Merge pull request #4024 from lonvia/remove-fallback-importance

Clean up importance computation

lib-sql/functions/placex_triggers.sql

@@ -731,6 +731,10 @@ BEGIN
 
   END IF;
 
+  IF NEW.importance IS NULL THEN
+    NEW.importance := 0.40001 - (NEW.rank_search::float / 75);
+  END IF;
+
   {% if debug %}RAISE WARNING 'placex_insert:END: % % % %',NEW.osm_type,NEW.osm_id,NEW.class,NEW.type;{% endif %}
 
 {% if not disable_diff_updates %}
@@ -1076,10 +1080,8 @@ BEGIN
       {% if debug %}RAISE WARNING 'Waterway processed';{% endif %}
   END IF;
 
-  NEW.importance := null;
+  SELECT wikipedia, importance INTO NEW.wikipedia, NEW.importance
-  SELECT wikipedia, importance
+    FROM compute_importance(NEW.extratags, NEW.country_code, NEW.rank_search, NEW.centroid);
-    FROM compute_importance(NEW.extratags, NEW.country_code, NEW.rank_search, NEW.centroid)
-    INTO NEW.wikipedia,NEW.importance;
 
 {% if debug %}RAISE WARNING 'Importance computed from wikipedia: %', NEW.importance;{% endif %}
 
@@ -1136,10 +1138,10 @@ BEGIN
                                        NEW.token_info, NEW.centroid);
 
         IF array_length(name_vector, 1) is not NULL THEN
-          INSERT INTO search_name (place_id, search_rank, address_rank,
+          INSERT INTO search_name (place_id, address_rank,
                                    importance, country_code, name_vector,
                                    nameaddress_vector, centroid)
-                 VALUES (NEW.place_id, NEW.rank_search, NEW.rank_address,
+                 VALUES (NEW.place_id, NEW.rank_address,
                          NEW.importance, NEW.country_code, name_vector,
                          nameaddress_vector, NEW.centroid);
           {% if debug %}RAISE WARNING 'Place added to search table';{% endif %}
@@ -1309,10 +1311,10 @@ BEGIN
     {% if debug %}RAISE WARNING 'added to search name (full)';{% endif %}
 
     {% if not db.reverse_only %}
-        INSERT INTO search_name (place_id, search_rank, address_rank,
+        INSERT INTO search_name (place_id, address_rank,
                                  importance, country_code, name_vector,
                                  nameaddress_vector, centroid)
-               VALUES (NEW.place_id, NEW.rank_search, NEW.rank_address,
+               VALUES (NEW.place_id, NEW.rank_address,
                        NEW.importance, NEW.country_code, name_vector,
                        nameaddress_vector, NEW.centroid);
     {% endif %}

lib-sql/tables/placex.sql

@@ -12,7 +12,7 @@ CREATE TABLE placex (
   place_id BIGINT NOT NULL,
   parent_place_id BIGINT,
   linked_place_id BIGINT,
-  importance FLOAT,
+  importance FLOAT NOT NULL,
   indexed_date TIMESTAMP,
   geometry_sector INTEGER NOT NULL,
   rank_address SMALLINT NOT NULL,

lib-sql/tables/search_name.sql

@@ -12,7 +12,6 @@ DROP TABLE IF EXISTS search_name;
 CREATE TABLE search_name (
   place_id BIGINT NOT NULL,
   importance FLOAT NOT NULL,
-  search_rank SMALLINT NOT NULL,
   address_rank SMALLINT NOT NULL,
   name_vector integer[] NOT NULL,
   nameaddress_vector integer[] NOT NULL,

src/nominatim_api/search/db_searches/address_search.py

@@ -2,7 +2,7 @@
 #
 # This file is part of Nominatim. (https://nominatim.org)
 #
-# Copyright (C) 2025 by the Nominatim developer community.
+# Copyright (C) 2026 by the Nominatim developer community.
 # For a full list of authors see the git log.
 """
 Implementation of search for an address (search with housenumber).
@@ -155,13 +155,7 @@ class AddressSearch(base.AbstractSearch):
         for ranking in self.rankings:
             penalty += ranking.sql_penalty(t)
 
