"""
Implementation of the acutal database accesses for forward search.
"""
-from typing import List, Tuple, AsyncIterator, Dict, Any, Callable
+from typing import List, Tuple, AsyncIterator, Dict, Any, Callable, cast
import abc
import sqlalchemy as sa
-from sqlalchemy.dialects.postgresql import ARRAY, array_agg
from nominatim.typing import SaFromClause, SaScalarSelect, SaColumn, \
SaExpression, SaSelect, SaLambdaSelect, SaRow, SaBind
from nominatim.api.types import SearchDetails, DataLayer, GeometryFormat, Bbox
import nominatim.api.results as nres
from nominatim.api.search.db_search_fields import SearchData, WeightedCategories
-from nominatim.db.sqlalchemy_types import Geometry
+from nominatim.db.sqlalchemy_types import Geometry, IntArray
#pylint: disable=singleton-comparison,not-callable
#pylint: disable=too-many-branches,too-many-arguments,too-many-locals,too-many-statements
NEAR_RADIUS_PARAM: SaBind = sa.bindparam('near_radius')
COUNTRIES_PARAM: SaBind = sa.bindparam('countries')
-def _within_near(t: SaFromClause) -> Callable[[], SaExpression]:
- return lambda: t.c.geometry.ST_DWithin(NEAR_PARAM, NEAR_RADIUS_PARAM)
+
+def filter_by_area(sql: SaSelect, t: SaFromClause,
+ details: SearchDetails, avoid_index: bool = False) -> SaSelect:
+ """ Apply SQL statements for filtering by viewbox and near point,
+ if applicable.
+ """
+ if details.near is not None and details.near_radius is not None:
+ if details.near_radius < 0.1 and not avoid_index:
+ sql = sql.where(t.c.geometry.within_distance(NEAR_PARAM, NEAR_RADIUS_PARAM))
+ else:
+ sql = sql.where(t.c.geometry.ST_Distance(NEAR_PARAM) <= NEAR_RADIUS_PARAM)
+ if details.viewbox is not None and details.bounded_viewbox:
+ sql = sql.where(t.c.geometry.intersects(VIEWBOX_PARAM,
+ use_index=not avoid_index and
+ details.viewbox.area < 0.2))
+
+ return sql
+
def _exclude_places(t: SaFromClause) -> Callable[[], SaExpression]:
return lambda: t.c.place_id.not_in(sa.bindparam('excluded'))
+
def _select_placex(t: SaFromClause) -> SaSelect:
return sa.select(t.c.place_id, t.c.osm_type, t.c.osm_id, t.c.name,
t.c.class_, t.c.type,
t.c.address, t.c.extratags,
t.c.housenumber, t.c.postcode, t.c.country_code,
- t.c.importance, t.c.wikipedia,
+ t.c.wikipedia,
t.c.parent_place_id, t.c.rank_address, t.c.rank_search,
t.c.linked_place_id, t.c.admin_level,
t.c.centroid,
def _make_interpolation_subquery(table: SaFromClause, inner: SaFromClause,
numerals: List[int], details: SearchDetails) -> SaScalarSelect:
- all_ids = array_agg(table.c.place_id) # type: ignore[no-untyped-call]
+ all_ids = sa.func.ArrayAgg(table.c.place_id)
sql = sa.select(all_ids).where(table.c.parent_place_id == inner.c.place_id)
if len(numerals) == 1:
orexpr.append(no_index(table.c.rank_address).between(1, 30))
elif layers & DataLayer.ADDRESS:
orexpr.append(no_index(table.c.rank_address).between(1, 29))
- orexpr.append(sa.and_(no_index(table.c.rank_address) == 30,
- sa.or_(table.c.housenumber != None,
- table.c.address.has_key('addr:housename'))))
+ orexpr.append(sa.func.IsAddressPoint(table))
elif layers & DataLayer.POI:
orexpr.append(sa.and_(no_index(table.c.rank_address) == 30,
table.c.class_.not_in(('place', 'building'))))
place_ids: List[int],
details: SearchDetails) -> AsyncIterator[nres.SearchResult]:
t = conn.t.placex
- sql = _select_placex(t).where(t.c.place_id.in_(place_ids))
+ sql = _select_placex(t).add_columns(t.c.importance)\
+ .where(t.c.place_id.in_(place_ids))
if details.geometry_output:
sql = _add_geometry_columns(sql, t.c.geometry, details)
yield result
+def _int_list_to_subquery(inp: List[int]) -> 'sa.Subquery':
+ """ Create a subselect that returns the given list of integers
+ as rows in the column 'nr'.
