class PlaceColumn:
def __init__(self, context, force_name):
- self.columns = { 'admin_level' : 100}
+ self.columns = { 'admin_level' : 15}
self.force_name = force_name
self.context = context
self.geometry = None
self.add_hstore('name', key[5:], value)
elif key.startswith('extra+'):
self.add_hstore('extratags', key[6:], value)
+ elif key.startswith('addr+'):
+ self.add_hstore('address', key[5:], value)
+ elif key in ('name', 'address', 'extratags'):
+ self.columns[key] = eval('{' + value + '}')
else:
- assert_in(key, ('class', 'type', 'street', 'addr_place',
- 'isin', 'postcode'))
+ assert_in(key, ('class', 'type'))
self.columns[key] = None if value == '' else value
def set_key_name(self, value):
self.columns['admin_level'] = int(value)
def set_key_housenr(self, value):
- self.columns['housenumber'] = None if value == '' else value
+ if value:
+ self.add_hstore('address', 'housenumber', value)
+
+ def set_key_postcode(self, value):
+ if value:
+ self.add_hstore('address', 'postcode', value)
+
+ def set_key_street(self, value):
+ if value:
+ self.add_hstore('address', 'street', value)
+
+ def set_key_addr_place(self, value):
+ if value:
+ self.add_hstore('address', 'place', value)
def set_key_country(self, value):
- self.columns['country_code'] = None if value == '' else value
+ if value:
+ self.add_hstore('address', 'country', value)
def set_key_geometry(self, value):
self.geometry = self.context.osm.parse_geometry(value, self.context.scene)
for _ in range(int(random.random()*30))))
if self.columns['osm_type'] == 'N' and self.geometry is None:
- self.geometry = "ST_SetSRID(ST_Point(%f, %f), 4326)" % (
- random.random()*360 - 180, random.random()*180 - 90)
+ pt = self.context.osm.grid_node(self.columns['osm_id'])
+ if pt is None:
+ pt = (random.random()*360 - 180, random.random()*180 - 90)
+
+ self.geometry = "ST_SetSRID(ST_Point(%f, %f), 4326)" % pt
else:
assert_is_not_none(self.geometry, "Geometry missing")
query = 'INSERT INTO place (%s, geometry) values(%s, %s)' % (
self.geometry)
cursor.execute(query, list(self.columns.values()))
+class LazyFmt(object):
+
+ def __init__(self, fmtstr, *args):
+ self.fmt = fmtstr
+ self.args = args
+
+ def __str__(self):
+ return self.fmt % self.args
+
+class PlaceObjName(object):
+
+ def __init__(self, placeid, conn):
+ self.pid = placeid
+ self.conn = conn
+
+ def __str__(self):
+ if self.pid is None:
+ return "<null>"
+
+ cur = self.conn.cursor()
+ cur.execute("""SELECT osm_type, osm_id, class
+ FROM placex WHERE place_id = %s""",
+ (self.pid, ))
+ eq_(1, cur.rowcount, "No entry found for place id %s" % self.pid)
+
+ return "%s%s:%s" % cur.fetchone()
+
+def compare_place_id(expected, result, column, context):
+ if expected == '0':
+ eq_(0, result,
+ LazyFmt("Bad place id in column %s. Expected: 0, got: %s.",
+ column, PlaceObjName(result, context.db)))
+ elif expected == '-':
+ assert_is_none(result,
+ LazyFmt("bad place id in column %s: %s.",
+ column, PlaceObjName(result, context.db)))
+ else:
+ eq_(NominatimID(expected).get_place_id(context.db.cursor()), result,
+ LazyFmt("Bad place id in column %s. Expected: %s, got: %s.",
+ column, expected, PlaceObjName(result, context.db)))
+
class NominatimID:
""" Splits a unique identifier for places into its components.
As place_ids cannot be used for testing, we use a unique
self.oid = m.group('id')
self.cls = m.group('cls')
+ def __str__(self):
+ if self.cls is None:
+ return self.typ + self.oid
+
+ return '%s%d:%s' % (self.typ, self.oid, self.cls)
+
def table_select(self):
""" Return where clause and parameter list to select the object
from a Nominatim table.
def get_place_id(self, cur):
where, params = self.table_select()
cur.execute("SELECT place_id FROM placex WHERE %s" % where, params)
- eq_(1, cur.rowcount, "Expected exactly 1 entry in placex found %s" % cur.rowcount)
+ eq_(1, cur.rowcount,
+ "Expected exactly 1 entry in placex for %s found %s"
+ % (str(self), cur.rowcount))
return cur.fetchone()[0]
context.nominatim.run_setup_script('create-functions', 'create-partition-functions')
cur = context.db.cursor()
cur.execute(
- """insert into placex (osm_type, osm_id, class, type, name, admin_level,
- housenumber, street, addr_place, isin, postcode, country_code, extratags,
- geometry)
- select * from place where not (class='place' and type='houses' and osm_type='W')""")
+ """insert into placex (osm_type, osm_id, class, type, name, admin_level, address, extratags, geometry)
+ select osm_type, osm_id, class, type, name, admin_level, address, extratags, geometry
+ from place where not (class='place' and type='houses' and osm_type='W')""")
cur.execute(
- """select insert_osmline (osm_id, housenumber, street, addr_place,
- postcode, country_code, geometry)
- from place where class='place' and type='houses' and osm_type='W'""")
+ """insert into location_property_osmline (osm_id, address, linegeo)
+ SELECT osm_id, address, geometry from place
+ WHERE class='place' and type='houses' and osm_type='W'
+ and ST_GeometryType(geometry) = 'ST_LineString'""")
context.db.commit()
- context.nominatim.run_setup_script('index', 'index-noanalyse')
+ context.nominatim.run_setup_script('calculate-postcodes', 'index', 'index-noanalyse')
@when("updating places")
def update_place_table(context):
col.db_insert(cur)
context.db.commit()
- context.nominatim.run_update_script('index')
+
+ while True:
+ context.nominatim.run_update_script('index')
+
+ cur = context.db.cursor()
+ cur.execute("SELECT 'a' FROM placex WHERE indexed_status != 0 LIMIT 1")
+ if cur.rowcount == 0:
+ break
@when("marking for delete (?P<oids>.*)")
def delete_places(context, oids):
where, params = NominatimID(oid).table_select()
cur.execute("DELETE FROM place WHERE " + where, params)
context.db.commit()
- context.nominatim.run_update_script('index')
+
+ while True:
+ context.nominatim.run_update_script('index')
+
+ cur = context.db.cursor()
+ cur.execute("SELECT 'a' FROM placex WHERE indexed_status != 0 LIMIT 1")
+ if cur.rowcount == 0:
+ break
@then("placex contains(?P<exact> exactly)?")
