]> git.openstreetmap.org Git - nominatim.git/commitdiff
query analyzer for ICU tokenizer
authorSarah Hoffmann <lonvia@denofr.de>
Mon, 22 May 2023 06:46:19 +0000 (08:46 +0200)
committerSarah Hoffmann <lonvia@denofr.de>
Mon, 22 May 2023 06:46:19 +0000 (08:46 +0200)
.pylintrc
nominatim/api/logging.py
nominatim/api/search/icu_tokenizer.py [new file with mode: 0644]
nominatim/api/search/query.py
test/python/api/search/test_icu_query_analyzer.py [new file with mode: 0644]

index 5159c51aac5f73beb4c947b2775a7bdf512d5021..f2d3491f27d60b6ac61bd44865844f3f78a38aa0 100644 (file)
--- a/.pylintrc
+++ b/.pylintrc
@@ -13,6 +13,6 @@ ignored-classes=NominatimArgs,closing
 # 'too-many-ancestors' is triggered already by deriving from UserDict
 # 'not-context-manager' disabled because it causes false positives once
 #   typed Python is enabled. See also https://github.com/PyCQA/pylint/issues/5273
-disable=too-few-public-methods,duplicate-code,too-many-ancestors,bad-option-value,no-self-use,not-context-manager,use-dict-literal,chained-comparison
+disable=too-few-public-methods,duplicate-code,too-many-ancestors,bad-option-value,no-self-use,not-context-manager,use-dict-literal,chained-comparison,attribute-defined-outside-init
 
-good-names=i,x,y,m,t,fd,db,cc,x1,x2,y1,y2,pt,k,v
+good-names=i,j,x,y,m,t,fd,db,cc,x1,x2,y1,y2,pt,k,v
index 05598660254bd0a42fe04f66c446174e5ce4a515..fdff73beb078ad14f0e6ad6104ece434e93e387a 100644 (file)
@@ -7,7 +7,7 @@
 """
 Functions for specialised logging with HTML output.
 """
-from typing import Any, cast
+from typing import Any, Iterator, Optional, List, cast
 from contextvars import ContextVar
 import textwrap
 import io
@@ -56,6 +56,11 @@ class BaseLogger:
         """
 
 
+    def table_dump(self, heading: str, rows: Iterator[Optional[List[Any]]]) -> None:
+        """ Print the table generated by the generator function.
+        """
+
+
     def sql(self, conn: AsyncConnection, statement: 'sa.Executable') -> None:
         """ Print the SQL for the given statement.
         """
@@ -101,9 +106,28 @@ class HTMLLogger(BaseLogger):
 
 
     def var_dump(self, heading: str, var: Any) -> None:
+        if callable(var):
+            var = var()
+
         self._write(f'<h5>{heading}</h5>{self._python_var(var)}')
 
 
+    def table_dump(self, heading: str, rows: Iterator[Optional[List[Any]]]) -> None:
+        head = next(rows)
+        assert head
+        self._write(f'<table><thead><tr><th colspan="{len(head)}">{heading}</th></tr><tr>')
+        for cell in head:
+            self._write(f'<th>{cell}</th>')
+        self._write('</tr></thead><tbody>')
+        for row in rows:
+            if row is not None:
+                self._write('<tr>')
+                for cell in row:
+                    self._write(f'<td>{cell}</td>')
+                self._write('</tr>')
+        self._write('</tbody></table>')
+
+
     def sql(self, conn: AsyncConnection, statement: 'sa.Executable') -> None:
         sqlstr = self.format_sql(conn, statement)
         if CODE_HIGHLIGHT:
@@ -155,9 +179,33 @@ class TextLogger(BaseLogger):
 
 
     def var_dump(self, heading: str, var: Any) -> None:
+        if callable(var):
+            var = var()
+
         self._write(f'{heading}:\n  {self._python_var(var)}\n\n')
 