+ """
+ vtab = sa.func.JsonArrayEach(sa.type_coerce(inp, sa.JSON))\
+ .table_valued(sa.column('value', type_=sa.JSON))
+ return sa.select(sa.cast(sa.cast(vtab.c.value, sa.Text), sa.Integer).label('nr')).subquery()
+
+
async def _get_osmline(conn: SearchConnection, place_ids: List[int],
numerals: List[int],
details: SearchDetails) -> AsyncIterator[nres.SearchResult]:
t = conn.t.osmline
- values = sa.values(sa.Column('nr', sa.Integer()), name='housenumber')\
- .data([(n,) for n in numerals])
+
+ values = _int_list_to_subquery(numerals)
sql = sa.select(t.c.place_id, t.c.osm_id,
t.c.parent_place_id, t.c.address,
values.c.nr.label('housenumber'),
numerals: List[int], osm_id: int,
details: SearchDetails) -> AsyncIterator[nres.SearchResult]:
t = conn.t.tiger
- values = sa.values(sa.Column('nr', sa.Integer()), name='housenumber')\
- .data([(n,) for n in numerals])
+ values = _int_list_to_subquery(numerals)
sql = sa.select(t.c.place_id, t.c.parent_place_id,
sa.literal('W').label('osm_type'),
sa.literal(osm_id).label('osm_id'),
class AbstractSearch(abc.ABC):
""" Encapuslation of a single lookup in the database.
"""
+ SEARCH_PRIO: int = 2
def __init__(self, penalty: float) -> None:
self.penalty = penalty
base.sort(key=lambda r: (r.accuracy, r.rank_search))
max_accuracy = base[0].accuracy + 0.5
+ if base[0].rank_address == 0:
+ min_rank = 0
+ max_rank = 0
+ elif base[0].rank_address < 26:
+ min_rank = 1
+ max_rank = min(25, base[0].rank_address + 4)
+ else:
+ min_rank = 26
+ max_rank = 30
base = nres.SearchResults(r for r in base if r.source_table == nres.SourceTable.PLACEX
and r.accuracy <= max_accuracy
- and r.bbox and r.bbox.area < 20)
+ and r.bbox and r.bbox.area < 20
+ and r.rank_address >= min_rank
+ and r.rank_address <= max_rank)
if base:
baseids = [b.place_id for b in base[:5] if b.place_id]
"""
table = await conn.get_class_table(*category)
- t = conn.t.placex
tgeom = conn.t.placex.alias('pgeom')
- sql = _select_placex(t).where(tgeom.c.place_id.in_(ids))\
- .where(t.c.class_ == category[0])\
- .where(t.c.type == category[1])
-
if table is None:
# No classtype table available, do a simplified lookup in placex.
- sql = sql.join(tgeom, t.c.geometry.ST_DWithin(tgeom.c.centroid, 0.01))\
- .order_by(tgeom.c.centroid.ST_Distance(t.c.centroid))
+ table = conn.t.placex
+ sql = sa.select(table.c.place_id,
+ sa.func.min(tgeom.c.centroid.ST_Distance(table.c.centroid))
+ .label('dist'))\
+ .join(tgeom, table.c.geometry.intersects(tgeom.c.centroid.ST_Expand(0.01)))\
+ .where(table.c.class_ == category[0])\
+ .where(table.c.type == category[1])
else:
