2 Tokenizer implementing normalisation as used before Nominatim 4.
4 from collections import OrderedDict
8 from textwrap import dedent
10 from icu import Transliterator
12 import psycopg2.extras
14 from nominatim.db.connection import connect
15 from nominatim.db import properties
16 from nominatim.db import utils as db_utils
17 from nominatim.db.sql_preprocessor import SQLPreprocessor
18 from nominatim.errors import UsageError
20 DBCFG_NORMALIZATION = "tokenizer_normalization"
21 DBCFG_MAXWORDFREQ = "tokenizer_maxwordfreq"
23 LOG = logging.getLogger()
25 def create(dsn, data_dir):
26 """ Create a new instance of the tokenizer provided by this module.
28 return LegacyTokenizer(dsn, data_dir)
31 def _install_module(config_module_path, src_dir, module_dir):
32 """ Copies the PostgreSQL normalisation module into the project
33 directory if necessary. For historical reasons the module is
34 saved in the '/module' subdirectory and not with the other tokenizer
37 The function detects when the installation is run from the
38 build directory. It doesn't touch the module in that case.
40 # Custom module locations are simply used as is.
41 if config_module_path:
42 LOG.info("Using custom path for database module at '%s'", config_module_path)
43 return config_module_path
45 # Compatibility mode for builddir installations.
46 if module_dir.exists() and src_dir.samefile(module_dir):
47 LOG.info('Running from build directory. Leaving database module as is.')
50 # In any other case install the module in the project directory.
51 if not module_dir.exists():
54 destfile = module_dir / 'nominatim.so'
55 shutil.copy(str(src_dir / 'nominatim.so'), str(destfile))
58 LOG.info('Database module installed at %s', str(destfile))
63 def _check_module(module_dir, conn):
64 """ Try to use the PostgreSQL module to confirm that it is correctly
65 installed and accessible from PostgreSQL.
67 with conn.cursor() as cur:
69 cur.execute("""CREATE FUNCTION nominatim_test_import_func(text)
70 RETURNS text AS '{}/nominatim.so', 'transliteration'
71 LANGUAGE c IMMUTABLE STRICT;
72 DROP FUNCTION nominatim_test_import_func(text)
73 """.format(module_dir))
74 except psycopg2.DatabaseError as err:
75 LOG.fatal("Error accessing database module: %s", err)
76 raise UsageError("Database module cannot be accessed.") from err
79 class LegacyTokenizer:
80 """ The legacy tokenizer uses a special PostgreSQL module to normalize
81 names and queries. The tokenizer thus implements normalization through
82 calls to the database.
85 def __init__(self, dsn, data_dir):
87 self.data_dir = data_dir
88 self.normalization = None
91 def init_new_db(self, config, init_db=True):
92 """ Set up a new tokenizer for the database.
94 This copies all necessary data in the project directory to make
95 sure the tokenizer remains stable even over updates.
97 module_dir = _install_module(config.DATABASE_MODULE_PATH,
98 config.lib_dir.module,
99 config.project_dir / 'module')
101 self.normalization = config.TERM_NORMALIZATION
103 self._install_php(config)
105 with connect(self.dsn) as conn:
106 _check_module(module_dir, conn)
107 self._save_config(conn, config)
111 self.update_sql_functions(config)
112 self._init_db_tables(config)
115 def init_from_project(self):
116 """ Initialise the tokenizer from the project directory.
118 with connect(self.dsn) as conn:
119 self.normalization = properties.get_property(conn, DBCFG_NORMALIZATION)
122 def finalize_import(self, config):
123 """ Do any required postprocessing to make the tokenizer data ready
126 with connect(self.dsn) as conn:
127 sqlp = SQLPreprocessor(conn, config)
128 sqlp.run_sql_file(conn, 'tokenizer/legacy_tokenizer_indices.sql')
131 def update_sql_functions(self, config):
132 """ Reimport the SQL functions for this tokenizer.
