]> git.openstreetmap.org Git - nominatim.git/blobdiff - nominatim/tokenizer/legacy_tokenizer.py
Merge pull request #2757 from lonvia/filter-postcodes
[nominatim.git] / nominatim / tokenizer / legacy_tokenizer.py
index 0aacb57f536f17cc1a07e10b74943991e281f71c..36fd5722441a12e92d24434d5dc1317497dd27bc 100644 (file)
@@ -1,3 +1,9 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# This file is part of Nominatim. (https://nominatim.org)
+#
+# Copyright (C) 2022 by the Nominatim developer community.
+# For a full list of authors see the git log.
 """
 Tokenizer implementing normalisation as used before Nominatim 4.
 """
@@ -5,7 +11,9 @@ from collections import OrderedDict
 import logging
 import re
 import shutil
+from textwrap import dedent
 
+from icu import Transliterator
 import psycopg2
 import psycopg2.extras
 
@@ -14,6 +22,7 @@ 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
+from nominatim.tokenizer.base import AbstractAnalyzer, AbstractTokenizer
 
 DBCFG_NORMALIZATION = "tokenizer_normalization"
 DBCFG_MAXWORDFREQ = "tokenizer_maxwordfreq"
@@ -65,16 +74,16 @@ def _check_module(module_dir, conn):
     with conn.cursor() as cur:
         try:
             cur.execute("""CREATE FUNCTION nominatim_test_import_func(text)
-                           RETURNS text AS '{}/nominatim.so', 'transliteration'
+                           RETURNS text AS %s, 'transliteration'
                            LANGUAGE c IMMUTABLE STRICT;
                            DROP FUNCTION nominatim_test_import_func(text)
-                        """.format(module_dir))
+                        """, (f'{module_dir}/nominatim.so', ))
         except psycopg2.DatabaseError as err:
             LOG.fatal("Error accessing database module: %s", err)
             raise UsageError("Database module cannot be accessed.") from err
 
 
-class LegacyTokenizer:
+class LegacyTokenizer(AbstractTokenizer):
     """ The legacy tokenizer uses a special PostgreSQL module to normalize
         names and queries. The tokenizer thus implements normalization through
         calls to the database.
@@ -86,7 +95,7 @@ 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
@@ -98,21 +107,39 @@ class LegacyTokenizer:
 
         self.normalization = config.TERM_NORMALIZATION
 
+        self._install_php(config, overwrite=True)
+
         with connect(self.dsn) as conn:
             _check_module(module_dir, conn)
             self._save_config(conn, config)
             conn.commit()
 
-        self.update_sql_functions(config)
-        self._init_db_tables(config)
+        if init_db:
+            self.update_sql_functions(config)
+            self._init_db_tables(config)
 
 
-    def init_from_project(self):
+    def init_from_project(self, config):
         """ Initialise the tokenizer from the project directory.
         """
         with connect(self.dsn) as conn:
             self.normalization = properties.get_property(conn, DBCFG_NORMALIZATION)
 
+        if not (config.project_dir / 'module' / 'nominatim.so').exists():
+            _install_module(config.DATABASE_MODULE_PATH,
+                            config.lib_dir.module,
+                            config.project_dir / 'module')
+
+        self._install_php(config, overwrite=False)
+
+    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.
@@ -127,6 +154,33 @@ class LegacyTokenizer:
                               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.
@@ -134,6 +188,7 @@ class LegacyTokenizer:
             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')
@@ -143,6 +198,32 @@ class LegacyTokenizer:
             self._save_config(conn, config)
 
 
+    def update_statistics(self):
+        """ Recompute the frequency of full words.
+        """
+        with connect(self.dsn) as conn:
+            if conn.table_exists('search_name'):
+                with conn.cursor() as cur:
+                    cur.drop_table("word_frequencies")
+                    LOG.info("Computing word frequencies")
+                    cur.execute("""CREATE TEMP TABLE word_frequencies AS
+                                     SELECT unnest(name_vector) as id, count(*)
+                                     FROM search_name GROUP BY id""")
+                    cur.execute("CREATE INDEX ON word_frequencies(id)")
+                    LOG.info("Update word table with recomputed frequencies")
+                    cur.execute("""UPDATE word SET search_name_count = count
+                                   FROM word_frequencies
+                                   WHERE word_token like ' %' and word_id = id""")
+                    cur.drop_table("word_frequencies")
+            conn.commit()
+
+
+    def update_word_tokens(self):
+        """ No house-keeping implemented for the legacy tokenizer.
+        """
+        LOG.info("No tokenizer clean-up available.")
+
+
     def name_analyzer(self):
         """ Create a new analyzer for tokenizing names and queries
             using this tokinzer. Analyzers are context managers and should
@@ -158,7 +239,23 @@ class LegacyTokenizer:
 