 
+    def table_dump(self, heading: str, rows: Iterator[Optional[List[Any]]]) -> None:
+        self._write(f'{heading}:\n')
+        data = [list(map(self._python_var, row)) if row else None for row in rows]
+        assert data[0] is not None
+        num_cols = len(data[0])
+
+        maxlens = [max(len(d[i]) for d in data if d) for i in range(num_cols)]
+        tablewidth = sum(maxlens) + 3 * num_cols + 1
+        row_format = '| ' +' | '.join(f'{{:<{l}}}' for l in maxlens) + ' |\n'
+        self._write('-'*tablewidth + '\n')
+        self._write(row_format.format(*data[0]))
+        self._write('-'*tablewidth + '\n')
+        for row in data[1:]:
+            if row:
+                self._write(row_format.format(*row))
+            else:
+                self._write('-'*tablewidth + '\n')
+        if data[-1]:
+            self._write('-'*tablewidth + '\n')
+
+
     def sql(self, conn: AsyncConnection, statement: 'sa.Executable') -> None:
         sqlstr = '\n| '.join(textwrap.wrap(self.format_sql(conn, statement), width=78))
         self._write(f"| {sqlstr}\n\n")
diff --git a/nominatim/api/search/icu_tokenizer.py b/nominatim/api/search/icu_tokenizer.py
new file mode 100644 (file)
index 0000000..14698a2
--- /dev/null
@@ -0,0 +1,294 @@
+# SPDX-License-Identifier: GPL-3.0-or-later
+#
+# This file is part of Nominatim. (https://nominatim.org)
+#
+# Copyright (C) 2023 by the Nominatim developer community.
+# For a full list of authors see the git log.
+"""
+Implementation of query analysis for the ICU tokenizer.
+"""
+from typing import Tuple, Dict, List, Optional, NamedTuple, Iterator, Any, cast
+from copy import copy
+from collections import defaultdict
+import dataclasses
+import difflib
+
+from icu import Transliterator
+
+import sqlalchemy as sa
+
+from nominatim.typing import SaRow
+from nominatim.api.connection import SearchConnection
+from nominatim.api.logging import log
+from nominatim.api.search import query as qmod
+
+# XXX: TODO
+class AbstractQueryAnalyzer:
+    pass
+
+
+DB_TO_TOKEN_TYPE = {
+    'W': qmod.TokenType.WORD,
+    'w': qmod.TokenType.PARTIAL,
+    'H': qmod.TokenType.HOUSENUMBER,
+    'P': qmod.TokenType.POSTCODE,
+    'C': qmod.TokenType.COUNTRY
+}
+
+
+class QueryPart(NamedTuple):
+    """ Normalized and transliterated form of a single term in the query.
+        When the term came out of a split during the transliteration,
+        the normalized string is the full word before transliteration.
+        The word number keeps track of the word before transliteration
+        and can be used to identify partial transliterated terms.
+    """
+    token: str
+    normalized: str
+    word_number: int
+
+
+QueryParts = List[QueryPart]
+WordDict = Dict[str, List[qmod.TokenRange]]
+
+def yield_words(terms: List[QueryPart], start: int) -> Iterator[Tuple[str, qmod.TokenRange]]:
+    """ Return all combinations of words in the terms list after the
+        given position.
+    """
+    total = len(terms)
+    for first in range(start, total):
+        word = terms[first].token
+        yield word, qmod.TokenRange(first, first + 1)
+        for last in range(first + 1, min(first + 20, total)):
+            word = ' '.join((word, terms[last].token))
+            yield word, qmod.TokenRange(first, last + 1)
+
+
+@dataclasses.dataclass
+class ICUToken(qmod.Token):
+    """ Specialised token for ICU tokenizer.
+    """
+    word_token: str
+    info: Optional[Dict[str, Any]]
+
+    def get_category(self) -> Tuple[str, str]:
+        assert self.info
+        return self.info.get('class', ''), self.info.get('type', '')
+
+
+    def rematch(self, norm: str) -> None:
+        """ Check how well the token matches the given normalized string
+            and add a penalty, if necessary.
+        """
+        if not self.lookup_word:
+            return
+
+        seq = difflib.SequenceMatcher(a=self.lookup_word, b=norm)
+        distance = 0
+        for tag, afrom, ato, bfrom, bto in seq.get_opcodes():
+            if tag == 'delete' and (afrom == 0 or ato == len(self.lookup_word)):
+                distance += 1
+            elif tag == 'replace':
+                distance += max((ato-afrom), (bto-bfrom))
+            elif tag != 'equal':
+                distance += abs((ato-afrom) - (bto-bfrom))
+        self.penalty += (distance/len(self.lookup_word))
