#pylint: disable=singleton-comparison,not-callable
#pylint: disable=too-many-branches,too-many-arguments,too-many-locals,too-many-statements
+def no_index(expr: SaColumn) -> SaColumn:
+ """ Wrap the given expression, so that the query planner will
+ refrain from using the expression for index lookup.
+ """
+ return sa.func.coalesce(sa.null(), expr) # pylint: disable=not-callable
+
+
def _details_to_bind_params(details: SearchDetails) -> Dict[str, Any]:
""" Create a dictionary from search parameters that can be used
as bind parameter for SQL execute.
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,
t.c.geometry.ST_Expand(0).label('bbox'))
col = sa.func.ST_SimplifyPreserveTopology(col, details.geometry_simplification)
if details.geometry_output & GeometryFormat.GEOJSON:
- out.append(sa.func.ST_AsGeoJSON(col).label('geometry_geojson'))
+ out.append(sa.func.ST_AsGeoJSON(col, 7).label('geometry_geojson'))
if details.geometry_output & GeometryFormat.TEXT:
out.append(sa.func.ST_AsText(col).label('geometry_text'))
if details.geometry_output & GeometryFormat.KML:
- out.append(sa.func.ST_AsKML(col).label('geometry_kml'))
+ out.append(sa.func.ST_AsKML(col, 7).label('geometry_kml'))
if details.geometry_output & GeometryFormat.SVG:
- out.append(sa.func.ST_AsSVG(col).label('geometry_svg'))
+ out.append(sa.func.ST_AsSVG(col, 0, 7).label('geometry_svg'))
return sql.add_columns(*out)
def _filter_by_layer(table: SaFromClause, layers: DataLayer) -> SaColumn:
orexpr: List[SaExpression] = []
if layers & DataLayer.ADDRESS and layers & DataLayer.POI:
- orexpr.append(table.c.rank_address.between(1, 30))
+ orexpr.append(no_index(table.c.rank_address).between(1, 30))
elif layers & DataLayer.ADDRESS:
- orexpr.append(table.c.rank_address.between(1, 29))
- orexpr.append(sa.and_(table.c.rank_address == 30,
+ 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('housename'))))
+ table.c.address.has_key('addr:housename'))))
elif layers & DataLayer.POI:
- orexpr.append(sa.and_(table.c.rank_address == 30,
+ orexpr.append(sa.and_(no_index(table.c.rank_address) == 30,
table.c.class_.not_in(('place', 'building'))))
if layers & DataLayer.MANMADE:
if not layers & DataLayer.NATURAL:
exclude.extend(('natural', 'water', 'waterway'))
orexpr.append(sa.and_(table.c.class_.not_in(tuple(exclude)),
- table.c.rank_address == 0))
+ no_index(table.c.rank_address) == 0))
else:
include = []
if layers & DataLayer.RAILWAY:
if layers & DataLayer.NATURAL:
include.extend(('natural', 'water', 'waterway'))
orexpr.append(sa.and_(table.c.class_.in_(tuple(include)),
- table.c.rank_address == 0))
+ no_index(table.c.rank_address) == 0))
if len(orexpr) == 1:
return orexpr[0]
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)
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.alias('p')
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.alias('inner')
+ 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))))
- sql = sql.where(t.c.rank_address.between(MIN_RANK_PARAM, MAX_RANK_PARAM))
+ 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:
sql = sql.where(t.c.country_code.in_(COUNTRIES_PARAM))
if details.excluded:
"""
def __init__(self, sdata: SearchData) -> None:
super().__init__(sdata.penalty)
- self.categories = sdata.qualifiers
+ self.qualifiers = sdata.qualifiers
self.countries = sdata.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)) \
.order_by(t.c.centroid.ST_Distance(NEAR_PARAM)) \
.limit(LIMIT_PARAM)
- classtype = self.categories.values
+ classtype = self.qualifiers.values
if len(classtype) == 1:
cclass, ctype = classtype[0]
sql: SaLambdaSelect = sa.lambda_stmt(lambda: _base_query()
rows.extend(await conn.execute(sql, bind_params))
else:
# use the class type tables
- for category in self.categories.values:
+ for category in self.qualifiers.values:
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])
for row in rows:
result = nres.create_from_placex_row(row, nres.SearchResult)
assert result
- result.accuracy = self.penalty + self.categories.get_penalty((row.class_, row.type))
+ result.accuracy = self.penalty + self.qualifiers.get_penalty((row.class_, row.type))
result.bbox = Bbox.from_wkb(row.bbox)
results.append(result)
t = conn.t.placex
ccodes = self.countries.values
- sql: SaLambdaSelect = sa.lambda_stmt(lambda: _select_placex(t)\
+ sql = _select_placex(t)\
+ .add_columns(t.c.importance)\
.where(t.c.country_code.in_(ccodes))\