134 with connect(self.dsn) as conn:
135 max_word_freq = properties.get_property(conn, DBCFG_MAXWORDFREQ)
136 modulepath = config.DATABASE_MODULE_PATH or \
137 str((config.project_dir / 'module').resolve())
138 sqlp = SQLPreprocessor(conn, config)
139 sqlp.run_sql_file(conn, 'tokenizer/legacy_tokenizer.sql',
140 max_word_freq=max_word_freq,
141 modulepath=modulepath)
144 def check_database(self):
145 """ Check that the tokenizer is set up correctly.
148 The Postgresql extension nominatim.so was not correctly loaded.
153 * Check the output of the CMmake/make installation step
154 * Does nominatim.so exist?
155 * Does nominatim.so exist on the database server?
156 * Can nominatim.so be accessed by the database user?
158 with connect(self.dsn) as conn:
159 with conn.cursor() as cur:
161 out = cur.scalar("SELECT make_standard_name('a')")
162 except psycopg2.Error as err:
163 return hint.format(error=str(err))
166 return hint.format(error='Unexpected result for make_standard_name()')
171 def migrate_database(self, config):
172 """ Initialise the project directory of an existing database for
173 use with this tokenizer.
175 This is a special migration function for updating existing databases
176 to new software versions.
178 self.normalization = config.TERM_NORMALIZATION
179 module_dir = _install_module(config.DATABASE_MODULE_PATH,
180 config.lib_dir.module,
181 config.project_dir / 'module')
183 with connect(self.dsn) as conn:
184 _check_module(module_dir, conn)
185 self._save_config(conn, config)
188 def name_analyzer(self):
189 """ Create a new analyzer for tokenizing names and queries
190 using this tokinzer. Analyzers are context managers and should
194 with tokenizer.name_analyzer() as analyzer:
198 When used outside the with construct, the caller must ensure to
199 call the close() function before destructing the analyzer.
201 Analyzers are not thread-safe. You need to instantiate one per thread.
203 normalizer = Transliterator.createFromRules("phrase normalizer",
205 return LegacyNameAnalyzer(self.dsn, normalizer)
208 def _install_php(self, config):
209 """ Install the php script for the tokenizer.
211 php_file = self.data_dir / "tokenizer.php"
212 php_file.write_text(dedent("""\
214 @define('CONST_Max_Word_Frequency', {0.MAX_WORD_FREQUENCY});
215 @define('CONST_Term_Normalization_Rules', "{0.TERM_NORMALIZATION}");
216 require_once('{0.lib_dir.php}/tokenizer/legacy_tokenizer.php');
220 def _init_db_tables(self, config):
221 """ Set up the word table and fill it with pre-computed word
224 with connect(self.dsn) as conn:
225 sqlp = SQLPreprocessor(conn, config)
226 sqlp.run_sql_file(conn, 'tokenizer/legacy_tokenizer_tables.sql')
229 LOG.warning("Precomputing word tokens")
230 db_utils.execute_file(self.dsn, config.lib_dir.data / 'words.sql')
233 def _save_config(self, conn, config):
234 """ Save the configuration that needs to remain stable for the given
235 database as database properties.
237 properties.set_property(conn, DBCFG_NORMALIZATION, self.normalization)
238 properties.set_property(conn, DBCFG_MAXWORDFREQ, config.MAX_WORD_FREQUENCY)
241 class LegacyNameAnalyzer:
242 """ The legacy analyzer uses the special Postgresql module for
245 Each instance opens a connection to the database to request the
249 def __init__(self, dsn, normalizer):
250 self.conn = connect(dsn).connection
251 self.conn.autocommit = True
252 self.normalizer = normalizer
253 psycopg2.extras.register_hstore(self.conn)
255 self._cache = _TokenCache(self.conn)
262 def __exit__(self, exc_type, exc_value, traceback):
267 """ Free all resources used by the analyzer.
275 def get_word_token_info(conn, words):
276 """ Return token information for the given list of words.
277 If a word starts with # it is assumed to be a full name
278 otherwise is a partial name.
280 The function returns a list of tuples with
281 (original word, word token, word id).
283 The function is used for testing and debugging only
284 and not necessarily efficient.
