-- score integer
--);
-
CREATE OR REPLACE FUNCTION getclasstypekey(c text, t text) RETURNS TEXT
AS $$
DECLARE
NEWgeometry geometry;
BEGIN
-- RAISE WARNING '%',place;
- NEWgeometry := place;
- IF ST_IsEmpty(NEWgeometry) OR NOT ST_IsValid(NEWgeometry) OR ST_X(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') THEN
- NEWgeometry := ST_buffer(NEWgeometry,0);
- IF ST_IsEmpty(NEWgeometry) OR NOT ST_IsValid(NEWgeometry) OR ST_X(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') THEN
- RETURN 0;
- END IF;
- END IF;
- RETURN (partition*1000000) + (500-ST_X(ST_Centroid(NEWgeometry))::integer)*1000 + (500-ST_Y(ST_Centroid(NEWgeometry))::integer);
-END;
-$$
-LANGUAGE plpgsql IMMUTABLE;
-
-CREATE OR REPLACE FUNCTION debug_geometry_sector(osmid integer, place geometry) RETURNS INTEGER
- AS $$
-DECLARE
- NEWgeometry geometry;
-BEGIN
--- RAISE WARNING '%',osmid;
- IF osmid = 61315 THEN
- return null;
- END IF;
- NEWgeometry := place;
- IF ST_IsEmpty(NEWgeometry) OR NOT ST_IsValid(NEWgeometry) OR ST_X(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') THEN
- NEWgeometry := ST_buffer(NEWgeometry,0);
- IF ST_IsEmpty(NEWgeometry) OR NOT ST_IsValid(NEWgeometry) OR ST_X(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') THEN
- RETURN NULL;
- END IF;
- END IF;
- RETURN (500-ST_X(ST_Centroid(NEWgeometry))::integer)*1000 + (500-ST_Y(ST_Centroid(NEWgeometry))::integer);
+ NEWgeometry := ST_PointOnSurface(place);
+-- IF ST_IsEmpty(NEWgeometry) OR NOT ST_IsValid(NEWgeometry) OR ST_X(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') THEN
+-- NEWgeometry := ST_buffer(NEWgeometry,0);
+-- IF ST_IsEmpty(NEWgeometry) OR NOT ST_IsValid(NEWgeometry) OR ST_X(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') OR ST_Y(ST_Centroid(NEWgeometry))::text in ('NaN','Infinity','-Infinity') THEN
+-- RETURN 0;
+-- END IF;
+-- END IF;
+ RETURN (partition*1000000) + (500-ST_X(NEWgeometry)::integer)*1000 + (500-ST_Y(NEWgeometry)::integer);
END;
$$
LANGUAGE plpgsql IMMUTABLE;
$$
LANGUAGE 'plpgsql' IMMUTABLE;
+-- returns NULL if the word is too common
CREATE OR REPLACE FUNCTION getorcreate_word_id(lookup_word TEXT)
RETURNS INTEGER
AS $$
DECLARE
lookup_token TEXT;
return_word_id INTEGER;
+ count INTEGER;
BEGIN
lookup_token := trim(lookup_word);
- SELECT min(word_id) FROM word WHERE word_token = lookup_token and class is null and type is null into return_word_id;
+ SELECT min(word_id), max(search_name_count) FROM word WHERE word_token = lookup_token and class is null and type is null into return_word_id, count;
IF return_word_id IS NULL THEN
return_word_id := nextval('seq_word');
INSERT INTO word VALUES (return_word_id, lookup_token, regexp_replace(lookup_token,E'([^0-9])\\1+',E'\\1','g'), null, null, null, null, 0, null);
+ ELSE
+ IF count > get_maxwordfreq() THEN
+ return_word_id := NULL;
+ END IF;
END IF;
RETURN return_word_id;
END;
END IF;
r := a;
FOR i IN 1..array_upper(b, 1) LOOP
- IF NOT (ARRAY[b[i]] && r) THEN
+ IF NOT (ARRAY[b[i]] <@ r) THEN
r := r || b[i];
END IF;
END LOOP;
$$
LANGUAGE plpgsql IMMUTABLE;
+CREATE OR REPLACE FUNCTION create_country(src HSTORE, lookup_country_code varchar(2)) RETURNS VOID
+ AS $$
+DECLARE
+ s TEXT;
+ w INTEGER;
+ words TEXT[];
+ item RECORD;
+ j INTEGER;
+BEGIN
+ FOR item IN SELECT (each(src)).* LOOP
+
+ s := make_standard_name(item.value);
+ w := getorcreate_country(s, lookup_country_code);
+
+ words := regexp_split_to_array(item.value, E'[,;()]');
+ IF array_upper(words, 1) != 1 THEN
+ FOR j IN 1..array_upper(words, 1) LOOP
+ s := make_standard_name(words[j]);
+ IF s != '' THEN
+ w := getorcreate_country(s, lookup_country_code);
+ END IF;
+ END LOOP;
+ END IF;
+ END LOOP;
+END;
+$$
+LANGUAGE plpgsql;
+
CREATE OR REPLACE FUNCTION make_keywords(src HSTORE) RETURNS INTEGER[]
AS $$
DECLARE
s := make_standard_name(item.value);
w := getorcreate_name_id(s, item.value);
- result := result | w;
+
+ IF not(ARRAY[w] <@ result) THEN
+ result := result || w;
+ END IF;
words := string_to_array(s, ' ');
IF array_upper(words, 1) IS NOT NULL THEN
FOR j IN 1..array_upper(words, 1) LOOP
IF (words[j] != '') THEN
w = getorcreate_word_id(words[j]);
- IF NOT (ARRAY[w] && result) THEN
- result := result | w;
+ IF w IS NOT NULL AND NOT (ARRAY[w] <@ result) THEN
+ result := result || w;
END IF;
END IF;
END LOOP;
s := make_standard_name(words[j]);
IF s != '' THEN
w := getorcreate_word_id(s);
- IF NOT (ARRAY[w] && result) THEN
- result := result | w;
+ IF w IS NOT NULL AND NOT (ARRAY[w] <@ result) THEN
+ result := result || w;
END IF;
END IF;
END LOOP;
s := make_standard_name(s);
IF s != '' THEN
w := getorcreate_name_id(s, item.value);
- IF NOT (ARRAY[w] && result) THEN
- result := result | w;
+ IF NOT (ARRAY[w] <@ result) THEN
+ result := result || w;
END IF;
END IF;
END IF;
result := '{}'::INTEGER[];
s := make_standard_name(src);
- w := getorcreate_name_id(s);
+ w := getorcreate_name_id(s, src);
- IF NOT (ARRAY[w] && result) THEN
+ IF NOT (ARRAY[w] <@ result) THEN
result := result || w;
END IF;
FOR j IN 1..array_upper(words, 1) LOOP
IF (words[j] != '') THEN
w = getorcreate_word_id(words[j]);
- IF NOT (ARRAY[w] && result) THEN
+ IF w IS NOT NULL AND NOT (ARRAY[w] <@ result) THEN
result := result || w;
END IF;
END IF;
END LOOP;
END IF;
+ words := regexp_split_to_array(src, E'[,;()]');
+ IF array_upper(words, 1) != 1 THEN
+ FOR j IN 1..array_upper(words, 1) LOOP
+ s := make_standard_name(words[j]);
+ IF s != '' THEN
+ w := getorcreate_word_id(s);
+ IF w IS NOT NULL AND NOT (ARRAY[w] <@ result) THEN
+ result := result || w;
+ END IF;
+ END IF;
+ END LOOP;
+ END IF;
+
+ s := regexp_replace(src, '市$', '');
+ IF s != src THEN
+ s := make_standard_name(s);
+ IF s != '' THEN
+ w := getorcreate_name_id(s, src);
+ IF NOT (ARRAY[w] <@ result) THEN
+ result := result || w;
+ END IF;
+ END IF;
+ END IF;
+
RETURN result;
END;
$$
place_centre GEOMETRY;
nearcountry RECORD;
BEGIN
- place_centre := ST_Centroid(place);
+ place_centre := ST_PointOnSurface(place);
+
+--DEBUG: RAISE WARNING 'get_country_code, start: %', ST_AsText(place_centre);
---RAISE WARNING 'start: %', ST_AsText(place_centre);
+--DEBUG: RAISE WARNING 'osm fallback: %', ST_AsText(place_centre);
- -- Try for a OSM polygon first
- FOR nearcountry IN select country_code from location_area_country where country_code is not null and not isguess and st_contains(geometry, place_centre) limit 1
+ -- Try for OSM fallback data
+ -- The order is to deal with places like HongKong that are 'states' within another polygon
+ FOR nearcountry IN select country_code from country_osm_grid where st_covers(geometry, place_centre) order by area asc limit 1
LOOP
RETURN nearcountry.country_code;
END LOOP;
---RAISE WARNING 'osm fallback: %', ST_AsText(place_centre);
-
- -- Try for OSM fallback data
- FOR nearcountry IN select country_code from country_osm_grid where st_contains(geometry, place_centre) limit 1
+ -- Try for a OSM polygon
+ FOR nearcountry IN select country_code from location_area_country where country_code is not null and not isguess and st_covers(geometry, place_centre) limit 1
LOOP
RETURN nearcountry.country_code;
END LOOP;
---RAISE WARNING 'natural earth: %', ST_AsText(place_centre);
+--DEBUG: RAISE WARNING 'natural earth: %', ST_AsText(place_centre);
- -- Natural earth data (first fallback)
- FOR nearcountry IN select country_code from country_naturalearthdata where st_contains(geometry, place_centre) limit 1
+ -- Natural earth data
+ FOR nearcountry IN select country_code from country_naturalearthdata where st_covers(geometry, place_centre) limit 1
LOOP
RETURN nearcountry.country_code;
END LOOP;
- -- Natural earth data (first fallback)
- FOR nearcountry IN select country_code from country_naturalearthdata where st_distance(geometry, place_centre) < 0.5 limit 1
+--DEBUG: 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