- .where(t.c.rank_address == 4))
+ .where(t.c.rank_address == 4)
if details.geometry_output:
sql = _add_geometry_columns(sql, t.c.geometry, details)
result = nres.create_from_placex_row(row, nres.SearchResult)
assert result
result.accuracy = self.penalty + self.countries.get_penalty(row.country_code, 5.0)
+ result.bbox = Bbox.from_wkb(row.bbox)
results.append(result)
return results or await self.lookup_in_country_table(conn, details)
sql = sa.select(tgrid.c.country_code,
tgrid.c.geometry.ST_Centroid().ST_Collect().ST_Centroid()
- .label('centroid'))\
+ .label('centroid'),
+ tgrid.c.geometry.ST_Collect().ST_Expand(0).label('bbox'))\
.where(tgrid.c.country_code.in_(self.countries.values))\
.group_by(tgrid.c.country_code)
+ sa.func.coalesce(t.c.derived_name,
sa.cast('', type_=conn.t.types.Composite))
).label('name'),
- sub.c.centroid)\
+ sub.c.centroid, sub.c.bbox)\
.join(sub, t.c.country_code == sub.c.country_code)
+ if details.geometry_output:
+ sql = _add_geometry_columns(sql, sub.c.centroid, details)
+
results = nres.SearchResults()
for row in await conn.execute(sql, _details_to_bind_params(details)):
result = nres.create_from_country_row(row, nres.SearchResult)
assert result
+ result.bbox = Bbox.from_wkb(row.bbox)
result.accuracy = self.penalty + self.countries.get_penalty(row.country_code, 5.0)
results.append(result)
t = conn.t.postcode
pcs = self.postcodes.values
- sql: SaLambdaSelect = sa.lambda_stmt(lambda:
- sa.select(t.c.place_id, t.c.parent_place_id,
+ sql = sa.select(t.c.place_id, t.c.parent_place_id,
t.c.rank_search, t.c.rank_address,
t.c.postcode, t.c.country_code,
- t.c.geometry.label('centroid'))
- .where(t.c.postcode.in_(pcs)))
+ t.c.geometry.label('centroid'))\
+ .where(t.c.postcode.in_(pcs))
if details.geometry_output:
sql = _add_geometry_columns(sql, t.c.geometry, details)
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), 1.0),
- else_=2.0)
+ (t.c.geometry.intersects(VIEWBOX2_PARAM), 0.5),
+ else_=1.0)
if details.near is not None:
if details.near_radius is not None:
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.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:
pcs = self.postcodes.values
if self.expected_count > 1000:
# Many results expected. Restrict by postcode.
- sql = sql.where(lambda: sa.select(tpc.c.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))
.exists())
if details.viewbox is not None:
if details.bounded_viewbox:
- sql = sql.where(tsearch.c.centroid.intersects(VIEWBOX_PARAM))
+ 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))
else:
penalty += sa.case((t.c.geometry.intersects(VIEWBOX_PARAM), 0.0),
- (t.c.geometry.intersects(VIEWBOX2_PARAM), 1.0),
- else_=2.0)
+ (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(tsearch.c.centroid.ST_DWithin(NEAR_PARAM, NEAR_RADIUS_PARAM))
- sql = sql.add_columns(-tsearch.c.centroid.ST_Distance(NEAR_PARAM)
+ if details.near_radius < 0.1:
+ sql = sql.where(tsearch.c.centroid.ST_DWithin(NEAR_PARAM, NEAR_RADIUS_PARAM))
+ else:
+ sql = sql.where(tsearch.c.centroid.ST_DWithin_no_index(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')))
else:
- sql = sql.order_by(penalty - sa.case((tsearch.c.importance > 0, tsearch.c.importance),
- else_=0.75001-(sa.cast(tsearch.c.search_rank, sa.Float())/40)))
+ if self.expected_count < 10000\
+ or (details.viewbox is not None and details.viewbox.area < 0.5):
+ sql = sql.order_by(
+ penalty - sa.case((tsearch.c.importance > 0, tsearch.c.importance),
+ else_=0.75001-(sa.cast(tsearch.c.search_rank, sa.Float())/40)))
sql = sql.add_columns(t.c.importance)
- sql = sql.add_columns(penalty.label('accuracy'))\
- .order_by(sa.text('accuracy'))
+ sql = sql.add_columns(penalty.label('accuracy'))
+
+ if self.expected_count < 10000:
+ sql = sql.order_by(sa.text('accuracy'))
if self.housenumbers:
hnr_regexp = f"\\m({'|'.join(self.housenumbers.values)})\\M"
.where(thnr.c.indexed_status == 0)
if details.excluded:
- place_sql = place_sql.where(_exclude_places(thnr))
+ place_sql = place_sql.where(thnr.c.place_id.not_in(sa.bindparam('excluded')))
if self.qualifiers:
place_sql = place_sql.where(self.qualifiers.sql_restrict(thnr))
- numerals = [int(n) for n in self.housenumbers.values if n.isdigit()]
+ numerals = [int(n) for n in self.housenumbers.values
+ if n.isdigit() and len(n) < 8]
interpol_sql: SaColumn
tiger_sql: SaColumn
if numerals and \
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