             Analyzers are not thread-safe. You need to instantiate one per thread.
         """
-        return LegacyNameAnalyzer(self.dsn)
+        normalizer = Transliterator.createFromRules("phrase normalizer",
+                                                    self.normalization)
+        return LegacyNameAnalyzer(self.dsn, normalizer)
+
+
+    def _install_php(self, config, overwrite=True):
+        """ Install the php script for the tokenizer.
+        """
+        php_file = self.data_dir / "tokenizer.php"
+
+        if not php_file.exists() or overwrite:
+            php_file.write_text(dedent(f"""\
+                <?php
+                @define('CONST_Max_Word_Frequency', {config.MAX_WORD_FREQUENCY});
+                @define('CONST_Term_Normalization_Rules', "{config.TERM_NORMALIZATION}");
+                require_once('{config.lib_dir.php}/tokenizer/legacy_tokenizer.php');
+                """), encoding='utf-8')
 
 
     def _init_db_tables(self, config):
@@ -182,8 +279,7 @@ class LegacyTokenizer:
         properties.set_property(conn, DBCFG_MAXWORDFREQ, config.MAX_WORD_FREQUENCY)
 
 
-
-class LegacyNameAnalyzer:
+class LegacyNameAnalyzer(AbstractAnalyzer):
     """ The legacy analyzer uses the special Postgresql module for
         splitting names.
 
@@ -191,22 +287,15 @@ class LegacyNameAnalyzer:
         normalization.
     """
 
-    def __init__(self, dsn):
+    def __init__(self, dsn, normalizer):
         self.conn = connect(dsn).connection
         self.conn.autocommit = True
+        self.normalizer = normalizer
         psycopg2.extras.register_hstore(self.conn)
 
         self._cache = _TokenCache(self.conn)
 
 
-    def __enter__(self):
-        return self
-
-
-    def __exit__(self, exc_type, exc_value, traceback):
-        self.close()
-
-
     def close(self):
         """ Free all resources used by the analyzer.
         """
@@ -215,13 +304,137 @@ class LegacyNameAnalyzer:
             self.conn = None
 
