]> git.openstreetmap.org Git - nominatim.git/blobdiff - nominatim/tokenizer/icu_tokenizer.py
move warm script to python code
[nominatim.git] / nominatim / tokenizer / icu_tokenizer.py
index 7b820c9da04173b4e585b816c828b597d775f87b..39c1cbc648ca33fe4ecce4ec77421db1de96b1c2 100644 (file)
@@ -8,42 +8,48 @@
 Tokenizer implementing normalisation as used before Nominatim 4 but using
 libICU instead of the PostgreSQL module.
 """
+from typing import Optional, Sequence, List, Tuple, Mapping, Any, cast, \
+                   Dict, Set, Iterable
 import itertools
 import json
 import logging
-import re
+from pathlib import Path
 from textwrap import dedent
 
-from nominatim.db.connection import connect
+from nominatim.db.connection import connect, Connection, Cursor
+from nominatim.config import Configuration
 from nominatim.db.utils import CopyBuffer
 from nominatim.db.sql_preprocessor import SQLPreprocessor
-from nominatim.indexer.place_info import PlaceInfo
+from nominatim.data.place_info import PlaceInfo
 from nominatim.tokenizer.icu_rule_loader import ICURuleLoader
+from nominatim.tokenizer.place_sanitizer import PlaceSanitizer
+from nominatim.data.place_name import PlaceName
+from nominatim.tokenizer.icu_token_analysis import ICUTokenAnalysis
 from nominatim.tokenizer.base import AbstractAnalyzer, AbstractTokenizer
 
 DBCFG_TERM_NORMALIZATION = "tokenizer_term_normalization"
 
 LOG = logging.getLogger()
 
-def create(dsn, data_dir):
+def create(dsn: str, data_dir: Path) -> 'ICUTokenizer':
     """ Create a new instance of the tokenizer provided by this module.
     """
-    return LegacyICUTokenizer(dsn, data_dir)
+    return ICUTokenizer(dsn, data_dir)
 
 
-class LegacyICUTokenizer(AbstractTokenizer):
-    """ This tokenizer uses libICU to covert names and queries to ASCII.
+class ICUTokenizer(AbstractTokenizer):
+    """ This tokenizer uses libICU to convert names and queries to ASCII.
         Otherwise it uses the same algorithms and data structures as the
         normalization routines in Nominatim 3.
     """
 
-    def __init__(self, dsn, data_dir):
+    def __init__(self, dsn: str, data_dir: Path) -> None:
         self.dsn = dsn
         self.data_dir = data_dir
-        self.loader = None
+        self.loader: Optional[ICURuleLoader] = None
 
 
-    def init_new_db(self, config, init_db=True):
+    def init_new_db(self, config: Configuration, init_db: bool = True) -> None:
         """ Set up a new tokenizer for the database.
 
             This copies all necessary data in the project directory to make
@@ -51,7 +57,7 @@ class LegacyICUTokenizer(AbstractTokenizer):
         """
         self.loader = ICURuleLoader(config)
 
-        self._install_php(config.lib_dir.php)
+        self._install_php(config.lib_dir.php, overwrite=True)
         self._save_config()
 
         if init_db:
@@ -59,7 +65,7 @@ class LegacyICUTokenizer(AbstractTokenizer):
             self._init_db_tables(config)
 
 
-    def init_from_project(self, config):
+    def init_from_project(self, config: Configuration) -> None:
         """ Initialise the tokenizer from the project directory.
         """
         self.loader = ICURuleLoader(config)
@@ -67,8 +73,10 @@ class LegacyICUTokenizer(AbstractTokenizer):
         with connect(self.dsn) as conn:
             self.loader.load_config_from_db(conn)
 
+        self._install_php(config.lib_dir.php, overwrite=False)
 
-    def finalize_import(self, config):
+
+    def finalize_import(self, config: Configuration) -> None:
         """ Do any required postprocessing to make the tokenizer data ready
             for use.
         """
@@ -77,7 +85,7 @@ class LegacyICUTokenizer(AbstractTokenizer):
             sqlp.run_sql_file(conn, 'tokenizer/legacy_tokenizer_indices.sql')
 
 
-    def update_sql_functions(self, config):
+    def update_sql_functions(self, config: Configuration) -> None:
         """ Reimport the SQL functions for this tokenizer.
         """
         with connect(self.dsn) as conn:
@@ -85,14 +93,14 @@ class LegacyICUTokenizer(AbstractTokenizer):
             sqlp.run_sql_file(conn, 'tokenizer/icu_tokenizer.sql')
 