LOOP
RETURN nearcountry.country_code;
END LOOP;
---RAISE WARNING 'in country: %', ST_AsText(place_centre);
+--DEBUG: RAISE WARNING 'near natural earth: %', ST_AsText(place_centre);
- -- WorldBoundaries data (second fallback - think there might be something broken in this data)
- FOR nearcountry IN select country_code from country where st_contains(geometry, place_centre) limit 1
+ -- Natural earth data
+ FOR nearcountry IN select country_code from country_naturalearthdata where st_dwithin(geometry, place_centre, 0.5) limit 1
LOOP
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;
+-- 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_language_codes(search_country_code VARCHAR(2)) RETURNS TEXT[]
+ AS $$
+DECLARE
+ nearcountry RECORD;
+BEGIN
+ FOR nearcountry IN select country_default_language_codes from country_name where country_code = search_country_code limit 1
+ LOOP
+ RETURN lower(nearcountry.country_default_language_codes);
+ END LOOP;
+ RETURN NULL;
+END;
+$$
+LANGUAGE plpgsql IMMUTABLE;
+
CREATE OR REPLACE FUNCTION get_partition(place geometry, in_country_code VARCHAR(10)) RETURNS INTEGER
AS $$
DECLARE
$$
LANGUAGE plpgsql IMMUTABLE;
-CREATE OR REPLACE FUNCTION delete_location(OLD_place_id INTEGER) RETURNS BOOLEAN
+CREATE OR REPLACE FUNCTION delete_location(OLD_place_id BIGINT) RETURNS BOOLEAN
AS $$
DECLARE
BEGIN
LANGUAGE plpgsql;
CREATE OR REPLACE FUNCTION add_location(
- place_id INTEGER,
+ place_id BIGINT,
country_code varchar(2),
partition INTEGER,
keywords INTEGER[],
DECLARE
locationid INTEGER;
isarea BOOLEAN;
- xmin INTEGER;
- ymin INTEGER;
- xmax INTEGER;
- ymax INTEGER;
- lon INTEGER;
- lat INTEGER;
centroid GEOMETRY;
- secgeo GEOMETRY;
- secbox GEOMETRY;
diameter FLOAT;
x BOOLEAN;
+ splitGeom RECORD;
+ secgeo GEOMETRY;
BEGIN
IF rank_search > 25 THEN
-- RAISE WARNING 'Adding location with rank > 25 (% rank %)', place_id, rank_search;
- x := deleteLocationArea(partition, place_id);
+ x := deleteLocationArea(partition, place_id, rank_search);
isarea := false;
IF (ST_GeometryType(geometry) in ('ST_Polygon','ST_MultiPolygon') AND ST_IsValid(geometry)) THEN
isArea := true;
centroid := ST_Centroid(geometry);
- xmin := floor(st_xmin(geometry));
- xmax := ceil(st_xmax(geometry));
- ymin := floor(st_ymin(geometry));
- ymax := ceil(st_ymax(geometry));
-
- IF xmin = xmax OR ymin = ymax OR (xmax-xmin < 2 AND ymax-ymin < 2) THEN
- x := insertLocationAreaLarge(partition, place_id, country_code, keywords, rank_search, rank_address, false, centroid, geometry);
- ELSE
--- RAISE WARNING 'Spliting geometry: % to %, % to %', xmin, xmax, ymin, ymax;
- FOR lon IN xmin..(xmax-1) LOOP
- FOR lat IN ymin..(ymax-1) LOOP
- secbox := ST_SetSRID(ST_MakeBox2D(ST_Point(lon,lat),ST_Point(lon+1,lat+1)),4326);
- IF st_intersects(geometry, secbox) THEN
- secgeo := st_intersection(geometry, secbox);
- IF NOT ST_IsEmpty(secgeo) AND ST_GeometryType(secgeo) in ('ST_Polygon','ST_MultiPolygon') THEN
- x := insertLocationAreaLarge(partition, place_id, country_code, keywords, rank_search, rank_address, false, centroid, secgeo);
- END IF;
- END IF;
- END LOOP;
- END LOOP;
- END IF;
+ 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);
+ END LOOP;
ELSEIF rank_search < 26 THEN
CREATE OR REPLACE FUNCTION update_location(
partition INTEGER,
- place_id INTEGER,
+ place_id BIGINT,
place_country_code varchar(2),
name hstore,
rank_search INTEGER,
DECLARE
b BOOLEAN;
BEGIN
- b := deleteLocationArea(partition, place_id);
+ b := deleteLocationArea(partition, place_id, rank_search);
-- result := add_location(NEW.place_id, NEW.country_code, NEW.partition, name_vector, NEW.rank_search, NEW.rank_address, NEW.geometry);
RETURN add_location(place_id, place_country_code, name, rank_search, rank_address, geometry);
END;
$$
LANGUAGE plpgsql;
-CREATE OR REPLACE FUNCTION search_name_add_words(parent_place_id INTEGER, to_add INTEGER[])
+CREATE OR REPLACE FUNCTION search_name_add_words(parent_place_id BIGINT, to_add INTEGER[])
RETURNS BOOLEAN
AS $$
DECLARE
and search_name.place_id = place_addressline.place_id
LOOP
delete from search_name where place_id = childplace.place_id;
- childplace.nameaddress_vector := uniq(sort_asc(childplace.nameaddress_vector + to_add));
- IF childplace.place_id = parent_place_id THEN
- childplace.name_vector := uniq(sort_asc(childplace.name_vector + to_add));
+ IF not (ARRAY[to_add] <@ childplace.nameaddress_vector) THEN
+ childplace.nameaddress_vector := childplace.nameaddress_vector || to_add;
+ END IF;
+ IF childplace.place_id = parent_place_id and not (ARRAY[to_add] <@ childplace.name_vector) THEN
+ childplace.name_vector := childplace.name_vector || to_add;
END IF;
insert into search_name (place_id, search_rank, address_rank, country_code, name_vector, nameaddress_vector, centroid)
values (childplace.place_id, childplace.search_rank, childplace.address_rank, childplace.country_code,
$$
LANGUAGE plpgsql;
-CREATE OR REPLACE FUNCTION update_location_nameonly(partition INTEGER, OLD_place_id INTEGER, name hstore) RETURNS BOOLEAN
+CREATE OR REPLACE FUNCTION update_location_nameonly(partition INTEGER, OLD_place_id BIGINT, name hstore) RETURNS BOOLEAN
AS $$
DECLARE
newkeywords INTEGER[];
LANGUAGE plpgsql;
-CREATE OR REPLACE FUNCTION create_interpolation(wayid INTEGER, interpolationtype TEXT) RETURNS INTEGER
+CREATE OR REPLACE FUNCTION create_interpolation(wayid BIGINT, interpolationtype TEXT) RETURNS INTEGER
AS $$
DECLARE
originalnumberrange INTEGER;
housenum INTEGER;
linegeo GEOMETRY;
- search_place_id INTEGER;
+ search_place_id BIGINT;
defpostalcode TEXT;
havefirstpoint BOOLEAN;
-- null record of right type
select * from placex where osm_type = 'N' and osm_id = waynodes[nodeidpos]::INTEGER and type = 'house' limit 1 INTO nextnode;
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;
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 THEN
+ 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;
country_code VARCHAR(2);
default_language VARCHAR(10);
diameter FLOAT;
+ classtable TEXT;
BEGIN
--- RAISE WARNING '%',NEW.osm_id;
+ --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
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;
+ RAISE WARNING 'invalid geometry %',NEW.osm_id;
RETURN NULL;
-- Dead code
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;
+ 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.country_code := lower(get_country_code(NEW.geometry, NEW.country_code));
- NEW.partition := get_partition(NEW.geometry, NEW.country_code);
+ NEW.calculated_country_code := lower(get_country_code(NEW.geometry, NEW.country_code));
+
+ NEW.partition := get_partition(NEW.geometry, NEW.calculated_country_code);
NEW.geometry_sector := geometry_sector(NEW.partition, NEW.geometry);
-- copy 'name' to or from the default language (if there is a default language)
- IF NEW.name is not null AND array_upper(%#NEW.name,1) > 1 THEN
- default_language := get_country_language_code(NEW.country_code);
+ 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 || (('name:'||default_language) => (NEW.name -> 'name'));
+ 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 || ('name' => (NEW.name -> 'name:'||default_language));
+ NEW.name := NEW.name || hstore('name', (NEW.name -> ('name:'||default_language)));
END IF;
END IF;
END IF;
NEW.rank_address := NEW.rank_search;
-- By doing in postgres we have the country available to us - currently only used for postcode
- IF NEW.class = 'place' THEN
+ IF NEW.class in ('place','boundary') AND NEW.type in ('postcode','postal_code') THEN
+
+ IF NEW.postcode IS NULL 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;
+
+ 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,}$' THEN
+ NEW.rank_search := 21;
+ NEW.rank_address := 11;
+ END IF;
+ END IF;
+ END IF;
+
+ ELSEIF NEW.class = 'place' THEN