 
-    def add_postcodes_from_db(self):
-        """ Add postcodes from the location_postcode table to the word table.
+    def get_word_token_info(self, words):
+        """ Return token information for the given list of words.
+            If a word starts with # it is assumed to be a full name
+            otherwise is a partial name.
+
+            The function returns a list of tuples with
+            (original word, word token, word id).
+
+            The function is used for testing and debugging only
+            and not necessarily efficient.
         """
         with self.conn.cursor() as cur:
-            cur.execute("""SELECT count(create_postcode_id(pc))
-                           FROM (SELECT distinct(postcode) as pc
-                                 FROM location_postcode) x""")
+            cur.execute("""SELECT t.term, word_token, word_id
+                           FROM word, (SELECT unnest(%s::TEXT[]) as term) t
+                           WHERE word_token = (CASE
+                                   WHEN left(t.term, 1) = '#' THEN
+                                     ' ' || make_standard_name(substring(t.term from 2))
+                                   ELSE
+                                     make_standard_name(t.term)
+                                   END)
+                                 and class is null and country_code is null""",
+                        (words, ))
+
+            return [(r[0], r[1], r[2]) for r in cur]
+
+
+    def normalize(self, phrase):
+        """ Normalize the given phrase, i.e. remove all properties that
+            are irrelevant for search.
+        """
+        return self.normalizer.transliterate(phrase)
+
+
+    def normalize_postcode(self, postcode):
+        """ Convert the postcode to a standardized form.
+
+            This function must yield exactly the same result as the SQL function
+            'token_normalized_postcode()'.
+        """
+        return postcode.strip().upper()
+
+
+    def update_postcodes_from_db(self):
+        """ Update postcode tokens in the word table from the location_postcode
+            table.
+        """
+        with self.conn.cursor() as cur:
+            # This finds us the rows in location_postcode and word that are
+            # missing in the other table.
+            cur.execute("""SELECT * FROM
+                            (SELECT pc, word FROM
+                              (SELECT distinct(postcode) as pc FROM location_postcode) p
+                              FULL JOIN
+                              (SELECT word FROM word
+                                WHERE class ='place' and type = 'postcode') w
+                              ON pc = word) x
+                           WHERE pc is null or word is null""")
+
+            to_delete = []
+            to_add = []
+
+            for postcode, word in cur:
+                if postcode is None:
+                    to_delete.append(word)
+                else:
+                    to_add.append(postcode)
+
+            if to_delete:
+                cur.execute("""DELETE FROM WORD
+                               WHERE class ='place' and type = 'postcode'
+                                     and word = any(%s)
+                            """, (to_delete, ))
+            if to_add:
+                cur.execute("""SELECT count(create_postcode_id(pc))
+                               FROM unnest(%s) as pc
+                            """, (to_add, ))
+
+
+
+    def update_special_phrases(self, phrases, should_replace):
+        """ Replace the search index for special phrases with the new phrases.
+        """
+        norm_phrases = set(((self.normalize(p[0]), p[1], p[2], p[3])
+                            for p in phrases))
+
+        with self.conn.cursor() as cur:
+            # Get the old phrases.
+            existing_phrases = set()
+            cur.execute("""SELECT word, class, type, operator FROM word
+                           WHERE class != 'place'
+                                 OR (type != 'house' AND type != 'postcode')""")
+            for label, cls, typ, oper in cur:
+                existing_phrases.add((label, cls, typ, oper or '-'))
+
+            to_add = norm_phrases - existing_phrases
+            to_delete = existing_phrases - norm_phrases
+
+            if to_add:
+                cur.execute_values(
+                    """ INSERT INTO word (word_id, word_token, word, class, type,
+                                          search_name_count, operator)
+                        (SELECT nextval('seq_word'), ' ' || make_standard_name(name), name,
+                                class, type, 0,
+                                CASE WHEN op in ('in', 'near') THEN op ELSE null END
+                           FROM (VALUES %s) as v(name, class, type, op))""",
+                    to_add)
+
+            if to_delete and should_replace:
+                cur.execute_values(
+                    """ DELETE FROM word USING (VALUES %s) as v(name, in_class, in_type, op)
+                        WHERE word = name and class = in_class and type = in_type
+                              and ((op = '-' and operator is null) or op = operator)""",
+                    to_delete)
+
+        LOG.info("Total phrases: %s. Added: %s. Deleted: %s",
+                 len(norm_phrases), len(to_add), len(to_delete))
+
+
+    def add_country_names(self, country_code, names):
+        """ Add names for the given country to the search index.
+        """
+        with self.conn.cursor() as cur:
+            cur.execute(
+                """INSERT INTO word (word_id, word_token, country_code)
+                   (SELECT nextval('seq_word'), lookup_token, %s
+                      FROM (SELECT DISTINCT ' ' || make_standard_name(n) as lookup_token
+                            FROM unnest(%s)n) y
+                      WHERE NOT EXISTS(SELECT * FROM word
+                                       WHERE word_token = lookup_token and country_code = %s))
+                """, (country_code, list(names.values()), country_code))
+
 
     def process_place(self, place):
         """ Determine tokenizer information about the given place.
@@ -231,31 +444,48 @@ class LegacyNameAnalyzer:
         """
         token_info = _TokenInfo(self._cache)
 
-        token_info.add_names(self.conn, place.get('name'), place.get('country_feature'))
+        names = place.name
 
-        address = place.get('address')
+        if names:
+            token_info.add_names(self.conn, names)
 
+            if place.is_country():
+                self.add_country_names(place.country_code, names)
+
+        address = place.address
         if address:
-            self._add_postcode(address.get('postcode'))
-            token_info.add_housenumbers(self.conn, address)
-            token_info.add_address_parent(self.conn, address.get('street'),
-                                          address.get('place'))
-            token_info.add_address_parts(self.conn, address)
+            self._process_place_address(token_info, address)
 
         return token_info.data
 
 
-    def _add_postcode(self, postcode):
-        """ Make sure the normalized postcode is present in the word table.
-        """
-        if not postcode or re.search(r'[:,;]', postcode) is not None:
-            return
+    def _process_place_address(self, token_info, address):
+        hnrs = []
+        addr_terms = []
 
-        def _create_postcode_from_db(pcode):
-            with self.conn.cursor() as cur:
-                cur.execute('SELECT create_postcode_id(%s)', (pcode, ))
+        for key, value in address.items():
+            if key == 'postcode':
+                # Make sure the normalized postcode is present in the word table.
+                if re.search(r'[:,;]', value) is None:
+                    norm_pc = self.normalize_postcode(value)
+                    token_info.set_postcode(norm_pc)
+                    self._cache.add_postcode(self.conn, norm_pc)
+            elif key in ('housenumber', 'streetnumber', 'conscriptionnumber'):
+                hnrs.append(value)
+            elif key == 'street':
+                token_info.add_street(self.conn, value)
+            elif key == 'place':
+                token_info.add_place(self.conn, value)
+            elif not key.startswith('_') \
+                 and key not in ('country', 'full', 'inclusion'):
+                addr_terms.append((key, value))
+
+        if hnrs:
+            token_info.add_housenumbers(self.conn, hnrs)
+
+        if addr_terms:
+            token_info.add_address_terms(self.conn, addr_terms)
 