286 with conn.cursor() as cur:
287 cur.execute("""SELECT t.term, word_token, word_id
288 FROM word, (SELECT unnest(%s::TEXT[]) as term) t
289 WHERE word_token = (CASE
290 WHEN left(t.term, 1) = '#' THEN
291 ' ' || make_standard_name(substring(t.term from 2))
293 make_standard_name(t.term)
295 and class is null and country_code is null""",
298 return [(r[0], r[1], r[2]) for r in cur]
301 def normalize(self, phrase):
302 """ Normalize the given phrase, i.e. remove all properties that
303 are irrelevant for search.
305 return self.normalizer.transliterate(phrase)
308 def add_postcodes_from_db(self):
309 """ Add postcodes from the location_postcode table to the word table.
311 with self.conn.cursor() as cur:
312 cur.execute("""SELECT count(create_postcode_id(pc))
313 FROM (SELECT distinct(postcode) as pc
314 FROM location_postcode) x""")
317 def update_special_phrases(self, phrases):
318 """ Replace the search index for special phrases with the new phrases.
320 norm_phrases = set(((self.normalize(p[0]), p[1], p[2], p[3])
323 with self.conn.cursor() as cur:
324 # Get the old phrases.
325 existing_phrases = set()
326 cur.execute("""SELECT word, class, type, operator FROM word
327 WHERE class != 'place'
328 OR (type != 'house' AND type != 'postcode')""")
329 for label, cls, typ, oper in cur:
330 existing_phrases.add((label, cls, typ, oper or '-'))
332 to_add = norm_phrases - existing_phrases
333 to_delete = existing_phrases - norm_phrases
336 psycopg2.extras.execute_values(
338 """ INSERT INTO word (word_id, word_token, word, class, type,
339 search_name_count, operator)
340 (SELECT nextval('seq_word'), make_standard_name(name), name,
342 CASE WHEN op in ('in', 'near') THEN op ELSE null END
343 FROM (VALUES %s) as v(name, class, type, op))""",
347 psycopg2.extras.execute_values(
349 """ DELETE FROM word USING (VALUES %s) as v(name, in_class, in_type, op)
350 WHERE word = name and class = in_class and type = in_type
351 and ((op = '-' and operator is null) or op = operator)""",
354 LOG.info("Total phrases: %s. Added: %s. Deleted: %s",
355 len(norm_phrases), len(to_add), len(to_delete))
358 def add_country_names(self, country_code, names):
359 """ Add names for the given country to the search index.
361 with self.conn.cursor() as cur:
363 """INSERT INTO word (word_id, word_token, country_code)
364 (SELECT nextval('seq_word'), lookup_token, %s
365 FROM (SELECT ' ' || make_standard_name(n) as lookup_token
367 WHERE NOT EXISTS(SELECT * FROM word
368 WHERE word_token = lookup_token and country_code = %s))
369 """, (country_code, names, country_code))
372 def process_place(self, place):
373 """ Determine tokenizer information about the given place.
375 Returns a JSON-serialisable structure that will be handed into
376 the database via the token_info field.
378 token_info = _TokenInfo(self._cache)
380 names = place.get('name')
383 token_info.add_names(self.conn, names)
385 country_feature = place.get('country_feature')
386 if country_feature and re.fullmatch(r'[A-Za-z][A-Za-z]', country_feature):
387 self.add_country_names(country_feature.lower(), list(names.values()))
389 address = place.get('address')
394 for key, value in address.items():
395 if key == 'postcode':
396 self._add_postcode(value)
397 elif key in ('housenumber', 'streetnumber', 'conscriptionnumber'):
399 elif key == 'street':
400 token_info.add_street(self.conn, value)
402 token_info.add_place(self.conn, value)
403 elif not key.startswith('_') and \
404 key not in ('country', 'full'):
405 addr_terms.append((key, value))
408 token_info.add_housenumbers(self.conn, hnrs)
411 token_info.add_address_terms(self.conn, addr_terms)
413 return token_info.data
416 def _add_postcode(self, postcode):
417 """ Make sure the normalized postcode is present in the word table.
419 def _create_postcode_from_db(pcode):
420 with self.conn.cursor() as cur:
421 cur.execute('SELECT create_postcode_id(%s)', (pcode, ))
423 if re.search(r'[:,;]', postcode) is None:
424 self._cache.postcodes.get(postcode.strip().upper(), _create_postcode_from_db)
428 """ Collect token information to be sent back to the database.