 
-    def check_database(self, config):
+    def check_database(self, config: Configuration) -> None:
         """ Check that the tokenizer is set up correctly.
         """
         # Will throw an error if there is an issue.
         self.init_from_project(config)
 
 
-    def update_statistics(self):
+    def update_statistics(self) -> None:
         """ Recompute frequencies for all name words.
         """
         with connect(self.dsn) as conn:
@@ -112,7 +120,50 @@ class LegacyICUTokenizer(AbstractTokenizer):
             conn.commit()
 
 
-    def name_analyzer(self):
+    def _cleanup_housenumbers(self) -> None:
+        """ Remove unused house numbers.
+        """
+        with connect(self.dsn) as conn:
+            if not conn.table_exists('search_name'):
+                return
+            with conn.cursor(name="hnr_counter") as cur:
+                cur.execute("""SELECT DISTINCT word_id, coalesce(info->>'lookup', word_token)
+                               FROM word
+                               WHERE type = 'H'
+                                 AND NOT EXISTS(SELECT * FROM search_name
+                                                WHERE ARRAY[word.word_id] && name_vector)
+                                 AND (char_length(coalesce(word, word_token)) > 6
+                                      OR coalesce(word, word_token) not similar to '\\d+')
+                            """)
+                candidates = {token: wid for wid, token in cur}
+            with conn.cursor(name="hnr_counter") as cur:
+                cur.execute("""SELECT housenumber FROM placex
+                               WHERE housenumber is not null
+                                     AND (char_length(housenumber) > 6
+                                          OR housenumber not similar to '\\d+')
+                            """)
+                for row in cur:
+                    for hnr in row[0].split(';'):
+                        candidates.pop(hnr, None)
+            LOG.info("There are %s outdated housenumbers.", len(candidates))
+            LOG.debug("Outdated housenumbers: %s", candidates.keys())
+            if candidates:
+                with conn.cursor() as cur:
+                    cur.execute("""DELETE FROM word WHERE word_id = any(%s)""",
+                                (list(candidates.values()), ))
+                conn.commit()
+
+
+
+    def update_word_tokens(self) -> None:
+        """ Remove unused tokens.
+        """
+        LOG.warning("Cleaning up housenumber tokens.")
+        self._cleanup_housenumbers()
+        LOG.warning("Tokenizer house-keeping done.")
+
+
+    def name_analyzer(self) -> 'ICUNameAnalyzer':
         """ Create a new analyzer for tokenizing names and queries
             using this tokinzer. Analyzers are context managers and should
             be used accordingly:
@@ -127,31 +178,48 @@ class LegacyICUTokenizer(AbstractTokenizer):
 
             Analyzers are not thread-safe. You need to instantiate one per thread.
         """
-        return LegacyICUNameAnalyzer(self.dsn, self.loader.make_sanitizer(),
-                                     self.loader.make_token_analysis())
+        assert self.loader is not None
+        return ICUNameAnalyzer(self.dsn, self.loader.make_sanitizer(),
+                               self.loader.make_token_analysis())
+
+
+    def most_frequent_words(self, conn: Connection, num: int) -> List[str]:
+        """ Return a list of the `num` most frequent full words
+            in the database.
+        """
+        with conn.cursor() as cur:
+            cur.execute("""SELECT word, sum((info->'count')::int) as count
+                             FROM word WHERE type = 'W'
+                             GROUP BY word
+                             ORDER BY count DESC LIMIT %s""", (num,))
+            return list(s[0].split('@')[0] for s in cur)
 
 
-    def _install_php(self, phpdir):
+    def _install_php(self, phpdir: Path, overwrite: bool = True) -> None:
         """ Install the php script for the tokenizer.
         """
+        assert self.loader is not None
         php_file = self.data_dir / "tokenizer.php"
-        php_file.write_text(dedent(f"""\
-            <?php
-            @define('CONST_Max_Word_Frequency', 10000000);
-            @define('CONST_Term_Normalization_Rules', "{self.loader.normalization_rules}");
-            @define('CONST_Transliteration', "{self.loader.get_search_rules()}");
-            require_once('{phpdir}/tokenizer/icu_tokenizer.php');"""))
 