# Use classtype table. We can afford to use a larger
# radius for the lookup.
- sql = sql.join(table, t.c.place_id == table.c.place_id)\
- .join(tgeom,
- table.c.centroid.ST_CoveredBy(
- sa.case((sa.and_(tgeom.c.rank_address < 9,
+ sql = sa.select(table.c.place_id,
+ sa.func.min(tgeom.c.centroid.ST_Distance(table.c.centroid))
+ .label('dist'))\
+ .join(tgeom,
+ table.c.centroid.ST_CoveredBy(
+ sa.case((sa.and_(tgeom.c.rank_address > 9,
tgeom.c.geometry.is_area()),
- tgeom.c.geometry),
- else_ = tgeom.c.centroid.ST_Expand(0.05))))\
- .order_by(tgeom.c.centroid.ST_Distance(table.c.centroid))
+ tgeom.c.geometry),
+ else_ = tgeom.c.centroid.ST_Expand(0.05))))
+
+ inner = sql.where(tgeom.c.place_id.in_(ids))\
+ .group_by(table.c.place_id).subquery()
+
+ t = conn.t.placex
+ sql = _select_placex(t).add_columns((-inner.c.dist).label('importance'))\
+ .join(inner, inner.c.place_id == t.c.place_id)\
+ .order_by(inner.c.dist)
sql = sql.where(no_index(t.c.rank_address).between(MIN_RANK_PARAM, MAX_RANK_PARAM))
if details.countries:
# simply search in placex table
def _base_query() -> SaSelect:
return _select_placex(t) \
+ .add_columns((-t.c.centroid.ST_Distance(NEAR_PARAM))
+ .label('importance'))\
.where(t.c.linked_place_id == None) \
- .where(t.c.geometry.ST_DWithin(NEAR_PARAM, NEAR_RADIUS_PARAM)) \
+ .where(t.c.geometry.within_distance(NEAR_PARAM, NEAR_RADIUS_PARAM)) \
.order_by(t.c.centroid.ST_Distance(NEAR_PARAM)) \
.limit(LIMIT_PARAM)
table = await conn.get_class_table(*category)
if table is not None:
sql = _select_placex(t)\
+ .add_columns(t.c.importance)\
.join(table, t.c.place_id == table.c.place_id)\
.where(t.c.class_ == category[0])\
.where(t.c.type == category[1])
if details.near and details.near_radius is not None:
sql = sql.order_by(table.c.centroid.ST_Distance(NEAR_PARAM))\
- .where(table.c.centroid.ST_DWithin(NEAR_PARAM,
- NEAR_RADIUS_PARAM))
+ .where(table.c.centroid.within_distance(NEAR_PARAM,
+ NEAR_RADIUS_PARAM))
if self.countries:
sql = sql.where(t.c.country_code.in_(self.countries.values))
class CountrySearch(AbstractSearch):
""" Search for a country name or country code.
"""
+ SEARCH_PRIO = 0
+
def __init__(self, sdata: SearchData) -> None:
super().__init__(sdata.penalty)
self.countries = sdata.countries
ccodes = self.countries.values
sql = _select_placex(t)\
+ .add_columns(t.c.importance)\
.where(t.c.country_code.in_(ccodes))\
.where(t.c.rank_address == 4)
if details.excluded:
sql = sql.where(_exclude_places(t))
- if details.viewbox is not None and details.bounded_viewbox:
- sql = sql.where(lambda: t.c.geometry.intersects(VIEWBOX_PARAM))
-
- if details.near is not None and details.near_radius is not None:
- sql = sql.where(_within_near(t))
+ sql = filter_by_area(sql, t, details)
results = nres.SearchResults()
for row in await conn.execute(sql, _details_to_bind_params(details)):
result.bbox = Bbox.from_wkb(row.bbox)
results.append(result)
- return results or await self.lookup_in_country_table(conn, details)
+ if not results:
+ results = await self.lookup_in_country_table(conn, details)
+
+ if results:
+ details.min_rank = min(5, details.max_rank)
