]> git.openstreetmap.org Git - nominatim.git/blob - nominatim/api/search/db_searches.py
Merge pull request #3260 from lonvia/improve-catgeory-search
[nominatim.git] / nominatim / api / search / db_searches.py
1 # SPDX-License-Identifier: GPL-3.0-or-later
2 #
3 # This file is part of Nominatim. (https://nominatim.org)
4 #
5 # Copyright (C) 2023 by the Nominatim developer community.
6 # For a full list of authors see the git log.
7 """
8 Implementation of the acutal database accesses for forward search.
9 """
10 from typing import List, Tuple, AsyncIterator, Dict, Any, Callable
11 import abc
12
13 import sqlalchemy as sa
14 from sqlalchemy.dialects.postgresql import ARRAY, array_agg
15
16 from nominatim.typing import SaFromClause, SaScalarSelect, SaColumn, \
17                              SaExpression, SaSelect, SaLambdaSelect, SaRow, SaBind
18 from nominatim.api.connection import SearchConnection
19 from nominatim.api.types import SearchDetails, DataLayer, GeometryFormat, Bbox
20 import nominatim.api.results as nres
21 from nominatim.api.search.db_search_fields import SearchData, WeightedCategories
22 from nominatim.db.sqlalchemy_types import Geometry
23
24 #pylint: disable=singleton-comparison,not-callable
25 #pylint: disable=too-many-branches,too-many-arguments,too-many-locals,too-many-statements
26
27 def no_index(expr: SaColumn) -> SaColumn:
28     """ Wrap the given expression, so that the query planner will
29         refrain from using the expression for index lookup.
30     """
31     return sa.func.coalesce(sa.null(), expr) # pylint: disable=not-callable
32
33
34 def _details_to_bind_params(details: SearchDetails) -> Dict[str, Any]:
35     """ Create a dictionary from search parameters that can be used
36         as bind parameter for SQL execute.
37     """
38     return {'limit': details.max_results,
39             'min_rank': details.min_rank,
40             'max_rank': details.max_rank,
41             'viewbox': details.viewbox,
42             'viewbox2': details.viewbox_x2,
43             'near': details.near,
44             'near_radius': details.near_radius,
45             'excluded': details.excluded,
46             'countries': details.countries}
47
48
49 LIMIT_PARAM: SaBind = sa.bindparam('limit')
50 MIN_RANK_PARAM: SaBind = sa.bindparam('min_rank')
51 MAX_RANK_PARAM: SaBind = sa.bindparam('max_rank')
52 VIEWBOX_PARAM: SaBind = sa.bindparam('viewbox', type_=Geometry)
53 VIEWBOX2_PARAM: SaBind = sa.bindparam('viewbox2', type_=Geometry)
54 NEAR_PARAM: SaBind = sa.bindparam('near', type_=Geometry)
55 NEAR_RADIUS_PARAM: SaBind = sa.bindparam('near_radius')
56 COUNTRIES_PARAM: SaBind = sa.bindparam('countries')
57
58 def _within_near(t: SaFromClause) -> Callable[[], SaExpression]:
59     return lambda: t.c.geometry.ST_DWithin(NEAR_PARAM, NEAR_RADIUS_PARAM)
60
61 def _exclude_places(t: SaFromClause) -> Callable[[], SaExpression]:
62     return lambda: t.c.place_id.not_in(sa.bindparam('excluded'))
63
64 def _select_placex(t: SaFromClause) -> SaSelect:
65     return sa.select(t.c.place_id, t.c.osm_type, t.c.osm_id, t.c.name,
66                      t.c.class_, t.c.type,
67                      t.c.address, t.c.extratags,
68                      t.c.housenumber, t.c.postcode, t.c.country_code,
69                      t.c.wikipedia,
70                      t.c.parent_place_id, t.c.rank_address, t.c.rank_search,
71                      t.c.linked_place_id, t.c.admin_level,
72                      t.c.centroid,
73                      t.c.geometry.ST_Expand(0).label('bbox'))
74
75
76 def _add_geometry_columns(sql: SaLambdaSelect, col: SaColumn, details: SearchDetails) -> SaSelect:
77     out = []
78
79     if details.geometry_simplification > 0.0:
80         col = sa.func.ST_SimplifyPreserveTopology(col, details.geometry_simplification)
81
82     if details.geometry_output & GeometryFormat.GEOJSON:
83         out.append(sa.func.ST_AsGeoJSON(col, 7).label('geometry_geojson'))
84     if details.geometry_output & GeometryFormat.TEXT:
85         out.append(sa.func.ST_AsText(col).label('geometry_text'))
86     if details.geometry_output & GeometryFormat.KML:
87         out.append(sa.func.ST_AsKML(col, 7).label('geometry_kml'))
88     if details.geometry_output & GeometryFormat.SVG:
89         out.append(sa.func.ST_AsSVG(col, 0, 7).label('geometry_svg'))
90
91     return sql.add_columns(*out)
92
93
94 def _make_interpolation_subquery(table: SaFromClause, inner: SaFromClause,