IF NEW.type in ('continent') THEN
NEW.rank_search := 2;
NEW.rank_address := NEW.rank_search;
+ NEW.calculated_country_code := NULL;
ELSEIF NEW.type in ('sea') THEN
NEW.rank_search := 2;
NEW.rank_address := 0;
+ NEW.calculated_country_code := NULL;
ELSEIF NEW.type in ('country') THEN
NEW.rank_search := 4;
NEW.rank_address := NEW.rank_search;
NEW.rank_search := 8;
NEW.rank_address := NEW.rank_search;
ELSEIF NEW.type in ('region') THEN
- NEW.rank_search := 10;
- NEW.rank_address := NEW.rank_search;
+ NEW.rank_search := 18; -- dropped from previous value of 10
+ NEW.rank_address := 0; -- So badly miss-used that better to just drop it!
ELSEIF NEW.type in ('county') THEN
NEW.rank_search := 12;
NEW.rank_address := NEW.rank_search;
ELSEIF NEW.type in ('suburb','croft','subdivision') THEN
NEW.rank_search := 20;
NEW.rank_address := NEW.rank_search;
- ELSEIF NEW.type in ('farm','locality','islet') THEN
+ ELSEIF NEW.type in ('farm','locality','islet','isolated_dwelling','mountain_pass') THEN
NEW.rank_search := 20;
NEW.rank_address := 0;
+ -- Irish townlands, tagged as place=locality and locality=townland
+ IF (NEW.extratags -> 'locality') = 'townland' THEN
+ NEW.rank_address := 20;
+ END IF;
ELSEIF NEW.type in ('hall_of_residence','neighbourhood','housing_estate','nature_reserve') THEN
NEW.rank_search := 22;
NEW.rank_address := 22;
- ELSEIF NEW.type in ('postcode') THEN
-
- NEW.name := 'ref'=>NEW.postcode;
-
- IF NEW.country_code = 'gb' THEN
-
- IF NEW.postcode ~ '^([A-Z][A-Z]?[0-9][0-9A-Z]? [0-9][A-Z][A-Z])$' 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.country_code = 'de' THEN
-
- IF NEW.postcode ~ '^([0-9]{5})$' THEN
- NEW.rank_search := 21;
- NEW.rank_address := 11;
- END IF;
-
- ELSE
- -- Guess at the postcode format and coverage (!)
- IF upper(NEW.postcode) ~ '^[A-Z0-9]{1,5}$' 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;
-
ELSEIF NEW.type in ('airport','street') THEN
NEW.rank_search := 26;
NEW.rank_address := NEW.rank_search;
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;
+ NEW.rank_address := 0;
-- 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 = 'natural' and NEW.type in ('coastline') THEN
RETURN NULL;
- ELSEIF NEW.class = 'natural' and NEW.type in ('peak','volcano') THEN
- NEW.rank_search := 18;
- NEW.rank_address := 0;
END IF;
END IF;
END IF;
IF (NEW.extratags -> 'capital') = 'yes' THEN
- NEW.rank_search := NEW.rank_search -1;
+ NEW.rank_search := NEW.rank_search - 1;
+ END IF;
+
+ -- a country code make no sense below rank 4 (country)
+ IF NEW.rank_address < 4 THEN
+ NEW.calculated_country_code := NULL;
END IF;
-- Block import below rank 22
-- RETURN NULL;
-- END IF;
- RETURN NEW; -- The following is not needed until doing diff updates, and slows the main index process down
+ --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
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
- update placex set indexed_status = 2 where (ST_Contains(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';
- update placex set indexed_status = 2 where (ST_Contains(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';
+
+ -- 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);
+ 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);
END IF;
ELSE
-- mark nearby items for re-indexing, where 'nearby' depends on the features rank_search and is a complete guess :(
END IF;
IF diameter > 0 THEN
-- RAISE WARNING 'placex point insert: % % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type,diameter;
- update placex set indexed_status = 2 where indexed_status = 0 and rank_search > NEW.rank_search and ST_DWithin(placex.geometry, NEW.geometry, diameter);
+ 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);
+ 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;
+ -- 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;
+ IF result THEN
+ EXECUTE 'INSERT INTO ' || classtable::regclass || ' (place_id, centroid) VALUES ($1,$2)'
+ USING NEW.place_id, ST_Centroid(NEW.geometry);
+ END IF;
+
+
-- IF NEW.rank_search < 26 THEN
-- RAISE WARNING 'placex insert: % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type;
-- END IF;
location RECORD;
way RECORD;
relation RECORD;
+ relation_members TEXT[];
+ relMember RECORD;
+ linkedplacex RECORD;
search_diameter FLOAT;
search_prevdiameter FLOAT;
search_maxrank INTEGER;
address_maxrank INTEGER;
address_street_word_id INTEGER;
- parent_place_id_rank INTEGER;
+ parent_place_id_rank BIGINT;
isin TEXT[];
isin_tokens INT[];
location_rank_search INTEGER;
location_distance FLOAT;
+ location_parent GEOMETRY;
+ location_isaddress BOOLEAN;
tagpairid INTEGER;
+ default_language TEXT;
name_vector INTEGER[];
nameaddress_vector INTEGER[];
+ linked_node_id BIGINT;
+
result BOOLEAN;
BEGIN
+ -- deferred delete
+ IF OLD.indexed_status = 100 THEN
+ --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;
+
+ IF NEW.indexed_status != 0 OR OLD.indexed_status = 0 OR NEW.linked_place_id is not null THEN
+ RETURN NEW;
+ END IF;
+
+ --DEBUG: RAISE WARNING 'placex_update % %',NEW.osm_type,NEW.osm_id;
+
--RAISE WARNING '%',NEW.place_id;
--RAISE WARNING '%', NEW;
RETURN NEW;
END IF;
- -- defered delete
- IF OLD.indexed_status = 100 THEN
- delete from placex where osm_type = OLD.osm_type and osm_id = OLD.osm_id and class = OLD.class and type = OLD.type;
- RETURN NULL;
- END IF;
-
- IF NEW.indexed_status = 0 and OLD.indexed_status != 0 THEN
+ IF OLD.indexed_status != 0 THEN
+ --DEBUG: RAISE WARNING 'placex_update_0 % %',NEW.osm_type,NEW.osm_id;
NEW.indexed_date = now();
RETURN NEW;
END IF;
- 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);
+ IF OLD.indexed_status > 0 THEN
+ result := deleteSearchName(NEW.partition, NEW.place_id);
+ DELETE FROM place_addressline WHERE place_id = NEW.place_id;
+ DELETE FROM place_boundingbox where place_id = NEW.place_id;
+ result := deleteRoad(NEW.partition, NEW.place_id);
+ result := deleteLocationArea(NEW.partition, NEW.place_id, NEW.rank_search);
+ UPDATE placex set linked_place_id = null where linked_place_id = NEW.place_id;
+ END IF;
- -- reclaculate country and partition (should probably have a country_code and calculated_country_code as seperate fields)
- SELECT country_code from place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type INTO NEW.country_code;
- NEW.country_code := lower(get_country_code(NEW.geometry, NEW.country_code));
- NEW.partition := get_partition(NEW.geometry, NEW.country_code);
- NEW.geometry_sector := geometry_sector(NEW.partition, NEW.geometry);
+ -- Speed up searches - just use the centroid of the feature
+ -- cheaper but less acurate
+ place_centroid := ST_PointOnSurface(NEW.geometry);
+ NEW.centroid := null;
+
+ -- reclaculate country and partition
+ IF NEW.rank_search >= 4 THEN
+ --NEW.calculated_country_code := lower(get_country_code(NEW.geometry, NEW.country_code));
+ NEW.calculated_country_code := lower(get_country_code(place_centroid));
+ 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;
- -- Speed up searches - just use the centroid of the feature
- -- cheaper but less acurate
- place_centroid := ST_Centroid(NEW.geometry);
+ -- 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;
+ END IF;
+ END IF;
-- Initialise the name vector using our name
name_vector := make_keywords(NEW.name);
address_havelevel[i] := false;
END LOOP;
---RAISE WARNING '% %', NEW.place_id, NEW.rank_search;
+ 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;
+
+--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
--RAISE WARNING 'finding street for %', NEW;
+ -- We won't get a better centroid, besides these places are too small to care
+ NEW.centroid := place_centroid;
+
NEW.parent_place_id := null;
-- to do that we have to find our parent road
IF NEW.parent_place_id IS NULL AND NEW.osm_type = 'N' THEN
-- Is this node part of a relation?
