]> git.openstreetmap.org Git - nominatim.git/blobdiff - nominatim/api/results.py
Merge pull request #3257 from lonvia/slow-queries
[nominatim.git] / nominatim / api / results.py
index 98b13380726e98361bc2571c85bd78928c1afc12..829008fb3ec316a09a753b39cb720404b1c8d98b 100644 (file)
@@ -11,7 +11,7 @@ Data classes are part of the public API while the functions are for
 internal use only. That's why they are implemented as free-standing functions
 instead of member functions.
 """
-from typing import Optional, Tuple, Dict, Sequence, TypeVar, Type, List
+from typing import Optional, Tuple, Dict, Sequence, TypeVar, Type, List, cast
 import enum
 import dataclasses
 import datetime as dt
@@ -19,6 +19,7 @@ import datetime as dt
 import sqlalchemy as sa
 
 from nominatim.typing import SaSelect, SaRow
+from nominatim.db.sqlalchemy_types import Geometry
 from nominatim.api.types import Point, Bbox, LookupDetails
 from nominatim.api.connection import SearchConnection
 from nominatim.api.logging import log
@@ -27,33 +28,107 @@ from nominatim.api.localization import Locales
 # This file defines complex result data classes.
 # pylint: disable=too-many-instance-attributes
 
+def _mingle_name_tags(names: Optional[Dict[str, str]]) -> Optional[Dict[str, str]]:
+    """ Mix-in names from linked places, so that they show up
+        as standard names where necessary.
+    """
+    if not names:
+        return None
+
+    out = {}
+    for k, v in names.items():
+        if k.startswith('_place_'):
+            outkey = k[7:]
+            out[k if outkey in names else outkey] = v
+        else:
+            out[k] = v
+
+    return out
+
+
 class SourceTable(enum.Enum):
-    """ Enumeration of kinds of results.
+    """ The `SourceTable` type lists the possible sources a result can have.
     """
     PLACEX = 1
+    """ The placex table is the main source for result usually containing
+        OSM data.
+    """
     OSMLINE = 2
+    """ The osmline table contains address interpolations from OSM data.
+        Interpolation addresses are always approximate. The OSM id in the
+        result refers to the OSM way with the interpolation line object.
+    """
     TIGER = 3
+    """ TIGER address data contains US addresses imported on the side,
+        see [Installing TIGER data](../customize/Tiger.md).
+        TIGER address are also interpolations. The addresses always refer
+        to a street from OSM data. The OSM id in the result refers to
+        that street.
+    """
     POSTCODE = 4
+    """ The postcode table contains artificial centroids for postcodes,
+        computed from the postcodes available with address points. Results
+        are always approximate.
+    """
     COUNTRY = 5
+    """ The country table provides a fallback, when country data is missing
+        in the OSM data.
+    """
 
 
 @dataclasses.dataclass
 class AddressLine:
-    """ Detailed information about a related place.
+    """ The `AddressLine` may contain the following fields about a related place
+        and its function as an address object. Most fields are optional.
+        Their presence depends on the kind and function of the address part.
     """
-    place_id: Optional[int]
-    osm_object: Optional[Tuple[str, int]]
     category: Tuple[str, str]
+    """ Main category of the place, described by a key-value pair.
+    """
     names: Dict[str, str]
-    extratags: Optional[Dict[str, str]]
-
-    admin_level: Optional[int]
+    """ All available names for the place including references, alternative
+        names and translations.
+    """
     fromarea: bool
+    """ If true, then the exact area of the place is known. Without area
+        information, Nominatim has to make an educated guess if an address
+        belongs to one place or another.
+    """
     isaddress: bool
+    """ If true, this place should be considered for the final address display.
+        Nominatim will sometimes include more than one candidate for
+        the address in the list when it cannot reliably determine where the
+        place belongs. It will consider names of all candidates when searching
+        but when displaying the result, only the most likely candidate should
+        be shown.
+    """
     rank_address: int
+    """ [Address rank](../customize/Ranking.md#address-rank) of the place.
+    """
     distance: float
+    """ Distance in degrees between the result place and this address part.
+    """
+    place_id: Optional[int] = None
+    """ Internal ID of the place.
+    """
+    osm_object: Optional[Tuple[str, int]] = None
+    """ OSM type and ID of the place, if such an object exists.
+    """
+    extratags: Optional[Dict[str, str]] = None
+    """ Any extra information available about the place. This is a dictionary
+        that usually contains OSM tag key-value pairs.
+    """
+
+    admin_level: Optional[int] = None
+    """ The administrative level of a boundary as tagged in the input data.
+        This field is only meaningful for places of the category
+        (boundary, administrative).
+    """
 