95                                  numerals: List[int], details: SearchDetails) -> SaScalarSelect:
96     all_ids = array_agg(table.c.place_id) # type: ignore[no-untyped-call]
97     sql = sa.select(all_ids).where(table.c.parent_place_id == inner.c.place_id)
98
99     if len(numerals) == 1:
100         sql = sql.where(sa.between(numerals[0], table.c.startnumber, table.c.endnumber))\
101                  .where((numerals[0] - table.c.startnumber) % table.c.step == 0)
102     else:
103         sql = sql.where(sa.or_(
104                 *(sa.and_(sa.between(n, table.c.startnumber, table.c.endnumber),
105                           (n - table.c.startnumber) % table.c.step == 0)
106                   for n in numerals)))
107
108     if details.excluded:
109         sql = sql.where(_exclude_places(table))
110
111     return sql.scalar_subquery()
112
113
114 def _filter_by_layer(table: SaFromClause, layers: DataLayer) -> SaColumn:
115     orexpr: List[SaExpression] = []
116     if layers & DataLayer.ADDRESS and layers & DataLayer.POI:
117         orexpr.append(no_index(table.c.rank_address).between(1, 30))
118     elif layers & DataLayer.ADDRESS:
119         orexpr.append(no_index(table.c.rank_address).between(1, 29))
120         orexpr.append(sa.and_(no_index(table.c.rank_address) == 30,
121                               sa.or_(table.c.housenumber != None,
122                                      table.c.address.has_key('addr:housename'))))
123     elif layers & DataLayer.POI:
124         orexpr.append(sa.and_(no_index(table.c.rank_address) == 30,
125                               table.c.class_.not_in(('place', 'building'))))
126
127     if layers & DataLayer.MANMADE:
128         exclude = []
129         if not layers & DataLayer.RAILWAY:
130             exclude.append('railway')
131         if not layers & DataLayer.NATURAL:
132             exclude.extend(('natural', 'water', 'waterway'))
133         orexpr.append(sa.and_(table.c.class_.not_in(tuple(exclude)),
134                               no_index(table.c.rank_address) == 0))
135     else:
136         include = []
137         if layers & DataLayer.RAILWAY:
138             include.append('railway')
139         if layers & DataLayer.NATURAL:
140             include.extend(('natural', 'water', 'waterway'))
141         orexpr.append(sa.and_(table.c.class_.in_(tuple(include)),
142                               no_index(table.c.rank_address) == 0))
143
144     if len(orexpr) == 1:
145         return orexpr[0]
146
147     return sa.or_(*orexpr)
148
149
150 def _interpolated_position(table: SaFromClause, nr: SaColumn) -> SaColumn:
151     pos = sa.cast(nr - table.c.startnumber, sa.Float) / (table.c.endnumber - table.c.startnumber)
152     return sa.case(
153             (table.c.endnumber == table.c.startnumber, table.c.linegeo.ST_Centroid()),
154             else_=table.c.linegeo.ST_LineInterpolatePoint(pos)).label('centroid')
155
156
157 async def _get_placex_housenumbers(conn: SearchConnection,
158                                    place_ids: List[int],
159                                    details: SearchDetails) -> AsyncIterator[nres.SearchResult]:
160     t = conn.t.placex
161     sql = _select_placex(t).add_columns(t.c.importance)\
162                            .where(t.c.place_id.in_(place_ids))
163
164     if details.geometry_output:
165         sql = _add_geometry_columns(sql, t.c.geometry, details)
166
167     for row in await conn.execute(sql):
168         result = nres.create_from_placex_row(row, nres.SearchResult)
169         assert result
170         result.bbox = Bbox.from_wkb(row.bbox)
171         yield result
172
173
174 async def _get_osmline(conn: SearchConnection, place_ids: List[int],
175                        numerals: List[int],
176                        details: SearchDetails) -> AsyncIterator[nres.SearchResult]:
177     t = conn.t.osmline
178     values = sa.values(sa.Column('nr', sa.Integer()), name='housenumber')\
179                .data([(n,) for n in numerals])
180     sql = sa.select(t.c.place_id, t.c.osm_id,
181                     t.c.parent_place_id, t.c.address,
182                     values.c.nr.label('housenumber'),
183                     _interpolated_position(t, values.c.nr),
184                     t.c.postcode, t.c.country_code)\
185             .where(t.c.place_id.in_(place_ids))\