-        sql = sa.select(t.c.place_id, t.c.search_rank, t.c.address_rank,
+        sql = sa.select(t.c.place_id, penalty.label('penalty'))
-                        t.c.country_code, t.c.centroid,
-                        t.c.name_vector, t.c.nameaddress_vector,
-                        sa.case((t.c.importance > 0, t.c.importance),
-                                else_=0.40001-(sa.cast(t.c.search_rank, sa.Float())/75))
-                          .label('importance'),
-                        penalty.label('penalty'))
 
         for lookup in self.lookups:
             sql = sql.where(lookup.sql_condition(t))
@@ -202,9 +196,7 @@ class AddressSearch(base.AbstractSearch):
 
         inner = sql.limit(10000).order_by(sa.desc(sa.text('importance'))).subquery()
 
-        sql = sa.select(inner.c.place_id, inner.c.search_rank, inner.c.address_rank,
+        sql = sa.select(inner.c.place_id, inner.c.penalty)
-                        inner.c.country_code, inner.c.centroid, inner.c.importance,
-                        inner.c.penalty)
 
         return sql.cte('searches')
 
@@ -240,12 +232,12 @@ class AddressSearch(base.AbstractSearch):
                                else_=1.0)
 
         if details.near is not None:
-            sql = sql.add_columns((-tsearch.c.centroid.ST_Distance(NEAR_PARAM))
+            sql = sql.add_columns((-t.c.centroid.ST_Distance(NEAR_PARAM))
                                   .label('importance'))
             sql = sql.order_by(sa.desc(sa.text('importance')))
         else:
-            sql = sql.order_by(penalty - tsearch.c.importance)
+            sql = sql.order_by(penalty - t.c.importance)
-            sql = sql.add_columns(tsearch.c.importance)
+            sql = sql.add_columns(t.c.importance)
 
         sql = sql.add_columns(penalty.label('accuracy'))\
                  .order_by(sa.text('accuracy'))
@@ -253,7 +245,7 @@ class AddressSearch(base.AbstractSearch):
         hnr_list = '|'.join(self.housenumbers.values)
 
         if self.has_address_terms:
-            sql = sql.where(sa.or_(tsearch.c.address_rank < 30,
+            sql = sql.where(sa.or_(t.c.rank_address < 30,
                                    sa.func.RegexpWord(hnr_list, t.c.housenumber)))
 
         inner = sql.subquery()

src/nominatim_api/search/db_searches/place_search.py

@@ -2,7 +2,7 @@
 #
 # This file is part of Nominatim. (https://nominatim.org)
 #
-# Copyright (C) 2025 by the Nominatim developer community.
+# Copyright (C) 2026 by the Nominatim developer community.
 # For a full list of authors see the git log.
 """
 Implementation of search for a named place (without housenumber).
@@ -58,12 +58,7 @@ class PlaceSearch(base.AbstractSearch):
         for ranking in self.rankings:
             penalty += ranking.sql_penalty(t)
 
-        sql = sa.select(t.c.place_id, t.c.search_rank, t.c.address_rank,
+        sql = sa.select(t.c.place_id, t.c.importance)
-                        t.c.country_code, t.c.centroid,
-                        t.c.name_vector, t.c.nameaddress_vector,
-                        sa.case((t.c.importance > 0, t.c.importance),
-                                else_=0.40001-(sa.cast(t.c.search_rank, sa.Float())/75))
-                          .label('importance'))
 
         for lookup in self.lookups:
             sql = sql.where(lookup.sql_condition(t))
@@ -103,12 +98,13 @@ class PlaceSearch(base.AbstractSearch):
 