     local_name: Optional[str] = None
+    """ Place holder for localization of this address part. See
+        [Localization](#localization) below.
+    """
 
 
 class AddressLines(List[AddressLine]):
@@ -62,7 +137,7 @@ class AddressLines(List[AddressLine]):
 
     def localize(self, locales: Locales) -> List[str]:
         """ Set the local name of address parts according to the chosen
-            locale. Return the list of local names without duplications.
+            locale. Return the list of local names without duplicates.
 
             Only address parts that are marked as isaddress are localized
             and returned.
@@ -81,11 +156,19 @@ class AddressLines(List[AddressLine]):
 
 @dataclasses.dataclass
 class WordInfo:
-    """ Detailed information about a search term.
+    """ Each entry in the list of search terms contains the
+        following detailed information.
     """
     word_id: int
+    """ Internal identifier for the word.
+    """
     word_token: str
+    """ Normalised and transliterated form of the word.
+        This form is used for searching.
+    """
     word: Optional[str] = None
+    """ Untransliterated form, if available.
+    """
 
 
 WordInfos = Sequence[WordInfo]
@@ -102,6 +185,12 @@ class BaseResult:
 
     place_id : Optional[int] = None
     osm_object: Optional[Tuple[str, int]] = None
+    parent_place_id: Optional[int] = None
+    linked_place_id: Optional[int] = None
+    admin_level: int = 15
+
+    locale_name: Optional[str] = None
+    display_name: Optional[str] = None
 
     names: Optional[Dict[str, str]] = None
     address: Optional[Dict[str, str]] = None
@@ -146,6 +235,19 @@ class BaseResult:
         """
         return self.importance or (0.7500001 - (self.rank_search/40.0))
 
+
+    def localize(self, locales: Locales) -> None:
+        """ Fill the locale_name and the display_name field for the
+            place and, if available, its address information.
+        """
+        self.locale_name = locales.display_name(self.names)
+        if self.address_rows:
+            self.display_name = ', '.join(self.address_rows.localize(locales))
+        else:
+            self.display_name = self.locale_name
+
+
+
 BaseResultT = TypeVar('BaseResultT', bound=BaseResult)
 
 @dataclasses.dataclass
@@ -153,9 +255,6 @@ class DetailedResult(BaseResult):
     """ A search result with more internal information from the database
         added.
     """
-    parent_place_id: Optional[int] = None
-    linked_place_id: Optional[int] = None
-    admin_level: int = 15
     indexed_date: Optional[dt.datetime] = None
 
 
@@ -178,6 +277,15 @@ class SearchResult(BaseResult):
     """ A search result for forward geocoding.
     """
     bbox: Optional[Bbox] = None
+    accuracy: float = 0.0
+
+
+    @property
+    def ranking(self) -> float:
+        """ Return the ranking, a combined measure of accuracy and importance.
+        """
+        return (self.accuracy if self.accuracy is not None else 1) \
+               - self.calculated_importance()
 
 
 class SearchResults(List[SearchResult]):
@@ -204,7 +312,10 @@ def create_from_placex_row(row: Optional[SaRow],
                       place_id=row.place_id,
                       osm_object=(row.osm_type, row.osm_id),
                       category=(row.class_, row.type),
-                      names=row.name,
+                      parent_place_id = row.parent_place_id,
+                      linked_place_id = getattr(row, 'linked_place_id', None),
+                      admin_level = getattr(row, 'admin_level', 15),
+                      names=_mingle_name_tags(row.name),
                       address=row.address,
                       extratags=row.extratags,
                       housenumber=row.housenumber,
@@ -214,7 +325,7 @@ def create_from_placex_row(row: Optional[SaRow],
                       rank_search=row.rank_search,
                       importance=row.importance,
                       country_code=row.country_code,
-                      centroid=Point.from_wkb(row.centroid.data),
+                      centroid=Point.from_wkb(row.centroid),
                       geometry=_filter_geometries(row))
 