186             .join(values, values.c.nr.between(t.c.startnumber, t.c.endnumber))
187
188     if details.geometry_output:
189         sub = sql.subquery()
190         sql = _add_geometry_columns(sa.select(sub), sub.c.centroid, details)
191
192     for row in await conn.execute(sql):
193         result = nres.create_from_osmline_row(row, nres.SearchResult)
194         assert result
195         yield result
196
197
198 async def _get_tiger(conn: SearchConnection, place_ids: List[int],
199                      numerals: List[int], osm_id: int,
200                      details: SearchDetails) -> AsyncIterator[nres.SearchResult]:
201     t = conn.t.tiger
202     values = sa.values(sa.Column('nr', sa.Integer()), name='housenumber')\
203                .data([(n,) for n in numerals])
204     sql = sa.select(t.c.place_id, t.c.parent_place_id,
205                     sa.literal('W').label('osm_type'),
206                     sa.literal(osm_id).label('osm_id'),
207                     values.c.nr.label('housenumber'),
208                     _interpolated_position(t, values.c.nr),
209                     t.c.postcode)\
210             .where(t.c.place_id.in_(place_ids))\
211             .join(values, values.c.nr.between(t.c.startnumber, t.c.endnumber))
212
213     if details.geometry_output:
214         sub = sql.subquery()
215         sql = _add_geometry_columns(sa.select(sub), sub.c.centroid, details)
216
217     for row in await conn.execute(sql):
218         result = nres.create_from_tiger_row(row, nres.SearchResult)
219         assert result
220         yield result
221
222
223 class AbstractSearch(abc.ABC):
224     """ Encapuslation of a single lookup in the database.
225     """
226
227     def __init__(self, penalty: float) -> None:
228         self.penalty = penalty
229
230     @abc.abstractmethod
231     async def lookup(self, conn: SearchConnection,
232                      details: SearchDetails) -> nres.SearchResults:
233         """ Find results for the search in the database.
234         """
235
236
237 class NearSearch(AbstractSearch):
238     """ Category search of a place type near the result of another search.
239     """
240     def __init__(self, penalty: float, categories: WeightedCategories,
241                  search: AbstractSearch) -> None:
242         super().__init__(penalty)
243         self.search = search
244         self.categories = categories
245
246
247     async def lookup(self, conn: SearchConnection,
248                      details: SearchDetails) -> nres.SearchResults:
249         """ Find results for the search in the database.
250         """
251         results = nres.SearchResults()
252         base = await self.search.lookup(conn, details)
253
254         if not base:
255             return results
256
257         base.sort(key=lambda r: (r.accuracy, r.rank_search))
258         max_accuracy = base[0].accuracy + 0.5
259         if base[0].rank_address == 0:
260             min_rank = 0
261             max_rank = 0
262         elif base[0].rank_address < 26:
263             min_rank = 1
264             max_rank = min(25, base[0].rank_address + 4)
265         else:
266             min_rank = 26
267             max_rank = 30
268         base = nres.SearchResults(r for r in base if r.source_table == nres.SourceTable.PLACEX
269                                                      and r.accuracy <= max_accuracy
270                                                      and r.bbox and r.bbox.area < 20
271                                                      and r.rank_address >= min_rank
272                                                      and r.rank_address <= max_rank)
273
274         if base:
275             baseids = [b.place_id for b in base[:5] if b.place_id]
276
277             for category, penalty in self.categories:
278                 await self.lookup_category(results, conn, baseids, category, penalty, details)
279                 if len(results) >= details.max_results:
280                     break
281
282         return results
283
284
285     async def lookup_category(self, results: nres.SearchResults,
286                               conn: SearchConnection, ids: List[int],
287                               category: Tuple[str, str], penalty: float,
288                               details: SearchDetails) -> None:
289         """ Find places of the given category near the list of
290             place ids and add the results to 'results'.
