X(ST_Centroid(NEW.geometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEW.geometry))::text in ('NaN','Infinity','-Infinity') THEN - -- block all invalid geometary - just not worth the risk. seg faults are causing serious problems. - RAISE WARNING 'invalid geometry %',NEW.osm_id; - RETURN NULL; - - -- Dead code - IF NEW.osm_type = 'R' THEN - -- invalid multipolygons can crash postgis, don't even bother to try! - RETURN NULL; - END IF; - NEW.geometry := ST_buffer(NEW.geometry,0); - IF ST_IsEmpty(NEW.geometry) OR NOT ST_IsValid(NEW.geometry) OR ST_X(ST_Centroid(NEW.geometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEW.geometry))::text in ('NaN','Infinity','-Infinity') THEN - RAISE WARNING 'Invalid geometary, rejecting: % %', NEW.osm_type, NEW.osm_id; - RETURN NULL; - END IF; - END IF; - --DEBUG: RAISE WARNING '% % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type; 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.country_code := lower(get_country_code(NEW.geometry)); - NEW.partition := get_partition(NEW.geometry, NEW.calculated_country_code); + NEW.partition := get_partition(NEW.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) 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); + default_language := get_country_language_code(NEW.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')); @@ -986,14 +823,6 @@ BEGIN END IF; END IF; - IF NEW.admin_level > 15 THEN - NEW.admin_level := 15; - END IF; - - IF NEW.housenumber IS NOT NULL THEN - i := getorcreate_housenumber_id(make_standard_name(NEW.housenumber)); - END IF; - IF NEW.osm_type = 'X' THEN -- E'X'ternal records should already be in the right format so do nothing ELSE @@ -1003,61 +832,25 @@ BEGIN -- By doing in postgres we have the country available to us - currently only used for postcode IF NEW.class in ('place','boundary') AND NEW.type in ('postcode','postal_code') THEN - IF NEW.postcode IS NULL THEN + IF NEW.address IS NULL OR NOT NEW.address ? 'postcode' THEN -- most likely just a part of a multipolygon postcode boundary, throw it away RETURN NULL; END IF; - NEW.name := hstore('ref', NEW.postcode); - - IF NEW.calculated_country_code = 'gb' THEN - - IF NEW.postcode ~ '^([A-Z][A-Z]?[0-9][0-9A-Z]? [0-9][A-Z][A-Z])$' THEN - NEW.rank_search := 25; - NEW.rank_address := 5; - ELSEIF NEW.postcode ~ '^([A-Z][A-Z]?[0-9][0-9A-Z]? [0-9])$' THEN - NEW.rank_search := 23; - NEW.rank_address := 5; - ELSEIF NEW.postcode ~ '^([A-Z][A-Z]?[0-9][0-9A-Z])$' THEN - NEW.rank_search := 21; - NEW.rank_address := 5; - END IF; - - ELSEIF NEW.calculated_country_code = 'de' THEN - - IF NEW.postcode ~ '^([0-9]{5})$' THEN - NEW.rank_search := 21; - NEW.rank_address := 11; - END IF; + NEW.name := hstore('ref', NEW.address->'postcode'); - ELSE - -- Guess at the postcode format and coverage (!) - IF upper(NEW.postcode) ~ '^[A-Z0-9]{1,5}$' THEN -- Probably too short to be very local - NEW.rank_search := 21; - NEW.rank_address := 11; - ELSE - -- Does it look splitable into and area and local code? - postcode := substring(upper(NEW.postcode) from '^([- :A-Z0-9]+)([- :][A-Z0-9]+)$'); - - IF postcode IS NOT NULL THEN - NEW.rank_search := 25; - NEW.rank_address := 11; - ELSEIF NEW.postcode ~ '^[- :A-Z0-9]{6,}$' THEN - NEW.rank_search := 21; - NEW.rank_address := 11; - END IF; - END IF; - END IF; + SELECT * FROM get_postcode_rank(NEW.country_code, NEW.address->'postcode') + INTO NEW.rank_search, NEW.rank_address; ELSEIF NEW.class = 'place' THEN IF NEW.type in ('continent') THEN NEW.rank_search := 2; NEW.rank_address := NEW.rank_search; - NEW.calculated_country_code := NULL; + NEW.country_code := NULL; ELSEIF NEW.type in ('sea') THEN NEW.rank_search := 2; NEW.rank_address := 0; - NEW.calculated_country_code := NULL; + NEW.country_code := NULL; ELSEIF NEW.type in ('country') THEN NEW.rank_search := 4; NEW.rank_address := NEW.rank_search; @@ -1082,18 +875,6 @@ 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; @@ -1104,19 +885,14 @@ 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; ELSEIF NEW.type in ('houses') THEN -- can't guarantee all required nodes loaded yet due to caching in osm2pgsql - -- insert new point into place for each derived building - --i := create_interpolation(NEW.osm_id, NEW.housenumber); NEW.rank_search := 28; NEW.rank_address := 0; END IF; @@ -1127,38 +903,46 @@ 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; - ELSEIF NEW.class = 'natural' and NEW.type in ('peak','volcano','mountain_range') THEN - NEW.rank_search := 18; + 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 = 'leisure' and NEW.type in ('park') THEN + NEW.rank_search := 24; NEW.rank_address := 0; + 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; -- 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; - NEW.rank_address := NEW.rank_search; - ELSEIF NEW.class = 'highway' AND NEW.name is NULL AND - NEW.type in ('service','cycleway','path','footway','steps','bridleway','track','byway','motorway_link','primary_link','trunk_link','secondary_link','tertiary_link') THEN --- RAISE WARNING 'unnamed minor feature %',NEW.osm_id; - RETURN NULL; + NEW.rank_address := 0; ELSEIF NEW.class = 'railway' AND NEW.type in ('rail') THEN RETURN NULL; - 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; @@ -1179,8 +963,8 @@ BEGIN END IF; -- a country code make no sense below rank 4 (country) - IF NEW.rank_address < 4 THEN - NEW.calculated_country_code := NULL; + IF NEW.rank_search < 4 THEN + NEW.country_code := NULL; END IF; -- Block import below rank 22 @@ -1190,79 +974,248 @@ BEGIN --DEBUG: RAISE WARNING 'placex_insert:END: % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type; - RETURN NEW; -- @DIFFUPDATES@ The following is not needed until doing diff updates, and slows the main index process down + 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; + 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; - -- 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 + -- 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 address ? 'place')); + 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 address ? 'place')); + 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); + -- reparenting also for OSM Interpolation Lines (and for Tiger?) + update location_property_osmline set indexed_status = 2 where indexed_status = 0 and ST_DWithin(location_property_osmline.linegeo, 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 address ? 'place'); + 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); END IF; + RETURN NEW; --- IF NEW.rank_search < 26 THEN --- RAISE WARNING 'placex insert: % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type; --- END IF; +END; +$$ +LANGUAGE plpgsql; + +CREATE OR REPLACE FUNCTION osmline_update() RETURNS +TRIGGER + AS $$ +DECLARE + place_centroid GEOMETRY; + waynodes BIGINT[]; + prevnode RECORD; + nextnode RECORD; + startnumber INTEGER; + endnumber INTEGER; + housenum INTEGER; + linegeo GEOMETRY; + splitline GEOMETRY; + sectiongeo GEOMETRY; + interpol_postcode TEXT; + postcode TEXT; +BEGIN + -- deferred delete + IF OLD.indexed_status = 100 THEN + delete from location_property_osmline where place_id = OLD.place_id; + RETURN NULL; + END IF; + + IF NEW.indexed_status != 0 OR OLD.indexed_status = 0 THEN + RETURN NEW; + END IF; + + NEW.interpolationtype = NEW.address->'interpolation'; + + place_centroid := ST_PointOnSurface(NEW.linegeo); + NEW.parent_place_id = get_interpolation_parent(NEW.osm_id, NEW.address->'street', + NEW.address->'place', + NEW.partition, place_centroid, NEW.linegeo); + + IF NEW.address is not NULL AND NEW.address ? 'postcode' AND NEW.address->'postcode' not similar to '%(,|;)%' THEN + interpol_postcode := NEW.address->'postcode'; + housenum := getorcreate_postcode_id(NEW.address->'postcode'); + ELSE + interpol_postcode := NULL; + END IF; + + -- if the line was newly inserted, split the line as necessary + IF OLD.indexed_status = 1 THEN + select nodes from planet_osm_ways where id = NEW.osm_id INTO waynodes; + + IF array_upper(waynodes, 1) IS NULL THEN + RETURN NEW; + END IF; + + linegeo := NEW.linegeo; + startnumber := NULL; + + FOR nodeidpos in 1..array_upper(waynodes, 1) LOOP + + select osm_id, address, geometry + from place where osm_type = 'N' and osm_id = waynodes[nodeidpos]::BIGINT + and address is not NULL and address ? 'housenumber' limit 1 INTO nextnode; + --RAISE NOTICE 'Nextnode.place_id: %s', nextnode.place_id; + IF nextnode.osm_id IS NOT NULL THEN + --RAISE NOTICE 'place_id is not null'; + IF nodeidpos > 1 and 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); + linegeo := ST_GeometryN(splitline, 2); + ELSE + sectiongeo = linegeo; + END IF; + endnumber := substring(nextnode.address->'housenumber','[0-9]+')::integer; + + IF startnumber IS NOT NULL AND endnumber IS NOT NULL + AND startnumber != endnumber + AND ST_GeometryType(sectiongeo) = 'ST_LineString' THEN + + IF (startnumber > endnumber) THEN + housenum := endnumber; + endnumber := startnumber; + startnumber := housenum; + sectiongeo := ST_Reverse(sectiongeo); + END IF; + + -- determine postcode + postcode := coalesce(interpol_postcode, + prevnode.address->'postcode', + nextnode.address->'postcode', + postcode); + + IF postcode is NULL THEN + SELECT placex.postcode FROM placex WHERE place_id = NEW.parent_place_id INTO postcode; + END IF; + IF postcode is NULL THEN + postcode := get_nearest_postcode(NEW.country_code, nextnode.geometry); + END IF; + + IF NEW.startnumber IS NULL THEN + NEW.startnumber := startnumber; + NEW.endnumber := endnumber; + NEW.linegeo := sectiongeo; + NEW.postcode := upper(trim(postcode)); + ELSE + insert into location_property_osmline + (linegeo, partition, osm_id, parent_place_id, + startnumber, endnumber, interpolationtype, + address, postcode, country_code, + geometry_sector, indexed_status) + values (sectiongeo, NEW.partition, NEW.osm_id, NEW.parent_place_id, + startnumber, endnumber, NEW.interpolationtype, + NEW.address, postcode, + NEW.country_code, NEW.geometry_sector, 0); + END IF; + END IF; + + -- early break if we are out of line string, + -- might happen when a line string loops back on itself + IF ST_GeometryType(linegeo) != 'ST_LineString' THEN + RETURN NEW; + END IF; + startnumber := substring(nextnode.address->'housenumber','[0-9]+')::integer; + prevnode := nextnode; + END IF; + END LOOP; + END IF; + + -- marking descendants for reparenting is not needed, because there are + -- actually no descendants for interpolation lines RETURN NEW; +END; +$$ +LANGUAGE plpgsql; + +-- Trigger for updates of location_postcode +-- +-- Computes the parent object the postcode most likely refers to. +-- This will be the place that determines the address displayed when +-- searching for this postcode. +CREATE OR REPLACE FUNCTION postcode_update() RETURNS +TRIGGER + AS $$ +DECLARE + partition SMALLINT; + location RECORD; +BEGIN + IF NEW.indexed_status != 0 OR OLD.indexed_status = 0 THEN + RETURN NEW; + END IF; + + NEW.indexed_date = now(); + + partition := get_partition(NEW.country_code); + + SELECT * FROM get_postcode_rank(NEW.country_code, NEW.postcode) + INTO NEW.rank_search, NEW.rank_address; + NEW.parent_place_id = 0; + FOR location IN + SELECT place_id + FROM getNearFeatures(partition, NEW.geometry, NEW.rank_search, '{}'::int[]) + WHERE NOT isguess ORDER BY rank_address DESC LIMIT 1 + LOOP + NEW.parent_place_id = location.place_id; + END LOOP; + + RETURN NEW; END; $$ LANGUAGE plpgsql; -CREATE OR REPLACE FUNCTION placex_update() RETURNS +CREATE OR REPLACE FUNCTION placex_update() RETURNS TRIGGER AS $$ DECLARE @@ -1272,8 +1225,6 @@ DECLARE search_maxdistance FLOAT[]; search_mindistance FLOAT[]; address_havelevel BOOLEAN[]; --- search_scores wordscore[]; --- search_scores_pos INTEGER; i INTEGER; iMax FLOAT; @@ -1283,13 +1234,18 @@ DECLARE relation_members TEXT[]; relMember RECORD; linkedplacex RECORD; + addr_item RECORD; search_diameter FLOAT; search_prevdiameter FLOAT; search_maxrank INTEGER; address_maxrank INTEGER; address_street_word_id INTEGER; + address_street_word_ids INTEGER[]; parent_place_id_rank BIGINT; - + + addr_street TEXT; + addr_place TEXT; + isin TEXT[]; isin_tokens INT[]; @@ -1297,21 +1253,21 @@ DECLARE location_distance FLOAT; location_parent GEOMETRY; location_isaddress BOOLEAN; - - tagpairid INTEGER; + location_keywords INTEGER[]; default_language TEXT; name_vector INTEGER[]; nameaddress_vector INTEGER[]; linked_node_id BIGINT; + linked_importance FLOAT; + linked_wikipedia TEXT; result BOOLEAN; BEGIN - -- deferred delete IF OLD.indexed_status = 100 THEN - --DEBUG: RAISE WARNING 'placex_update_delete % %',NEW.osm_type,NEW.osm_id; + --DEBUG: RAISE WARNING 'placex_update delete % %',NEW.osm_type,NEW.osm_id; delete from placex where place_id = OLD.place_id; RETURN NULL; END IF; @@ -1320,609 +1276,666 @@ BEGIN RETURN NEW; END IF; - --DEBUG: RAISE WARNING 'placex_update % %',NEW.osm_type,NEW.osm_id; + --DEBUG: RAISE WARNING 'placex_update % % (%)',NEW.osm_type,NEW.osm_id,NEW.place_id; + + NEW.indexed_date = now(); ---RAISE WARNING '%',NEW.place_id; ---RAISE WARNING '%', NEW; + result := deleteSearchName(NEW.partition, NEW.place_id); + DELETE FROM place_addressline 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, indexed_status = 2 + where linked_place_id = NEW.place_id; + -- update not necessary for osmline, cause linked_place_id does not exist - IF NEW.class = 'place' AND NEW.type = 'postcodearea' THEN - -- Silently do nothing + IF NEW.linked_place_id is not null THEN + --DEBUG: RAISE WARNING 'place already linked to %', NEW.linked_place_id; RETURN NEW; END IF; - -- TODO: this test is now redundant? - IF OLD.indexed_status != 0 THEN + --DEBUG: RAISE WARNING 'Copy over address tags'; + IF NEW.address is not NULL THEN + IF NEW.address ? 'conscriptionnumber' THEN + i := getorcreate_housenumber_id(make_standard_name(NEW.address->'conscriptionnumber')); + IF NEW.address ? 'streetnumber' THEN + i := getorcreate_housenumber_id(make_standard_name(NEW.address->'streetnumber')); + NEW.housenumber := (NEW.address->'conscriptionnumber') || '/' || (NEW.address->'streetnumber'); + ELSE + NEW.housenumber := NEW.address->'conscriptionnumber'; + END IF; + ELSEIF NEW.address ? 'streetnumber' THEN + NEW.housenumber := NEW.address->'streetnumber'; + i := getorcreate_housenumber_id(make_standard_name(NEW.address->'streetnumber')); + ELSEIF NEW.address ? 'housenumber' THEN + NEW.housenumber := NEW.address->'housenumber'; + i := getorcreate_housenumber_id(make_standard_name(NEW.housenumber)); + END IF; - NEW.indexed_date = now(); + addr_street := NEW.address->'street'; + addr_place := NEW.address->'place'; - 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.address ? 'postcode' and NEW.address->'postcode' not similar to '%(,|;)%' THEN + i := getorcreate_postcode_id(NEW.address->'postcode'); + END IF; + END IF; - IF NEW.linked_place_id is not null THEN - RETURN NEW; + -- Speed up searches - just use the centroid of the feature + -- cheaper but less acurate + place_centroid := ST_PointOnSurface(NEW.geometry); + NEW.centroid := null; + NEW.postcode := null; + --DEBUG: RAISE WARNING 'Computing preliminary centroid at %',ST_AsText(place_centroid); + + -- recalculate country and partition + IF NEW.rank_search = 4 AND NEW.address is not NULL AND NEW.address ? 'country' THEN + -- for countries, believe the mapped country code, + -- so that we remain in the right partition if the boundaries + -- suddenly expand. + NEW.country_code := lower(NEW.address->'country'); + NEW.partition := get_partition(lower(NEW.country_code)); + IF NEW.partition = 0 THEN + NEW.country_code := lower(get_country_code(place_centroid)); + NEW.partition := get_partition(NEW.country_code); END IF; + ELSE + IF NEW.rank_search >= 4 THEN + NEW.country_code := lower(get_country_code(place_centroid)); + ELSE + NEW.country_code := NULL; + END IF; + NEW.partition := get_partition(NEW.country_code); + END IF; + --DEBUG: RAISE WARNING 'Country updated: "%"', NEW.country_code; + + -- 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_members IN select members 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_members[i]; + FOR linked_node_id IN SELECT place_id FROM placex + WHERE osm_type = 'W' and osm_id = substring(relation_members[i],2,200)::bigint + and class = NEW.class and type in ('river', 'stream', 'canal', 'drain', 'ditch') + 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 = linked_node_id; + END LOOP; + END IF; + END LOOP; + END LOOP; + --DEBUG: RAISE WARNING 'Waterway processed'; + END IF; - IF NEW.class = 'place' AND NEW.type = 'houses' THEN - i := create_interpolation(NEW.osm_id, NEW.housenumber); - RETURN NEW; - END IF; + -- Adding ourselves to the list simplifies address calculations later + INSERT INTO place_addressline (place_id, address_place_id, fromarea, isaddress, distance, cached_rank_address) + VALUES (NEW.place_id, NEW.place_id, true, true, 0, NEW.rank_address); - -- Speed up searches - just use the centroid of the feature - -- cheaper but less acurate - place_centroid := ST_PointOnSurface(NEW.geometry); - NEW.centroid := null; + -- What level are we searching from + search_maxrank := NEW.rank_search; - -- 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)); - ELSE - NEW.calculated_country_code := NULL; - 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 - INSERT INTO place_addressline VALUES (NEW.place_id, NEW.place_id, true, true, 0, NEW.rank_address); - - -- What level are we searching from - search_maxrank := NEW.rank_search; - - -- Thought this wasn't needed but when we add new languages to the country_name table - -- we need to update the existing names - 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; + -- Thought this wasn't needed but when we add new languages to the country_name table + -- we need to update the existing names + IF NEW.name is not null AND array_upper(akeys(NEW.name),1) > 1 THEN + default_language := get_country_language_code(NEW.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; + --DEBUG: RAISE WARNING 'Local names updated'; - -- Initialise the name vector using our name - name_vector := make_keywords(NEW.name); - nameaddress_vector := '{}'::int[]; + -- Initialise the name vector using our name + name_vector := make_keywords(NEW.name); + nameaddress_vector := '{}'::int[]; - -- some tag combinations add a special id for search - tagpairid := get_tagpair(NEW.class,NEW.type); - IF tagpairid IS NOT NULL THEN - name_vector := name_vector + tagpairid; - END IF; + FOR i IN 1..28 LOOP + address_havelevel[i] := false; + END LOOP; - FOR i IN 1..28 LOOP - address_havelevel[i] := false; - END LOOP; + NEW.importance := null; + select language||':'||title,importance from get_wikipedia_match(NEW.extratags, NEW.country_code) INTO NEW.wikipedia,NEW.importance; + IF NEW.importance IS NULL THEN + select language||':'||title,importance from wikipedia_article where osm_type = NEW.osm_type and osm_id = NEW.osm_id order by importance desc limit 1 INTO NEW.wikipedia,NEW.importance; + END IF; - NEW.importance := null; - select language||':'||title,importance from get_wikipedia_match(NEW.extratags, NEW.calculated_country_code) INTO NEW.wikipedia,NEW.importance; - IF NEW.importance IS NULL THEN - select language||':'||title,importance from wikipedia_article where osm_type = NEW.osm_type and osm_id = NEW.osm_id order by importance desc limit 1 INTO NEW.wikipedia,NEW.importance; - END IF; +--DEBUG: RAISE WARNING 'Importance computed from wikipedia: %', NEW.importance; ---RAISE WARNING 'before low level% %', NEW.place_id, NEW.rank_search; + -- --------------------------------------------------------------------------- + -- For low level elements we inherit from our parent road + IF (NEW.rank_search > 27 OR (NEW.type = 'postcode' AND NEW.rank_search = 25)) THEN - -- For low level elements we inherit from our parent road - IF (NEW.rank_search > 27 OR (NEW.type = 'postcode' AND NEW.rank_search = 25)) THEN + --DEBUG: RAISE WARNING 'finding street for % %', NEW.osm_type, NEW.osm_id; ---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; - -- We won't get a better centroid, besides these places are too small to care - NEW.centroid := place_centroid; + NEW.parent_place_id := null; - NEW.parent_place_id := null; + -- if we have a POI and there is no address information, + -- see if we can get it from a surrounding building + IF NEW.osm_type = 'N' AND addr_street IS NULL AND addr_place IS NULL + AND NEW.housenumber IS NULL THEN + FOR location IN select address from placex where ST_Covers(geometry, place_centroid) + and address is not null + and (address ? 'housenumber' or address ? 'street' or address ? 