-        self._cache.postcodes.get(postcode.strip().upper(), _create_postcode_from_db)
 
 
 class _TokenInfo:
@@ -266,32 +496,18 @@ class _TokenInfo:
         self.data = {}
 
 
-    def add_names(self, conn, names, country_feature):
+    def add_names(self, conn, names):
         """ Add token information for the names of the place.
         """
-        if not names:
-            return
-
         with conn.cursor() as cur:
             # Create the token IDs for all names.
             self.data['names'] = cur.scalar("SELECT make_keywords(%s)::text",
                                             (names, ))
 
-            # Add country tokens to word table if necessary.
-            if country_feature and re.fullmatch(r'[A-Za-z][A-Za-z]', country_feature):
-                cur.execute("SELECT create_country(%s, %s)",
-                            (names, country_feature.lower()))
 
-
-    def add_housenumbers(self, conn, address):
+    def add_housenumbers(self, conn, hnrs):
         """ Extract housenumber information from the address.
         """
-        hnrs = [v for k, v in address.items()
-                if k in ('housenumber', 'streetnumber', 'conscriptionnumber')]
-
-        if not hnrs:
-            return
-
         if len(hnrs) == 1:
             token = self.cache.get_housenumber(hnrs[0])
             if token is not None:
@@ -308,31 +524,43 @@ class _TokenInfo:
             simple_list = list(set(simple_list))
 
         with conn.cursor() as cur:
-            cur.execute("SELECT (create_housenumbers(%s)).* ", (simple_list, ))
+            cur.execute("SELECT * FROM create_housenumbers(%s)", (simple_list, ))
             self.data['hnr_tokens'], self.data['hnr'] = cur.fetchone()
 
 
-    def add_address_parent(self, conn, street, place):
-        """ Extract the tokens for street and place terms.
+    def set_postcode(self, postcode):
+        """ Set or replace the postcode token with the given value.
+        """
+        self.data['postcode'] = postcode
+
+    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, ))
+
+        tokens = self.cache.streets.get(street, _get_street)
+        if tokens:
+            self.data['street'] = tokens
+
+
+    def add_place(self, conn, place):
+        """ Add addr:place search and match terms.
         """
-        def _get_streetplace(name):
+        def _get_place(name):
             with conn.cursor() as cur:
-                cur.execute("""SELECT (addr_ids_from_name(%s) || getorcreate_name_id(make_standard_name(%s), ''))::text,
+                cur.execute("""SELECT make_keywords(hstore('name' , %s))::text,
                                       word_ids_from_name(%s)::text""",
-                            (name, name, name))
+                            (name, name))
                 return cur.fetchone()
 
-        if street:
-            self.data['street_search'], self.data['street_match'] = \
-                self.cache.streets.get(street, _get_streetplace)
+        self.data['place_search'], self.data['place_match'] = \
+            self.cache.places.get(place, _get_place)
 
-        if place:
-            self.data['place_search'], self.data['place_match'] = \
-                self.cache.streets.get(place, _get_streetplace)
 
-
-    def add_address_parts(self, conn, address):
-        """ Extract address terms.
+    def add_address_terms(self, conn, terms):
+        """ Add additional address terms.
         """
         def _get_address_term(name):
             with conn.cursor() as cur:
@@ -342,11 +570,10 @@ class _TokenInfo:
                 return cur.fetchone()
 
         tokens = {}
-        for key, value in address.items():
-            if not key.startswith('_') and \
-               key not in ('country', 'street', 'place', 'postcode', 'full',
-                           'housenumber', 'streetnumber', 'conscriptionnumber'):
-                tokens[key] = self.cache.address_terms.get(value, _get_address_term)
+        for key, value in terms:
+            items = self.cache.address_terms.get(value, _get_address_term)
+            if items[0] or items[1]:
+                tokens[key] = items
 
         if tokens:
             self.data['addr'] = tokens
@@ -357,9 +584,11 @@ class _LRU:
         produce the item when there is a cache miss.
     """
 
-    def __init__(self, maxsize=128):
-        self.data = OrderedDict()
+    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
@@ -386,7 +615,6 @@ class _TokenCache:
     """
     def __init__(self, conn):
         # various LRU caches
-        self.postcodes = _LRU(maxsize=32)
         self.streets = _LRU(maxsize=256)
         self.places = _LRU(maxsize=128)
         self.address_terms = _LRU(maxsize=1024)
@@ -395,10 +623,21 @@ class _TokenCache:
         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}
+            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)