+        if not php_file.exists() or overwrite:
+            php_file.write_text(dedent(f"""\
+                <?php
+                @define('CONST_Max_Word_Frequency', 10000000);
+                @define('CONST_Term_Normalization_Rules', "{self.loader.normalization_rules}");
+                @define('CONST_Transliteration', "{self.loader.get_search_rules()}");
+                require_once('{phpdir}/tokenizer/icu_tokenizer.php');"""), encoding='utf-8')
 
-    def _save_config(self):
+
+    def _save_config(self) -> None:
         """ Save the configuration that needs to remain stable for the given
             database as database properties.
         """
+        assert self.loader is not None
         with connect(self.dsn) as conn:
             self.loader.save_config_to_db(conn)
 
 
-    def _init_db_tables(self, config):
+    def _init_db_tables(self, config: Configuration) -> None:
         """ Set up the word table and fill it with pre-computed word
             frequencies.
         """
@@ -161,15 +229,16 @@ class LegacyICUTokenizer(AbstractTokenizer):
             conn.commit()
 
 
-class LegacyICUNameAnalyzer(AbstractAnalyzer):
-    """ The legacy analyzer uses the ICU library for splitting names.
+class ICUNameAnalyzer(AbstractAnalyzer):
+    """ The ICU analyzer uses the ICU library for splitting names.
 
         Each instance opens a connection to the database to request the
         normalization.
     """
 
-    def __init__(self, dsn, sanitizer, token_analysis):
-        self.conn = connect(dsn).connection
+    def __init__(self, dsn: str, sanitizer: PlaceSanitizer,
+                 token_analysis: ICUTokenAnalysis) -> None:
+        self.conn: Optional[Connection] = connect(dsn).connection
         self.conn.autocommit = True
         self.sanitizer = sanitizer
         self.token_analysis = token_analysis
@@ -177,7 +246,7 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
         self._cache = _TokenCache()
 
 
-    def close(self):
+    def close(self) -> None:
         """ Free all resources used by the analyzer.
         """
         if self.conn:
@@ -185,20 +254,20 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
             self.conn = None
 
 
-    def _search_normalized(self, name):
+    def _search_normalized(self, name: str) -> str:
         """ Return the search token transliteration of the given name.
         """
-        return self.token_analysis.search.transliterate(name).strip()
+        return cast(str, self.token_analysis.search.transliterate(name)).strip()
 
 
-    def _normalized(self, name):
+    def _normalized(self, name: str) -> str:
         """ Return the normalized version of the given name with all
             non-relevant information removed.
         """
-        return self.token_analysis.normalizer.transliterate(name).strip()
+        return cast(str, self.token_analysis.normalizer.transliterate(name)).strip()
 
 
-    def get_word_token_info(self, words):
+    def get_word_token_info(self, words: Sequence[str]) -> List[Tuple[str, str, int]]:
         """ 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.
@@ -209,6 +278,7 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
             The function is used for testing and debugging only
             and not necessarily efficient.
         """
+        assert self.conn is not None
         full_tokens = {}
         partial_tokens = {}
         for word in words:
@@ -231,8 +301,7 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
                + [(k, v, part_ids.get(v, None)) for k, v in partial_tokens.items()]
 
 
-    @staticmethod
-    def normalize_postcode(postcode):
+    def normalize_postcode(self, postcode: str) -> str:
         """ Convert the postcode to a standardized form.
 
             This function must yield exactly the same result as the SQL function
@@ -241,52 +310,91 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
         return postcode.strip().upper()
 