'place') + and rank_search > 28 AND ST_GeometryType(geometry) in ('ST_Polygon','ST_MultiPolygon') + limit 1 + LOOP + NEW.housenumber := location.address->'housenumber'; + addr_street := location.address->'street'; + addr_place := location.address->'place'; + --DEBUG: RAISE WARNING 'Found surrounding building % %', location.osm_type, location.osm_id; + END LOOP; + END IF; - -- to do that we have to find our parent road - -- Copy data from linked items (points on ways, addr:street links, relations) - -- 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 + -- We have to find our parent road. + -- Copy data from linked items (points on ways, addr:street links, relations) - -- 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 - -- At the moment we only process one type of relation - associatedStreet - IF relation.tags @> ARRAY['associatedStreet'] AND array_upper(relation.members, 1) IS NOT NULL THEN - 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 + -- Is this object part of a relation? + FOR relation IN select * from planet_osm_rels where parts @> ARRAY[NEW.osm_id] and members @> ARRAY[lower(NEW.osm_type)||NEW.osm_id] + LOOP + -- At the moment we only process one type of relation - associatedStreet + IF relation.tags @> ARRAY['associatedStreet'] THEN + 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)::bigint - and rank_search = 26 INTO NEW.parent_place_id; - END IF; - END LOOP; + SELECT place_id from placex where osm_type = 'W' + and osm_id = substring(relation.members[i],2,200)::bigint + and rank_search = 26 and name is not null INTO NEW.parent_place_id; END IF; - END LOOP; + END LOOP; + END IF; + END LOOP; + --DEBUG: RAISE WARNING 'Checked for street relation (%)', NEW.parent_place_id; ---RAISE WARNING 'x1'; - -- Is this node part of a way? - FOR way IN select id from planet_osm_ways where nodes @> ARRAY[NEW.osm_id] LOOP ---RAISE WARNING '%', way; - FOR location IN select * from placex where osm_type = 'W' and osm_id = way.id - LOOP ---RAISE WARNING '%', location; - -- Way IS a road then we are on it - that must be our road - IF location.rank_search = 26 AND NEW.parent_place_id IS NULL THEN ---RAISE WARNING 'node in way that is a street %',location; - NEW.parent_place_id := location.place_id; - END IF; + -- Note that addr:street links can only be indexed once the street itself is indexed + IF NEW.parent_place_id IS NULL AND addr_street IS NOT NULL THEN + address_street_word_ids := get_name_ids(make_standard_name(addr_street)); + IF address_street_word_ids IS NOT NULL THEN + SELECT place_id from getNearestNamedRoadFeature(NEW.partition, place_centroid, address_street_word_ids) INTO NEW.parent_place_id; + END IF; + END IF; + --DEBUG: RAISE WARNING 'Checked for addr:street (%)', NEW.parent_place_id; - -- Is the WAY part of a relation - IF NEW.parent_place_id IS NULL THEN - FOR relation IN select * from planet_osm_rels where parts @> ARRAY[location.osm_id] and members @> ARRAY['w'||location.osm_id] - LOOP - -- At the moment we only process one type of relation - associatedStreet - IF relation.tags @> ARRAY['associatedStreet'] AND array_upper(relation.members, 1) IS NOT NULL THEN - 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)::bigint - and rank_search = 26 INTO NEW.parent_place_id; - END IF; - END LOOP; - END IF; - END LOOP; - 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 ---RAISE WARNING 'node in way that has a streetname %',location; - NEW.street := location.street; - END IF; + IF NEW.parent_place_id IS NULL AND addr_place IS NOT NULL THEN + address_street_word_ids := get_name_ids(make_standard_name(addr_place)); + IF address_street_word_ids IS NOT NULL THEN + SELECT place_id from getNearestNamedPlaceFeature(NEW.partition, place_centroid, address_street_word_ids) INTO NEW.parent_place_id; + END IF; + END IF; + --DEBUG: RAISE WARNING 'Checked for addr:place (%)', NEW.parent_place_id; - -- 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; + -- Is this node part of an interpolation? + IF NEW.parent_place_id IS NULL AND NEW.osm_type = 'N' THEN + SELECT q.parent_place_id FROM location_property_osmline q, planet_osm_ways x + WHERE q.linegeo && NEW.geometry and x.id = q.osm_id and NEW.osm_id = any(x.nodes) + LIMIT 1 INTO NEW.parent_place_id; + END IF; + --DEBUG: RAISE WARNING 'Checked for interpolation (%)', NEW.parent_place_id; - -- 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 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 - LOOP ---RAISE WARNING 'using nearest street to address interpolation line,0.001 %',relation; - NEW.parent_place_id := relation.place_id; - END LOOP; - END IF; + -- Is this node part of a way? + IF NEW.parent_place_id IS NULL AND NEW.osm_type = 'N' THEN - END LOOP; - END LOOP; - - END IF; + FOR location IN + SELECT p.place_id, p.osm_id, p.rank_search, p.address from placex p, planet_osm_ways w + WHERE p.osm_type = 'W' and p.rank_search >= 26 and p.geometry && NEW.geometry and w.id = p.osm_id and NEW.osm_id = any(w.nodes) + LOOP + --DEBUG: RAISE WARNING 'Node is part of way % ', location.osm_id; ---RAISE WARNING 'x2'; + -- Way IS a road then we are on it - that must be our road + IF location.rank_search < 28 THEN +--RAISE WARNING 'node in way that is a street %',location; + NEW.parent_place_id := location.place_id; + EXIT; + END IF; + --DEBUG: RAISE WARNING 'Checked if way is street (%)', NEW.parent_place_id; + + -- If the way mentions a street or place address, try that for parenting. + IF location.address is not null THEN + IF location.address ? 'street' THEN + address_street_word_ids := get_name_ids(make_standard_name(location.address->'street')); + IF address_street_word_ids IS NOT NULL THEN + SELECT place_id from getNearestNamedRoadFeature(NEW.partition, place_centroid, address_street_word_ids) INTO NEW.parent_place_id; + EXIT WHEN NEW.parent_place_id is not NULL; + END IF; + END IF; + --DEBUG: RAISE WARNING 'Checked for addr:street in way (%)', NEW.parent_place_id; + + IF location.address ? 'place' THEN + address_street_word_ids := get_name_ids(make_standard_name(location.address->'place')); + IF address_street_word_ids IS NOT NULL THEN + SELECT place_id from getNearestNamedPlaceFeature(NEW.partition, place_centroid, address_street_word_ids) INTO NEW.parent_place_id; + EXIT WHEN NEW.parent_place_id is not NULL; + END IF; + END IF; + --DEBUG: RAISE WARNING 'Checked for addr:place in way (%)', NEW.parent_place_id; + END IF; - IF NEW.parent_place_id IS NULL AND NEW.osm_type = 'W' THEN - -- Is this way part of a relation? - FOR relation IN select * from planet_osm_rels where parts @> ARRAY[NEW.osm_id] and members @> ARRAY['w'||NEW.osm_id] + -- Is the WAY part of a relation + FOR relation IN select * from planet_osm_rels where parts @> ARRAY[location.osm_id] and members @> ARRAY['w'||location.osm_id] LOOP -- At the moment we only process one type of relation - associatedStreet IF relation.tags @> ARRAY['associatedStreet'] AND array_upper(relation.members, 1) IS NOT NULL THEN 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)::bigint - and rank_search = 26 INTO NEW.parent_place_id; +--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 and name is not null INTO NEW.parent_place_id; END IF; END LOOP; END IF; END LOOP; - END IF; - ---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 - NEW.parent_place_id := location.place_id; - END LOOP; - END IF; - END IF; + EXIT WHEN NEW.parent_place_id is not null; + --DEBUG: RAISE WARNING 'Checked for street relation in way (%)', NEW.parent_place_id; - 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; - END IF; - ---RAISE WARNING 'x4 %',NEW.parent_place_id; - -- Still nothing, just use the nearest road - IF NEW.parent_place_id IS NULL THEN - FOR location IN SELECT place_id FROM getNearestRoadFeature(NEW.partition, place_centroid) LOOP - NEW.parent_place_id := location.place_id; - END LOOP; - END IF; - ---return NEW; ---RAISE WARNING 'x6 %',NEW.parent_place_id; - - -- If we didn't find any road fallback to standard method - IF NEW.parent_place_id IS NOT NULL THEN - - -- Add the street to the address as zero distance to force to front of list --- INSERT INTO place_addressline VALUES (NEW.place_id, NEW.parent_place_id, true, true, 0, 26); - address_havelevel[26] := true; - - -- Import address details from parent, reclculating distance in process --- INSERT INTO place_addressline select NEW.place_id, x.address_place_id, x.fromarea, x.isaddress, ST_distance(NEW.geometry, placex.geometry), placex.rank_address --- from place_addressline as x join placex on (address_place_id = placex.place_id) --- where x.place_id = NEW.parent_place_id and x.address_place_id != NEW.parent_place_id; + END LOOP; + END IF; - -- Get the details of the parent road - select * from search_name where place_id = NEW.parent_place_id INTO location; - NEW.calculated_country_code := location.country_code; + -- Still nothing, just use the nearest road + IF NEW.parent_place_id IS NULL THEN + SELECT place_id FROM getNearestRoadFeature(NEW.partition, place_centroid) INTO NEW.parent_place_id; + END IF; + --DEBUG: RAISE WARNING 'Checked for nearest way (%)', NEW.parent_place_id; ---RAISE WARNING '%', NEW.name; - -- If there is no name it isn't searchable, don't bother to create a search record - IF NEW.name is NULL THEN - return NEW; - END IF; - -- 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 + -- If we didn't find any road fallback to standard method + IF NEW.parent_place_id IS NOT NULL THEN - IF NEW.rank_search <= 25 THEN - result := add_location(NEW.place_id, NEW.calculated_country_code, NEW.partition, name_vector, NEW.rank_search, NEW.rank_address, NEW.geometry); - END IF; + -- Get the details of the parent road + select s.country_code, s.name_vector, s.nameaddress_vector from search_name s + where s.place_id = NEW.parent_place_id INTO location; + NEW.country_code := location.country_code; + --DEBUG: RAISE WARNING 'Got parent details from search name'; - 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); + -- determine postcode + IF NEW.rank_search > 4 THEN + IF NEW.address is not null AND NEW.address ? 'postcode' THEN + NEW.postcode = upper(trim(NEW.address->'postcode')); + ELSE + SELECT postcode FROM placex WHERE place_id = NEW.parent_place_id INTO NEW.postcode; + END IF; + IF NEW.postcode is null THEN + NEW.postcode := get_nearest_postcode(NEW.country_code, place_centroid); + END IF; + END IF; + -- If there is no name it isn't searchable, don't bother to create a search record + IF NEW.name is NULL THEN + --DEBUG: RAISE WARNING 'Not a searchable place % %', NEW.osm_type, NEW.osm_id; return NEW; END IF; - END IF; + -- Merge address from parent + nameaddress_vector := array_merge(nameaddress_vector, location.nameaddress_vector); + nameaddress_vector := array_merge(nameaddress_vector, location.name_vector); --- RAISE WARNING ' INDEXING Started:'; --- RAISE WARNING ' INDEXING: %',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 - IF NEW.osm_type = 'R' AND NEW.rank_search < 26 THEN - - -- see if we have any special relation members - select members from planet_osm_rels where id = NEW.osm_id INTO relation_members; + 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); + --DEBUG: RAISE WARNING 'Place added to location table'; + END IF; --- RAISE WARNING 'get_osm_rel_members, label'; - IF relation_members IS NOT NULL THEN - FOR relMember IN select get_osm_rel_members(relation_members,ARRAY['label']) as member LOOP + result := insertSearchName(NEW.partition, NEW.place_id, NEW.country_code, name_vector, nameaddress_vector, NEW.rank_search, NEW.rank_address, NEW.importance, place_centroid, NEW.geometry); + --DEBUG: RAISE WARNING 'Place added to search table'; - FOR linkedPlacex IN select * from placex where osm_type = upper(substring(relMember.member,1,1))::char(1) - and osm_id = substring(relMember.member,2,10000)::bigint order by rank_search