291         """
292         table = await conn.get_class_table(*category)
293
294         tgeom = conn.t.placex.alias('pgeom')
295
296         if table is None:
297             # No classtype table available, do a simplified lookup in placex.
298             table = conn.t.placex.alias('inner')
299             sql = sa.select(table.c.place_id,
300                             sa.func.min(tgeom.c.centroid.ST_Distance(table.c.centroid))
301                               .label('dist'))\
302                     .join(tgeom, table.c.geometry.intersects(tgeom.c.centroid.ST_Expand(0.01)))\
303                     .where(table.c.class_ == category[0])\
304                     .where(table.c.type == category[1])
305         else:
306             # Use classtype table. We can afford to use a larger
307             # radius for the lookup.
308             sql = sa.select(table.c.place_id,
309                             sa.func.min(tgeom.c.centroid.ST_Distance(table.c.centroid))
310                               .label('dist'))\
311                     .join(tgeom,
312                           table.c.centroid.ST_CoveredBy(
313                               sa.case((sa.and_(tgeom.c.rank_address > 9,
314                                                 tgeom.c.geometry.is_area()),
315                                        tgeom.c.geometry),
316                                       else_ = tgeom.c.centroid.ST_Expand(0.05))))
317
318         inner = sql.where(tgeom.c.place_id.in_(ids))\
319                    .group_by(table.c.place_id).subquery()
320
321         t = conn.t.placex
322         sql = _select_placex(t).add_columns((-inner.c.dist).label('importance'))\
323                                .join(inner, inner.c.place_id == t.c.place_id)\
324                                .order_by(inner.c.dist)
325
326         sql = sql.where(no_index(t.c.rank_address).between(MIN_RANK_PARAM, MAX_RANK_PARAM))
327         if details.countries:
328             sql = sql.where(t.c.country_code.in_(COUNTRIES_PARAM))
329         if details.excluded:
330             sql = sql.where(_exclude_places(t))
331         if details.layers is not None:
332             sql = sql.where(_filter_by_layer(t, details.layers))
333
334         sql = sql.limit(LIMIT_PARAM)
335         for row in await conn.execute(sql, _details_to_bind_params(details)):
336             result = nres.create_from_placex_row(row, nres.SearchResult)
337             assert result
338             result.accuracy = self.penalty + penalty
339             result.bbox = Bbox.from_wkb(row.bbox)
340             results.append(result)
341
342
343
344 class PoiSearch(AbstractSearch):
345     """ Category search in a geographic area.
346     """
347     def __init__(self, sdata: SearchData) -> None:
348         super().__init__(sdata.penalty)
349         self.qualifiers = sdata.qualifiers
350         self.countries = sdata.countries
351
352
353     async def lookup(self, conn: SearchConnection,
354                      details: SearchDetails) -> nres.SearchResults:
355         """ Find results for the search in the database.
356         """
357         bind_params = _details_to_bind_params(details)
358         t = conn.t.placex
359
360         rows: List[SaRow] = []
361
362         if details.near and details.near_radius is not None and details.near_radius < 0.2:
363             # simply search in placex table
364             def _base_query() -> SaSelect:
365                 return _select_placex(t) \
366                            .add_columns((-t.c.centroid.ST_Distance(NEAR_PARAM))
367                                          .label('importance'))\
368                            .where(t.c.linked_place_id == None) \
369                            .where(t.c.geometry.ST_DWithin(NEAR_PARAM, NEAR_RADIUS_PARAM)) \
370                            .order_by(t.c.centroid.ST_Distance(NEAR_PARAM)) \
371                            .limit(LIMIT_PARAM)
372
373             classtype = self.qualifiers.values
374             if len(classtype) == 1:
375                 cclass, ctype = classtype[0]
376                 sql: SaLambdaSelect = sa.lambda_stmt(lambda: _base_query()
377                                                  .where(t.c.class_ == cclass)
378                                                  .where(t.c.type == ctype))
379             else:
380                 sql = _base_query().where(sa.or_(*(sa.and_(t.c.class_ == cls, t.c.type == typ)
381                                                    for cls, typ in classtype)))
382
383             if self.countries:
384                 sql = sql.where(t.c.country_code.in_(self.countries.values))
385
386             if details.viewbox is not None and details.bounded_viewbox:
387                 sql = sql.where(t.c.geometry.intersects(VIEWBOX_PARAM))
388
389             rows.extend(await conn.execute(sql, bind_params))
390         else:
391             # use the class type tables
392             for category in self.qualifiers.values:
393                 table = await conn.get_class_table(*category)
394                 if table is not None:
395                     sql = _select_placex(t)\
396                                .add_columns(t.c.importance)\
397                                .join(table, t.c.place_id == table.c.place_id)\
398                                .where(t.c.class_ == category[0])\
399                                .where(t.c.type == category[1])