def check_placex_contents(context, exact):
if exact:
expected_content.add((res['osm_type'], res['osm_id'], res['class']))
for h in row.headings:
- if h.startswith('name'):
+ if h in ('extratags', 'address'):
+ if row[h] == '-':
+ assert_is_none(res[h])
+ else:
+ vdict = eval('{' + row[h] + '}')
+ assert_equals(vdict, res[h])
+ elif h.startswith('name'):
name = h[5:] if h.startswith('name+') else 'name'
assert_in(name, res['name'])
eq_(res['name'][name], row[h])
elif h.startswith('extratags+'):
eq_(res['extratags'][h[10:]], row[h])
- elif h in ('linked_place_id', 'parent_place_id'):
- if row[h] == '0':
- eq_(0, res[h])
- elif row[h] == '-':
- assert_is_none(res[h])
+ elif h.startswith('addr+'):
+ if row[h] == '-':
+ if res['address'] is not None:
+ assert_not_in(h[5:], res['address'])
else:
- eq_(NominatimID(row[h]).get_place_id(context.db.cursor()),
- res[h])
+ assert_in(h[5:], res['address'], "column " + h)
+ assert_equals(res['address'][h[5:]], row[h],
+ "column %s" % h)
+ elif h in ('linked_place_id', 'parent_place_id'):
+ compare_place_id(row[h], res[h], h, context)
else:
assert_db_column(res, h, row[h], context)
expected_content.add((res['osm_type'], res['osm_id'], res['class']))
for h in row.headings:
msg = "%s: %s" % (row['object'], h)
- if h in ('name', 'extratags'):
+ if h in ('name', 'extratags', 'address'):
if row[h] == '-':
assert_is_none(res[h], msg)
else:
assert_equals(res['name'][h[5:]], row[h], msg)
elif h.startswith('extratags+'):
assert_equals(res['extratags'][h[10:]], row[h], msg)
- elif h in ('linked_place_id', 'parent_place_id'):
- if row[h] == '0':
- assert_equals(0, res[h], msg)
- elif row[h] == '-':
- assert_is_none(res[h], msg)
+ elif h.startswith('addr+'):
+ if row[h] == '-':
+ if res['address'] is not None:
+ assert_not_in(h[5:], res['address'])
else:
- assert_equals(NominatimID(row[h]).get_place_id(context.db.cursor()),
- res[h], msg)
+ assert_equals(res['address'][h[5:]], row[h], msg)
+ elif h in ('linked_place_id', 'parent_place_id'):
+ compare_place_id(row[h], res[h], h, context)
else:
assert_db_column(res, h, row[h], context)
FROM word, (SELECT unnest(%s) as term) t
WHERE word_token = make_standard_name(t.term)""",
(terms,))
- ok_(subcur.rowcount >= len(terms))
+ ok_(subcur.rowcount >= len(terms),
+ "No word entry found for " + row[h])
for wid in subcur:
assert_in(wid[0], res[h],
"Missing term for %s/%s: %s" % (pid, h, wid[1]))
context.db.commit()
+@then("place_addressline contains")
+def check_place_addressline(context):
+ cur = context.db.cursor(cursor_factory=psycopg2.extras.DictCursor)
+
+ for row in context.table:
+ pid = NominatimID(row['object']).get_place_id(cur)
+ apid = NominatimID(row['address']).get_place_id(cur)
+ cur.execute(""" SELECT * FROM place_addressline
+ WHERE place_id = %s AND address_place_id = %s""",
+ (pid, apid))
+ assert_less(0, cur.rowcount,
+ "No rows found for place %s and address %s"
+ % (row['object'], row['address']))
+
+ for res in cur:
+ for h in row.headings:
+ if h not in ('address', 'object'):
+ assert_db_column(res, h, row[h], context)
+
+ context.db.commit()
+
@then("(?P<oid>\w+) expands to(?P<neg> no)? interpolation")
def check_location_property_osmline(context, oid, neg):
cur = context.db.cursor(cursor_factory=psycopg2.extras.DictCursor)
eq_('W', nid.typ, "interpolation must be a way")
cur.execute("""SELECT *, ST_AsText(linegeo) as geomtxt
- FROM location_property_osmline WHERE osm_id = %s""",
+ FROM location_property_osmline
+ WHERE osm_id = %s AND startnumber IS NOT NULL""",
(nid.oid, ))
if neg:
if h in ('start', 'end'):
continue
elif h == 'parent_place_id':
- if row[h] == '0':
- eq_(0, res[h])
- elif row[h] == '-':
- assert_is_none(res[h])
- else:
- eq_(NominatimID(row[h]).get_place_id(context.db.cursor()),
- res[h])
+ compare_place_id(row[h], res[h], h, context)
else:
assert_db_column(res, h, row[h], context)