 
@@ -234,12 +345,13 @@ def create_from_osmline_row(row: Optional[SaRow],
 
     res = class_type(source_table=SourceTable.OSMLINE,
                      place_id=row.place_id,
+                     parent_place_id = row.parent_place_id,
                      osm_object=('W', row.osm_id),
                      category=('place', 'houses' if hnr is None else 'house'),
                      address=row.address,
                      postcode=row.postcode,
                      country_code=row.country_code,
-                     centroid=Point.from_wkb(row.centroid.data),
+                     centroid=Point.from_wkb(row.centroid),
                      geometry=_filter_geometries(row))
 
     if hnr is None:
@@ -253,7 +365,9 @@ def create_from_osmline_row(row: Optional[SaRow],
 
 
 def create_from_tiger_row(row: Optional[SaRow],
-                          class_type: Type[BaseResultT]) -> Optional[BaseResultT]:
+                          class_type: Type[BaseResultT],
+                          osm_type: Optional[str] = None,
+                          osm_id: Optional[int] = None) -> Optional[BaseResultT]:
     """ Construct a new result and add the data from the result row
         from the Tiger data interpolation table. 'class_type' defines
         the type of result to return. Returns None if the row is None.
@@ -268,11 +382,12 @@ def create_from_tiger_row(row: Optional[SaRow],
 
     res = class_type(source_table=SourceTable.TIGER,
                      place_id=row.place_id,
-                     osm_object=(row.osm_type, row.osm_id),
+                     parent_place_id = row.parent_place_id,
+                     osm_object=(osm_type or row.osm_type, osm_id or row.osm_id),
                      category=('place', 'houses' if hnr is None else 'house'),
                      postcode=row.postcode,
                      country_code='us',
-                     centroid=Point.from_wkb(row.centroid.data),
+                     centroid=Point.from_wkb(row.centroid),
                      geometry=_filter_geometries(row))
 
     if hnr is None:
@@ -296,48 +411,74 @@ def create_from_postcode_row(row: Optional[SaRow],
 
     return class_type(source_table=SourceTable.POSTCODE,
                       place_id=row.place_id,
+                      parent_place_id = row.parent_place_id,
                       category=('place', 'postcode'),
                       names={'ref': row.postcode},
                       rank_search=row.rank_search,
                       rank_address=row.rank_address,
                       country_code=row.country_code,
-                      centroid=Point.from_wkb(row.centroid.data),
+                      centroid=Point.from_wkb(row.centroid),
                       geometry=_filter_geometries(row))
 
 
-async def add_result_details(conn: SearchConnection, result: BaseResult,
+def create_from_country_row(row: Optional[SaRow],
+                        class_type: Type[BaseResultT]) -> Optional[BaseResultT]:
+    """ Construct a new result and add the data from the result row
+        from the fallback country tables. 'class_type' defines
+        the type of result to return. Returns None if the row is None.
+    """
+    if row is None:
+        return None
+
+    return class_type(source_table=SourceTable.COUNTRY,
+                      category=('place', 'country'),
+                      centroid=Point.from_wkb(row.centroid),
+                      names=row.name,
+                      rank_address=4, rank_search=4,
+                      country_code=row.country_code,
+                      geometry=_filter_geometries(row))
+
+
+async def add_result_details(conn: SearchConnection, results: List[BaseResultT],
                              details: LookupDetails) -> None:
     """ Retrieve more details from the database according to the
         parameters specified in 'details'.
     """
-    log().section('Query details for result')
-    if details.address_details:
-        log().comment('Query address details')
-        await complete_address_details(conn, result)
-    if details.linked_places:
-        log().comment('Query linked places')
-        await complete_linked_places(conn, result)
-    if details.parented_places:
-        log().comment('Query parent places')
-        await complete_parented_places(conn, result)
-    if details.keywords:
-        log().comment('Query keywords')
-        await complete_keywords(conn, result)
-
-
-def _result_row_to_address_row(row: SaRow) -> AddressLine:
+    if results:
+        log().section('Query details for result')
+        if details.address_details:
+            log().comment('Query address details')
+            await complete_address_details(conn, results)
+        if details.linked_places:
+            log().comment('Query linked places')
+            for result in results:
+                await complete_linked_places(conn, result)
+        if details.parented_places:
+            log().comment('Query parent places')
+            for result in results:
+                await complete_parented_places(conn, result)
+        if details.keywords:
+            log().comment('Query keywords')
+            for result in results:
+                await complete_keywords(conn, result)
+        for result in results:
+            result.localize(details.locales)
+
+
+def _result_row_to_address_row(row: SaRow, isaddress: Optional[bool] = None) -> AddressLine:
     """ Create a new AddressLine from the results of a datbase query.
     """
-    extratags: Dict[str, str] = getattr(row, 'extratags', {})
-    if hasattr(row, 'place_type') and row.place_type:
-        extratags['place'] = row.place_type
+    extratags: Dict[str, str] = getattr(row, 'extratags', {}) or {}
+    if 'linked_place' in extratags:
+        extratags['place'] = extratags['linked_place']
 