desc limit 1 LOOP + return NEW; + END IF; - -- 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)); - END IF; + END IF; - -- merge in the label name, re-init word vector - IF NOT linkedPlacex.name IS NULL THEN - NEW.name := linkedPlacex.name || NEW.name; - name_vector := array_merge(name_vector, make_keywords(linkedPlacex.name)); - END IF; + -- --------------------------------------------------------------------------- + -- Full indexing + --DEBUG: RAISE WARNING 'Using full index mode for % %', NEW.osm_type, NEW.osm_id; - -- merge in extra tags - IF NOT linkedPlacex.extratags IS NULL THEN - NEW.extratags := linkedPlacex.extratags || NEW.extratags; - END IF; + IF NEW.osm_type = 'R' AND NEW.rank_search < 26 THEN - IF NOT NEW.extratags ? linkedPlacex.class THEN - NEW.extratags := NEW.extratags || hstore(linkedPlacex.class, linkedPlacex.type); - END IF; + -- see if we have any special relation members + select members from planet_osm_rels where id = NEW.osm_id INTO relation_members; + --DEBUG: RAISE WARNING 'Got relation members'; - -- mark the linked place (excludes from search results) - UPDATE placex set linked_place_id = NEW.place_id where place_id = linkedPlacex.place_id; + IF relation_members IS NOT NULL THEN + FOR relMember IN select get_osm_rel_members(relation_members,ARRAY['label']) as member LOOP + --DEBUG: RAISE WARNING 'Found label member %', relMember.member; - END LOOP; + FOR linkedPlacex IN select * from placex where osm_type = upper(substring(relMember.member,1,1))::char(1) + and osm_id = substring(relMember.member,2,10000)::bigint + and class = 'place' order by rank_search desc limit 1 LOOP - END LOOP; + -- 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)); + END IF; - IF NEW.centroid IS NULL THEN + -- merge in the label name, re-init word vector + IF NOT linkedPlacex.name IS NULL THEN + NEW.name := linkedPlacex.name || NEW.name; + name_vector := array_merge(name_vector, make_keywords(linkedPlacex.name)); + END IF; - FOR relMember IN select get_osm_rel_members(relation_members,ARRAY['admin_center','admin_centre']) as member LOOP + -- merge in extra tags + NEW.extratags := hstore(linkedPlacex.class, linkedPlacex.type) || coalesce(linkedPlacex.extratags, ''::hstore) || coalesce(NEW.extratags, ''::hstore); - FOR linkedPlacex IN select * from placex where osm_type = upper(substring(relMember.member,1,1))::char(1) - and osm_id = substring(relMember.member,2,10000)::bigint order by rank_search desc limit 1 LOOP + -- mark the linked place (excludes from search results) + UPDATE placex set linked_place_id = NEW.place_id where place_id = linkedPlacex.place_id; - -- For an admin centre we also want a name match - still not perfect, for example 'new york, new york' - -- But that can be fixed by explicitly setting the label in the data - IF make_standard_name(NEW.name->'name') = make_standard_name(linkedPlacex.name->'name') - AND NEW.rank_address = linkedPlacex.rank_address THEN + -- keep a note of the node id in case we need it for wikipedia in a bit + linked_node_id := linkedPlacex.osm_id; + select language||':'||title,importance from get_wikipedia_match(linkedPlacex.extratags, NEW.country_code) INTO linked_wikipedia,linked_importance; + --DEBUG: RAISE WARNING 'Linked label member'; + END LOOP; - -- 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)); - END IF; + END LOOP; - -- merge in the name, re-init word vector - IF NOT linkedPlacex.name IS NULL THEN - NEW.name := linkedPlacex.name || NEW.name; - name_vector := make_keywords(NEW.name); - END IF; + IF NEW.centroid IS NULL THEN - -- merge in extra tags - IF NOT linkedPlacex.extratags IS NULL THEN - NEW.extratags := linkedPlacex.extratags || NEW.extratags; - END IF; + FOR relMember IN select get_osm_rel_members(relation_members,ARRAY['admin_center','admin_centre']) as member LOOP + --DEBUG: RAISE WARNING 'Found admin_center member %', relMember.member; - IF NOT NEW.extratags ? linkedPlacex.class THEN - NEW.extratags := NEW.extratags || hstore(linkedPlacex.class, linkedPlacex.type); - END IF; + FOR linkedPlacex IN select * from placex where osm_type = upper(substring(relMember.member,1,1))::char(1) + and osm_id = substring(relMember.member,2,10000)::bigint + and class = 'place' order by rank_search desc limit 1 LOOP - -- mark the linked place (excludes from search results) - UPDATE placex set linked_place_id = NEW.place_id where place_id = linkedPlacex.place_id; + -- For an admin centre we also want a name match - still not perfect, for example 'new york, new york' + -- But that can be fixed by explicitly setting the label in the data + IF make_standard_name(NEW.name->'name') = make_standard_name(linkedPlacex.name->'name') + AND NEW.rank_address = linkedPlacex.rank_address THEN - -- keep a note of the node id in case we need it for wikipedia in a bit - linked_node_id := linkedPlacex.osm_id; + -- 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)); END IF; - END LOOP; - - END LOOP; - - END IF; - END IF; - - -- 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 - - FOR linkedPlacex IN select placex.* from placex WHERE - make_standard_name(name->'name') = make_standard_name(NEW.name->'name') - AND placex.rank_address = NEW.rank_address - AND placex.place_id != NEW.place_id - AND placex.osm_type = 'N'::char(1) AND placex.rank_search < 26 - AND st_covers(NEW.geometry, placex.geometry) - LOOP - - -- 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)); - END IF; + -- merge in the name, re-init word vector + IF NOT linkedPlacex.name IS NULL THEN + NEW.name := linkedPlacex.name || NEW.name; + name_vector := make_keywords(NEW.name); + END IF; - -- merge in the name, re-init word vector - NEW.name := linkedPlacex.name || NEW.name; - name_vector := make_keywords(NEW.name); + -- merge in extra tags + NEW.extratags := hstore(linkedPlacex.class, linkedPlacex.type) || coalesce(linkedPlacex.extratags, ''::hstore) || coalesce(NEW.extratags, ''::hstore); - -- merge in extra tags - NEW.extratags := linkedPlacex.extratags || NEW.extratags; + -- mark the linked place (excludes from search results) + UPDATE placex set linked_place_id = NEW.place_id where place_id = linkedPlacex.place_id; - IF NOT NEW.extratags ? linkedPlacex.class THEN - NEW.extratags := NEW.extratags || hstore(linkedPlacex.class, linkedPlacex.type); - END IF; + -- keep a note of the node id in case we need it for wikipedia in a bit + linked_node_id := linkedPlacex.osm_id; + select language||':'||title,importance from get_wikipedia_match(linkedPlacex.extratags, NEW.country_code) INTO linked_wikipedia,linked_importance; + --DEBUG: RAISE WARNING 'Linked admin_center'; + 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; + END LOOP; - -- 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 IF; - 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; + END IF; - -- Did we gain a wikipedia tag in the process? then we need to recalculate our importance - IF NEW.importance is null THEN - select language||':'||title,importance from get_wikipedia_match(NEW.extratags, NEW.calculated_country_code) INTO NEW.wikipedia,NEW.importance; - END IF; - -- Still null? how about looking it up by the node id - IF NEW.importance IS NULL THEN - select language||':'||title,importance from wikipedia_article where osm_type = 'N'::char(1) and osm_id = linked_node_id order by importance desc limit 1 INTO NEW.wikipedia,NEW.importance; - 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 AND NEW.rank_address > 0 THEN - -- make sure all names are in the word table - IF NEW.admin_level = 2 AND NEW.class = 'boundary' AND NEW.type = 'administrative' AND NEW.country_code IS NOT NULL THEN - perform create_country(NEW.name, lower(NEW.country_code)); - END IF; + -- 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 - NEW.parent_place_id = 0; - parent_place_id_rank = 0; - - -- convert isin to array of tokenids - isin_tokens := '{}'::int[]; - IF NEW.isin IS NOT NULL THEN - isin := regexp_split_to_array(NEW.isin, E'[;,]'); - IF array_upper(isin, 1) IS NOT NULL THEN - 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; + --DEBUG: RAISE WARNING 'Looking for nodes with matching names'; + FOR linkedPlacex IN select placex.* from placex WHERE + make_standard_name(name->'name') = make_standard_name(NEW.name->'name') + AND placex.rank_address = NEW.rank_address + AND placex.place_id != NEW.place_id + AND placex.osm_type = 'N'::char(1) AND placex.rank_search < 26 + AND st_covers(NEW.geometry, placex.geometry) + LOOP + --DEBUG: RAISE WARNING 'Found matching place node %', linkedPlacex.osm_id; + -- 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)); + 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; + -- merge in the name, re-init word vector + NEW.name := linkedPlacex.name || NEW.name; + name_vector := make_keywords(NEW.name); + + -- merge in extra tags + 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; + select language||':'||title,importance from get_wikipedia_match(linkedPlacex.extratags, NEW.country_code) INTO linked_wikipedia,linked_importance; + --DEBUG: RAISE WARNING 'Linked named place'; + END LOOP; END IF; - IF NEW.postcode IS NOT NULL THEN - isin := regexp_split_to_array(NEW.postcode, E'[;,]'); - IF array_upper(isin, 1) IS NOT NULL THEN - 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]); + 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.