400
401                     if details.viewbox is not None and details.bounded_viewbox:
402                         sql = sql.where(table.c.centroid.intersects(VIEWBOX_PARAM))
403
404                     if details.near and details.near_radius is not None:
405                         sql = sql.order_by(table.c.centroid.ST_Distance(NEAR_PARAM))\
406                                  .where(table.c.centroid.ST_DWithin(NEAR_PARAM,
407                                                                     NEAR_RADIUS_PARAM))
408
409                     if self.countries:
410                         sql = sql.where(t.c.country_code.in_(self.countries.values))
411
412                     sql = sql.limit(LIMIT_PARAM)
413                     rows.extend(await conn.execute(sql, bind_params))
414
415         results = nres.SearchResults()
416         for row in rows:
417             result = nres.create_from_placex_row(row, nres.SearchResult)
418             assert result
419             result.accuracy = self.penalty + self.qualifiers.get_penalty((row.class_, row.type))
420             result.bbox = Bbox.from_wkb(row.bbox)
421             results.append(result)
422
423         return results
424
425
426 class CountrySearch(AbstractSearch):
427     """ Search for a country name or country code.
428     """
429     def __init__(self, sdata: SearchData) -> None:
430         super().__init__(sdata.penalty)
431         self.countries = sdata.countries
432
433
434     async def lookup(self, conn: SearchConnection,
435                      details: SearchDetails) -> nres.SearchResults:
436         """ Find results for the search in the database.
437         """
438         t = conn.t.placex
439
440         ccodes = self.countries.values
441         sql = _select_placex(t)\
442                 .add_columns(t.c.importance)\
443                 .where(t.c.country_code.in_(ccodes))\
444                 .where(t.c.rank_address == 4)
445
446         if details.geometry_output:
447             sql = _add_geometry_columns(sql, t.c.geometry, details)
448
449         if details.excluded:
450             sql = sql.where(_exclude_places(t))
451
452         if details.viewbox is not None and details.bounded_viewbox:
453             sql = sql.where(lambda: t.c.geometry.intersects(VIEWBOX_PARAM))
454
455         if details.near is not None and details.near_radius is not None:
456             sql = sql.where(_within_near(t))
457
458         results = nres.SearchResults()
459         for row in await conn.execute(sql, _details_to_bind_params(details)):
460             result = nres.create_from_placex_row(row, nres.SearchResult)
461             assert result
462             result.accuracy = self.penalty + self.countries.get_penalty(row.country_code, 5.0)
463             result.bbox = Bbox.from_wkb(row.bbox)
464             results.append(result)
465
466         return results or await self.lookup_in_country_table(conn, details)
467
468
469     async def lookup_in_country_table(self, conn: SearchConnection,
470                                       details: SearchDetails) -> nres.SearchResults:
471         """ Look up the country in the fallback country tables.
472         """
473         # Avoid the fallback search when this is a more search. Country results
474         # usually are in the first batch of results and it is not possible
475         # to exclude these fallbacks.
476         if details.excluded:
477             return nres.SearchResults()
478
479         t = conn.t.country_name
480         tgrid = conn.t.country_grid
481
482         sql = sa.select(tgrid.c.country_code,
483                         tgrid.c.geometry.ST_Centroid().ST_Collect().ST_Centroid()
484                               .label('centroid'),
485                         tgrid.c.geometry.ST_Collect().ST_Expand(0).label('bbox'))\
486                 .where(tgrid.c.country_code.in_(self.countries.values))\
487                 .group_by(tgrid.c.country_code)
488
489         if details.viewbox is not None and details.bounded_viewbox:
490             sql = sql.where(tgrid.c.geometry.intersects(VIEWBOX_PARAM))
491         if details.near is not None and details.near_radius is not None:
492             sql = sql.where(_within_near(tgrid))
493
494         sub = sql.subquery('grid')
495
496         sql = sa.select(t.c.country_code,
497                         (t.c.name
498                          + sa.func.coalesce(t.c.derived_name,
499                                             sa.cast('', type_=conn.t.types.Composite))
500                         ).label('name'),
501                         sub.c.centroid, sub.c.bbox)\
502                 .join(sub, t.c.country_code == sub.c.country_code)
503
504         if details.geometry_output:
505             sql = _add_geometry_columns(sql, sub.c.centroid, details)
506
507         results = nres.SearchResults()
508         for row in await conn.execute(sql, _details_to_bind_params(details)):
509             result = nres.create_from_country_row(row, nres.SearchResult)
510             assert result
511             result.bbox = Bbox.from_wkb(row.bbox)
512             result.accuracy = self.penalty + self.countries.get_penalty(row.country_code, 5.0)
513             results.append(result)
514
515         return results
516
517
518
519 class PostcodeSearch(AbstractSearch):
520     """ Search for a postcode.
