X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/296a66558ffb2ee2c43e653b00ff24097ee3525c..73953cbac95847740339d7a5ee3c0e5f142d93c5:/nominatim/tokenizer/legacy_tokenizer.py diff --git a/nominatim/tokenizer/legacy_tokenizer.py b/nominatim/tokenizer/legacy_tokenizer.py index bd7b7709..c19dce2f 100644 --- a/nominatim/tokenizer/legacy_tokenizer.py +++ b/nominatim/tokenizer/legacy_tokenizer.py @@ -1,16 +1,24 @@ """ Tokenizer implementing normalisation as used before Nominatim 4. """ +from collections import OrderedDict import logging +import re import shutil +from textwrap import dedent +from icu import Transliterator import psycopg2 +import psycopg2.extras from nominatim.db.connection import connect from nominatim.db import properties +from nominatim.db import utils as db_utils +from nominatim.db.sql_preprocessor import SQLPreprocessor from nominatim.errors import UsageError DBCFG_NORMALIZATION = "tokenizer_normalization" +DBCFG_MAXWORDFREQ = "tokenizer_maxwordfreq" LOG = logging.getLogger() @@ -20,7 +28,7 @@ def create(dsn, data_dir): return LegacyTokenizer(dsn, data_dir) -def _install_module(src_dir, module_dir): +def _install_module(config_module_path, src_dir, module_dir): """ Copies the PostgreSQL normalisation module into the project directory if necessary. For historical reasons the module is saved in the '/module' subdirectory and not with the other tokenizer @@ -29,10 +37,17 @@ def _install_module(src_dir, module_dir): The function detects when the installation is run from the build directory. It doesn't touch the module in that case. """ + # Custom module locations are simply used as is. + if config_module_path: + LOG.info("Using custom path for database module at '%s'", config_module_path) + return config_module_path + + # Compatibility mode for builddir installations. if module_dir.exists() and src_dir.samefile(module_dir): LOG.info('Running from build directory. Leaving database module as is.') - return + return module_dir + # In any other case install the module in the project directory. if not module_dir.exists(): module_dir.mkdir() @@ -42,8 +57,13 @@ def _install_module(src_dir, module_dir): LOG.info('Database module installed at %s', str(destfile)) + return module_dir + def _check_module(module_dir, conn): + """ Try to use the PostgreSQL module to confirm that it is correctly + installed and accessible from PostgreSQL. + """ with conn.cursor() as cur: try: cur.execute("""CREATE FUNCTION nominatim_test_import_func(text) @@ -68,34 +88,513 @@ class LegacyTokenizer: self.normalization = None - def init_new_db(self, config): + def init_new_db(self, config, init_db=True): """ Set up a new tokenizer for the database. This copies all necessary data in the project directory to make sure the tokenizer remains stable even over updates. """ - # Find and optionally install the PsotgreSQL normalization module. - if config.DATABASE_MODULE_PATH: - LOG.info("Using custom path for database module at '%s'", - config.DATABASE_MODULE_PATH) - module_dir = config.DATABASE_MODULE_PATH - else: - _install_module(config.lib_dir.module, config.project_dir / 'module') - module_dir = config.project_dir / 'module' + module_dir = _install_module(config.DATABASE_MODULE_PATH, + config.lib_dir.module, + config.project_dir / 'module') self.normalization = config.TERM_NORMALIZATION + self._install_php(config) + with connect(self.dsn) as conn: _check_module(module_dir, conn) - - # Stable configuration is saved in the database. - properties.set_property(conn, DBCFG_NORMALIZATION, self.normalization) - + self._save_config(conn, config) conn.commit() + if init_db: + self.update_sql_functions(config) + self._init_db_tables(config) + def init_from_project(self): """ Initialise the tokenizer from the project directory. """ with connect(self.dsn) as conn: self.normalization = properties.get_property(conn, DBCFG_NORMALIZATION) + + + def finalize_import(self, config): + """ Do any required postprocessing to make the tokenizer data ready + for use. + """ + with connect(self.dsn) as conn: + sqlp = SQLPreprocessor(conn, config) + sqlp.run_sql_file(conn, 'tokenizer/legacy_tokenizer_indices.sql') + + + def update_sql_functions(self, config): + """ Reimport the SQL functions for this tokenizer. + """ + with connect(self.dsn) as conn: + max_word_freq = properties.get_property(conn, DBCFG_MAXWORDFREQ) + modulepath = config.DATABASE_MODULE_PATH or \ + str((config.project_dir / 'module').resolve()) + sqlp = SQLPreprocessor(conn, config) + sqlp.run_sql_file(conn, 'tokenizer/legacy_tokenizer.sql', + max_word_freq=max_word_freq, + modulepath=modulepath) + + + def check_database(self): + """ Check that the tokenizer is set up correctly. + """ + hint = """\ + The Postgresql extension nominatim.so was not correctly loaded. + + Error: {error} + + Hints: + * Check the output of the CMmake/make installation step + * Does nominatim.so