 
-    def _make_standard_hnr(self, hnr):
-        """ Create a normalised version of a housenumber.
-
-            This function takes minor shortcuts on transliteration.
-        """
-        return self._search_normalized(hnr)
-
-    def update_postcodes_from_db(self):
+    def update_postcodes_from_db(self) -> None:
         """ Update postcode tokens in the word table from the location_postcode
             table.
         """
-        to_delete = []
+        assert self.conn is not None
+        analyzer = self.token_analysis.analysis.get('@postcode')
+
         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 type = 'P') w
-                              ON pc = word) x
-                           WHERE pc is null or word is null""")
-
-            with CopyBuffer() as copystr:
-                for postcode, word in cur:
-                    if postcode is None:
-                        to_delete.append(word)
-                    else:
-                        copystr.add(self._search_normalized(postcode),
-                                    'P', postcode)
-
-                if to_delete:
-                    cur.execute("""DELETE FROM WORD
-                                   WHERE type ='P' and word = any(%s)
-                                """, (to_delete, ))
-
-                copystr.copy_out(cur, 'word',
-                                 columns=['word_token', 'type', 'word'])
-
-
-    def update_special_phrases(self, phrases, should_replace):
+            # First get all postcode names currently in the word table.
+            cur.execute("SELECT DISTINCT word FROM word WHERE type = 'P'")
+            word_entries = set((entry[0] for entry in cur))
+
+            # Then compute the required postcode names from the postcode table.
+            needed_entries = set()
+            cur.execute("SELECT country_code, postcode FROM location_postcode")
+            for cc, postcode in cur:
+                info = PlaceInfo({'country_code': cc,
+                                  'class': 'place', 'type': 'postcode',
+                                  'address': {'postcode': postcode}})
+                address = self.sanitizer.process_names(info)[1]
+                for place in address:
+                    if place.kind == 'postcode':
+                        if analyzer is None:
+                            postcode_name = place.name.strip().upper()
+                            variant_base = None
+                        else:
+                            postcode_name = analyzer.get_canonical_id(place)
+                            variant_base = place.get_attr("variant")
+
+                        if variant_base:
+                            needed_entries.add(f'{postcode_name}@{variant_base}')
+                        else:
+                            needed_entries.add(postcode_name)
+                        break
+
+        # Now update the word table.
+        self._delete_unused_postcode_words(word_entries - needed_entries)
+        self._add_missing_postcode_words(needed_entries - word_entries)
+
+    def _delete_unused_postcode_words(self, tokens: Iterable[str]) -> None:
+        assert self.conn is not None
+        if tokens:
+            with self.conn.cursor() as cur:
+                cur.execute("DELETE FROM word WHERE type = 'P' and word = any(%s)",
+                            (list(tokens), ))
+
+    def _add_missing_postcode_words(self, tokens: Iterable[str]) -> None:
+        assert self.conn is not None
+        if not tokens:
+            return
+
+        analyzer = self.token_analysis.analysis.get('@postcode')
+        terms = []
+
+        for postcode_name in tokens:
+            if '@' in postcode_name:
+                term, variant = postcode_name.split('@', 2)
+                term = self._search_normalized(term)
+                if analyzer is None:
+                    variants = [term]
+                else:
+                    variants = analyzer.compute_variants(variant)
+                    if term not in variants:
+                        variants.append(term)
+            else:
+                variants = [self._search_normalized(postcode_name)]
+            terms.append((postcode_name, variants))
+
+        if terms:
+            with self.conn.cursor() as cur:
+                cur.execute_values("""SELECT create_postcode_word(pc, var)
+                                      FROM (VALUES %s) AS v(pc, var)""",
+                                   terms)
+
+
+
+
+    def update_special_phrases(self, phrases: Iterable[Tuple[str, str, str, str]],
+                               should_replace: bool) -> None:
         """ Replace the search index for special phrases with the new phrases.
             If `should_replace` is True, then the previous set of will be
             completely replaced. Otherwise the phrases are added to the
             already existing ones.
         """
+        assert self.conn is not None
         norm_phrases = set(((self._normalized(p[0]), p[1], p[2], p[3])
                             for p in phrases))
 
@@ -309,7 +417,9 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
                  len(norm_phrases), added, deleted)
 
 
-    def _add_special_phrases(self, cursor, new_phrases, existing_phrases):
+    def _add_special_phrases(self, cursor: Cursor,
+                             new_phrases: Set[Tuple[str, str, str, str]],
+                             existing_phrases: Set[Tuple[str, str, str, str]]) -> int:
         """ Add all phrases to the database that are not yet there.
         """
         to_add = new_phrases - existing_phrases
@@ -330,9 +440,10 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
         return added
 
 
-    @staticmethod
-    def _remove_special_phrases(cursor, new_phrases, existing_phrases):
-        """ Remove all phrases from the databse that are no longer in the
+    def _remove_special_phrases(self, cursor: Cursor,
+                             new_phrases: Set[Tuple[str, str, str, str]],
+                             existing_phrases: Set[Tuple[str, str, str, str]]) -> int:
+        """ Remove all phrases from the database that are no longer in the
             new phrase list.
         """
         to_delete = existing_phrases - new_phrases
@@ -348,21 +459,24 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
         return len(to_delete)
 