- FOR relation IN select * from planet_osm_rels where parts @> ARRAY[NEW.osm_id::integer] and members @> ARRAY['n'||NEW.osm_id]
+ 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
--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::integer] LOOP
+ 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
END IF;
-- Is the WAY part of a relation
- FOR relation IN select * from planet_osm_rels where parts @> ARRAY[location.osm_id::integer] 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)::integer
- and rank_search = 26 INTO NEW.parent_place_id;
+ 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)::integer
+ and rank_search = 26 INTO NEW.parent_place_id;
+ END IF;
+ END LOOP;
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 location.street IS NOT NULL THEN
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::integer] and members @> ARRAY['w'||NEW.osm_id]
+ FOR relation IN select * from planet_osm_rels where parts @> ARRAY[NEW.osm_id] and members @> ARRAY['w'||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
-- Get the details of the parent road
select * from search_name where place_id = NEW.parent_place_id INTO location;
- NEW.country_code := location.country_code;
+ NEW.calculated_country_code := location.country_code;
--RAISE WARNING '%', NEW.name;
-- If there is no name it isn't searchable, don't bother to create a search record
-- 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 NEW.rank_search <= 25 THEN
- result := add_location(NEW.place_id, NEW.country_code, NEW.partition, name_vector, NEW.rank_search, NEW.rank_address, NEW.geometry);
+ result := add_location(NEW.place_id, NEW.calculated_country_code, NEW.partition, name_vector, NEW.rank_search, NEW.rank_address, NEW.geometry);
END IF;
- result := insertSearchName(NEW.partition, NEW.place_id, NEW.country_code, name_vector, nameaddress_vector, NEW.rank_search, NEW.rank_address, NEW.importance, place_centroid);
+ result := insertSearchName(NEW.partition, NEW.place_id, NEW.calculated_country_code, name_vector, nameaddress_vector, NEW.rank_search, NEW.rank_address, NEW.importance, place_centroid);
return NEW;
END IF;
END IF;
+-- RAISE WARNING ' INDEXING Started:';
-- RAISE WARNING ' INDEXING: %',NEW;
+ 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;
+
+-- 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
+
+ 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
+
+ -- 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 label name, re-init word vector
+ IF NOT linkedPlacex.name IS NULL THEN
+ NEW.name := linkedPlacex.name || NEW.name;
+ name_vector := make_keywords(NEW.name);
+ END IF;
+
+ -- merge in extra tags
+ IF NOT linkedPlacex.extratags IS NULL THEN
+ NEW.extratags := linkedPlacex.extratags || NEW.extratags;
+ 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;
+
+ END LOOP;
+
+ IF NEW.centroid IS NULL THEN
+
+ FOR relMember IN select get_osm_rel_members(relation_members,ARRAY['admin_center','admin_centre']) as member 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 order by rank_search desc limit 1 LOOP
+
+ -- 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
+
+
+ -- 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 extra tags
+ IF NOT linkedPlacex.extratags IS NULL THEN
+ NEW.extratags := linkedPlacex.extratags || NEW.extratags;
+ 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;
+
+ -- keep a note of the node id in case we need it for wikipedia in a bit
+ linked_node_id := linkedPlacex.osm_id;
+ 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
+ NEW.name := linkedPlacex.name || NEW.name;
+ name_vector := make_keywords(NEW.name);
+
+ -- 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;
+
+ -- 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;
+ 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;
+
+ -- 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;
+
NEW.parent_place_id = 0;
parent_place_id_rank = 0;
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 THEN
- isin_tokens := isin_tokens + address_street_word_id;
+ 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;
END LOOP;
END IF;
- isin_tokens := uniq(sort(isin_tokens));
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 THEN
- isin_tokens := isin_tokens + address_street_word_id;
+ 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;
END LOOP;
END IF;
- isin_tokens := uniq(sort(isin_tokens));
END IF;
---RAISE WARNING 'ISIN: %', isin_tokens;
+
+ -- 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;
+ END LOOP;
+ END IF;
+
+-- RAISE WARNING 'ISIN: %', isin_tokens;
-- Process area matches
- location_rank_search := 100;
+ location_rank_search := 0;
location_distance := 0;
---RAISE WARNING ' getNearFeatures(%,%,%,%)',NEW.partition, place_centroid, search_maxrank, isin_tokens;
+ 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
--RAISE WARNING ' AREA: %',location;
- IF location.rank_search < location_rank_search THEN
- location_rank_search := location.rank_search;
+ IF location.rank_address != location_rank_search THEN
+ location_rank_search := location.rank_address;
location_distance := location.distance * 1.5;
END IF;
- IF location.distance < location_distance THEN
+ 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);
+ END IF;
+ -- RAISE WARNING '% isaddress: %', location.place_id, location_isaddress;
-- 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, NOT address_havelevel[location.rank_address], location.distance, location.rank_address);
- address_havelevel[location.rank_address] := true;
+ IF location.rank_search > 4 THEN
+ nameaddress_vector := array_merge(nameaddress_vector, location.keywords::integer[]);
+ END IF;
+ INSERT INTO place_addressline VALUES (NEW.place_id, location.place_id, true, location_isaddress, location.distance, location.rank_address);
+
+ IF location_isaddress THEN
+ address_havelevel[location.rank_address] := true;
+ IF NOT location.isguess THEN
+ SELECT geometry FROM placex WHERE place_id = location.place_id INTO location_parent;
+ END IF;
+ END IF;
+
+--RAISE WARNING ' Terms: (%) %',location, nameaddress_vector;
IF location.rank_address > parent_place_id_rank THEN
NEW.parent_place_id = location.place_id;
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
- FOR location IN SELECT * from getNearestNamedFeature(NEW.partition, place_centroid, search_maxrank, isin_tokens[i]) LOOP
+ FOR location IN SELECT * from getNearestNamedFeature(NEW.partition, place_centroid, search_maxrank, isin_tokens[i]) LOOP
---RAISE WARNING ' ISIN: %',location;
+ --RAISE WARNING ' ISIN: %',location;
- 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_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 LOOP;
+ END IF;
END LOOP;
END IF;
-- for long ways we should add search terms for the entire length
IF st_length(NEW.geometry) > 0.05 THEN
- location_rank_search := 100;
+ location_rank_search := 0;
location_distance := 0;
FOR location IN SELECT * from getNearFeatures(NEW.partition, NEW.geometry, search_maxrank, isin_tokens) LOOP
- IF location.rank_search < location_rank_search THEN
- location_rank_search := location.rank_search;