-    names = row.name
+    names = _mingle_name_tags(row.name) or {}
     if getattr(row, 'housenumber', None) is not None:
-        if names is None:
-            names = {}
         names['housenumber'] = row.housenumber
 
+    if isaddress is None:
+        isaddress = getattr(row, 'isaddress', True)
+
     return AddressLine(place_id=row.place_id,
                        osm_object=None if row.osm_type is None else (row.osm_type, row.osm_id),
                        category=(getattr(row, 'class'), row.type),
@@ -345,56 +486,211 @@ def _result_row_to_address_row(row: SaRow) -> AddressLine:
                        extratags=extratags,
                        admin_level=row.admin_level,
                        fromarea=row.fromarea,
-                       isaddress=getattr(row, 'isaddress', True),
+                       isaddress=isaddress,
                        rank_address=row.rank_address,
                        distance=row.distance)
 
 
-async def complete_address_details(conn: SearchConnection, result: BaseResult) -> None:
+def _get_address_lookup_id(result: BaseResultT) -> int:
+    assert result.place_id
+    if result.source_table != SourceTable.PLACEX or result.rank_search > 27:
+        return result.parent_place_id or result.place_id
+
+    return result.linked_place_id or result.place_id
+
+
+async def _finalize_entry(conn: SearchConnection, result: BaseResultT) -> None:
+    assert result.address_rows is not None
+    postcode = result.postcode
+    if not postcode and result.address:
+        postcode = result.address.get('postcode')
+    if postcode and ',' not in postcode and ';' not in postcode:
+        result.address_rows.append(AddressLine(
+            category=('place', 'postcode'),
+            names={'ref': postcode},
+            fromarea=False, isaddress=True, rank_address=5,
+            distance=0.0))
+    if result.country_code:
+        async def _get_country_names() -> Optional[Dict[str, str]]:
+            t = conn.t.country_name
+            sql = sa.select(t.c.name, t.c.derived_name)\
+                    .where(t.c.country_code == result.country_code)
+            for cres in await conn.execute(sql):
+                names = cast(Dict[str, str], cres[0])
+                if cres[1]:
+                    names.update(cast(Dict[str, str], cres[1]))
+                return names
+            return None
+
+        country_names = await conn.get_cached_value('COUNTRY_NAME',
+                                                    result.country_code,
+                                                    _get_country_names)
+        if country_names:
+            result.address_rows.append(AddressLine(
+                category=('place', 'country'),
+                names=country_names,
+                fromarea=False, isaddress=True, rank_address=4,
+                distance=0.0))
+        result.address_rows.append(AddressLine(
+            category=('place', 'country_code'),
+            names={'ref': result.country_code}, extratags = {},
+            fromarea=True, isaddress=False, rank_address=4,
+            distance=0.0))
+
+
+def _setup_address_details(result: BaseResultT) -> None:
     """ Retrieve information about places that make up the address of the result.
     """
-    housenumber = -1
-    if result.source_table in (SourceTable.TIGER, SourceTable.OSMLINE):
-        if result.housenumber is not None:
-            housenumber = int(result.housenumber)
-        elif result.extratags is not None and 'startnumber' in result.extratags:
-            # details requests do not come with a specific house number
-            housenumber = int(result.extratags['startnumber'])
-
-    sfn = sa.func.get_addressdata(result.place_id, housenumber)\
-            .table_valued( # type: ignore[no-untyped-call]
-                sa.column('place_id', type_=sa.Integer),
-                'osm_type',
-                sa.column('osm_id', type_=sa.BigInteger),
-                sa.column('name', type_=conn.t.types.Composite),
-                'class', 'type', 'place_type',
-                sa.column('admin_level', type_=sa.Integer),
-                sa.column('fromarea', type_=sa.Boolean),
-                sa.column('isaddress', type_=sa.Boolean),
-                sa.column('rank_address', type_=sa.SmallInteger),
-                sa.column('distance', type_=sa.Float))
-    sql = sa.select(sfn).order_by(sa.column('rank_address').desc(),
-                                  sa.column('isaddress').desc())
-
     result.address_rows = AddressLines()
+    if result.names:
+        result.address_rows.append(AddressLine(
+            place_id=result.place_id,
+            osm_object=result.osm_object,
+            category=result.category,
+            names=result.names,
+            extratags=result.extratags or {},
+            admin_level=result.admin_level,
+            fromarea=True, isaddress=True,
+            rank_address=result.rank_address or 100, distance=0.0))
+    if result.source_table == SourceTable.PLACEX and result.address:
+        housenumber = result.address.get('housenumber')\
+                      or result.address.get('streetnumber')\
+                      or result.address.get('conscriptionnumber')
+    elif result.housenumber:
+        housenumber = result.housenumber
+    else:
+        housenumber = None
+    if housenumber:
+        result.address_rows.append(AddressLine(
+            category=('place', 'house_number'),
+            names={'ref': housenumber},
+            fromarea=True, isaddress=True, rank_address=28, distance=0))