exist? + * Does nominatim.so exist on the database server? + * Can nominatim.so be accessed by the database user? + """ + with connect(self.dsn) as conn: + with conn.cursor() as cur: + try: + out = cur.scalar("SELECT make_standard_name('a')") + except psycopg2.Error as err: + return hint.format(error=str(err)) + + if out != 'a': + return hint.format(error='Unexpected result for make_standard_name()') + + return None + + + def migrate_database(self, config): + """ Initialise the project directory of an existing database for + use with this tokenizer. + + This is a special migration function for updating existing databases + to new software versions. + """ + self.normalization = config.TERM_NORMALIZATION + module_dir = _install_module(config.DATABASE_MODULE_PATH, + config.lib_dir.module, + config.project_dir / 'module') + + with connect(self.dsn) as conn: + _check_module(module_dir, conn) + self._save_config(conn, config) + + + def name_analyzer(self): + """ Create a new analyzer for tokenizing names and queries + using this tokinzer. Analyzers are context managers and should + be used accordingly: + + ``` + with tokenizer.name_analyzer() as analyzer: + analyser.tokenize() + ``` + + When used outside the with construct, the caller must ensure to + call the close() function before destructing the analyzer. + + Analyzers are not thread-safe. You need to instantiate one per thread. + """ + normalizer = Transliterator.createFromRules("phrase normalizer", + self.normalization) + return LegacyNameAnalyzer(self.dsn, normalizer) + + + def _install_php(self, config): + """ Install the php script for the tokenizer. + """ + php_file = self.data_dir / "tokenizer.php" + php_file.write_text(dedent("""\ + 1: + simple_list = list(set(simple_list)) + + with conn.cursor() as cur: + cur.execute("SELECT (create_housenumbers(%s)).* ", (simple_list, )) + self.data['hnr_tokens'], self.data['hnr'] = cur.fetchone() + + + def add_street(self, conn, street): + """ Add addr:street match terms. + """ + def _get_street(name): + with conn.cursor() as cur: + return cur.scalar("SELECT word_ids_from_name(%s)::text", (name, )) + + self.data['street'] = self.cache.streets.get(street, _get_street) + + + def add_place(self, conn, place): + """ Add addr:place search and match terms. + """ + def _get_place(name): + with conn.cursor() as cur: + cur.execute("""SELECT make_keywords(hstore('name' , %s))::text, + word_ids_from_name(%s)::text""", + (name, name)) + return cur.fetchone() + + self.data['place_search'], self.data['place_match'] = \ + self.cache.places.get(place, _get_place) + + + def add_address_terms(self, conn, terms): + """ Add additional address terms. + """ + def _get_address_term(name): + with conn.cursor() as cur: + cur.execute("""SELECT addr_ids_from_name(%s)::text, + word_ids_from_name(%s)::text""", + (name, name)) + return cur.fetchone() + + tokens = {} + for key, value in terms: + tokens[key] = self.cache.address_terms.get(value, _get_address_term) + + self.data['addr'] = tokens + + +class _LRU: + """ Least recently used cache that accepts a generator function to + produce the item when there is a cache miss. + """ + + def __init__(self, maxsize=128, init_data=None): + self.data = init_data or OrderedDict() + self.maxsize = maxsize + if init_data is not None and len(init_data) > maxsize: + self.maxsize = len(init_data) + + def get(self, key, generator): + """ Get the item with the given key from the cache. If nothing + is found in the cache, generate the value through the + generator function and store it in the cache. + """ + value = self.data.get(key) + if value is not None: + self.data.move_to_end(key) + else: + value = generator(key) + if len(self.data) >= self.maxsize: + self.data.popitem(last=False) + self.data[key] = value + + return value + + +class _TokenCache: + """ Cache for token information to avoid repeated database queries. + + This cache is not thread-safe and needs to be instantiated per + analyzer. + """ + def __init__(self, conn): + # various LRU caches + self.streets = _LRU(maxsize=256) + self.places = _LRU(maxsize=128) + self.address_terms = _LRU(maxsize=1024) + + # Lookup houseunumbers up to 100 and cache them + with conn.cursor() as cur: + cur.execute("""SELECT i, ARRAY[getorcreate_housenumber_id(i::text)]::text + FROM generate_series(1, 100) as i""") + self._cached_housenumbers = {str(r[0]): r[1] for r in cur} + + # For postcodes remember the ones that have already been added + self.postcodes = set() + + def get_housenumber(self, number): + """ Get a housenumber token from the cache. + """ + return self._cached_housenumbers.get(number) + + + def add_postcode(self, conn, postcode): + """ Make sure the given postcode is in the database. + """ + if postcode not in self.postcodes: + with conn.cursor() as cur: + cur.execute('SELECT create_postcode_id(%s)', (postcode, )) + self.postcodes.add(postcode)