+ IF location.rank_address != location_rank_search THEN
+ location_rank_search := location.rank_address;
location_distance := location.distance * 1.5;
END IF;
- IF location.distance < location_distance THEN
+ IF location.rank_search > 4 AND location.distance < location_distance THEN
-- Add it to the list of search terms
nameaddress_vector := array_merge(nameaddress_vector, location.keywords::integer[]);
IF NEW.name IS NOT NULL THEN
IF NEW.rank_search <= 25 THEN
- result := add_location(NEW.place_id, NEW.country_code, NEW.partition, name_vector, NEW.rank_search, NEW.rank_address, NEW.geometry);
+ 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.rank_search between 26 and 27 and NEW.class = 'highway' THEN
- result := insertLocationRoad(NEW.partition, NEW.place_id, NEW.country_code, NEW.geometry);
+ result := insertLocationRoad(NEW.partition, NEW.place_id, NEW.calculated_country_code, NEW.geometry);
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);
+ 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);
--- INSERT INTO search_name values (NEW.place_id, NEW.rank_search, NEW.rank_search, 0, NEW.country_code, name_vector, nameaddress_vector, place_centroid);
+-- INSERT INTO search_name values (NEW.place_id, NEW.rank_search, NEW.rank_search, 0, NEW.calculated_country_code, name_vector, nameaddress_vector, place_centroid);
+ END IF;
+
+ -- If we've not managed to pick up a better one - default centroid
+ IF NEW.centroid IS NULL THEN
+ NEW.centroid := place_centroid;
END IF;
END IF;
AS $$
DECLARE
b BOOLEAN;
+ classtable TEXT;
BEGIN
+ -- RAISE WARNING 'placex_delete % %',OLD.osm_type,OLD.osm_id;
+
+ update placex set linked_place_id = null, indexed_status = 2 where linked_place_id = OLD.place_id and indexed_status = 0;
+ --DEBUG: RAISE WARNING 'placex_delete:01 % %',OLD.osm_type,OLD.osm_id;
+ update placex set linked_place_id = null where linked_place_id = OLD.place_id;
+ --DEBUG: RAISE WARNING 'placex_delete:02 % %',OLD.osm_type,OLD.osm_id;
IF OLD.rank_address < 30 THEN
-- 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;
+ --DEBUG: RAISE WARNING 'placex_delete:04 % %',OLD.osm_type,OLD.osm_id;
DELETE FROM place_addressline where address_place_id = OLD.place_id;
+ --DEBUG: RAISE WARNING 'placex_delete:05 % %',OLD.osm_type,OLD.osm_id;
b := deleteRoad(OLD.partition, OLD.place_id);
+ --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;
END IF;
+ --DEBUG: RAISE WARNING 'placex_delete:08 % %',OLD.osm_type,OLD.osm_id;
+
IF OLD.rank_address < 26 THEN
- b := deleteLocationArea(OLD.partition, OLD.place_id);
+ b := deleteLocationArea(OLD.partition, OLD.place_id, OLD.rank_search);
END IF;
+ --DEBUG: RAISE WARNING 'placex_delete:09 % %',OLD.osm_type,OLD.osm_id;
+
IF OLD.name is not null THEN
b := deleteSearchName(OLD.partition, OLD.place_id);
END IF;
+ --DEBUG: RAISE WARNING 'placex_delete:10 % %',OLD.osm_type,OLD.osm_id;
+
DELETE FROM place_addressline where place_id = OLD.place_id;
+ --DEBUG: RAISE WARNING 'placex_delete:11 % %',OLD.osm_type,OLD.osm_id;
+
+ -- remove from tables for special search
+ classtable := 'place_classtype_' || OLD.class || '_' || OLD.type;
+ SELECT count(*)>0 FROM pg_tables WHERE tablename = classtable INTO b;
+ IF b THEN
+ EXECUTE 'DELETE FROM ' || classtable::regclass || ' WHERE place_id = $1' USING OLD.place_id;
+ END IF;
+
+ --DEBUG: RAISE WARNING 'placex_delete:12 % %',OLD.osm_type,OLD.osm_id;
+
RETURN OLD;
END;
CREATE OR REPLACE FUNCTION place_delete() RETURNS TRIGGER
AS $$
DECLARE
- placeid INTEGER;
+ placeid BIGINT;
BEGIN
--- RAISE WARNING 'delete: % % % %',OLD.osm_type,OLD.osm_id,OLD.class,OLD.type;
+ --DEBUG: RAISE WARNING 'delete: % % % %',OLD.osm_type,OLD.osm_id,OLD.class,OLD.type;
- -- deleting large polygons can have a massive effect ont he system - require manual intervention to let them through
- IF st_area(OLD.geometry) > 2 THEN
+ -- 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;
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;
--- delete from placex where osm_type = OLD.osm_type and osm_id = OLD.osm_id and class = OLD.class and type = OLD.type;
RETURN OLD;
END;
existing RECORD;
existingplacex RECORD;
existinggeometry GEOMETRY;
- existingplace_id INTEGER;
+ existingplace_id BIGINT;
result BOOLEAN;
partition INTEGER;
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
- RAISE WARNING '-----------------------------------------------------------------------------------';
- RAISE WARNING 'place_insert: % % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type,st_area(NEW.geometry);
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;
- RAISE WARNING '%', existingplacex;
+ --DEBUG: RAISE WARNING '%', existingplacex;
END IF;
-- Just block these - lots and pointless
-- Patch in additional country names
IF NEW.admin_level = 2 AND NEW.type = 'administrative' AND NEW.country_code is not null THEN
- select country_name.name || NEW.name from country_name where country_name.country_code = lower(NEW.country_code) INTO NEW.name;
+ 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 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 tags
+ -- 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;
+ 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;
--- RAISE WARNING 'Existing: %',existing.place_id;
+ --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)
DELETE from import_polygon_delete where osm_type = NEW.osm_type and osm_id = NEW.osm_id;
-- To paraphrase, if there isn't an existing item, OR if the admin level has changed, OR if it is a major change in geometry
- IF existing.osm_type IS NULL
- OR existingplacex.osm_type IS NULL
- OR coalesce(existing.admin_level, 100) != coalesce(NEW.admin_level, 100)
- OR coalesce(existing.country_code, '') != coalesce(NEW.country_code, '')
- OR (existing.geometry::text != NEW.geometry::text AND ST_Distance(ST_Centroid(existing.geometry),ST_Centroid(NEW.geometry)) > 0.01 AND NOT
- (ST_GeometryType(existing.geometry) in ('ST_Polygon','ST_MultiPolygon') AND ST_GeometryType(NEW.geometry) in ('ST_Polygon','ST_MultiPolygon')))
- THEN
-
--- IF existing.osm_type IS NULL THEN
--- RAISE WARNING 'no existing place';
--- END IF;
--- IF existingplacex.osm_type IS NULL THEN
--- RAISE WARNING 'no existing placex %', existingplacex;
--- END IF;
-
--- RAISE WARNING 'delete and replace';
+ IF existingplacex.osm_type IS NULL THEN
IF existing.osm_type IS NOT NULL THEN
--- RAISE WARNING 'insert delete % % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type,ST_Distance(ST_Centroid(existing.geometry),ST_Centroid(NEW.geometry)),existing;
- 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;
-
--- RAISE WARNING 'delete and replace2';
+ DELETE from place where osm_type = NEW.osm_type and osm_id = NEW.osm_id and class = NEW.class and type = NEW.type;
+ END IF;
-- No - process it as a new insertion (hopefully of low rank or it will be slow)
insert into placex (osm_type, osm_id, class, type, name, admin_level, housenumber,
,NEW.geometry
);
--- RAISE WARNING 'insert done % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type;
+ --DEBUG: RAISE WARNING 'insert done % % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type,NEW.name;