 
-    def add_country_names(self, country_code, names):
-        """ Add names for the given country to the search index.
+    def add_country_names(self, country_code: str, names: Mapping[str, str]) -> None:
+        """ Add default names for the given country to the search index.
         """
         # Make sure any name preprocessing for country names applies.
         info = PlaceInfo({'name': names, 'country_code': country_code,
                           'rank_address': 4, 'class': 'boundary',
                           'type': 'administrative'})
         self._add_country_full_names(country_code,
-                                     self.sanitizer.process_names(info)[0])
+                                     self.sanitizer.process_names(info)[0],
+                                     internal=True)
 
 
-    def _add_country_full_names(self, country_code, names):
+    def _add_country_full_names(self, country_code: str, names: Sequence[PlaceName],
+                                internal: bool = False) -> None:
         """ Add names for the given country from an already sanitized
             name list.
         """
+        assert self.conn is not None
         word_tokens = set()
         for name in names:
             norm_name = self._search_normalized(name.name)
@@ -371,80 +485,127 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
 
         with self.conn.cursor() as cur:
             # Get existing names
-            cur.execute("""SELECT word_token FROM word
-                            WHERE type = 'C' and word = %s""",
+            cur.execute("""SELECT word_token, coalesce(info ? 'internal', false) as is_internal
+                             FROM word
+                             WHERE type = 'C' and word = %s""",
                         (country_code, ))
-            word_tokens.difference_update((t[0] for t in cur))
+            # internal/external names
+            existing_tokens: Dict[bool, Set[str]] = {True: set(), False: set()}
+            for word in cur:
+                existing_tokens[word[1]].add(word[0])
+
+            # Delete names that no longer exist.
+            gone_tokens = existing_tokens[internal] - word_tokens
+            if internal:
+                gone_tokens.update(existing_tokens[False] & word_tokens)
+            if gone_tokens:
+                cur.execute("""DELETE FROM word
+                               USING unnest(%s) as token
+                               WHERE type = 'C' and word = %s
+                                     and word_token = token""",
+                            (list(gone_tokens), country_code))
 
             # Only add those names that are not yet in the list.
-            if word_tokens:
-                cur.execute("""INSERT INTO word (word_token, type, word)
-                               (SELECT token, 'C', %s
-                                FROM unnest(%s) as token)
-                            """, (country_code, list(word_tokens)))
-
-            # No names are deleted at the moment.
-            # If deletion is made possible, then the static names from the
-            # initial 'country_name' table should be kept.
-
-
-    def process_place(self, place):
+            new_tokens = word_tokens - existing_tokens[True]
+            if not internal:
+                new_tokens -= existing_tokens[False]
+            if new_tokens:
+                if internal:
+                    sql = """INSERT INTO word (word_token, type, word, info)
+                               (SELECT token, 'C', %s, '{"internal": "yes"}'
+                                  FROM unnest(%s) as token)
+                           """
+                else:
+                    sql = """INSERT INTO word (word_token, type, word)
+                                   (SELECT token, 'C', %s
+                                    FROM unnest(%s) as token)
+                          """
+                cur.execute(sql, (country_code, list(new_tokens)))
+
+
+    def process_place(self, place: PlaceInfo) -> Mapping[str, Any]:
         """ Determine tokenizer information about the given place.
 
             Returns a JSON-serializable structure that will be handed into
             the database via the token_info field.
         """
-        token_info = _TokenInfo(self._cache)
+        token_info = _TokenInfo()
 
         names, address = self.sanitizer.process_names(place)
 
         if names:
-            fulls, partials = self._compute_name_tokens(names)
-
-            token_info.add_names(fulls, partials)
+            token_info.set_names(*self._compute_name_tokens(names))
 
             if place.is_country():
+                assert place.country_code is not None
                 self._add_country_full_names(place.country_code, names)
 
         if address:
             self._process_place_address(token_info, address)
 
-        return token_info.data
+        return token_info.to_dict()
 