521     """
522     def __init__(self, extra_penalty: float, sdata: SearchData) -> None:
523         super().__init__(sdata.penalty + extra_penalty)
524         self.countries = sdata.countries
525         self.postcodes = sdata.postcodes
526         self.lookups = sdata.lookups
527         self.rankings = sdata.rankings
528
529
530     async def lookup(self, conn: SearchConnection,
531                      details: SearchDetails) -> nres.SearchResults:
532         """ Find results for the search in the database.
533         """
534         t = conn.t.postcode
535         pcs = self.postcodes.values
536
537         sql = sa.select(t.c.place_id, t.c.parent_place_id,
538                         t.c.rank_search, t.c.rank_address,
539                         t.c.postcode, t.c.country_code,
540                         t.c.geometry.label('centroid'))\
541                 .where(t.c.postcode.in_(pcs))
542
543         if details.geometry_output:
544             sql = _add_geometry_columns(sql, t.c.geometry, details)
545
546         penalty: SaExpression = sa.literal(self.penalty)
547
548         if details.viewbox is not None:
549             if details.bounded_viewbox:
550                 sql = sql.where(t.c.geometry.intersects(VIEWBOX_PARAM))
551             else:
552                 penalty += sa.case((t.c.geometry.intersects(VIEWBOX_PARAM), 0.0),
553                                    (t.c.geometry.intersects(VIEWBOX2_PARAM), 0.5),
554                                    else_=1.0)
555
556         if details.near is not None:
557             if details.near_radius is not None:
558                 sql = sql.where(_within_near(t))
559             sql = sql.order_by(t.c.geometry.ST_Distance(NEAR_PARAM))
560
561         if self.countries:
562             sql = sql.where(t.c.country_code.in_(self.countries.values))
563
564         if details.excluded:
565             sql = sql.where(_exclude_places(t))
566
567         if self.lookups:
568             assert len(self.lookups) == 1
569             assert self.lookups[0].lookup_type == 'restrict'
570             tsearch = conn.t.search_name
571             sql = sql.where(tsearch.c.place_id == t.c.parent_place_id)\
572                      .where(sa.func.array_cat(tsearch.c.name_vector,
573                                               tsearch.c.nameaddress_vector,
574                                               type_=ARRAY(sa.Integer))
575                                     .contains(self.lookups[0].tokens))
576
577         for ranking in self.rankings:
578             penalty += ranking.sql_penalty(conn.t.search_name)
579         penalty += sa.case(*((t.c.postcode == v, p) for v, p in self.postcodes),
580                        else_=1.0)
581
582
583         sql = sql.add_columns(penalty.label('accuracy'))
584         sql = sql.order_by('accuracy').limit(LIMIT_PARAM)
585
586         results = nres.SearchResults()
587         for row in await conn.execute(sql, _details_to_bind_params(details)):
588             result = nres.create_from_postcode_row(row, nres.SearchResult)
589             assert result
590             result.accuracy = row.accuracy
591             results.append(result)
592
593         return results
594
595
596
597 class PlaceSearch(AbstractSearch):
598     """ Generic search for an address or named place.
599     """
600     def __init__(self, extra_penalty: float, sdata: SearchData, expected_count: int) -> None:
601         super().__init__(sdata.penalty + extra_penalty)
602         self.countries = sdata.countries
603         self.postcodes = sdata.postcodes
604         self.housenumbers = sdata.housenumbers
605         self.qualifiers = sdata.qualifiers
606         self.lookups = sdata.lookups
607         self.rankings = sdata.rankings
608         self.expected_count = expected_count
609
610
611     async def lookup(self, conn: SearchConnection,
612                      details: SearchDetails) -> nres.SearchResults:
613         """ Find results for the search in the database.