430 def __init__(self, cache):
435 def add_names(self, conn, names):
436 """ Add token information for the names of the place.
438 with conn.cursor() as cur:
439 # Create the token IDs for all names.
440 self.data['names'] = cur.scalar("SELECT make_keywords(%s)::text",
444 def add_housenumbers(self, conn, hnrs):
445 """ Extract housenumber information from the address.
448 token = self.cache.get_housenumber(hnrs[0])
449 if token is not None:
450 self.data['hnr_tokens'] = token
451 self.data['hnr'] = hnrs[0]
454 # split numbers if necessary
457 simple_list.extend((x.strip() for x in re.split(r'[;,]', hnr)))
459 if len(simple_list) > 1:
460 simple_list = list(set(simple_list))
462 with conn.cursor() as cur:
463 cur.execute("SELECT (create_housenumbers(%s)).* ", (simple_list, ))
464 self.data['hnr_tokens'], self.data['hnr'] = cur.fetchone()
467 def add_street(self, conn, street):
468 """ Add addr:street match terms.
470 def _get_street(name):
471 with conn.cursor() as cur:
472 return cur.scalar("SELECT word_ids_from_name(%s)::text", (name, ))
474 self.data['street'] = self.cache.streets.get(street, _get_street)
477 def add_place(self, conn, place):
478 """ Add addr:place search and match terms.
480 def _get_place(name):
481 with conn.cursor() as cur:
482 cur.execute("""SELECT (addr_ids_from_name(%s)
483 || getorcreate_name_id(make_standard_name(%s), ''))::text,
484 word_ids_from_name(%s)::text""",
486 return cur.fetchone()
488 self.data['place_search'], self.data['place_match'] = \
489 self.cache.places.get(place, _get_place)
492 def add_address_terms(self, conn, terms):
493 """ Add additional address terms.
495 def _get_address_term(name):
496 with conn.cursor() as cur:
497 cur.execute("""SELECT addr_ids_from_name(%s)::text,
498 word_ids_from_name(%s)::text""",
500 return cur.fetchone()
503 for key, value in terms:
504 tokens[key] = self.cache.address_terms.get(value, _get_address_term)
506 self.data['addr'] = tokens
510 """ Least recently used cache that accepts a generator function to
511 produce the item when there is a cache miss.
514 def __init__(self, maxsize=128, init_data=None):
515 self.data = init_data or OrderedDict()
516 self.maxsize = maxsize
517 if init_data is not None and len(init_data) > maxsize:
518 self.maxsize = len(init_data)
520 def get(self, key, generator):
521 """ Get the item with the given key from the cache. If nothing
522 is found in the cache, generate the value through the
523 generator function and store it in the cache.
525 value = self.data.get(key)
526 if value is not None:
527 self.data.move_to_end(key)
529 value = generator(key)
530 if len(self.data) >= self.maxsize:
531 self.data.popitem(last=False)
532 self.data[key] = value
538 """ Cache for token information to avoid repeated database queries.
540 This cache is not thread-safe and needs to be instantiated per
543 def __init__(self, conn):
545 self.streets = _LRU(maxsize=256)
546 self.places = _LRU(maxsize=128)
547 self.address_terms = _LRU(maxsize=1024)
549 # Lookup houseunumbers up to 100 and cache them
550 with conn.cursor() as cur:
551 cur.execute("""SELECT i, ARRAY[getorcreate_housenumber_id(i::text)]::text
552 FROM generate_series(1, 100) as i""")
553 self._cached_housenumbers = {str(r[0]) : r[1] for r in cur}
555 # Get postcodes that are already saved
556 postcodes = OrderedDict()
557 with conn.cursor() as cur:
558 cur.execute("""SELECT word FROM word
559 WHERE class ='place' and type = 'postcode'""")
561 postcodes[row[0]] = None
562 self.postcodes = _LRU(maxsize=32, init_data=postcodes)
564 def get_housenumber(self, number):
565 """ Get a housenumber token from the cache.
567 return self._cached_housenumbers.get(number)