 
-    def _process_place_address(self, token_info, address):
-        hnrs = []
-        addr_terms = []
-        streets = []
+    def _process_place_address(self, token_info: '_TokenInfo',
+                               address: Sequence[PlaceName]) -> None:
         for item in address:
             if item.kind == 'postcode':
-                self._add_postcode(item.name)
-            elif item.kind in ('housenumber', 'streetnumber', 'conscriptionnumber'):
-                hnrs.append(item.name)
+                token_info.set_postcode(self._add_postcode(item))
+            elif item.kind == 'housenumber':
+                token_info.add_housenumber(*self._compute_housenumber_token(item))
             elif item.kind == 'street':
-                streets.extend(self._retrieve_full_tokens(item.name))
+                token_info.add_street(self._retrieve_full_tokens(item.name))
             elif item.kind == 'place':
                 if not item.suffix:
                     token_info.add_place(self._compute_partial_tokens(item.name))
             elif not item.kind.startswith('_') and not item.suffix and \
-                 item.kind not in ('country', 'full'):
-                addr_terms.append((item.kind, self._compute_partial_tokens(item.name)))
-
-        if hnrs:
-            hnrs = self._split_housenumbers(hnrs)
-            token_info.add_housenumbers(self.conn, [self._make_standard_hnr(n) for n in hnrs])
-
-        if addr_terms:
-            token_info.add_address_terms(addr_terms)
+                 item.kind not in ('country', 'full', 'inclusion'):
+                token_info.add_address_term(item.kind, self._compute_partial_tokens(item.name))
 
-        if streets:
-            token_info.add_street(streets)
 
+    def _compute_housenumber_token(self, hnr: PlaceName) -> Tuple[Optional[int], Optional[str]]:
+        """ Normalize the housenumber and return the word token and the
+            canonical form.
+        """
+        assert self.conn is not None
+        analyzer = self.token_analysis.analysis.get('@housenumber')
+        result: Tuple[Optional[int], Optional[str]] = (None, None)
 
-    def _compute_partial_tokens(self, name):
+        if analyzer is None:
+            # When no custom analyzer is set, simply normalize and transliterate
+            norm_name = self._search_normalized(hnr.name)
+            if norm_name:
+                result = self._cache.housenumbers.get(norm_name, result)
+                if result[0] is None:
+                    with self.conn.cursor() as cur:
+                        hid = cur.scalar("SELECT getorcreate_hnr_id(%s)", (norm_name, ))
+
+                        result = hid, norm_name
+                        self._cache.housenumbers[norm_name] = result
+        else:
+            # Otherwise use the analyzer to determine the canonical name.
+            # Per convention we use the first variant as the 'lookup name', the
+            # name that gets saved in the housenumber field of the place.
+            word_id = analyzer.get_canonical_id(hnr)
+            if word_id:
+                result = self._cache.housenumbers.get(word_id, result)
+                if result[0] is None:
+                    variants = analyzer.compute_variants(word_id)
+                    if variants:
+                        with self.conn.cursor() as cur:
+                            hid = cur.scalar("SELECT create_analyzed_hnr_id(%s, %s)",
+                                             (word_id, list(variants)))
+                            result = hid, variants[0]
+                            self._cache.housenumbers[word_id] = result
+
+        return result
+
+
+    def _compute_partial_tokens(self, name: str) -> List[int]:
         """ Normalize the given term, split it into partial words and return
             then token list for them.
         """
+        assert self.conn is not None
         norm_name = self._search_normalized(name)
 
         tokens = []
@@ -463,16 +624,18 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
                             (need_lookup, ))
 
                 for partial, token in cur:
+                    assert token is not None
                     tokens.append(token)
                     self._cache.partials[partial] = token
 
         return tokens
 
 
-    def _retrieve_full_tokens(self, name):
+    def _retrieve_full_tokens(self, name: str) -> List[int]:
         """ Get the full name token for the given name, if it exists.
-            The name is only retrived for the standard analyser.
+            The name is only retrieved for the standard analyser.
         """
+        assert self.conn is not None
         norm_name = self._search_normalized(name)
 
         # return cached if possible
@@ -489,127 +652,163 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
         return full
 
