]> git.openstreetmap.org Git - nominatim.git/blobdiff - nominatim/tokenizer/legacy_tokenizer.py
port legacy tokenizer to new postcode handling
[nominatim.git] / nominatim / tokenizer / legacy_tokenizer.py
index 438a5aff9ed3861995606c5d8409ff0c7ac13c35..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.
 """
@@ -16,6 +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"
@@ -67,16 +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.
@@ -100,7 +107,7 @@ class LegacyTokenizer:
 
         self.normalization = config.TERM_NORMALIZATION
 
-        self._install_php(config)
+        self._install_php(config, overwrite=True)
 
         with connect(self.dsn) as conn:
             _check_module(module_dir, conn)
@@ -112,12 +119,18 @@ class LegacyTokenizer:
             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
@@ -141,7 +154,7 @@ class LegacyTokenizer:
                               modulepath=modulepath)
 
 
-    def check_database(self):
+    def check_database(self, _):
         """ Check that the tokenizer is set up correctly.
         """
         hint = """\
@@ -185,6 +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
@@ -205,16 +244,18 @@ class LegacyTokenizer:
         return LegacyNameAnalyzer(self.dsn, normalizer)
 
 
-    def _install_php(self, config):
+    def _install_php(self, config, overwrite=True):
         """ Install the php script for the tokenizer.
         """
         php_file = self.data_dir / "tokenizer.php"
-        php_file.write_text(dedent("""\
-            <?php
-            @define('CONST_Max_Word_Frequency', {0.MAX_WORD_FREQUENCY});
-            @define('CONST_Term_Normalization_Rules', "{0.TERM_NORMALIZATION}");
-            require_once('{0.lib_dir.php}/tokenizer/legacy_tokenizer.php');
-            """.format(config)))
+
+        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):
@@ -238,7 +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.
 
@@ -255,14 +296,6 @@ class LegacyNameAnalyzer:
         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.
         """
@@ -271,8 +304,7 @@ class LegacyNameAnalyzer:
             self.conn = None
 
 
-    @staticmethod
-    def get_word_token_info(conn, words):
+    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.
@@ -283,7 +315,7 @@ class LegacyNameAnalyzer:
             The function is used for testing and debugging only
             and not necessarily efficient.
         """
-        with conn.cursor() as cur:
+        with self.conn.cursor() as cur:
             cur.execute("""SELECT t.term, word_token, word_id
                            FROM word, (SELECT unnest(%s::TEXT[]) as term) t
                            WHERE word_token = (CASE
@@ -305,16 +337,53 @@ class LegacyNameAnalyzer:
         return self.normalizer.transliterate(phrase)
 
 
-    def add_postcodes_from_db(self):
-        """ Add postcodes from the location_postcode table to the word table.
+    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:
-            cur.execute("""SELECT count(create_postcode_id(pc))
-                           FROM (SELECT distinct(postcode) as pc
-                                 FROM location_postcode) x""")
+            # 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):
+    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])
@@ -333,19 +402,17 @@ class LegacyNameAnalyzer:
             to_delete = existing_phrases - norm_phrases
 
             if to_add:
-                psycopg2.extras.execute_values(
-                    cur,
+                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,
+                        (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:
-                psycopg2.extras.execute_values(
-                    cur,
+            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)""",
@@ -362,11 +429,11 @@ class LegacyNameAnalyzer:
             cur.execute(
                 """INSERT INTO word (word_id, word_token, country_code)
                    (SELECT nextval('seq_word'), lookup_token, %s
-                      FROM (SELECT ' ' || make_standard_name(n) as lookup_token
+                      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, names, country_code))
+                """, (country_code, list(names.values()), country_code))
 
 
     def process_place(self, place):
@@ -377,51 +444,48 @@ class LegacyNameAnalyzer:
         """
         token_info = _TokenInfo(self._cache)
 
-        names = place.get('name')
+        names = place.name
 
         if names:
             token_info.add_names(self.conn, names)
 
-            country_feature = place.get('country_feature')
-            if country_feature and re.fullmatch(r'[A-Za-z][A-Za-z]', country_feature):
-                self.add_country_names(country_feature.lower(), list(names.values()))
-
-        address = place.get('address')
+            if place.is_country():
+                self.add_country_names(place.country_code, names)
 
+        address = place.address
         if address:
-            hnrs = []
-            addr_terms = []
-            for key, value in address.items():
-                if key == 'postcode':
-                    self._add_postcode(value)
-                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'):
-                    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._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.
-        """
-        def _create_postcode_from_db(pcode):
-            with self.conn.cursor() as cur:
-                cur.execute('SELECT create_postcode_id(%s)', (pcode, ))
+    def _process_place_address(self, token_info, address):
+        hnrs = []
+        addr_terms = []
+
+        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)
 
-        if re.search(r'[:,;]', postcode) is None:
-            self._cache.postcodes.get(postcode.strip().upper(), _create_postcode_from_db)
 
 
 class _TokenInfo:
@@ -460,10 +524,15 @@ 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 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.
         """
@@ -471,7 +540,9 @@ class _TokenInfo:
             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)
+        tokens = self.cache.streets.get(street, _get_street)
+        if tokens:
+            self.data['street'] = tokens
 
 
     def add_place(self, conn, place):
@@ -479,10 +550,9 @@ class _TokenInfo:
         """
         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()
 
         self.data['place_search'], self.data['place_match'] = \
@@ -501,9 +571,12 @@ class _TokenInfo:
 
         tokens = {}
         for key, value in terms:
-            tokens[key] = self.cache.address_terms.get(value, _get_address_term)
+            items = self.cache.address_terms.get(value, _get_address_term)
+            if items[0] or items[1]:
+                tokens[key] = items
 
-        self.data['addr'] = tokens
+        if tokens:
+            self.data['addr'] = tokens
 
 
 class _LRU:
@@ -550,18 +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}
 
-        # Get postcodes that are already saved
-        postcodes = OrderedDict()
-        with conn.cursor() as cur:
-            cur.execute("""SELECT word FROM word
-                           WHERE class ='place' and type = 'postcode'""")
-            for row in cur:
-                postcodes[row[0]] = None
-        self.postcodes = _LRU(maxsize=32, init_data=postcodes)
+        # 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)