614         """
615         t = conn.t.placex
616         tsearch = conn.t.search_name
617
618         sql: SaLambdaSelect = sa.lambda_stmt(lambda:
619                   _select_placex(t).where(t.c.place_id == tsearch.c.place_id))
620
621
622         if details.geometry_output:
623             sql = _add_geometry_columns(sql, t.c.geometry, details)
624
625         penalty: SaExpression = sa.literal(self.penalty)
626         for ranking in self.rankings:
627             penalty += ranking.sql_penalty(tsearch)
628
629         for lookup in self.lookups:
630             sql = sql.where(lookup.sql_condition(tsearch))
631
632         if self.countries:
633             sql = sql.where(tsearch.c.country_code.in_(self.countries.values))
634
635         if self.postcodes:
636             # if a postcode is given, don't search for state or country level objects
637             sql = sql.where(tsearch.c.address_rank > 9)
638             tpc = conn.t.postcode
639             pcs = self.postcodes.values
640             if self.expected_count > 1000:
641                 # Many results expected. Restrict by postcode.
642                 sql = sql.where(sa.select(tpc.c.postcode)
643                                   .where(tpc.c.postcode.in_(pcs))
644                                   .where(tsearch.c.centroid.ST_DWithin(tpc.c.geometry, 0.12))
645                                   .exists())
646
647             # Less results, only have a preference for close postcodes
648             pc_near = sa.select(sa.func.min(tpc.c.geometry.ST_Distance(tsearch.c.centroid)))\
649                       .where(tpc.c.postcode.in_(pcs))\
650                       .scalar_subquery()
651             penalty += sa.case((t.c.postcode.in_(pcs), 0.0),
652                                else_=sa.func.coalesce(pc_near, 2.0))
653
654         if details.viewbox is not None:
655             if details.bounded_viewbox:
656                 if details.viewbox.area < 0.2:
657                     sql = sql.where(tsearch.c.centroid.intersects(VIEWBOX_PARAM))
658                 else:
659                     sql = sql.where(tsearch.c.centroid.ST_Intersects_no_index(VIEWBOX_PARAM))
660             elif self.expected_count >= 10000:
661                 if details.viewbox.area < 0.5:
662                     sql = sql.where(tsearch.c.centroid.intersects(VIEWBOX2_PARAM))
663                 else:
664                     sql = sql.where(tsearch.c.centroid.ST_Intersects_no_index(VIEWBOX2_PARAM))
665             else:
666                 penalty += sa.case((t.c.geometry.intersects(VIEWBOX_PARAM), 0.0),
667                                    (t.c.geometry.intersects(VIEWBOX2_PARAM), 0.5),
668                                    else_=1.0)
669
670         if details.near is not None:
671             if details.near_radius is not None:
672                 if details.near_radius < 0.1:
673                     sql = sql.where(tsearch.c.centroid.ST_DWithin(NEAR_PARAM, NEAR_RADIUS_PARAM))
674                 else:
675                     sql = sql.where(tsearch.c.centroid.ST_DWithin_no_index(NEAR_PARAM,
676                                                                            NEAR_RADIUS_PARAM))
677             sql = sql.add_columns((-tsearch.c.centroid.ST_Distance(NEAR_PARAM))
678                                       .label('importance'))
679             sql = sql.order_by(sa.desc(sa.text('importance')))
680         else:
681             if self.expected_count < 10000\
682                or (details.viewbox is not None and details.viewbox.area < 0.5):
683                 sql = sql.order_by(
684                         penalty - sa.case((tsearch.c.importance > 0, tsearch.c.importance),
685                                     else_=0.75001-(sa.cast(tsearch.c.search_rank, sa.Float())/40)))
686             sql = sql.add_columns(t.c.importance)
687
688
689         sql = sql.add_columns(penalty.label('accuracy'))
690
691         if self.expected_count < 10000:
692             sql = sql.order_by(sa.text('accuracy'))
693
694         if self.housenumbers:
695             hnr_regexp = f"\\m({'|'.join(self.housenumbers.values)})\\M"
696             sql = sql.where(tsearch.c.address_rank.between(16, 30))\
697                      .where(sa.or_(tsearch.c.address_rank < 30,
698                                    t.c.housenumber.op('~*')(hnr_regexp)))
699
700             # Cross check for housenumbers, need to do that on a rather large
701             # set. Worst case there are 40.000 main streets in OSM.