+ details.max_rank = min(25, details.max_rank)
+
+ return results
async def lookup_in_country_table(self, conn: SearchConnection,
.where(tgrid.c.country_code.in_(self.countries.values))\
.group_by(tgrid.c.country_code)
- if details.viewbox is not None and details.bounded_viewbox:
- sql = sql.where(tgrid.c.geometry.intersects(VIEWBOX_PARAM))
- if details.near is not None and details.near_radius is not None:
- sql = sql.where(_within_near(tgrid))
+ sql = filter_by_area(sql, tgrid, details, avoid_index=True)
sub = sql.subquery('grid')
sql = sa.select(t.c.country_code,
- (t.c.name
- + sa.func.coalesce(t.c.derived_name,
- sa.cast('', type_=conn.t.types.Composite))
- ).label('name'),
+ t.c.name.merge(t.c.derived_name).label('name'),
sub.c.centroid, sub.c.bbox)\
.join(sub, t.c.country_code == sub.c.country_code)
penalty: SaExpression = sa.literal(self.penalty)
- if details.viewbox is not None:
- if details.bounded_viewbox:
- sql = sql.where(t.c.geometry.intersects(VIEWBOX_PARAM))
- else:
- penalty += sa.case((t.c.geometry.intersects(VIEWBOX_PARAM), 0.0),
- (t.c.geometry.intersects(VIEWBOX2_PARAM), 0.5),
- else_=1.0)
+ if details.viewbox is not None and not details.bounded_viewbox:
+ penalty += sa.case((t.c.geometry.intersects(VIEWBOX_PARAM), 0.0),
+ (t.c.geometry.intersects(VIEWBOX2_PARAM), 0.5),
+ else_=1.0)
if details.near is not None:
- if details.near_radius is not None:
- sql = sql.where(_within_near(t))
sql = sql.order_by(t.c.geometry.ST_Distance(NEAR_PARAM))
+ sql = filter_by_area(sql, t, details)
+
if self.countries:
sql = sql.where(t.c.country_code.in_(self.countries.values))
if self.lookups:
assert len(self.lookups) == 1
- assert self.lookups[0].lookup_type == 'restrict'
tsearch = conn.t.search_name
sql = sql.where(tsearch.c.place_id == t.c.parent_place_id)\
- .where(sa.func.array_cat(tsearch.c.name_vector,
- tsearch.c.nameaddress_vector,
- type_=ARRAY(sa.Integer))
- .contains(self.lookups[0].tokens))
+ .where((tsearch.c.name_vector + tsearch.c.nameaddress_vector)
+ .contains(sa.type_coerce(self.lookups[0].tokens,
+ IntArray)))
for ranking in self.rankings:
penalty += ranking.sql_penalty(conn.t.search_name)
class PlaceSearch(AbstractSearch):
""" Generic search for an address or named place.
"""
+ SEARCH_PRIO = 1
+
def __init__(self, extra_penalty: float, sdata: SearchData, expected_count: int) -> None:
super().__init__(sdata.penalty + extra_penalty)
self.countries = sdata.countries
tsearch = conn.t.search_name
sql: SaLambdaSelect = sa.lambda_stmt(lambda:
- sa.select(t.c.place_id, t.c.osm_type, t.c.osm_id, t.c.name,
- t.c.class_, t.c.type,
- t.c.address, t.c.extratags, t.c.admin_level,
- t.c.housenumber, t.c.postcode, t.c.country_code,
- t.c.wikipedia,
- t.c.parent_place_id, t.c.rank_address, t.c.rank_search,
- t.c.centroid,
- t.c.geometry.ST_Expand(0).label('bbox'))
- .where(t.c.place_id == tsearch.c.place_id))
+ _select_placex(t).where(t.c.place_id == tsearch.c.place_id))
if details.geometry_output:
sql = sql.where(tsearch.c.address_rank > 9)
tpc = conn.t.postcode
pcs = self.postcodes.values
- if self.expected_count > 1000:
+ if self.expected_count > 5000:
# Many results expected. Restrict by postcode.
sql = sql.where(sa.select(tpc.c.postcode)
.where(tpc.c.postcode.in_(pcs))
- .where(tsearch.c.centroid.ST_DWithin(tpc.c.geometry, 0.12))
+ .where(tsearch.c.centroid.within_distance(tpc.c.geometry, 0.12))
.exists())
# Less results, only have a preference for close postcodes
.where(tpc.c.postcode.in_(pcs))\
.scalar_subquery()
penalty += sa.case((t.c.postcode.in_(pcs), 0.0),
- else_=sa.func.coalesce(pc_near, 2.0))
+ else_=sa.func.coalesce(pc_near, cast(SaColumn, 2.0)))
if details.viewbox is not None:
if details.bounded_viewbox:
- if details.viewbox.area < 0.2:
- sql = sql.where(tsearch.c.centroid.intersects(VIEWBOX_PARAM))
- else:
- sql = sql.where(tsearch.c.centroid.ST_Intersects_no_index(VIEWBOX_PARAM))
- elif self.expected_count >= 10000:
- if details.viewbox.area < 0.5:
- sql = sql.where(tsearch.c.centroid.intersects(VIEWBOX2_PARAM))
- else:
- sql = sql.where(tsearch.c.centroid.ST_Intersects_no_index(VIEWBOX2_PARAM))
+ sql = sql.where(tsearch.c.centroid
+ .intersects(VIEWBOX_PARAM,
+ use_index=details.viewbox.area < 0.2))
+ elif not self.postcodes and not self.housenumbers and self.expected_count >= 10000:
+ sql = sql.where(tsearch.c.centroid
+ .intersects(VIEWBOX2_PARAM,
+ use_index=details.viewbox.area < 0.5))
else:
- penalty += sa.case((t.c.geometry.intersects(VIEWBOX_PARAM), 0.0),
- (t.c.geometry.intersects(VIEWBOX2_PARAM), 0.5),
+ penalty += sa.case((t.c.geometry.intersects(VIEWBOX_PARAM, use_index=False), 0.0),
+ (t.c.geometry.intersects(VIEWBOX2_PARAM, use_index=False), 0.5),
else_=1.0)
if details.near is not None:
if details.near_radius is not None:
if details.near_radius < 0.1:
- sql = sql.where(tsearch.c.centroid.ST_DWithin(NEAR_PARAM, NEAR_RADIUS_PARAM))
+ sql = sql.where(tsearch.c.centroid.within_distance(NEAR_PARAM,
+ NEAR_RADIUS_PARAM))
else:
- sql = sql.where(tsearch.c.centroid.ST_DWithin_no_index(NEAR_PARAM,
- NEAR_RADIUS_PARAM))
+ sql = sql.where(tsearch.c.centroid
+ .ST_Distance(NEAR_PARAM) < NEAR_RADIUS_PARAM)
sql = sql.add_columns((-tsearch.c.centroid.ST_Distance(NEAR_PARAM))
.label('importance'))
sql = sql.order_by(sa.desc(sa.text('importance')))
sql = sql.order_by(sa.text('accuracy'))
if self.housenumbers:
- hnr_regexp = f"\\m({'|'.join(self.housenumbers.values)})\\M"
+ hnr_list = '|'.join(self.housenumbers.values)
sql = sql.where(tsearch.c.address_rank.between(16, 30))\
.where(sa.or_(tsearch.c.address_rank < 30,
- t.c.housenumber.op('~*')(hnr_regexp)))
+ sa.func.RegexpWord(hnr_list, t.c.housenumber)))
# Cross check for housenumbers, need to do that on a rather large
# set. Worst case there are 40.000 main streets in OSM.
# Housenumbers from placex
thnr = conn.t.placex.alias('hnr')
- pid_list = array_agg(thnr.c.place_id) # type: ignore[no-untyped-call]
+ pid_list = sa.func.ArrayAgg(thnr.c.place_id)
place_sql = sa.select(pid_list)\
.where(thnr.c.parent_place_id == inner.c.place_id)\
- .where(thnr.c.housenumber.op('~*')(hnr_regexp))\
+ .where(sa.func.RegexpWord(hnr_list, thnr.c.housenumber))\
.where(thnr.c.linked_place_id == None)\
.where(thnr.c.indexed_status == 0)
assert result
result.bbox = Bbox.from_wkb(row.bbox)
result.accuracy = row.accuracy
- if not details.excluded or not result.place_id in details.excluded:
- results.append(result)
-
if self.housenumbers and row.rank_address < 30:
if row.placex_hnr:
subs = _get_placex_housenumbers(conn, row.placex_hnr, details)
sub.accuracy += 0.6
results.append(sub)
- result.accuracy += 1.0 # penalty for missing housenumber
+ # Only add the street as a result, if it meets all other
+ # filter conditions.
+ if (not details.excluded or result.place_id not in details.excluded)\
+ and (not self.qualifiers or result.category in self.qualifiers.values)\
+ and result.rank_address >= details.min_rank:
+ result.accuracy += 1.0 # penalty for missing housenumber
+ results.append(result)
+ else:
+ results.append(result)
return results
projects / nominatim.git / blobdiff