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 LOOP; + END IF; END IF; + --DEBUG: RAISE WARNING 'Names updated from linked places'; 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; + -- Use the maximum importance if a one could be computed from the linked object. + IF linked_importance is not null AND + (NEW.importance is null or NEW.importance < linked_importance) THEN + NEW.importance = linked_importance; END IF; --- RAISE WARNING 'ISIN: %', isin_tokens; + -- Still null? how about looking it up by the node id + IF NEW.importance IS NULL THEN + --DEBUG: RAISE WARNING 'Looking up importance by linked node id'; + select language||':'||title,importance from wikipedia_article where osm_type = 'N'::char(1) and osm_id = linked_node_id order by importance desc limit 1 INTO NEW.wikipedia,NEW.importance; + END IF; - -- Process area matches - location_rank_search := 0; - location_distance := 0; - location_parent := NULL; - -- added ourself as address already - address_havelevel[NEW.rank_address] := true; - -- RAISE WARNING ' getNearFeatures(%,''%'',%,''%'')',NEW.partition, place_centroid, search_maxrank, isin_tokens; - FOR location IN SELECT * from getNearFeatures(NEW.partition, place_centroid, search_maxrank, isin_tokens) LOOP + END IF; ---RAISE WARNING ' AREA: %',location; + -- make sure all names are in the word table + IF NEW.admin_level = 2 AND NEW.class = 'boundary' AND NEW.type = 'administrative' AND NEW.country_code IS NOT NULL AND NEW.osm_type = 'R' THEN + perform create_country(NEW.name, lower(NEW.country_code)); + --DEBUG: RAISE WARNING 'Country names updated'; + END IF; - IF location.rank_address != location_rank_search THEN - location_rank_search := location.rank_address; - location_distance := location.distance * 1.5; - END IF; + NEW.parent_place_id = 0; + parent_place_id_rank = 0; - IF location.distance < location_distance OR NOT location.isguess THEN - location_isaddress := NOT address_havelevel[location.rank_address]; - IF location_isaddress AND location.isguess AND location_parent IS NOT NULL THEN - location_isaddress := ST_Contains(location_parent,location.centroid); + -- convert address store to array of tokenids + --DEBUG: RAISE WARNING 'Starting address search'; + isin_tokens := '{}'::int[]; + IF NEW.address IS NOT NULL THEN + FOR addr_item IN SELECT * FROM each(NEW.address) + LOOP + IF addr_item.key IN ('city', 'tiger:county', 'state', 'suburb', 'province', 'district', 'region', 'county', 'municipality', 'hamlet', 'village', 'subdistrict', 'town', 'neighbourhood', 'quarter', 'parish') THEN + address_street_word_id := get_name_id(make_standard_name(addr_item.value)); + IF address_street_word_id IS NOT NULL AND NOT(ARRAY[address_street_word_id] <@ isin_tokens) THEN + isin_tokens := isin_tokens || address_street_word_id; END IF; - - -- RAISE WARNING '% isaddress: %', location.place_id, location_isaddress; - -- Add it to the list of search terms - IF location.rank_search > 4 THEN - nameaddress_vector := array_merge(nameaddress_vector, location.keywords::integer[]); + address_street_word_id := get_word_id(make_standard_name(addr_item.value)); + IF address_street_word_id IS NOT NULL THEN + nameaddress_vector := array_merge(nameaddress_vector, ARRAY[address_street_word_id]); END IF; - INSERT INTO place_addressline VALUES (NEW.place_id, location.place_id, true, location_isaddress, location.distance, location.rank_address); + END IF; + IF addr_item.key = 'is_in' THEN + -- is_in items need splitting + isin := regexp_split_to_array(addr_item.value, E'[;,]'); + IF array_upper(isin, 1) IS NOT NULL THEN + 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 + isin_tokens := isin_tokens || address_street_word_id; + END IF; - 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; + -- 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; + END LOOP; + END IF; + nameaddress_vector := array_merge(nameaddress_vector, isin_tokens); ---RAISE WARNING ' Terms: (%) %',location, nameaddress_vector; +-- RAISE WARNING 'ISIN: %', isin_tokens; - IF location.rank_address > parent_place_id_rank THEN - NEW.parent_place_id = location.place_id; - parent_place_id_rank = location.rank_address; + -- Process area matches + location_rank_search := 0; + location_distance := 0; + location_parent := NULL; + -- added ourself as address already + address_havelevel[NEW.rank_address] := true; + --DEBUG: RAISE WARNING ' getNearFeatures(%,''%'',%,''%'')',NEW.partition, place_centroid, search_maxrank, isin_tokens; + FOR location IN + SELECT * from getNearFeatures(NEW.partition, + CASE WHEN NEW.rank_search >= 26 + AND NEW.rank_search < 30 + THEN NEW.geometry + ELSE place_centroid END, + search_maxrank, isin_tokens) + LOOP + IF location.rank_address != location_rank_search THEN + location_rank_search := 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; - END IF; + ELSE + CONTINUE WHEN location.keywords <@ location_keywords; + END IF; - END LOOP; + IF location.distance < location_distance OR NOT location.isguess THEN + location_keywords := location.keywords; - -- try using the isin value to find parent places - IF array_upper(isin_tokens, 1) IS NOT NULL THEN - FOR i IN 1..array_upper(isin_tokens, 1) LOOP ---RAISE WARNING ' getNearestNamedFeature: % % % %',NEW.partition, place_centroid, search_maxrank, isin_tokens[i]; - IF NOT ARRAY[isin_tokens[i]] <@ nameaddress_vector THEN + location_isaddress := NOT address_havelevel[location.rank_address]; + IF location_isaddress AND location.isguess AND location_parent IS NOT NULL THEN + location_isaddress := ST_Contains(location_parent,location.centroid); + END IF; - FOR location IN SELECT * from getNearestNamedFeature(NEW.partition, place_centroid, search_maxrank, isin_tokens[i]) LOOP + -- RAISE WARNING '% isaddress: %', location.place_id, location_isaddress; + -- Add it to the list of search terms + IF location.rank_search > 4 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 (NEW.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 + NEW.postcode = location.postcode; + END IF; - --RAISE WARNING ' ISIN: %',location; + 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_search > 4 THEN - nameaddress_vector := array_merge(nameaddress_vector, location.keywords::integer[]); - INSERT INTO place_addressline VALUES (NEW.place_id, location.place_id, false, NOT address_havelevel[location.rank_address], location.distance, location.rank_address); - address_havelevel[location.rank_address] := true; + 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; - END IF; - END LOOP; + END IF; - END IF; + --DEBUG: RAISE WARNING ' Terms: (%) %',location, nameaddress_vector; - END LOOP; END IF; - -- for long ways we should add search terms for the entire length - IF st_length(NEW.geometry) > 0.05 THEN + END LOOP; + --DEBUG: RAISE WARNING 'address computed'; - location_rank_search := 0; - location_distance := 0; + -- for long ways we should add search terms for the entire length + IF st_length(NEW.geometry) > 0.05 THEN - FOR location IN SELECT * from getNearFeatures(NEW.partition, NEW.geometry, search_maxrank, isin_tokens) LOOP + location_rank_search := 0; + location_distance := 0; - IF location.rank_address != location_rank_search THEN - location_rank_search := location.rank_address; - location_distance := location.distance * 1.5; - END IF; + FOR location IN SELECT * from getNearFeatures(NEW.partition, NEW.geometry, search_maxrank, isin_tokens) LOOP - IF location.rank_search > 4 AND location.distance < location_distance THEN + IF location.rank_address != location_rank_search THEN + location_rank_search := location.rank_address; + location_distance := location.distance * 1.5; + END IF; - -- Add it to the list of search terms - nameaddress_vector := array_merge(nameaddress_vector, location.keywords::integer[]); - INSERT INTO place_addressline VALUES (NEW.place_id, location.place_id, true, false, location.distance, location.rank_address); + IF location.rank_search > 4 AND location.distance < location_distance THEN - END IF; + -- Add it to the list of search terms + nameaddress_vector := array_merge(nameaddress_vector, location.keywords::integer[]); + INSERT INTO place_addressline (place_id, address_place_id, fromarea, isaddress, distance, cached_rank_address) + VALUES (NEW.place_id, location.place_id, true, false, location.distance, location.rank_address); - END LOOP; + END IF; - END IF; + END LOOP; - -- if we have a name add this to the name search table - IF NEW.name IS NOT NULL THEN + END IF; + --DEBUG: RAISE WARNING 'search terms for long ways added'; - IF NEW.rank_search <= 25 THEN - result := add_location(NEW.place_id, NEW.calculated_country_code, NEW.partition, name_vector, NEW.rank_search, NEW.rank_address, NEW.geometry); - END IF; + IF NEW.address is not null AND NEW.address ? 'postcode' + AND NEW.address->'postcode' not similar to '%(,|;)%' THEN + NEW.postcode := upper(trim(NEW.address->'postcode')); + END IF; - IF NEW.rank_search between 26 and 27 and NEW.class = 'highway' THEN - result := insertLocationRoad(NEW.partition, NEW.place_id, NEW.calculated_country_code, NEW.geometry); - END IF; + IF NEW.postcode is null AND NEW.rank_search > 8 THEN + NEW.postcode := get_nearest_postcode(NEW.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, NEW.geometry); + -- if we have a name add this to the name search table + 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); + --DEBUG: RAISE WARNING 'added to location (full)'; END IF; - -- If we've not managed to pick up a better one - default centroid - IF NEW.centroid IS NULL THEN - NEW.centroid := place_centroid; + IF NEW.rank_search between 26 and 27 and NEW.class = 'highway' THEN + result := insertLocationRoad(NEW.partition, NEW.place_id, NEW.country_code, NEW.geometry); + --DEBUG: RAISE WARNING 'insert into road location table (full)'; END IF; + result := insertSearchName(NEW.partition, NEW.place_id, NEW.country_code, name_vector, nameaddress_vector, NEW.rank_search, NEW.rank_address, NEW.importance, place_centroid, NEW.geometry); + --DEBUG: RAISE WARNING 'added to serach name (full)'; + END IF; + -- If we've not managed to pick up a better one - default centroid + IF NEW.centroid IS NULL THEN + NEW.centroid := place_centroid; + END IF; + + --DEBUG: RAISE WARNING 'place update % % finsihed.', NEW.osm_type, NEW.osm_id; + RETURN NEW; END; $$ @@ -1946,7 +1959,7 @@ BEGIN -- mark everything linked to this place for re-indexing --DEBUG: RAISE WARNING 'placex_delete:03 % %',OLD.osm_type,OLD.osm_id; UPDATE placex set indexed_status = 2 from place_addressline where address_place_id = OLD.place_id - and placex.place_id = place_addressline.place_id and indexed_status = 0; + and placex.place_id = place_addressline.place_id and indexed_status = 0 and place_addressline.isaddress; --DEBUG: RAISE WARNING 'placex_delete:04 % %',OLD.osm_type,OLD.osm_id; DELETE FROM place_addressline where address_place_id = OLD.place_id; @@ -1957,6 +1970,8 @@ BEGIN --DEBUG: RAISE WARNING 'placex_delete:06 % %',OLD.osm_type,OLD.osm_id; update placex set indexed_status = 2 where parent_place_id = OLD.place_id and indexed_status = 0; --DEBUG: RAISE WARNING 'placex_delete:07 % %',OLD.osm_type,OLD.osm_id; + -- reparenting also for OSM Interpolation Lines (and for Tiger?) + update location_property_osmline set indexed_status = 2 where indexed_status = 0 and parent_place_id = OLD.place_id; END IF; @@ -1980,7 +1995,7 @@ BEGIN -- remove from tables for special search classtable := 'place_classtype_' || OLD.class || '_' || OLD.type; - SELECT count(*)>0 FROM pg_tables WHERE tablename = classtable INTO b; + SELECT count(*)>0 FROM pg_tables WHERE tablename = classtable and schemaname = current_schema() INTO b; IF b THEN EXECUTE 'DELETE FROM ' || classtable::regclass || ' WHERE place_id = $1' USING OLD.place_id; END IF; @@ -1996,20 +2011,28 @@ LANGUAGE plpgsql; CREATE OR REPLACE FUNCTION place_delete() RETURNS TRIGGER AS $$ 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 (osm_type, osm_id, class, type) values (OLD.osm_type,OLD.osm_id,OLD.class,OLD.type); + RETURN NULL; + END IF; END IF; -- mark for delete UPDATE placex set indexed_status = 100 where osm_type = OLD.osm_type and osm_id = OLD.osm_id and class = OLD.class and type = OLD.type; + -- interpolations are special + IF OLD.osm_type='W' and OLD.class = 'place' and OLD.type = 'houses' THEN + UPDATE location_property_osmline set indexed_status = 100 where osm_id = OLD.osm_id; -- osm_id = wayid (=old.osm_id) + END IF; + RETURN OLD; END; @@ -2022,6 +2045,7 @@ DECLARE i INTEGER; existing RECORD; existingplacex RECORD; + existingline RECORD; existinggeometry GEOMETRY; existingplace_id BIGINT; result BOOLEAN; @@ -2030,327 +2054,268 @@ BEGIN --DEBUG: RAISE WARNING '-----------------------------------------------------------------------------------'; --DEBUG: RAISE WARNING 'place_insert: % % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type,st_area(NEW.geometry); - - IF FALSE and NEW.osm_type = 'R' THEN - select * from placex where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type INTO existingplacex; - --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; - END IF; - IF NEW.class in ('landuse','natural') and NEW.name is null THEN - RETURN null; - END IF; - + -- filter wrong tupels IF ST_IsEmpty(NEW.geometry) OR NOT ST_IsValid(NEW.geometry) OR ST_X(ST_Centroid(NEW.geometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEW.geometry))::text in ('NaN','Infinity','-Infinity') THEN - INSERT INTO import_polygon_error values (NEW.osm_type, NEW.osm_id, NEW.class, NEW.type, NEW.name, NEW.country_code, - now(), ST_IsValidReason(NEW.geometry), null, NEW.geometry); + INSERT INTO import_polygon_error (osm_type, osm_id, class, type, name, country_code, updated, errormessage, prevgeometry, newgeometry) + VALUES (NEW.osm_type, NEW.osm_id, NEW.class, NEW.type, NEW.name, NEW.address->'country', now(), ST_IsValidReason(NEW.geometry), null, NEW.geometry); -- RAISE WARNING 'Invalid Geometry: % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type; RETURN null; END IF; - -- Patch in additional country names - IF NEW.admin_level = 2 AND NEW.type = 'administrative' AND NEW.country_code is not null THEN - select coalesce(country_name.name || NEW.name,NEW.name) from country_name where country_name.country_code = lower(NEW.country_code) INTO NEW.name; - END IF; - - -- Have we already done this place? - select * from place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type INTO existing; + -- decide, whether it is an osm interpolation line => insert intoosmline, or else just placex + IF NEW.class='place' and NEW.type='houses' and NEW.osm_type='W' and ST_GeometryType(NEW.geometry) = 'ST_LineString' THEN + -- Have we already done this place? + select * from place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type INTO existing; - -- Get the existing place_id - select * from placex where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type INTO existingplacex; + -- Get the existing place_id + select * from location_property_osmline where osm_id = NEW.osm_id INTO existingline; - -- 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'); - END IF; + -- Handle a place changing type by removing the old data (this trigger is executed BEFORE INSERT of the NEW tupel) + 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; - --DEBUG: RAISE WARNING 'Existing PlaceX: %',existingplacex.place_id; + 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; - -- Log and discard - IF existing.geometry is not null AND st_isvalid(existing.geometry) - AND st_area(existing.geometry) > 0.02 - AND ST_GeometryType(NEW.geometry) in ('ST_Polygon','ST_MultiPolygon') - AND st_area(NEW.geometry) < st_area(existing.geometry)*0.5 - THEN - INSERT INTO import_polygon_error values (NEW.osm_type, NEW.osm_id, NEW.class, NEW.type, NEW.name, NEW.country_code, now(), - 'Area reduced from '||st_area(existing.geometry)||' to '||st_area(NEW.geometry), existing.geometry, NEW.geometry); - RETURN null; - END IF; + -- update method for interpolation lines: delete all old interpolation lines with same osm_id (update on place) and insert the new one(s) (they can be split up, if they have > 2 nodes) + IF existingline.osm_id IS NOT NULL THEN + delete from location_property_osmline where osm_id = NEW.osm_id; + END IF; - 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; + -- for interpolations invalidate all nodes on the line + update placex p set indexed_status = 2 + from planet_osm_ways w + where w.id = NEW.osm_id and p.osm_type = 'N' and p.osm_id = any(w.nodes); - -- 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 - IF existing.osm_type IS NOT NULL THEN - DELETE from place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type; + INSERT INTO location_property_osmline (osm_id, address, linegeo) + VALUES (NEW.osm_id, NEW.address, NEW.geometry); + + + IF existing.osm_type IS NULL THEN + return NEW; 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, addr_place, isin, postcode, country_code, extratags, geometry) - values (NEW.osm_type - ,NEW.osm_id - ,NEW.class - ,NEW.type - ,NEW.name - ,NEW.admin_level - ,NEW.housenumber - ,NEW.street - ,NEW.addr_place - ,NEW.isin - ,NEW.postcode - ,NEW.country_code - ,NEW.extratags - ,NEW.geometry - ); + IF coalesce(existing.address, ''::hstore) != coalesce(NEW.address, ''::hstore) + OR (coalesce(existing.extratags, ''::hstore) != coalesce(NEW.extratags, ''::hstore)) + OR existing.geometry::text != NEW.geometry::text + THEN - --DEBUG: RAISE WARNING 'insert done % % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type,NEW.name; + update place set + name = NEW.name, + address = NEW.address, + 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; + END IF; - RETURN NEW; - END IF; + RETURN NULL; - -- Various ways to do the update + ELSE -- insert to placex - -- Debug, what's changed? - IF FALSE THEN - IF coalesce(existing.name::text, '') != coalesce(NEW.name::text, '') THEN - RAISE WARNING 'update details, name: % % % %',NEW.osm_type,NEW.osm_id,existing.name::text,NEW.name::text; - END IF; - IF coalesce(existing.housenumber, '') != coalesce(NEW.housenumber, '') THEN - RAISE WARNING 'update details, housenumber: % % % %',NEW.osm_type,NEW.osm_id,existing.housenumber,NEW.housenumber; - END IF; - 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; + -- Patch in additional country names + IF NEW.admin_level = 2 AND NEW.type = 'administrative' + AND NEW.address is not NULL AND NEW.address ? 'country' THEN + SELECT name FROM country_name WHERE country_code = lower(NEW.address->'country') INTO existing; + IF existing.name IS NOT NULL THEN + NEW.name = existing.name || NEW.name; + END IF; END IF; - IF coalesce(existing.postcode, '') != coalesce(NEW.postcode, '') THEN - RAISE WARNING 'update details, postcode: % % % %',NEW.osm_type,NEW.osm_id,existing.postcode,NEW.postcode; + + -- Have we already done this place? + select * from place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type INTO existing; + + -- Get the existing place_id + select * from placex where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type INTO existingplacex; + + -- 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 THEN + DELETE FROM place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class; END IF; - IF coalesce(existing.country_code, '') != coalesce(NEW.country_code, '') THEN - RAISE WARNING 'update details, country_code: % % % %',NEW.osm_type,NEW.osm_id,existing.country_code,NEW.country_code; + + --DEBUG: RAISE WARNING 'Existing: %',existing.osm_id; + --DEBUG: RAISE WARNING 'Existing PlaceX: %',existingplacex.place_id; + + -- Log and discard + IF existing.geometry is not null AND st_isvalid(existing.geometry) + AND st_area(existing.geometry) > 0.02 + AND ST_GeometryType(NEW.geometry) in ('ST_Polygon','ST_MultiPolygon') + AND st_area(NEW.geometry) < st_area(existing.geometry)*0.5 + THEN + INSERT INTO import_polygon_error (osm_type, osm_id, class, type, name, country_code, updated, errormessage, prevgeometry, newgeometry) + VALUES (NEW.osm_type, NEW.osm_id, NEW.class, NEW.type, NEW.name, NEW.address->'country', now(), + 'Area reduced from '||st_area(existing.geometry)||' to '||st_area(NEW.geometry), existing.geometry, NEW.geometry); + RETURN null; END IF; - END IF; - -- Special case for polygon shape changes because they tend to be large and we can be a bit clever about how we handle them - IF existing.geometry::text != NEW.geometry::text - AND ST_GeometryType(existing.geometry) in ('ST_Polygon','ST_MultiPolygon') - AND ST_GeometryType(NEW.geometry) in ('ST_Polygon','ST_MultiPolygon') - THEN + 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 + IF existingplacex.osm_type IS NULL OR + (existingplacex.class = 'boundary' AND + ((coalesce(existingplacex.admin_level, 15) != coalesce(NEW.admin_level, 15) AND existingplacex.type = 'administrative') OR + (existingplacex.type != NEW.type))) + THEN - -- Get the version of the geometry actually used (in placex table) - select geometry from placex where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type into existinggeometry; + 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; - -- Performance limit - IF st_area(NEW.geometry) < 0.000000001 AND st_area(existinggeometry) < 1 THEN + 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; - -- re-index points that have moved in / out of the polygon, could be done as a single query but postgres gets the index usage wrong - update placex set indexed_status = 2 where indexed_status = 0 and - (st_covers(NEW.geometry, placex.geometry) OR ST_Intersects(NEW.geometry, placex.geometry)) - AND NOT (st_covers(existinggeometry, placex.geometry) OR ST_Intersects(existinggeometry, placex.geometry)) - AND rank_search > existingplacex.rank_search AND (rank_search < 28 or name is not null); + -- 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, address, extratags, geometry) + values (NEW.osm_type, NEW.osm_id, NEW.class, NEW.type, NEW.name, + NEW.admin_level, NEW.address, NEW.extratags, NEW.geometry); - update placex set indexed_status = 2 where indexed_status = 0 and - (st_covers(existinggeometry, placex.geometry) OR ST_Intersects(existinggeometry, placex.geometry)) - AND NOT (st_covers(NEW.geometry, placex.geometry) OR ST_Intersects(NEW.geometry, placex.geometry)) - AND rank_search > existingplacex.rank_search AND (rank_search < 28 or name is not null); + --DEBUG: RAISE WARNING 'insert done % % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type,NEW.name; + RETURN NEW; END IF; - END IF; + -- Special case for polygon shape changes because they tend to be large and we can be a bit clever about how we handle them + IF existing.geometry::text != NEW.geometry::text + AND ST_GeometryType(existing.geometry) in ('ST_Polygon','ST_MultiPolygon') + AND ST_GeometryType(NEW.geometry) in ('ST_Polygon','ST_MultiPolygon') + THEN - -- Special case - if we are just adding extra words we hack them into the search_name table rather than reindexing - 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, '') - AND coalesce(existing.name::text, '') != coalesce(NEW.name::text, '') - THEN + -- Get the version of the geometry actually used (in placex table) + select geometry from placex where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type into existinggeometry; - IF NOT update_location_nameonly(existingplacex.place_id, NEW.name) THEN + -- Performance limit + IF st_area(NEW.geometry) < 0.000000001 AND st_area(existinggeometry) < 1 THEN + + -- re-index points that have moved in / out of the polygon, could be done as a single query but postgres gets the index usage wrong + update placex set indexed_status = 2 where indexed_status = 0 and + (st_covers(NEW.geometry, placex.geometry) OR ST_Intersects(NEW.geometry, placex.geometry)) + AND NOT (st_covers(existinggeometry, placex.geometry) OR ST_Intersects(existinggeometry, placex.geometry)) + AND rank_search > existingplacex.rank_search AND (rank_search < 28 or name is not null); + + update placex set indexed_status = 2 where indexed_status = 0 and + (st_covers(existinggeometry, placex.geometry) OR ST_Intersects(existinggeometry, placex.geometry)) + AND NOT (st_covers(NEW.geometry, placex.geometry) OR ST_Intersects(NEW.geometry, placex.geometry)) + AND rank_search > existingplacex.rank_search AND (rank_search < 28 or name is not null); - IF st_area(NEW.geometry) < 0.5 THEN - UPDATE placex set indexed_status = 2 from place_addressline where address_place_id = existingplacex.place_id - and placex.place_id = place_addressline.place_id and indexed_status = 0 - and (rank_search < 28 or name is not null); END IF; END IF; - - ELSE - -- Anything else has changed - reindex the lot + 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 - - -- performance, can't take the load of re-indexing a whole country / huge area - IF st_area(NEW.geometry) < 0.5 THEN --- UPDATE placex set indexed_status = 2 from place_addressline where address_place_id = existingplacex.place_id --- and placex.place_id = place_addressline.place_id and indexed_status = 0; + OR coalesce(existing.extratags::text, '') != coalesce(NEW.extratags::text, '') + OR coalesce(existing.address, ''::hstore) != coalesce(NEW.address, ''::hstore) + OR coalesce(existing.admin_level, 15) != coalesce(NEW.admin_level, 15) + OR existing.geometry::text != NEW.geometry::text + THEN + + update place set + name = NEW.name, + address = NEW.address, + 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; + + + IF NEW.class in ('place','boundary') AND NEW.type in ('postcode','postal_code') THEN + IF NEW.address is NULL OR NOT NEW.address ? 'postcode' 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.address->'postcode'); + END IF; + + IF NEW.class in ('boundary') + AND ST_GeometryType(NEW.geometry) not in ('ST_Polygon','ST_MultiPolygon') THEN + DELETE FROM placex where place_id = existingplacex.place_id; + RETURN NULL; + END IF; + + update placex set + name = NEW.name, + address = NEW.address, + parent_place_id = null, + extratags = NEW.extratags, + admin_level = NEW.admin_level, + indexed_status = 2, + geometry = NEW.geometry + where place_id = existingplacex.place_id; + + -- if a node(=>house), which is part of a interpolation line, changes (e.g. the street attribute) => mark this line for reparenting + -- (already here, because interpolation lines are reindexed before nodes, so in the second call it would be too late) + IF NEW.osm_type='N' and NEW.class='place' and NEW.type='house' THEN + -- Is this node part of an interpolation line? search for it in location_property_osmline and mark the interpolation line for reparenting + update location_property_osmline p set indexed_status = 2 from planet_osm_ways w where p.linegeo && NEW.geometry and p.osm_id = w.id and NEW.osm_id = any(w.nodes); + END IF; + + -- linked places should get potential new naming and addresses + IF existingplacex.linked_place_id is not NULL THEN + update placex x set + name = p.name, + extratags = p.extratags, + indexed_status = 2 + from place p + where x.place_id = existingplacex.linked_place_id + and x.indexed_status = 0 + and x.osm_type = p.osm_type + and x.osm_id = p.osm_id + and x.class = p.class; END IF; END IF; + -- Abort the add (we modified the existing place instead) + 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, '') - OR coalesce(existing.admin_level, 15) != coalesce(NEW.admin_level, 15) - OR existing.geometry::text != NEW.geometry::text - THEN - - update place set - 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; - - update placex set - 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 = NEW.admin_level, - indexed_status = 2, - geometry = NEW.geometry - where place_id = existingplacex.place_id; - - END IF; - - -- Abort the add (we modified the existing place instead) - RETURN NULL; - -END; +END; $$ LANGUAGE plpgsql; + CREATE OR REPLACE FUNCTION get_name_by_language(name hstore, languagepref TEXT[]) RETURNS TEXT AS $$ DECLARE - search TEXT[]; - found BOOLEAN; + result TEXT; BEGIN - IF name is null THEN RETURN null; END IF; - search := languagepref; - - FOR j IN 1..array_upper(search, 1) LOOP - IF name ? search[j] AND trim(name->search[j]) != '' THEN - return trim(name->search[j]); + FOR j IN 1..array_upper(languagepref,1) LOOP + IF name ? languagepref[j] THEN + result := trim(name->languagepref[j]); + IF result != '' THEN + return result; + END IF; END IF; END LOOP; -- anything will do as a fallback - just take the first name type thing there is - search := avals(name); - RETURN search[1]; -END; -$$ -LANGUAGE plpgsql IMMUTABLE; - -CREATE OR REPLACE FUNCTION get_connected_ways(way_ids INTEGER[]) RETURNS SETOF planet_osm_ways - AS $$ -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; + RETURN trim((avals(name))[1]); END; $$ LANGUAGE plpgsql IMMUTABLE; -CREATE OR REPLACE FUNCTION get_address_postcode(for_place_id BIGINT) RETURNS TEXT - AS $$ -DECLARE - result TEXT[]; - search TEXT[]; - for_postcode TEXT; - found INTEGER; - location RECORD; -BEGIN - - found := 1000; - search := ARRAY['ref']; - result := '{}'; - - select postcode from placex where place_id = for_place_id limit 1 into for_postcode; - - FOR location IN - select rank_address,name,distance,length(name::text) as namelength - from place_addressline join placex on (address_place_id = placex.place_id) - where place_addressline.place_id = for_place_id and rank_address in (5,11) - order by rank_address desc,rank_search desc,fromarea desc,distance asc,namelength desc - LOOP - IF array_upper(search, 1) IS NOT NULL AND array_upper(location.name, 1) IS NOT NULL THEN - FOR j IN 1..array_upper(search, 1) LOOP - FOR k IN 1..array_upper(location.name, 1) LOOP - IF (found > location.rank_address AND location.name[k].key = search[j] AND location.name[k].value != '') AND NOT result @> ARRAY[trim(location.name[k].value)] AND (for_postcode IS NULL OR location.name[k].value ilike for_postcode||'%') THEN - result[(100 - location.rank_address)] := trim(location.name[k].value); - found := location.rank_address; - END IF; - END LOOP; - END LOOP; - END IF; - END LOOP; - - RETURN array_to_string(result,', '); -END; -$$ -LANGUAGE plpgsql; - -CREATE OR REPLACE FUNCTION get_address_by_language(for_place_id BIGINT, languagepref TEXT[]) RETURNS TEXT +--housenumber only needed for tiger data +CREATE OR REPLACE FUNCTION get_address_by_language(for_place_id BIGINT, housenumber INTEGER, languagepref TEXT[]) RETURNS TEXT AS $$ DECLARE result TEXT[]; @@ -2362,7 +2327,7 @@ BEGIN result := '{}'; prevresult := ''; - FOR location IN select * from get_addressdata(for_place_id) where isaddress order by rank_address desc LOOP + FOR location IN select * from get_addressdata(for_place_id, housenumber) where isaddress order by rank_address desc LOOP currresult := trim(get_name_by_language(location.name, languagepref)); IF currresult != prevresult AND currresult IS NOT NULL AND result[(100 - location.rank_address)] IS NULL THEN result[(100 - location.rank_address)] := trim(get_name_by_language(location.name, languagepref)); @@ -2390,7 +2355,7 @@ create type addressline as ( distance FLOAT ); -CREATE OR REPLACE FUNCTION get_addressdata(in_place_id BIGINT) RETURNS setof addressline +CREATE OR REPLACE FUNCTION get_addressdata(in_place_id BIGINT, in_housenumber INTEGER) RETURNS setof addressline AS $$ DECLARE for_place_id BIGINT; @@ -2409,27 +2374,53 @@ DECLARE countryname HSTORE; hadcountry BOOLEAN; BEGIN + -- first query osmline (interpolation lines) + select parent_place_id, country_code, 30, postcode, null, 'place', 'house' from location_property_osmline + WHERE place_id = in_place_id AND in_housenumber>=startnumber AND in_housenumber <= endnumber + INTO for_place_id,searchcountrycode, searchrankaddress, searchpostcode, searchhousename, searchclass, searchtype; + IF for_place_id IS NOT NULL THEN + searchhousenumber = in_housenumber::text; + END IF; - select parent_place_id,'us', housenumber, 30, postcode, null, 'place', 'house' from location_property_tiger - WHERE place_id = in_place_id - INTO for_place_id,searchcountrycode, searchhousenumber, searchrankaddress, searchpostcode, searchhousename, searchclass, searchtype; + --then query tiger data + -- %NOTIGERDATA% IF 0 THEN + IF for_place_id IS NULL THEN + select parent_place_id,'us', 30, postcode, null, 'place', 'house' from location_property_tiger + WHERE place_id = in_place_id AND in_housenumber>=startnumber AND in_housenumber <= endnumber + INTO for_place_id,searchcountrycode, searchrankaddress, searchpostcode, searchhousename, searchclass, searchtype; + IF for_place_id IS NOT NULL THEN + searchhousenumber = in_housenumber::text; + END IF; + END IF; + -- %NOTIGERDATA% END IF; + -- %NOAUXDATA% IF 0 THEN IF for_place_id IS NULL THEN select parent_place_id,'us', housenumber, 30, postcode, null, 'place', 'house' from location_property_aux - WHERE place_id = in_place_id + WHERE place_id = in_place_id INTO for_place_id,searchcountrycode, searchhousenumber, searchrankaddress, searchpostcode, searchhousename, searchclass, searchtype; END IF; + -- %NOAUXDATA% END IF; + + -- postcode table + IF for_place_id IS NULL THEN + select parent_place_id, country_code, rank_address, postcode, 'place', 'postcode' + FROM location_postcode + WHERE place_id = in_place_id + INTO for_place_id, searchcountrycode, searchrankaddress, searchpostcode, searchclass, searchtype; + END IF; IF for_place_id IS NULL THEN - select parent_place_id, calculated_country_code, housenumber, rank_search, postcode, name, class, type from placex - WHERE place_id = in_place_id and rank_address = 30 + select parent_place_id, country_code, housenumber, rank_search, postcode, name, class, type from placex + WHERE place_id = in_place_id and rank_search > 27 INTO for_place_id, searchcountrycode, searchhousenumber, searchrankaddress, searchpostcode, searchhousename, searchclass, searchtype; END IF; IF for_place_id IS NULL THEN - for_place_id := in_place_id; - select calculated_country_code, housenumber, rank_search, postcode, null from