702             inner = sql.limit(10000).subquery()
703
704             # Housenumbers from placex
705             thnr = conn.t.placex.alias('hnr')
706             pid_list = array_agg(thnr.c.place_id) # type: ignore[no-untyped-call]
707             place_sql = sa.select(pid_list)\
708                           .where(thnr.c.parent_place_id == inner.c.place_id)\
709                           .where(thnr.c.housenumber.op('~*')(hnr_regexp))\
710                           .where(thnr.c.linked_place_id == None)\
711                           .where(thnr.c.indexed_status == 0)
712
713             if details.excluded:
714                 place_sql = place_sql.where(thnr.c.place_id.not_in(sa.bindparam('excluded')))
715             if self.qualifiers:
716                 place_sql = place_sql.where(self.qualifiers.sql_restrict(thnr))
717
718             numerals = [int(n) for n in self.housenumbers.values
719                         if n.isdigit() and len(n) < 8]
720             interpol_sql: SaColumn
721             tiger_sql: SaColumn
722             if numerals and \
723                (not self.qualifiers or ('place', 'house') in self.qualifiers.values):
724                 # Housenumbers from interpolations
725                 interpol_sql = _make_interpolation_subquery(conn.t.osmline, inner,
726                                                             numerals, details)
727                 # Housenumbers from Tiger
728                 tiger_sql = sa.case((inner.c.country_code == 'us',
729                                      _make_interpolation_subquery(conn.t.tiger, inner,
730                                                                   numerals, details)
731                                     ), else_=None)
732             else:
733                 interpol_sql = sa.null()
734                 tiger_sql = sa.null()
735
736             unsort = sa.select(inner, place_sql.scalar_subquery().label('placex_hnr'),
737                                interpol_sql.label('interpol_hnr'),
738                                tiger_sql.label('tiger_hnr')).subquery('unsort')
739             sql = sa.select(unsort)\
740                     .order_by(sa.case((unsort.c.placex_hnr != None, 1),
741                                       (unsort.c.interpol_hnr != None, 2),
742                                       (unsort.c.tiger_hnr != None, 3),
743                                       else_=4),
744                               unsort.c.accuracy)
745         else:
746             sql = sql.where(t.c.linked_place_id == None)\
747                      .where(t.c.indexed_status == 0)
748             if self.qualifiers:
749                 sql = sql.where(self.qualifiers.sql_restrict(t))
750             if details.excluded:
751                 sql = sql.where(_exclude_places(tsearch))
752             if details.min_rank > 0:
753                 sql = sql.where(sa.or_(tsearch.c.address_rank >= MIN_RANK_PARAM,
754                                        tsearch.c.search_rank >= MIN_RANK_PARAM))
755             if details.max_rank < 30:
756                 sql = sql.where(sa.or_(tsearch.c.address_rank <= MAX_RANK_PARAM,
757                                        tsearch.c.search_rank <= MAX_RANK_PARAM))
758             if details.layers is not None:
759                 sql = sql.where(_filter_by_layer(t, details.layers))
760
761         sql = sql.limit(LIMIT_PARAM)
762
763         results = nres.SearchResults()
764         for row in await conn.execute(sql, _details_to_bind_params(details)):
765             result = nres.create_from_placex_row(row, nres.SearchResult)
766             assert result
767             result.bbox = Bbox.from_wkb(row.bbox)
768             result.accuracy = row.accuracy
769             if not details.excluded or not result.place_id in details.excluded:
770                 results.append(result)
771
772             if self.housenumbers and row.rank_address < 30:
773                 if row.placex_hnr:
774                     subs = _get_placex_housenumbers(conn, row.placex_hnr, details)
775                 elif row.interpol_hnr:
776                     subs = _get_osmline(conn, row.interpol_hnr, numerals, details)
777                 elif row.tiger_hnr:
778                     subs = _get_tiger(conn, row.tiger_hnr, numerals, row.osm_id, details)
779                 else:
780                     subs = None
781
782                 if subs is not None:
783                     async for sub in subs:
784                         assert sub.housenumber
785                         sub.accuracy = result.accuracy
786                         if not any(nr in self.housenumbers.values
787                                    for nr in sub.housenumber.split(';')):
788                             sub.accuracy += 0.6
789                         results.append(sub)
790
791                 result.accuracy += 1.0 # penalty for missing housenumber
792
793         return results