+    if result.address and '_unlisted_place' in result.address:
+        result.address_rows.append(AddressLine(
+            category=('place', 'locality'),
+            names={'name': result.address['_unlisted_place']},
+            fromarea=False, isaddress=True, rank_address=25, distance=0))
+
+
+async def complete_address_details(conn: SearchConnection, results: List[BaseResultT]) -> None:
+    """ Retrieve information about places that make up the address of the result.
+    """
+    for result in results:
+        _setup_address_details(result)
+
+    ### Lookup entries from place_address line
+
+    lookup_ids = [{'pid': r.place_id,
+                   'lid': _get_address_lookup_id(r),
+                   'names': list(r.address.values()) if r.address else [],
+                   'c': ('SRID=4326;' + r.centroid.to_wkt()) if r.centroid else '' }
+                  for r in results if r.place_id]
+
+    if not lookup_ids:
+        return
+
+    ltab = sa.func.JsonArrayEach(sa.type_coerce(lookup_ids, sa.JSON))\
+             .table_valued(sa.column('value', type_=sa.JSON)) # type: ignore[no-untyped-call]
+
+    t = conn.t.placex
+    taddr = conn.t.addressline
+
+    sql = sa.select(ltab.c.value['pid'].as_integer().label('src_place_id'),
+                    t.c.place_id, t.c.osm_type, t.c.osm_id, t.c.name,
+                    t.c.class_, t.c.type, t.c.extratags,
+                    t.c.admin_level, taddr.c.fromarea,
+                    sa.case((t.c.rank_address == 11, 5),
+                            else_=t.c.rank_address).label('rank_address'),
+                    taddr.c.distance, t.c.country_code, t.c.postcode)\
+            .join(taddr, sa.or_(taddr.c.place_id == ltab.c.value['pid'].as_integer(),
+                                taddr.c.place_id == ltab.c.value['lid'].as_integer()))\
+            .join(t, taddr.c.address_place_id == t.c.place_id)\
+            .order_by('src_place_id')\
+            .order_by(sa.column('rank_address').desc())\
+            .order_by((taddr.c.place_id == ltab.c.value['pid'].as_integer()).desc())\
+            .order_by(sa.case((sa.func.CrosscheckNames(t.c.name, ltab.c.value['names']), 2),
+                              (taddr.c.isaddress, 0),
+                              (sa.and_(taddr.c.fromarea,
+                                       t.c.geometry.ST_Contains(
+                                           sa.func.ST_GeomFromEWKT(
+                                               ltab.c.value['c'].as_string()))), 1),
+                              else_=-1).desc())\
+            .order_by(taddr.c.fromarea.desc())\
+            .order_by(taddr.c.distance.desc())\
+            .order_by(t.c.rank_search.desc())
+
+
+    current_result = None
+    current_rank_address = -1
     for row in await conn.execute(sql):
-        result.address_rows.append(_result_row_to_address_row(row))
+        if current_result is None or row.src_place_id != current_result.place_id:
+            current_result = next((r for r in results if r.place_id == row.src_place_id), None)
+            assert current_result is not None
+            current_rank_address = -1
+
+        location_isaddress = row.rank_address != current_rank_address
+
+        if current_result.country_code is None and row.country_code:
+            current_result.country_code = row.country_code
+
+        if row.type in ('postcode', 'postal_code') and location_isaddress:
+            if not row.fromarea or \
+               (current_result.address and 'postcode' in current_result.address):
+                location_isaddress = False
+            else:
+                current_result.postcode = None
+
+        assert current_result.address_rows is not None
+        current_result.address_rows.append(_result_row_to_address_row(row, location_isaddress))
+        current_rank_address = row.rank_address
+
+    for result in results:
+        await _finalize_entry(conn, result)
+
+
+    ### Finally add the record for the parent entry where necessary.
+
+    parent_lookup_ids = list(filter(lambda e: e['pid'] != e['lid'], lookup_ids))
+    if parent_lookup_ids:
+        ltab = sa.func.JsonArrayEach(sa.type_coerce(parent_lookup_ids, sa.JSON))\
+                 .table_valued(sa.column('value', type_=sa.JSON)) # type: ignore[no-untyped-call]
+        sql = sa.select(ltab.c.value['pid'].as_integer().label('src_place_id'),
+                        t.c.place_id, t.c.osm_type, t.c.osm_id, t.c.name,
+                        t.c.class_, t.c.type, t.c.extratags,
+                        t.c.admin_level,
+                        t.c.rank_address)\
+                 .where(t.c.place_id == ltab.c.value['lid'].as_integer())
+
+        for row in await conn.execute(sql):
+            current_result = next((r for r in results if r.place_id == row.src_place_id), None)
+            assert current_result is not None
+            assert current_result.address_rows is not None
+
+            current_result.address_rows.append(AddressLine(
+                    place_id=row.place_id,
+                    osm_object=(row.osm_type, row.osm_id),
+                    category=(row.class_, row.type),
+                    names=row.name, extratags=row.extratags or {},
+                    admin_level=row.admin_level,
+                    fromarea=True, isaddress=True,
+                    rank_address=row.rank_address, distance=0.0))
+
+    ### Now sort everything
+    for result in results:
+        assert result.address_rows is not None
+        result.address_rows.sort(key=lambda a: (-a.rank_address, a.isaddress))
 
