+
+-- Insert address of a place into the place_addressline table.
+--
+-- \param obj_place_id Place_id of the place to compute the address for.
+-- \param partition Partition number where the place is in.
+-- \param maxrank Rank of the place. All address features must have
+-- a search rank lower than the given rank.
+-- \param address Address terms for the place.
+-- \param geoemtry Geometry to which the address objects should be close.
+--
+-- \retval parent_place_id Place_id of the address object that is the direct
+-- ancestor.
+-- \retval postcode Postcode computed from the address. This is the
+-- addr:postcode of one of the address objects. If
+-- more than one of has a postcode, the highest ranking
+-- one is used. May be NULL.
+-- \retval nameaddress_vector Search terms for the address. This is the sum
+-- of name terms of all address objects.
+CREATE OR REPLACE FUNCTION insert_addresslines(obj_place_id BIGINT,
+ partition SMALLINT,
+ maxrank SMALLINT,
+ address HSTORE,
+ geometry GEOMETRY,
+ OUT parent_place_id BIGINT,
+ OUT postcode TEXT,
+ OUT nameaddress_vector INT[])
+ AS $$
+DECLARE
+ current_rank_address INTEGER := 0;
+ location_distance FLOAT := 0;
+ location_parent GEOMETRY := NULL;
+ parent_place_id_rank SMALLINT := 0;
+
+ location_isaddress BOOLEAN;
+
+ address_havelevel BOOLEAN[];
+ location_keywords INT[];
+
+ location RECORD;
+ addr_item RECORD;
+
+ isin_tokens INT[];
+ isin TEXT[];
+BEGIN
+ parent_place_id := 0;
+ nameaddress_vector := '{}'::int[];
+ isin_tokens := '{}'::int[];
+
+ ---- convert address store to array of tokenids
+ IF address IS NOT NULL THEN
+ FOR addr_item IN SELECT * FROM each(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
+ isin_tokens := array_merge(isin_tokens,
+ word_ids_from_name(addr_item.value));
+ IF NOT %REVERSE-ONLY% THEN
+ nameaddress_vector := array_merge(nameaddress_vector,
+ addr_ids_from_name(addr_item.value));
+ END IF;
+ END IF;
+ END LOOP;
+
+ IF address ? 'is_in' THEN
+ -- is_in items need splitting
+ isin := regexp_split_to_array(address->'is_in', E'[;,]');
+ IF array_upper(isin, 1) IS NOT NULL THEN
+ FOR i IN 1..array_upper(isin, 1) LOOP
+ isin_tokens := array_merge(isin_tokens,
+ word_ids_from_name(isin[i]));
+
+ -- merge word into address vector
+ IF NOT %REVERSE-ONLY% THEN
+ nameaddress_vector := array_merge(nameaddress_vector,
+ addr_ids_from_name(isin[i]));
+ END IF;
+ END LOOP;
+ END IF;
+ END IF;
+ END IF;
+ IF NOT %REVERSE-ONLY% THEN
+ nameaddress_vector := array_merge(nameaddress_vector, isin_tokens);
+ END IF;
+
+ ---- now compute the address terms
+ FOR i IN 1..28 LOOP
+ address_havelevel[i] := false;
+ END LOOP;
+
+ FOR location IN
+ SELECT * FROM getNearFeatures(partition, geometry, maxrank, isin_tokens)
+ LOOP
+ IF location.rank_address != current_rank_address THEN
+ current_rank_address := location.rank_address;
+ IF location.isguess THEN
+ location_distance := location.distance * 1.5;
+ ELSE
+ IF location.rank_address <= 12 THEN
+ -- for county and above, if we have an area consider that exact
+ -- (It would be nice to relax the constraint for places close to
+ -- the boundary but we'd need the exact geometry for that. Too
+ -- expensive.)
+ location_distance = 0;
+ ELSE
+ -- Below county level remain slightly fuzzy.
+ location_distance := location.distance * 0.5;
+ END IF;
+ END IF;
+ ELSE
+ CONTINUE WHEN location.keywords <@ location_keywords;
+ END IF;
+
+ IF location.distance < location_distance OR NOT location.isguess THEN
+ location_keywords := location.keywords;
+
+ 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
+ IF NOT %REVERSE-ONLY% THEN
+ nameaddress_vector := array_merge(nameaddress_vector,
+ location.keywords::integer[]);
+ END IF;
+
+ INSERT INTO place_addressline (place_id, address_place_id, fromarea,
+ isaddress, distance, cached_rank_address)
+ VALUES (obj_place_id, location.place_id, true,
+ location_isaddress, location.distance, location.rank_address);
+
+ IF location_isaddress THEN
+ -- add postcode if we have one
+ -- (If multiple postcodes are available, we end up with the highest ranking one.)
+ IF location.postcode is not null THEN
+ postcode = location.postcode;
+ END IF;
+
+ address_havelevel[location.rank_address] := true;
+ IF NOT location.isguess THEN
+ SELECT placex.geometry FROM placex
+ WHERE obj_place_id = location.place_id INTO location_parent;
+ END IF;
+
+ IF location.rank_address > parent_place_id_rank THEN
+ parent_place_id = location.place_id;
+ parent_place_id_rank = location.rank_address;
+ END IF;
+ END IF;
+ --DEBUG: RAISE WARNING ' Terms: (%) %',location, nameaddress_vector;
+ END IF;
+
+ END LOOP;
+END;
+$$
+LANGUAGE plpgsql;
+
+