RETURN NEW;
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_Contains(NEW.geometry, placex.geometry) OR ST_Intersects(NEW.geometry, placex.geometry))
- AND NOT (ST_Contains(existinggeometry, placex.geometry) OR ST_Intersects(existinggeometry, placex.geometry))
- AND rank_search > existingplacex.rank_search;
+ (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_Contains(existinggeometry, placex.geometry) OR ST_Intersects(existinggeometry, placex.geometry))
- AND NOT (ST_Contains(NEW.geometry, placex.geometry) OR ST_Intersects(NEW.geometry, placex.geometry))
- AND rank_search > existingplacex.rank_search;
+ (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);
END IF;
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 placex.place_id = place_addressline.place_id and indexed_status = 0
+ and (rank_search < 28 or name is not null);
END IF;
END IF;
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
geometry = NEW.geometry
where place_id = existingplacex.place_id;
--- now done as part of insert
--- partition := get_partition(NEW.geometry, existingplacex.country_code);
--- result := update_location(partition, existingplacex.place_id, existingplacex.country_code, NEW.name, existingplacex.rank_search, existingplacex.rank_address, NEW.geometry);
-
END IF;
-- Abort the add (we modified the existing place instead)
END IF;
END LOOP;
- RETURN null;
+ -- anything will do as a fallback - just take the first name type thing there is
+ search := avals(name);
+ RETURN search[1];
END;
$$
LANGUAGE plpgsql IMMUTABLE;
FOR location IN
select nodes from planet_osm_ways where id = way_ids[j] LIMIT 1
LOOP
- searchnodes := searchnodes | location.nodes;
+ IF not (ARRAY[location.nodes] <@ searchnodes) THEN
+ searchnodes := searchnodes || location.nodes;
+ END IF;
END LOOP;
END LOOP;
$$
LANGUAGE plpgsql IMMUTABLE;
-CREATE OR REPLACE FUNCTION get_address_postcode(for_place_id INTEGER) RETURNS TEXT
+CREATE OR REPLACE FUNCTION get_address_postcode(for_place_id BIGINT) RETURNS TEXT
AS $$
DECLARE
result TEXT[];
IF array_upper(search, 1) IS NOT NULL AND array_upper(location.name, 1) IS NOT NULL THEN
FOR j IN 1..array_upper(search, 1) LOOP
FOR k IN 1..array_upper(location.name, 1) LOOP
- IF (found > location.rank_address AND location.name[k].key = search[j] AND location.name[k].value != '') AND NOT result && ARRAY[trim(location.name[k].value)] AND (for_postcode IS NULL OR location.name[k].value ilike for_postcode||'%') THEN
+ IF (found > location.rank_address AND location.name[k].key = search[j] AND location.name[k].value != '') AND NOT result @> ARRAY[trim(location.name[k].value)] AND (for_postcode IS NULL OR location.name[k].value ilike for_postcode||'%') THEN
result[(100 - location.rank_address)] := trim(location.name[k].value);
found := location.rank_address;
END IF;
$$
LANGUAGE plpgsql;
-CREATE OR REPLACE FUNCTION get_address_by_language(for_place_id INTEGER, languagepref TEXT[]) RETURNS TEXT
+CREATE OR REPLACE FUNCTION get_address_by_language(for_place_id BIGINT, languagepref TEXT[]) RETURNS TEXT
AS $$
DECLARE
result TEXT[];
FOR location IN select * from get_addressdata(for_place_id) where isaddress order by rank_address desc LOOP
currresult := trim(get_name_by_language(location.name, languagepref));
- IF currresult != prevresult AND currresult IS NOT NULL THEN
+ IF currresult != prevresult AND currresult IS NOT NULL AND result[(100 - location.rank_address)] IS NULL THEN
result[(100 - location.rank_address)] := trim(get_name_by_language(location.name, languagepref));
prevresult := currresult;
END IF;
DROP TYPE addressline CASCADE;
create type addressline as (
- place_id INTEGER,
+ place_id BIGINT,
osm_type CHAR(1),
osm_id INTEGER,
name HSTORE,
distance FLOAT
);
-CREATE OR REPLACE FUNCTION get_addressdata(in_place_id INTEGER) RETURNS setof addressline
+CREATE OR REPLACE FUNCTION get_addressdata(in_place_id BIGINT) RETURNS setof addressline
AS $$
DECLARE
- for_place_id INTEGER;
+ for_place_id BIGINT;
result TEXT[];
search TEXT[];
found INTEGER;
END IF;
IF for_place_id IS NULL THEN
- select parent_place_id, country_code, housenumber, rank_address, postcode, name, class, type from placex
+ 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
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 country_code, housenumber, rank_address, postcode, null from placex where place_id = for_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;
END IF;
hadcountry := false;
FOR location IN
select placex.place_id, osm_type, osm_id,
- CASE WHEN class = 'place' and type = 'postcode' THEN 'name' => postcode ELSE name END as name,
+ CASE WHEN class = 'place' and type = 'postcode' THEN hstore('name', postcode) ELSE name END as name,
class, type, admin_level, true as fromarea, true as isaddress,
CASE WHEN rank_address = 0 THEN 100 WHEN rank_address = 11 THEN 5 ELSE rank_address END as rank_address,
- 0 as distance, country_code
+ 0 as distance, calculated_country_code
from placex
where place_id = for_place_id
LOOP
--RAISE WARNING '%',location;
- IF searchcountrycode IS NULL AND location.country_code IS NOT NULL THEN
- searchcountrycode := location.country_code;
+ IF searchcountrycode IS NULL AND location.calculated_country_code IS NOT NULL THEN
+ searchcountrycode := location.calculated_country_code;
END IF;
IF searchpostcode IS NOT NULL and location.type = 'postcode' THEN
location.isaddress := FALSE;
FOR location IN
select placex.place_id, osm_type, osm_id,
- CASE WHEN class = 'place' and type = 'postcode' THEN 'name' => postcode ELSE name END as name,
+ CASE WHEN class = 'place' and type = 'postcode' THEN hstore('name', postcode) ELSE name END as name,
class, type, admin_level, fromarea, isaddress,
CASE WHEN address_place_id = for_place_id AND rank_address = 0 THEN 100 WHEN rank_address = 11 THEN 5 ELSE rank_address END as rank_address,
- distance,country_code
+ distance,calculated_country_code
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.country_code IS NULL OR searchcountrycode IS NULL OR placex.country_code = searchcountrycode OR rank_address < 4)
+ and (placex.calculated_country_code IS NULL OR searchcountrycode IS NULL OR placex.calculated_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.country_code IS NOT NULL THEN
- searchcountrycode := location.country_code;
+ IF searchcountrycode IS NULL AND location.calculated_country_code IS NOT NULL THEN
+ searchcountrycode := location.calculated_country_code;
END IF;
IF searchpostcode IS NOT NULL and location.type = 'postcode' THEN
location.isaddress := FALSE;
END IF;
IF searchcountrycode IS NOT NULL THEN
- location := ROW(null, null, null, 'ref'=>searchcountrycode, 'place', 'country_code', null, true, false, 4, 0)::addressline;
+ location := ROW(null, null, null, hstore('ref', searchcountrycode), 'place', 'country_code', null, true, false, 4, 0)::addressline;
RETURN NEXT location;
END IF;
END IF;
IF searchhousenumber IS NOT NULL THEN
- location := ROW(in_place_id, null, null, 'ref'=>searchhousenumber, 'place', 'house_number', null, true, true, 28, 0)::addressline;