 
-    def _compute_name_tokens(self, names):
+    def _compute_name_tokens(self, names: Sequence[PlaceName]) -> Tuple[Set[int], Set[int]]:
         """ Computes the full name and partial name tokens for the given
             dictionary of names.
         """
-        full_tokens = set()
-        partial_tokens = set()
+        assert self.conn is not None
+        full_tokens: Set[int] = set()
+        partial_tokens: Set[int] = set()
 
         for name in names:
             analyzer_id = name.get_attr('analyzer')
-            norm_name = self._normalized(name.name)
+            analyzer = self.token_analysis.get_analyzer(analyzer_id)
+            word_id = analyzer.get_canonical_id(name)
             if analyzer_id is None:
-                token_id = norm_name
+                token_id = word_id
             else:
-                token_id = f'{norm_name}@{analyzer_id}'
+                token_id = f'{word_id}@{analyzer_id}'
 
             full, part = self._cache.names.get(token_id, (None, None))
             if full is None:
-                variants = self.token_analysis.analysis[analyzer_id].get_variants_ascii(norm_name)
+                variants = analyzer.compute_variants(word_id)
                 if not variants:
                     continue
 
                 with self.conn.cursor() as cur:
-                    cur.execute("SELECT (getorcreate_full_word(%s, %s)).*",
+                    cur.execute("SELECT * FROM getorcreate_full_word(%s, %s)",
                                 (token_id, variants))
-                    full, part = cur.fetchone()
+                    full, part = cast(Tuple[int, List[int]], cur.fetchone())
 
                 self._cache.names[token_id] = (full, part)
 
+            assert part is not None
+
             full_tokens.add(full)
             partial_tokens.update(part)
 
         return full_tokens, partial_tokens
 
 
-    def _add_postcode(self, postcode):
+    def _add_postcode(self, item: PlaceName) -> Optional[str]:
         """ Make sure the normalized postcode is present in the word table.
         """
-        if re.search(r'[:,;]', postcode) is None:
-            postcode = self.normalize_postcode(postcode)
+        assert self.conn is not None
+        analyzer = self.token_analysis.analysis.get('@postcode')
 
-            if postcode not in self._cache.postcodes:
-                term = self._search_normalized(postcode)
-                if not term:
-                    return
+        if analyzer is None:
+            postcode_name = item.name.strip().upper()
+            variant_base = None
+        else:
+            postcode_name = analyzer.get_canonical_id(item)
+            variant_base = item.get_attr("variant")
 
-                with self.conn.cursor() as cur:
-                    # no word_id needed for postcodes
-                    cur.execute("""INSERT INTO word (word_token, type, word)
-                                   (SELECT %s, 'P', pc FROM (VALUES (%s)) as v(pc)
-                                    WHERE NOT EXISTS
-                                     (SELECT * FROM word
-                                      WHERE type = 'P' and word = pc))
-                                """, (term, postcode))
-                self._cache.postcodes.add(postcode)
-
-
-    @staticmethod
-    def _split_housenumbers(hnrs):
-        if len(hnrs) > 1 or ',' in hnrs[0] or ';' in hnrs[0]:
-            # split numbers if necessary
-            simple_list = []
-            for hnr in hnrs:
-                simple_list.extend((x.strip() for x in re.split(r'[;,]', hnr)))
-
-            if len(simple_list) > 1:
-                hnrs = list(set(simple_list))
-            else:
-                hnrs = simple_list
+        if variant_base:
+            postcode = f'{postcode_name}@{variant_base}'
+        else:
+            postcode = postcode_name
+
+        if postcode not in self._cache.postcodes:
+            term = self._search_normalized(postcode_name)
+            if not term:
+                return None
 
-        return hnrs
+            variants = {term}
+            if analyzer is not None and variant_base:
+                variants.update(analyzer.compute_variants(variant_base))
 
+            with self.conn.cursor() as cur:
+                cur.execute("SELECT create_postcode_word(%s, %s)",
+                            (postcode, list(variants)))
+            self._cache.postcodes.add(postcode)
 
+        return postcode_name
 
 
 class _TokenInfo:
     """ Collect token information to be sent back to the database.
     """
-    def __init__(self, cache):
-        self._cache = cache
-        self.data = {}
+    def __init__(self) -> None:
+        self.names: Optional[str] = None
+        self.housenumbers: Set[str] = set()
+        self.housenumber_tokens: Set[int] = set()
+        self.street_tokens: Optional[Set[int]] = None
+        self.place_tokens: Set[int] = set()
+        self.address_tokens: Dict[str, str] = {}
+        self.postcode: Optional[str] = None
+
+
+    def _mk_array(self, tokens: Iterable[Any]) -> str:
+        return f"{{{','.join((str(s) for s in tokens))}}}"
+
+
+    def to_dict(self) -> Dict[str, Any]:
+        """ Return the token information in database importable format.
+        """
+        out: Dict[str, Any] = {}
+
+        if self.names:
+            out['names'] = self.names
+
+        if self.housenumbers:
+            out['hnr'] = ';'.join(self.housenumbers)
+            out['hnr_tokens'] = self._mk_array(self.housenumber_tokens)
+
+        if self.street_tokens is not None:
+            out['street'] = self._mk_array(self.street_tokens)
+
+        if self.place_tokens:
+            out['place'] = self._mk_array(self.place_tokens)
 