         if details.excluded:
             sql = sql.where(base.exclude_places(t))
-        if details.min_rank > 0:
+        # Do not restrict ranks too much yet because rank restriction
-            sql = sql.where(sa.or_(t.c.address_rank >= MIN_RANK_PARAM,
+        # currently also depends on search_rank to account for state-cities
-                                   t.c.search_rank >= MIN_RANK_PARAM))
+        # like Berlin.
-        if details.max_rank < 30:
+        if details.max_rank < 26:
-            sql = sql.where(sa.or_(t.c.address_rank <= MAX_RANK_PARAM,
+            sql = sql.where(t.c.address_rank < 26)
-                                   t.c.search_rank <= MAX_RANK_PARAM))
+        elif details.max_rank < 30:
+            sql = sql.where(t.c.address_rank < MAX_RANK_PARAM)
 
         sql = sql.add_columns(penalty.label('penalty'))
 
@@ -116,11 +112,9 @@ class PlaceSearch(base.AbstractSearch):
                    .order_by(sa.desc(sa.text('importance')))\
                    .subquery()
 
-        sql = sa.select(inner.c.place_id, inner.c.search_rank, inner.c.address_rank,
+        sql = sa.select(inner.c.place_id, inner.c.importance, inner.c.penalty)
-                        inner.c.country_code, inner.c.centroid, inner.c.importance,
-                        inner.c.penalty)
 
-        # If the query is not an address search or has a geographic preference,
+        # If the query has no geographic preference,
         # preselect most important items to restrict the number of places
         # that need to be looked up in placex.
         if (details.viewbox is None or not details.bounded_viewbox)\
@@ -132,9 +126,7 @@ class PlaceSearch(base.AbstractSearch):
 
             inner = sql.subquery()
 
-            sql = sa.select(inner.c.place_id, inner.c.search_rank, inner.c.address_rank,
+            sql = sa.select(inner.c.place_id, inner.c.penalty)\
-                            inner.c.country_code, inner.c.centroid, inner.c.importance,
-                            inner.c.penalty)\
                     .where(inner.c.penalty - inner.c.importance < inner.c.min_penalty + 0.5)
 
         return sql.cte('searches')
@@ -169,12 +161,19 @@ class PlaceSearch(base.AbstractSearch):
                 penalty += sa.case((t.c.postcode.in_(self.postcodes.values), 0.0), else_=1.0)
 
         if details.near is not None:
-            sql = sql.add_columns((-tsearch.c.centroid.ST_Distance(NEAR_PARAM))
+            sql = sql.add_columns((-t.c.centroid.ST_Distance(NEAR_PARAM))
                                   .label('importance'))
             sql = sql.order_by(sa.desc(sa.text('importance')))
         else:
-            sql = sql.order_by(penalty - tsearch.c.importance)
+            sql = sql.order_by(penalty - t.c.importance)
-            sql = sql.add_columns(tsearch.c.importance)
+            sql = sql.add_columns(t.c.importance)
+
+        if details.min_rank > 0:
+            sql = sql.where(sa.or_(t.c.rank_address >= MIN_RANK_PARAM,
+                                   t.c.rank_search >= MIN_RANK_PARAM))
+        if details.max_rank < 30:
+            sql = sql.where(sa.or_(t.c.rank_address <= MAX_RANK_PARAM,
+                                   t.c.rank_search <= MAX_RANK_PARAM))
 
         sql = sql.add_columns(penalty.label('accuracy'))\
                  .order_by(sa.text('accuracy'))

src/nominatim_api/sql/sqlalchemy_schema.py

@@ -2,7 +2,7 @@
 #
 # This file is part of Nominatim. (https://nominatim.org)
 #
-# Copyright (C) 2025 by the Nominatim developer community.
+# Copyright (C) 2026 by the Nominatim developer community.
 # For a full list of authors see the git log.
 """
 SQLAlchemy definitions for all tables used by the frontend.
@@ -112,7 +112,6 @@ class SearchTables:
             'search_name', meta,
             sa.Column('place_id', sa.BigInteger),
             sa.Column('importance', sa.Float),
-            sa.Column('search_rank', sa.SmallInteger),
             sa.Column('address_rank', sa.SmallInteger),
             sa.Column('name_vector', IntArray),
             sa.Column('nameaddress_vector', IntArray),