 
-# pylint: disable=consider-using-f-string
 def _placex_select_address_row(conn: SearchConnection,
                                centroid: Point) -> SaSelect:
     t = conn.t.placex
     return sa.select(t.c.place_id, t.c.osm_type, t.c.osm_id, t.c.name,
                      t.c.class_.label('class'), t.c.type,
                      t.c.admin_level, t.c.housenumber,
-                     sa.literal_column("""ST_GeometryType(geometry) in
-                                        ('ST_Polygon','ST_MultiPolygon')""").label('fromarea'),
+                     t.c.geometry.is_area().label('fromarea'),
                      t.c.rank_address,
-                     sa.literal_column(
-                         """ST_DistanceSpheroid(geometry, 'SRID=4326;POINT(%f %f)'::geometry,
-                              'SPHEROID["WGS 84",6378137,298.257223563, AUTHORITY["EPSG","7030"]]')
-                         """ % centroid).label('distance'))
+                     t.c.geometry.distance_spheroid(
+                       sa.bindparam('centroid', value=centroid, type_=Geometry)).label('distance'))
 
 
 async def complete_linked_places(conn: SearchConnection, result: BaseResult) -> None:
@@ -413,6 +709,9 @@ async def complete_linked_places(conn: SearchConnection, result: BaseResult) ->
 
 async def complete_keywords(conn: SearchConnection, result: BaseResult) -> None:
     """ Retrieve information about the search terms used for this place.
+
+        Requires that the query analyzer was initialised to get access to
+        the word table.
     """
     t = conn.t.search_name
     sql = sa.select(t.c.name_vector, t.c.nameaddress_vector)\
@@ -420,14 +719,15 @@ async def complete_keywords(conn: SearchConnection, result: BaseResult) -> None:
 
     result.name_keywords = []
     result.address_keywords = []
-    for name_tokens, address_tokens in await conn.execute(sql):
-        t = conn.t.word
-        sel = sa.select(t.c.word_id, t.c.word_token, t.c.word)
 
-        for row in await conn.execute(sel.where(t.c.word_id == sa.any_(name_tokens))):
+    t = conn.t.meta.tables['word']
+    sel = sa.select(t.c.word_id, t.c.word_token, t.c.word)
+
+    for name_tokens, address_tokens in await conn.execute(sql):
+        for row in await conn.execute(sel.where(t.c.word_id.in_(name_tokens))):
             result.name_keywords.append(WordInfo(*row))
 
-        for row in await conn.execute(sel.where(t.c.word_id == sa.any_(address_tokens))):
+        for row in await conn.execute(sel.where(t.c.word_id.in_(address_tokens))):
             result.address_keywords.append(WordInfo(*row))