]> git.openstreetmap.org Git - nominatim.git/blobdiff - test/python/api/conftest.py
Merge pull request #3328 from lonvia/word-count-into-new-table
[nominatim.git] / test / python / api / conftest.py
index 2fc7120254d1e4a2488f44bd770c64b8f9c1d9ac..05eaddf5fc0f182cfc501504e48b8865fdb9af95 100644 (file)
@@ -12,13 +12,18 @@ import pytest
 import time
 import datetime as dt
 
+import sqlalchemy as sa
+
 import nominatim.api as napi
 from nominatim.db.sql_preprocessor import SQLPreprocessor
+from nominatim.api.search.query_analyzer_factory import make_query_analyzer
+from nominatim.tools import convert_sqlite
+import nominatim.api.logging as loglib
 
 class APITester:
 
     def __init__(self):
-        self.api = napi.NominatimAPI(Path('/invalid'), {})
+        self.api = napi.NominatimAPI(Path('/invalid'))
         self.async_to_sync(self.api._async_api.setup_database())
 
 
@@ -41,6 +46,9 @@ class APITester:
         if isinstance(name, str):
             name = {'name': name}
 
+        centroid = kw.get('centroid', (23.0, 34.0))
+        geometry = kw.get('geometry', 'POINT(%f %f)' % centroid)
+
         self.add_data('placex',
                      {'place_id': kw.get('place_id', 1000),
                       'osm_type': kw.get('osm_type', 'W'),
@@ -60,10 +68,11 @@ class APITester:
                       'rank_search': kw.get('rank_search', 30),
                       'rank_address': kw.get('rank_address', 30),
                       'importance': kw.get('importance'),
-                      'centroid': 'SRID=4326;POINT(%f %f)' % kw.get('centroid', (23.0, 34.0)),
+                      'centroid': 'POINT(%f %f)' % centroid,
+                      'indexed_status': kw.get('indexed_status', 0),
                       'indexed_date': kw.get('indexed_date',
                                              dt.datetime(2022, 12, 7, 14, 14, 46, 0)),
-                      'geometry': 'SRID=4326;' + kw.get('geometry', 'POINT(23 34)')})
+                      'geometry': geometry})
 
 
     def add_address_placex(self, object_id, **kw):
@@ -77,6 +86,97 @@ class APITester:
                        'isaddress': kw.get('isaddress', True)})
 
 
+    def add_osmline(self, **kw):
+        self.add_data('osmline',
+                     {'place_id': kw.get('place_id', 10000),
+                      'osm_id': kw.get('osm_id', 4004),
+                      'parent_place_id': kw.get('parent_place_id'),
+                      'indexed_date': kw.get('indexed_date',
+                                             dt.datetime(2022, 12, 7, 14, 14, 46, 0)),
+                      'startnumber': kw.get('startnumber', 2),
+                      'endnumber': kw.get('endnumber', 6),
+                      'step': kw.get('step', 2),
+                      'address': kw.get('address'),
+                      'postcode': kw.get('postcode'),
+                      'country_code': kw.get('country_code'),
+                      'linegeo': kw.get('geometry', 'LINESTRING(1.1 -0.2, 1.09 -0.22)')})
+
+
+    def add_tiger(self, **kw):
+        self.add_data('tiger',
+                     {'place_id': kw.get('place_id', 30000),
+                      'parent_place_id': kw.get('parent_place_id'),
+                      'startnumber': kw.get('startnumber', 2),
+                      'endnumber': kw.get('endnumber', 6),
+                      'step': kw.get('step', 2),
+                      'postcode': kw.get('postcode'),
+                      'linegeo': kw.get('geometry', 'LINESTRING(1.1 -0.2, 1.09 -0.22)')})
+
+
+    def add_postcode(self, **kw):
+        self.add_data('postcode',
+                     {'place_id': kw.get('place_id', 1000),
+                      'parent_place_id': kw.get('parent_place_id'),
+                      'country_code': kw.get('country_code'),
+                      'postcode': kw.get('postcode'),
+                      'rank_search': kw.get('rank_search', 20),
+                      'rank_address': kw.get('rank_address', 22),
+                      'indexed_date': kw.get('indexed_date',
+                                             dt.datetime(2022, 12, 7, 14, 14, 46, 0)),
+                      'geometry': kw.get('geometry', 'POINT(23 34)')})
+
+
+    def add_country(self, country_code, geometry):
+        self.add_data('country_grid',
+                      {'country_code': country_code,
+                       'area': 0.1,
+                       'geometry': geometry})
+
+
+    def add_country_name(self, country_code, names, partition=0):
+        self.add_data('country_name',
+                      {'country_code': country_code,
+                       'name': names,
+                       'partition': partition})
+
+
+    def add_search_name(self, place_id, **kw):
+        centroid = kw.get('centroid', (23.0, 34.0))
+        self.add_data('search_name',
+                      {'place_id': place_id,
+                       'importance': kw.get('importance', 0.00001),
+                       'search_rank': kw.get('search_rank', 30),
+                       'address_rank': kw.get('address_rank', 30),
+                       'name_vector': kw.get('names', []),
+                       'nameaddress_vector': kw.get('address', []),
+                       'country_code': kw.get('country_code', 'xx'),
+                       'centroid': 'POINT(%f %f)' % centroid})
+
+
+    def add_class_type_table(self, cls, typ):
+        self.async_to_sync(
+            self.exec_async(sa.text(f"""CREATE TABLE place_classtype_{cls}_{typ}
+                                         AS (SELECT place_id, centroid FROM placex
+                                             WHERE class = '{cls}' AND type = '{typ}')
+                                     """)))
+
+
+    def add_word_table(self, content):
+        data = [dict(zip(['word_id', 'word_token', 'type', 'word', 'info'], c))
+                for c in content]
+
+        async def _do_sql():
+            async with self.api._async_api.begin() as conn:
+                if 'word' not in conn.t.meta.tables:
+                    await make_query_analyzer(conn)
+                    word_table = conn.t.meta.tables['word']
+                    await conn.connection.run_sync(word_table.create)
+                if data:
+                    await conn.execute(conn.t.meta.tables['word'].insert(), data)
+
+        self.async_to_sync(_do_sql())
+
+
     async def exec_async(self, sql, *args, **kwargs):
         async with self.api._async_api.begin() as conn:
             return await conn.execute(sql, *args, **kwargs)
@@ -88,15 +188,59 @@ class APITester:
 