src/nominatim_db/tools/migration.py

@@ -14,7 +14,7 @@ from ..errors import UsageError
 from ..config import Configuration
 from ..db import properties
 from ..db.connection import connect, Connection, \
-                            table_exists, register_hstore
+                            table_exists, register_hstore, table_has_column
 from ..db.sql_preprocessor import SQLPreprocessor
 from ..version import NominatimVersion, NOMINATIM_VERSION, parse_version
 from ..tokenizer import factory as tokenizer_factory
@@ -420,3 +420,20 @@ def create_place_interpolation_table(conn: Connection, config: Configuration, **
               SET type = coalesce(address->'interpolation', 'all'),
                   address = address - 'interpolation'::TEXT;
             """)
+
+
+@_migration(5, 2, 99, 4)
+def backfill_importance(conn: Connection, **_: Any) -> None:
+    """ Backfill missing importance values.
+    """
+    conn.execute("""UPDATE placex
+                    SET importance = 0.40001 - (rank_search::float / 75)
+                    WHERE importance is NULL OR importance <= 0
+                 """)
+    if table_exists(conn, 'search_name')\
+       and table_has_column(conn, 'search_name', 'search_rank'):
+        conn.execute("""UPDATE search_name
+                        SET importance = 0.40001 - (search_rank::float / 75)
+                        WHERE importance is NULL OR importance <= 0
+                     """)
+        conn.execute("ALTER TABLE search_name DROP COLUMN search_rank")

src/nominatim_db/version.py

@@ -55,7 +55,7 @@ def parse_version(version: str) -> NominatimVersion:
     return NominatimVersion(*[int(x) for x in parts[:2] + parts[2].split('-')])
 
 
-NOMINATIM_VERSION = parse_version('5.2.99-3')
+NOMINATIM_VERSION = parse_version('5.2.99-4')
 
 POSTGRESQL_REQUIRED_VERSION = (12, 0)
 POSTGIS_REQUIRED_VERSION = (3, 0)

test/python/conftest.py

@@ -293,13 +293,14 @@ def placex_row(placex_table, temp_db_cursor):
     def _add(osm_type='N', osm_id=None, cls='amenity', typ='cafe', names=None,
              admin_level=None, address=None, extratags=None, geom='POINT(10 4)',
              country=None, housenumber=None, rank_search=30, rank_address=30,
-             centroid='POINT(10 4)', indexed_status=0, indexed_date=None):
+             centroid='POINT(10 4)', indexed_status=0, indexed_date=None,
+             importance=0.00001):
         args = {'place_id': pysql.SQL("nextval('seq_place')"),
                 'osm_type': osm_type, 'osm_id': osm_id or next(idseq),
                 'class': cls, 'type': typ, 'name': names, 'admin_level': admin_level,
                 'address': address, 'housenumber': housenumber,
                 'rank_search': rank_search, 'rank_address': rank_address,
-                'extratags': extratags,
+                'extratags': extratags, 'importance': importance,
                 'centroid': _with_srid(centroid), 'geometry': _with_srid(geom),
                 'country_code': country,
                 'indexed_status': indexed_status, 'indexed_date': indexed_date,

test/python/tools/test_database_import.py

@@ -176,6 +176,7 @@ async def test_load_data(dsn, place_row, place_interpolation_row, placex_table,
         BEGIN
           NEW.place_id := nextval('seq_place');
           NEW.indexed_status := 1;
+          NEW.importance := 0.00001;
           NEW.centroid := ST_Centroid(NEW.geometry);
           NEW.partition := 0;
           NEW.geometry_sector := 2424;