+
+
+    @staticmethod
+    def from_db_row(row: SaRow) -> 'ICUToken':
+        """ Create a ICUToken from the row of the word table.
+        """
+        count = 1 if row.info is None else row.info.get('count', 1)
+
+        penalty = 0.0
+        if row.type == 'w':
+            penalty = 0.3
+        elif row.type == 'H':
+            penalty = sum(0.1 for c in row.word_token if c != ' ' and not c.isdigit())
+            if all(not c.isdigit() for c in row.word_token):
+                penalty += 0.2 * (len(row.word_token) - 1)
+
+        if row.info is None:
+            lookup_word = row.word
+        else:
+            lookup_word = row.info.get('lookup', row.word)
+        if lookup_word:
+            lookup_word = lookup_word.split('@', 1)[0]
+        else:
+            lookup_word = row.word_token
+
+        return ICUToken(penalty=penalty, token=row.word_id, count=count,
+                        lookup_word=lookup_word, is_indexed=True,
+                        word_token=row.word_token, info=row.info)
+
+
+
+class ICUQueryAnalyzer(AbstractQueryAnalyzer):
+    """ Converter for query strings into a tokenized query
+        using the tokens created by a ICU tokenizer.
+    """
+
+    def __init__(self, conn: SearchConnection) -> None:
+        self.conn = conn
+
+
+    async def setup(self) -> None:
+        """ Set up static data structures needed for the analysis.
+        """
+        rules = await self.conn.get_property('tokenizer_import_normalisation')
+        self.normalizer = Transliterator.createFromRules("normalization", rules)
+        rules = await self.conn.get_property('tokenizer_import_transliteration')
+        self.transliterator = Transliterator.createFromRules("transliteration", rules)
+
+        if 'word' not in self.conn.t.meta.tables:
+            sa.Table('word', self.conn.t.meta,
+                     sa.Column('word_id', sa.Integer),
+                     sa.Column('word_token', sa.Text, nullable=False),
+                     sa.Column('type', sa.Text, nullable=False),
+                     sa.Column('word', sa.Text),
+                     sa.Column('info', self.conn.t.types.Json))
+
+
+    async def analyze_query(self, phrases: List[qmod.Phrase]) -> qmod.QueryStruct:
+        """ Analyze the given list of phrases and return the
+            tokenized query.
+        """
+        log().section('Analyze query (using ICU tokenizer)')
+        normalized = list(filter(lambda p: p.text,
+                                 (qmod.Phrase(p.ptype, self.normalizer.transliterate(p.text))
+                                  for p in phrases)))
+        query = qmod.QueryStruct(normalized)
+        log().var_dump('Normalized query', query.source)
+        if not query.source:
+            return query
+
+        parts, words = self.split_query(query)
+        log().var_dump('Transliterated query', lambda: _dump_transliterated(query, parts))
+
+        for row in await self.lookup_in_db(list(words.keys())):
+            for trange in words[row.word_token]:
+                token = ICUToken.from_db_row(row)
+                if row.type == 'S':
+                    if row.info['op'] in ('in', 'near'):
+                        if trange.start == 0:
+                            query.add_token(trange, qmod.TokenType.CATEGORY, token)
+                    else:
+                        query.add_token(trange, qmod.TokenType.QUALIFIER, token)
+                        if trange.start == 0 or trange.end == query.num_token_slots():
+                            token = copy(token)
+                            token.penalty += 0.1 * (query.num_token_slots())
+                            query.add_token(trange, qmod.TokenType.CATEGORY, token)
+                else:
+                    query.add_token(trange, DB_TO_TOKEN_TYPE[row.type], token)
+
+        self.add_extra_tokens(query, parts)
+        self.rerank_tokens(query, parts)
+
+        log().table_dump('Word tokens', _dump_word_tokens(query))
+
+        return query
+
+
+    def split_query(self, query: qmod.QueryStruct) -> Tuple[QueryParts, WordDict]:
+        """ Transliterate the phrases and split them into tokens.
+
+            Returns the list of transliterated tokens together with their
+            normalized form and a dictionary of words for lookup together
+            with their position.
+        """