+ location := ROW(in_place_id, null, null, hstore('ref', searchhousenumber), 'place', 'house_number', null, true, true, 28, 0)::addressline;
RETURN NEXT location;
END IF;
IF searchpostcode IS NOT NULL THEN
- location := ROW(null, null, null, 'ref'=>searchpostcode, 'place', 'postcode', null, true, true, 5, 0)::addressline;
+ location := ROW(null, null, null, hstore('ref', searchpostcode), 'place', 'postcode', null, true, true, 5, 0)::addressline;
RETURN NEXT location;
END IF;
$$
LANGUAGE plpgsql;
-CREATE OR REPLACE FUNCTION get_place_boundingbox(search_place_id INTEGER) RETURNS place_boundingbox
+CREATE OR REPLACE FUNCTION get_place_boundingbox(search_place_id BIGINT) RETURNS place_boundingbox
AS $$
DECLARE
result place_boundingbox;
LANGUAGE plpgsql;
-- don't do the operation if it would be slow
-CREATE OR REPLACE FUNCTION get_place_boundingbox_quick(search_place_id INTEGER) RETURNS place_boundingbox
+CREATE OR REPLACE FUNCTION get_place_boundingbox_quick(search_place_id BIGINT) RETURNS place_boundingbox
AS $$
DECLARE
result place_boundingbox;
$$
LANGUAGE plpgsql;
-CREATE OR REPLACE FUNCTION update_place(search_place_id INTEGER) RETURNS BOOLEAN
+CREATE OR REPLACE FUNCTION update_place(search_place_id BIGINT) RETURNS BOOLEAN
AS $$
DECLARE
result place_boundingbox;
$$
LANGUAGE plpgsql;
-CREATE AGGREGATE array_agg(INT[])
-(
- sfunc = array_cat,
- stype = INT[],
- initcond = '{}'
-);
-
-CREATE OR REPLACE FUNCTION tigger_create_interpolation(linegeo GEOMETRY, in_startnumber INTEGER,
- in_endnumber INTEGER, interpolationtype TEXT,
- in_street TEXT, in_isin TEXT, in_postcode TEXT) RETURNS INTEGER
+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 $$
DECLARE
-
- startnumber INTEGER;
- endnumber INTEGER;
- stepsize INTEGER;
- housenum INTEGER;
+
newpoints INTEGER;
- numberrange INTEGER;
- rangestartnumber INTEGER;
place_centroid GEOMETRY;
- partition INTEGER;
- parent_place_id INTEGER;
+ out_partition INTEGER;
+ out_parent_place_id BIGINT;
location RECORD;
address_street_word_id INTEGER;
+ out_postcode TEXT;
BEGIN
- IF in_endnumber > in_startnumber THEN
- startnumber = in_startnumber;
- endnumber = in_endnumber;
- ELSE
- startnumber = in_endnumber;
- endnumber = in_startnumber;
- END IF;
-
- numberrange := endnumber - startnumber;
- rangestartnumber := startnumber;
-
- 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
- stepsize := 2;
- ELSE -- everything else assumed to be 'all'
- stepsize := 1;
- END IF;
- END IF;
-
- -- Filter out really broken tiger data
- IF numberrange > 0 AND (numberrange::float/stepsize::float > 500)
- AND ST_length(linegeo)/(numberrange::float/stepsize::float) < 0.000001 THEN
- RAISE WARNING 'Road too short for number range % to % on %, % (%)',startnumber,endnumber,in_street,in_isin,
- ST_length(linegeo)/(numberrange::float/stepsize::float);
- RETURN 0;
- END IF;
-
- place_centroid := ST_Centroid(linegeo);
- partition := get_partition(place_centroid, 'us');
- parent_place_id := null;
+ place_centroid := ST_Centroid(pointgeo);
+ out_partition := get_partition(place_centroid, in_countrycode);
+ out_parent_place_id := null;
address_street_word_id := get_name_id(make_standard_name(in_street));
IF address_street_word_id IS NOT NULL THEN
- FOR location IN SELECT * from getNearestNamedRoadFeature(partition, place_centroid, address_street_word_id) LOOP
- parent_place_id := location.place_id;
+ FOR location IN SELECT * from getNearestNamedRoadFeature(out_partition, place_centroid, address_street_word_id) LOOP
+ out_parent_place_id := location.place_id;
END LOOP;
END IF;
- IF parent_place_id IS NULL THEN
- FOR location IN SELECT place_id FROM getNearestParellelRoadFeature(partition, linegeo) LOOP
- parent_place_id := location.place_id;
+ 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 IF;
- IF parent_place_id IS NULL THEN
- FOR location IN SELECT place_id FROM getNearestRoadFeature(partition, place_centroid) LOOP
- parent_place_id := location.place_id;
- END LOOP;
+ 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;
newpoints := 0;
- FOR housenum IN startnumber..endnumber BY stepsize LOOP
- insert into location_property_tiger (place_id, partition, parent_place_id, housenumber, postcode, centroid)
- values (nextval('seq_place'), partition, parent_place_id, housenum, in_postcode,
- ST_Line_Interpolate_Point(linegeo, (housenum::float-rangestartnumber::float)/numberrange::float));
- newpoints := newpoints + 1;
- END LOOP;
+ insert into location_property_aux (place_id, partition, parent_place_id, housenumber, postcode, centroid)
+ values (nextval('seq_place'), out_partition, out_parent_place_id, in_housenumber, out_postcode, place_centroid);
+ newpoints := newpoints + 1;
RETURN newpoints;
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
+CREATE OR REPLACE FUNCTION get_osm_rel_members(members TEXT[], member TEXT) RETURNS TEXT[]
AS $$
DECLARE
+ result TEXT[];
+ i INTEGER;
+BEGIN
- newpoints INTEGER;
- place_centroid GEOMETRY;
- partition INTEGER;
- parent_place_id INTEGER;
- location RECORD;
- address_street_word_id INTEGER;
+ FOR i IN 1..ARRAY_UPPER(members,1) BY 2 LOOP
+ IF members[i+1] = member THEN
+ result := result || members[i];
+ END IF;
+ END LOOP;
+ return result;
+END;
+$$
+LANGUAGE plpgsql;
+
+CREATE OR REPLACE FUNCTION get_osm_rel_members(members TEXT[], memberLabels TEXT[]) RETURNS SETOF TEXT
+ AS $$
+DECLARE
+ i INTEGER;
BEGIN
- place_centroid := ST_Centroid(pointgeo);
- partition := get_partition(place_centroid, in_countrycode);
- parent_place_id := null;
+ FOR i IN 1..ARRAY_UPPER(members,1) BY 2 LOOP
+ IF members[i+1] = ANY(memberLabels) THEN
+ RETURN NEXT members[i];
+ END IF;
+ END LOOP;
- address_street_word_id := get_name_id(make_standard_name(in_street));
- IF address_street_word_id IS NOT NULL THEN
- FOR location IN SELECT * from getNearestNamedRoadFeature(partition, place_centroid, address_street_word_id) LOOP
- parent_place_id := location.place_id;
- END LOOP;
+ RETURN;
+END;
+$$
+LANGUAGE plpgsql;
+
+-- See: http://stackoverflow.com/questions/6410088/how-can-i-mimic-the-php-urldecode-function-in-postgresql
+CREATE OR REPLACE FUNCTION decode_url_part(p varchar) RETURNS varchar
+ AS $$
+SELECT convert_from(CAST(E'\\x' || array_to_string(ARRAY(
+ SELECT CASE WHEN length(r.m[1]) = 1 THEN encode(convert_to(r.m[1], 'SQL_ASCII'), 'hex') ELSE substring(r.m[1] from 2 for 2) END
+ FROM regexp_matches($1, '%[0-9a-f][0-9a-f]|.', 'gi') AS r(m)
+), '') AS bytea), 'UTF8');
+$$
+LANGUAGE SQL IMMUTABLE STRICT;
+
+CREATE OR REPLACE FUNCTION catch_decode_url_part(p varchar) RETURNS varchar
+ AS $$
+DECLARE
+BEGIN
+ RETURN decode_url_part(p);
+EXCEPTION
+ WHEN others THEN return null;
+END;
+$$
+LANGUAGE plpgsql IMMUTABLE;
+
+DROP TYPE wikipedia_article_match CASCADE;
+create type wikipedia_article_match as (
+ language TEXT,
+ title TEXT,
+ importance FLOAT
+);
+
+CREATE OR REPLACE FUNCTION get_wikipedia_match(extratags HSTORE, country_code varchar(2)) RETURNS wikipedia_article_match
+ AS $$
+DECLARE
+ langs TEXT[];
+ i INT;
+ wiki_article TEXT;
+ wiki_article_title TEXT;
+ wiki_article_language TEXT;
+ result wikipedia_article_match;
+BEGIN