-    @staticmethod
-    def _mk_array(tokens):
-        return '{%s}' % ','.join((str(s) for s in tokens))
+        if self.address_tokens:
+            out['addr'] = self.address_tokens
 
+        if self.postcode:
+            out['postcode'] = self.postcode
 
-    def add_names(self, fulls, partials):
+        return out
+
+
+    def set_names(self, fulls: Iterable[int], partials: Iterable[int]) -> None:
         """ Adds token information for the normalised names.
         """
-        self.data['names'] = self._mk_array(itertools.chain(fulls, partials))
+        self.names = self._mk_array(itertools.chain(fulls, partials))
 
 
-    def add_housenumbers(self, conn, hnrs):
+    def add_housenumber(self, token: Optional[int], hnr: Optional[str]) -> None:
         """ Extract housenumber information from a list of normalised
             housenumbers.
         """
-        self.data['hnr_tokens'] = self._mk_array(self._cache.get_hnr_tokens(conn, hnrs))
-        self.data['hnr'] = ';'.join(hnrs)
+        if token:
+            assert hnr is not None
+            self.housenumbers.add(hnr)
+            self.housenumber_tokens.add(token)
 
 
-    def add_street(self, tokens):
+    def add_street(self, tokens: Iterable[int]) -> None:
         """ Add addr:street match terms.
         """
-        self.data['street'] = self._mk_array(tokens)
+        if self.street_tokens is None:
+            self.street_tokens = set()
+        self.street_tokens.update(tokens)
 
 
-    def add_place(self, tokens):
+    def add_place(self, tokens: Iterable[int]) -> None:
         """ Add addr:place search and match terms.
         """
-        if tokens:
-            self.data['place'] = self._mk_array(tokens)
+        self.place_tokens.update(tokens)
 
 
-    def add_address_terms(self, terms):
+    def add_address_term(self, key: str, partials: Iterable[int]) -> None:
         """ Add additional address terms.
         """
-        tokens = {key: self._mk_array(partials)
-                  for key, partials in terms if partials}
+        if partials:
+            self.address_tokens[key] = self._mk_array(partials)
 
-        if tokens:
-            self.data['addr'] = tokens
+    def set_postcode(self, postcode: Optional[str]) -> None:
+        """ Set the postcode to the given one.
+        """
+        self.postcode = postcode
 
 
 class _TokenCache:
@@ -618,35 +817,9 @@ class _TokenCache:
         This cache is not thread-safe and needs to be instantiated per
         analyzer.
     """
-    def __init__(self):
-        self.names = {}
-        self.partials = {}
-        self.fulls = {}
-        self.postcodes = set()
-        self.housenumbers = {}
-
-
-    def get_hnr_tokens(self, conn, terms):
-        """ Get token ids for a list of housenumbers, looking them up in the
-            database if necessary. `terms` is an iterable of normalized
-            housenumbers.
-        """
-        tokens = []
-        askdb = []
-
-        for term in terms:
-            token = self.housenumbers.get(term)
-            if token is None:
-                askdb.append(term)
-            else:
-                tokens.append(token)
-
-        if askdb:
-            with conn.cursor() as cur:
-                cur.execute("SELECT nr, getorcreate_hnr_id(nr) FROM unnest(%s) as nr",
-                            (askdb, ))
-                for term, tid in cur:
-                    self.housenumbers[term] = tid
-                    tokens.append(tid)
-
-        return tokens
+    def __init__(self) -> None:
+        self.names: Dict[str, Tuple[int, List[int]]] = {}
+        self.partials: Dict[str, int] = {}
+        self.fulls: Dict[str, List[int]] = {}
+        self.postcodes: Set[str] = set()
+        self.housenumbers: Dict[str, Tuple[Optional[int], Optional[str]]] = {}