placex where place_id = for_place_id - INTO searchcountrycode, searchhousenumber, searchrankaddress, searchpostcode, searchhousename; + select coalesce(linked_place_id, place_id), country_code, + housenumber, rank_search, postcode, null + from placex where place_id = in_place_id + INTO for_place_id, searchcountrycode, searchhousenumber, searchrankaddress, searchpostcode, searchhousename; END IF; --RAISE WARNING '% % % %',searchcountrycode, searchhousenumber, searchrankaddress, searchpostcode; @@ -2437,22 +2428,20 @@ BEGIN found := 1000; hadcountry := false; 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, true as fromarea, true as isaddress, + select placex.place_id, osm_type, osm_id, name, + class, type, admin_level, 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, country_code, postcode from placex where place_id = for_place_id LOOP --RAISE WARNING '%',location; - IF searchcountrycode IS NULL AND location.calculated_country_code IS NOT NULL THEN - searchcountrycode := location.calculated_country_code; + IF searchcountrycode IS NULL AND location.country_code IS NOT NULL THEN + searchcountrycode := location.country_code; END IF; - IF searchpostcode IS NOT NULL and location.type = 'postcode' THEN + IF location.type in ('postcode', 'postal_code') THEN location.isaddress := FALSE; - END IF; - IF location.rank_address = 4 AND location.isaddress THEN + ELSEIF location.rank_address = 4 THEN hadcountry := true; END IF; IF location.rank_address < 4 AND NOT hadcountry THEN @@ -2463,30 +2452,31 @@ BEGIN END IF; END IF; countrylocation := ROW(location.place_id, location.osm_type, location.osm_id, location.name, location.class, - location.type, location.admin_level, location.fromarea, location.isaddress, location.rank_address, + location.type, location.admin_level, true, location.isaddress, location.rank_address, location.distance)::addressline; RETURN NEXT countrylocation; found := location.rank_address; END LOOP; 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, + select placex.place_id, osm_type, osm_id, name, + 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,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) - and address_place_id != for_place_id - and (placex.calculated_country_code IS NULL OR searchcountrycode IS NULL OR placex.calculated_country_code = searchcountrycode) + and address_place_id != for_place_id and linked_place_id is null + and (placex.country_code IS NULL OR searchcountrycode IS NULL OR placex.country_code = searchcountrycode) order by rank_address desc,isaddress desc,fromarea desc,distance asc,rank_search desc LOOP --RAISE WARNING '%',location; - IF searchcountrycode IS NULL AND location.calculated_country_code IS NOT NULL THEN - searchcountrycode := location.calculated_country_code; + IF searchcountrycode IS NULL AND location.country_code IS NOT NULL THEN + searchcountrycode := location.country_code; END IF; - IF searchpostcode IS NOT NULL and location.type = 'postcode' THEN + IF location.type in ('postcode', 'postal_code') THEN location.isaddress := FALSE; END IF; IF location.rank_address = 4 AND location.isaddress THEN @@ -2522,7 +2512,6 @@ BEGIN IF searchhousename IS NOT NULL THEN location := ROW(in_place_id, null, null, searchhousename, searchclass, searchtype, null, true, true, 29, 0)::addressline; --- location := ROW(in_place_id, null, null, searchhousename, 'place', 'house_name', null, true, true, 29, 0)::addressline; RETURN NEXT location; END IF; @@ -2541,107 +2530,6 @@ END; $$ LANGUAGE plpgsql; -CREATE OR REPLACE FUNCTION get_place_boundingbox(search_place_id BIGINT) RETURNS place_boundingbox - AS $$ -DECLARE - result place_boundingbox; - numfeatures integer; -BEGIN - select * from place_boundingbox into result where place_id = search_place_id; - IF result.place_id IS NULL THEN --- remove isaddress = true because if there is a matching polygon it always wins - select count(*) from place_addressline where address_place_id = search_place_id into numfeatures; - insert into place_boundingbox select place_id, - ST_Y(ST_PointN(ExteriorRing(ST_Box2D(geometry)),4)),ST_Y(ST_PointN(ExteriorRing(ST_Box2D(geometry)),2)), - ST_X(ST_PointN(ExteriorRing(ST_Box2D(geometry)),1)),ST_X(ST_PointN(ExteriorRing(ST_Box2D(geometry)),3)), - numfeatures, ST_Area(geometry), - geometry as area from location_area where place_id = search_place_id; - select * from place_boundingbox into result where place_id = search_place_id; - END IF; - IF result.place_id IS NULL THEN --- TODO 0.0001 - insert into place_boundingbox select address_place_id, - min(ST_Y(ST_Centroid(geometry))) as minlon,max(ST_Y(ST_Centroid(geometry))) as maxlon, - min(ST_X(ST_Centroid(geometry))) as minlat,max(ST_X(ST_Centroid(geometry))) as maxlat, - count(*), ST_Area(ST_Buffer(ST_Convexhull(ST_Collect(geometry)),0.0001)) as area, - ST_Buffer(ST_Convexhull(ST_Collect(geometry)),0.0001) as boundary - from (select * from place_addressline where address_place_id = search_place_id order by cached_rank_address limit 4000) as place_addressline join placex using (place_id) - where address_place_id = search_place_id --- and (isaddress = true OR place_id = search_place_id) - and (st_length(geometry) < 0.01 or place_id = search_place_id) - group by address_place_id limit 1; - select * from place_boundingbox into result where place_id = search_place_id; - END IF; - return result; -END; -$$ -LANGUAGE plpgsql; - --- don't do the operation if it would be slow -CREATE OR REPLACE FUNCTION get_place_boundingbox_quick(search_place_id BIGINT) RETURNS place_boundingbox - AS $$ -DECLARE - result place_boundingbox; - numfeatures integer; - rank integer; -BEGIN - select * from place_boundingbox into result where place_id = search_place_id; - IF result IS NULL AND rank > 14 THEN - select count(*) from place_addressline where address_place_id = search_place_id and isaddress = true into numfeatures; - insert into place_boundingbox select place_id, - ST_Y(ST_PointN(ExteriorRing(ST_Box2D(geometry)),4)),ST_Y(ST_PointN(ExteriorRing(ST_Box2D(geometry)),2)), - ST_X(ST_PointN(ExteriorRing(ST_Box2D(geometry)),1)),ST_X(ST_PointN(ExteriorRing(ST_Box2D(geometry)),3)), - numfeatures, ST_Area(geometry), - geometry as area from location_area where place_id = search_place_id; - select * from place_boundingbox into result where place_id = search_place_id; - END IF; - IF result IS NULL THEN - select rank_search from placex where place_id = search_place_id into rank; - IF rank > 20 THEN --- TODO 0.0001 - insert into place_boundingbox select address_place_id, - min(ST_Y(ST_Centroid(geometry))) as minlon,max(ST_Y(ST_Centroid(geometry))) as maxlon, - min(ST_X(ST_Centroid(geometry))) as minlat,max(ST_X(ST_Centroid(geometry))) as maxlat, - count(*), ST_Area(ST_Buffer(ST_Convexhull(ST_Collect(geometry)),0.0001)) as area, - ST_Buffer(ST_Convexhull(ST_Collect(geometry)),0.0001) as boundary - from place_addressline join placex using (place_id) - where address_place_id = search_place_id - and (isaddress = true OR place_id = search_place_id) - and (st_length(geometry) < 0.01 or place_id = search_place_id) - group by address_place_id limit 1; - select * from place_boundingbox into result where place_id = search_place_id; - END IF; - END IF; - return result; -END; -$$ -LANGUAGE plpgsql; - -CREATE OR REPLACE FUNCTION update_place(search_place_id BIGINT) RETURNS BOOLEAN - AS $$ -DECLARE - result place_boundingbox; - numfeatures integer; -BEGIN - update placex set - name = place.name, - housenumber = place.housenumber, - street = place.street, - addr_place = place.addr_place, - isin = place.isin, - postcode = place.postcode, - country_code = place.country_code, - parent_place_id = null - from place - where placex.place_id = search_place_id - and place.osm_type = placex.osm_type and place.osm_id = placex.osm_id - and place.class = placex.class and place.type = placex.type; - update placex set indexed_status = 2 where place_id = search_place_id; - update placex set indexed_status = 0 where place_id = search_place_id; - return true; -END; -$$ -LANGUAGE plpgsql; CREATE OR REPLACE FUNCTION get_searchrank_label(rank INTEGER) RETURNS TEXT AS $$ @@ -2733,42 +2621,6 @@ END; $$ LANGUAGE plpgsql; -CREATE OR REPLACE FUNCTION get_word_suggestion(srcword TEXT) RETURNS TEXT - AS $$ -DECLARE - trigramtoken TEXT; - result TEXT; -BEGIN - - trigramtoken := regexp_replace(make_standard_name(srcword),E'([^0-9])\\1+',E'\\1','g'); - SELECT word FROM word WHERE word_trigram like ' %' and word_trigram % trigramtoken ORDER BY similarity(word_trigram, trigramtoken) DESC, word limit 1 into result; - - return result; -END; -$$ -LANGUAGE plpgsql; - -CREATE OR REPLACE FUNCTION get_word_suggestions(srcword TEXT) RETURNS TEXT[] - AS $$ -DECLARE - trigramtoken TEXT; - result TEXT[]; - r RECORD; -BEGIN - - trigramtoken := regexp_replace(make_standard_name(srcword),E'([^0-9])\\1+',E'\\1','g'); - - FOR r IN SELECT word,similarity(word_trigram, trigramtoken) as score FROM word - WHERE word_trigram like ' %' and word_trigram % trigramtoken ORDER BY similarity(word_trigram, trigramtoken) DESC, word limit 4 - LOOP - result[coalesce(array_upper(result,1)+1,1)] := r.word; - END LOOP; - - return result; -END; -$$ -LANGUAGE plpgsql; - CREATE OR REPLACE FUNCTION aux_create_property(pointgeo GEOMETRY, in_housenumber TEXT, in_street TEXT, in_isin TEXT, in_postcode TEXT, in_countrycode char(2)) RETURNS INTEGER AS $$ @@ -2785,7 +2637,7 @@ 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)); @@ -2798,16 +2650,14 @@ BEGIN IF out_parent_place_id IS NULL THEN FOR location IN SELECT place_id FROM getNearestRoadFeature(out_partition, place_centroid) LOOP out_parent_place_id := location.place_id; - END LOOP; + END LOOP; END IF; out_postcode := in_postcode; IF out_postcode IS NULL THEN SELECT postcode from placex where place_id = out_parent_place_id INTO out_postcode; END IF; - IF out_postcode IS NULL THEN - out_postcode := getNearestPostcode(out_partition, place_centroid); - END IF; + -- XXX look into postcode table newpoints := 0; insert into location_property_aux (place_id, partition, parent_place_id, housenumber, postcode, centroid) @@ -2900,14 +2750,14 @@ 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 @@ -3059,14 +2909,15 @@ 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 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 or (rank >= 16 and addr_place 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 address ? 'place')); 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 or (rank >= 16 and addr_place 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 address ? 'place')); END LOOP; ELSE diameter := 0; @@ -3091,7 +2942,7 @@ BEGIN 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); + 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 address ? 'place'); 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);