+ langs := ARRAY['english','country','ar','bg','ca','cs','da','de','en','es','eo','eu','fa','fr','ko','hi','hr','id','it','he','lt','hu','ms','nl','ja','no','pl','pt','kk','ro','ru','sk','sl','sr','fi','sv','tr','uk','vi','vo','war','zh'];
+ i := 1;
+ WHILE langs[i] IS NOT NULL LOOP
+ wiki_article := extratags->(case when langs[i] in ('english','country') THEN 'wikipedia' ELSE 'wikipedia:'||langs[i] END);
+ IF wiki_article is not null THEN
+ 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 := replace(wiki_article,' ','_');
+ wiki_article_title := trim(split_part(wiki_article, ':', 2));
+ IF wiki_article_title IS NULL OR wiki_article_title = '' THEN
+ 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
+ from wikipedia_article
+ where language = wiki_article_language and
+ (title = wiki_article_title OR title = catch_decode_url_part(wiki_article_title) OR title = replace(catch_decode_url_part(wiki_article_title),E'\\',''))
+ UNION ALL
+ select wikipedia_article.language,wikipedia_article.title,wikipedia_article.importance
+ from wikipedia_redirect join wikipedia_article on (wikipedia_redirect.language = wikipedia_article.language and wikipedia_redirect.to_title = wikipedia_article.title)
+ where wikipedia_redirect.language = wiki_article_language and
+ (from_title = wiki_article_title OR from_title = catch_decode_url_part(wiki_article_title) OR from_title = replace(catch_decode_url_part(wiki_article_title),E'\\',''))
+ order by importance desc limit 1 INTO result;
+
+ IF result.language is not null THEN
+ return result;
+ END IF;
+ END IF;
+ i := i + 1;
+ END LOOP;
+ RETURN NULL;
+END;
+$$
+LANGUAGE plpgsql;
+
+CREATE OR REPLACE FUNCTION quad_split_geometry(geometry GEOMETRY, maxarea FLOAT, maxdepth INTEGER)
+ RETURNS SETOF GEOMETRY
+ AS $$
+DECLARE
+ xmin FLOAT;
+ ymin FLOAT;
+ xmax FLOAT;
+ ymax FLOAT;
+ xmid FLOAT;
+ ymid FLOAT;
+ secgeo GEOMETRY;
+ secbox GEOMETRY;
+ seg INTEGER;
+ geo RECORD;
+ area FLOAT;
+ remainingdepth INTEGER;
+ added INTEGER;
+
+BEGIN
+
+-- RAISE WARNING 'quad_split_geometry: maxarea=%, depth=%',maxarea,maxdepth;
+
+ IF (ST_GeometryType(geometry) not in ('ST_Polygon','ST_MultiPolygon') OR NOT ST_IsValid(geometry)) THEN
+ RETURN NEXT geometry;
+ RETURN;
END IF;
- IF parent_place_id IS NULL THEN
- FOR location IN SELECT place_id FROM getNearestRoadFeature(partition, place_centroid) LOOP
- parent_place_id := location.place_id;
- END LOOP;
+ remainingdepth := maxdepth - 1;
+ area := ST_AREA(geometry);
+ IF remainingdepth < 1 OR area < maxarea THEN
+ RETURN NEXT geometry;
+ RETURN;
END IF;
- newpoints := 0;
- insert into location_property_aux (place_id, partition, parent_place_id, housenumber, postcode, centroid)
- values (nextval('seq_place'), partition, parent_place_id, in_housenumber, in_postcode, place_centroid);
- newpoints := newpoints + 1;
+ xmin := st_xmin(geometry);
+ xmax := st_xmax(geometry);
+ ymin := st_ymin(geometry);
+ ymax := st_ymax(geometry);
+ secbox := ST_SetSRID(ST_MakeBox2D(ST_Point(ymin,xmin),ST_Point(ymax,xmax)),4326);
- RETURN newpoints;
+ -- if the geometry completely covers the box don't bother to slice any more
+ IF ST_AREA(secbox) = area THEN
+ RETURN NEXT geometry;
+ RETURN;
+ END IF;
+
+ xmid := (xmin+xmax)/2;
+ ymid := (ymin+ymax)/2;
+
+ added := 0;
+ FOR seg IN 1..4 LOOP
+
+ IF seg = 1 THEN
+ secbox := ST_SetSRID(ST_MakeBox2D(ST_Point(xmin,ymin),ST_Point(xmid,ymid)),4326);
+ END IF;
+ IF seg = 2 THEN
+ secbox := ST_SetSRID(ST_MakeBox2D(ST_Point(xmin,ymid),ST_Point(xmid,ymax)),4326);
+ END IF;
+ IF seg = 3 THEN
+ secbox := ST_SetSRID(ST_MakeBox2D(ST_Point(xmid,ymin),ST_Point(xmax,ymid)),4326);
+ END IF;
+ IF seg = 4 THEN
+ secbox := ST_SetSRID(ST_MakeBox2D(ST_Point(xmid,ymid),ST_Point(xmax,ymax)),4326);
+ END IF;
+
+ IF st_intersects(geometry, secbox) THEN
+ secgeo := st_intersection(geometry, secbox);
+ IF NOT ST_IsEmpty(secgeo) AND ST_GeometryType(secgeo) in ('ST_Polygon','ST_MultiPolygon') THEN
+ FOR geo IN select quad_split_geometry(secgeo, maxarea, remainingdepth) as geom LOOP
+ IF NOT ST_IsEmpty(geo.geom) AND ST_GeometryType(geo.geom) in ('ST_Polygon','ST_MultiPolygon') THEN
+ added := added + 1;
+ RETURN NEXT geo.geom;
+ END IF;
+ END LOOP;
+ END IF;
+ END IF;
+ END LOOP;
+
+ RETURN;
+END;
+$$
+LANGUAGE plpgsql;
+
+CREATE OR REPLACE FUNCTION split_geometry(geometry GEOMETRY)
+ RETURNS SETOF GEOMETRY
+ AS $$
+DECLARE
+ geo RECORD;
+BEGIN
+ -- 10000000000 is ~~ 1x1 degree
+ FOR geo IN select quad_split_geometry(geometry, 0.25, 20) as geom LOOP
+ RETURN NEXT geo.geom;
+ END LOOP;
+ RETURN;
+END;
+$$
+LANGUAGE plpgsql;
+
+
+CREATE OR REPLACE FUNCTION place_force_delete(placeid BIGINT) RETURNS BOOLEAN
+ AS $$
+DECLARE
+ osmid BIGINT;
+ osmtype character(1);
+ pclass text;
+ ptype text;
+BEGIN
+ SELECT osm_type, osm_id, class, type FROM placex WHERE place_id = placeid INTO osmtype, osmid, pclass, ptype;
+ DELETE FROM import_polygon_delete where osm_type = osmtype and osm_id = osmid and class = pclass and type = ptype;
+ DELETE FROM import_polygon_error where osm_type = osmtype and osm_id = osmid and class = pclass and type = ptype;
+ -- force delete from place/placex by making it a very small geometry
+ UPDATE place set geometry = ST_SetSRID(ST_Point(0,0), 4326) where osm_type = osmtype and osm_id = osmid and class = pclass and type = ptype;
+ DELETE FROM place where osm_type = osmtype and osm_id = osmid and class = pclass and type = ptype;
+
+ RETURN TRUE;
+END;
+$$
+LANGUAGE plpgsql;
+
+CREATE OR REPLACE FUNCTION place_force_update(placeid BIGINT) RETURNS BOOLEAN
+ AS $$
+DECLARE
+ placegeom GEOMETRY;
+ geom GEOMETRY;
+ diameter FLOAT;
+ rank INTEGER;
+BEGIN
+ SELECT geometry, rank_search FROM placex WHERE place_id = placeid INTO placegeom, rank;
+ IF placegeom IS NOT NULL AND ST_IsValid(placegeom) THEN
+ IF ST_GeometryType(placegeom) in ('ST_Polygon','ST_MultiPolygon') THEN
+ FOR geom IN select split_geometry(placegeom) FROM placex WHERE place_id = placeid LOOP
+ update placex set indexed_status = 2 where (st_covers(geom, placex.geometry) OR ST_Intersects(geom, placex.geometry))
+ AND rank_search > rank and indexed_status = 0 and ST_geometrytype(placex.geometry) = 'ST_Point' and (rank_search < 28 or name is not null);
+ update placex set indexed_status = 2 where (st_covers(geom, placex.geometry) OR ST_Intersects(geom, placex.geometry))
+ AND rank_search > rank and indexed_status = 0 and ST_geometrytype(placex.geometry) != 'ST_Point' and (rank_search < 28 or name is not null);
+ END LOOP;
+ ELSE
+ diameter := 0;
+ IF rank = 11 THEN
+ diameter := 0.05;
+ ELSEIF rank < 18 THEN
+ diameter := 0.1;
+ ELSEIF rank < 20 THEN
+ diameter := 0.05;
+ ELSEIF rank = 21 THEN
+ diameter := 0.001;
+ ELSEIF rank < 24 THEN
+ diameter := 0.02;
+ ELSEIF rank < 26 THEN
+ diameter := 0.002; -- 100 to 200 meters
+ ELSEIF rank < 28 THEN
+ diameter := 0.001; -- 50 to 100 meters
+ END IF;
+ IF diameter > 0 THEN
+ update placex set indexed_status = 2 where indexed_status = 0 and rank_search > rank and ST_DWithin(placex.geometry, placegeom, diameter) and (rank_search < 28 or name is not null);
+ END IF;
+ END IF;
+ RETURN TRUE;
+ END IF;
+
+ RETURN FALSE;
END;
$$
LANGUAGE plpgsql;