X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/784dad866fab9f5b2ccb72228d5eaacc7d5a070a..32d3eb46d5dfa4fd5486dc8abfb6afc1dcb0a360:/lib-sql/functions/partition-functions.sql diff --git a/lib-sql/functions/partition-functions.sql b/lib-sql/functions/partition-functions.sql index 5eb1aef1..595e4a61 100644 --- a/lib-sql/functions/partition-functions.sql +++ b/lib-sql/functions/partition-functions.sql @@ -17,28 +17,6 @@ CREATE TYPE nearfeaturecentr AS ( centroid GEOMETRY ); --- feature intersects geoemtry --- for areas and linestrings they must touch at least along a line -CREATE OR REPLACE FUNCTION is_relevant_geometry(de9im TEXT, geom_type TEXT) -RETURNS BOOLEAN -AS $$ -BEGIN - IF substring(de9im from 1 for 2) != 'FF' THEN - RETURN TRUE; - END IF; - - IF geom_type = 'ST_Point' THEN - RETURN substring(de9im from 4 for 1) = '0'; - END IF; - - IF geom_type in ('ST_LineString', 'ST_MultiLineString') THEN - RETURN substring(de9im from 4 for 1) = '1'; - END IF; - - RETURN substring(de9im from 4 for 1) = '2'; -END -$$ LANGUAGE plpgsql IMMUTABLE; - CREATE OR REPLACE function getNearFeatures(in_partition INTEGER, feature GEOMETRY, feature_centroid GEOMETRY, maxrank INTEGER) @@ -51,11 +29,20 @@ BEGIN IF in_partition = {{ partition }} THEN FOR r IN SELECT place_id, keywords, rank_address, rank_search, - min(ST_Distance(feature_centroid, geometry)) as distance, + CASE WHEN isguess THEN ST_Distance(feature, centroid) + ELSE min(ST_Distance(feature_centroid, geometry)) + -- tie breaker when distance is the same (i.e. way is on boundary) + + 0.00001 * ST_Distance(feature, centroid) + END as distance, isguess, postcode, centroid FROM location_area_large_{{ partition }} WHERE geometry && feature - AND is_relevant_geometry(ST_Relate(geometry, feature), ST_GeometryType(feature)) + AND CASE WHEN ST_Dimension(feature) = 0 + THEN _ST_Covers(geometry, feature) + WHEN ST_Dimension(feature) = 2 + THEN ST_Relate(geometry, feature, 'T********') + ELSE ST_NPoints(ST_Intersection(geometry, feature)) > 1 + END AND rank_address < maxrank -- Postcodes currently still use rank_search to define for which -- features they are relevant. @@ -92,7 +79,7 @@ BEGIN AND rank_address between from_rank and to_rank AND token_matches_address(token_info, key, keywords) GROUP BY place_id, keywords, rank_address, rank_search, isguess, postcode, centroid - ORDER BY bool_or(ST_Intersects(geometry, feature)), distance LIMIT 1; + ORDER BY bool_or(ST_Intersects(geometry, feature)) DESC, distance LIMIT 1; RETURN r; END IF; {% endfor %} @@ -138,14 +125,16 @@ BEGIN IF in_rank_search <= 4 and not in_estimate THEN INSERT INTO location_area_country (place_id, country_code, geometry) - values (in_place_id, in_country_code, in_geometry); + (SELECT in_place_id, in_country_code, geom + FROM split_geometry(in_geometry) as geom); RETURN TRUE; END IF; {% for partition in db.partitions %} IF in_partition = {{ partition }} THEN INSERT INTO location_area_large_{{ partition }} (partition, place_id, country_code, keywords, rank_search, rank_address, isguess, postcode, centroid, geometry) - values (in_partition, in_place_id, in_country_code, in_keywords, in_rank_search, in_rank_address, in_estimate, postcode, in_centroid, in_geometry); + (SELECT in_partition, in_place_id, in_country_code, in_keywords, in_rank_search, in_rank_address, in_estimate, postcode, in_centroid, geom + FROM split_geometry(in_geometry) as geom); RETURN TRUE; END IF; {% endfor %}