 
 @pytest.fixture
-def apiobj(temp_db_with_extensions, temp_db_conn):
+def apiobj(temp_db_with_extensions, temp_db_conn, monkeypatch):
     """ Create an asynchronous SQLAlchemy engine for the test DB.
     """
+    monkeypatch.setenv('NOMINATIM_USE_US_TIGER_DATA', 'yes')
     testapi = APITester()
     testapi.async_to_sync(testapi.create_tables())
 
-    SQLPreprocessor(temp_db_conn, testapi.api.config)\
-        .run_sql_file(temp_db_conn, 'functions/address_lookup.sql')
+    proc = SQLPreprocessor(temp_db_conn, testapi.api.config)
+    proc.run_sql_file(temp_db_conn, 'functions/ranking.sql')
 
+    loglib.set_log_output('text')
     yield testapi
+    print(loglib.get_and_disable())
 
     testapi.api.close()
+
+
+@pytest.fixture(params=['postgres_db', 'sqlite_db'])
+def frontend(request, event_loop, tmp_path):
+    testapis = []
+    if request.param == 'sqlite_db':
+        db = str(tmp_path / 'test_nominatim_python_unittest.sqlite')
+
+        def mkapi(apiobj, options={'reverse'}):
+            apiobj.add_data('properties',
+                        [{'property': 'tokenizer', 'value': 'icu'},
+                         {'property': 'tokenizer_import_normalisation', 'value': ':: lower();'},
+                         {'property': 'tokenizer_import_transliteration', 'value': "'1' > '/1/'; 'ä' > 'ä '"},
+                        ])
+
+            async def _do_sql():
+                async with apiobj.api._async_api.begin() as conn:
+                    if 'word' in conn.t.meta.tables:
+                        return
+                    await make_query_analyzer(conn)
+                    word_table = conn.t.meta.tables['word']
+                    await conn.connection.run_sync(word_table.create)
+
+            apiobj.async_to_sync(_do_sql())
+
+            event_loop.run_until_complete(convert_sqlite.convert(Path('/invalid'),
+                                                                 db, options))
+            outapi = napi.NominatimAPI(Path('/invalid'),
+                                       {'NOMINATIM_DATABASE_DSN': f"sqlite:dbname={db}",
+                                        'NOMINATIM_USE_US_TIGER_DATA': 'yes'})
+            testapis.append(outapi)
+
+            return outapi
+    elif request.param == 'postgres_db':
+        def mkapi(apiobj, options=None):
+            return apiobj.api
+
+    yield mkapi
+
+    for api in testapis:
+        api.close()