+        parts: QueryParts = []
+        phrase_start = 0
+        words = defaultdict(list)
+        wordnr = 0
+        for phrase in query.source:
+            query.nodes[-1].ptype = phrase.ptype
+            for word in phrase.text.split(' '):
+                trans = self.transliterator.transliterate(word)
+                if trans:
+                    for term in trans.split(' '):
+                        if term:
+                            parts.append(QueryPart(term, word, wordnr))
+                            query.add_node(qmod.BreakType.TOKEN, phrase.ptype)
+                    query.nodes[-1].btype = qmod.BreakType.WORD
+                wordnr += 1
+            query.nodes[-1].btype = qmod.BreakType.PHRASE
+
+            for word, wrange in yield_words(parts, phrase_start):
+                words[word].append(wrange)
+
+            phrase_start = len(parts)
+        query.nodes[-1].btype = qmod.BreakType.END
+
+        return parts, words
+
+
+    async def lookup_in_db(self, words: List[str]) -> 'sa.Result[Any]':
+        """ Return the token information from the database for the
+            given word tokens.
+        """
+        t = self.conn.t.meta.tables['word']
+        return await self.conn.execute(t.select().where(t.c.word_token.in_(words)))
+
+
+    def add_extra_tokens(self, query: qmod.QueryStruct, parts: QueryParts) -> None:
+        """ Add tokens to query that are not saved in the database.
+        """
+        for part, node, i in zip(parts, query.nodes, range(1000)):
+            if len(part.token) <= 4 and part[0].isdigit()\
+               and not node.has_tokens(i+1, qmod.TokenType.HOUSENUMBER):
+                query.add_token(qmod.TokenRange(i, i+1), qmod.TokenType.HOUSENUMBER,
+                                ICUToken(0.5, 0, 1, part.token, True, part.token, None))
+
+
+    def rerank_tokens(self, query: qmod.QueryStruct, parts: QueryParts) -> None:
+        """ Add penalties to tokens that depend on presence of other token.
+        """
+        for i, node, tlist in query.iter_token_lists():
+            if tlist.ttype == qmod.TokenType.POSTCODE:
+                for repl in node.starting:
+                    if repl.end == tlist.end and repl.ttype != qmod.TokenType.POSTCODE \
+                       and (repl.ttype != qmod.TokenType.HOUSENUMBER
+                            or len(tlist.tokens[0].lookup_word) > 4):
+                        repl.add_penalty(0.39)
+            elif tlist.ttype == qmod.TokenType.HOUSENUMBER:
+                if any(c.isdigit() for c in tlist.tokens[0].lookup_word):
+                    for repl in node.starting:
+                        if repl.end == tlist.end and repl.ttype != qmod.TokenType.HOUSENUMBER \
+                           and (repl.ttype != qmod.TokenType.HOUSENUMBER
+                                or len(tlist.tokens[0].lookup_word) <= 3):
+                            repl.add_penalty(0.5 - tlist.tokens[0].penalty)
+            elif tlist.ttype not in (qmod.TokenType.COUNTRY, qmod.TokenType.PARTIAL):
+                norm = parts[i].normalized
+                for j in range(i + 1, tlist.end):
+                    if parts[j - 1].word_number != parts[j].word_number:
+                        norm += '  ' + parts[j].normalized
+                for token in tlist.tokens:
+                    cast(ICUToken, token).rematch(norm)
+
+
+def _dump_transliterated(query: qmod.QueryStruct, parts: QueryParts) -> str:
+    out = query.nodes[0].btype.value
+    for node, part in zip(query.nodes[1:], parts):
+        out += part.token + node.btype.value
+    return out
+
+
+def _dump_word_tokens(query: qmod.QueryStruct) -> Iterator[List[Any]]:
+    yield ['type', 'token', 'word_token', 'lookup_word', 'penalty', 'count', 'info']
+    for node in query.nodes:
+        for tlist in node.starting:
+            for token in tlist.tokens:
+                t = cast(ICUToken, token)
+                yield [tlist.ttype.name, t.token, t.word_token or '',
+                       t.lookup_word or '', t.penalty, t.count, t.info]
+
+
+async def create_query_analyzer(conn: SearchConnection) -> AbstractQueryAnalyzer:
+    """ Create and set up a new query analyzer for a database based
+        on the ICU tokenizer.
+    """
+    out = ICUQueryAnalyzer(conn)
+    await out.setup()
+
+    return out
index bc1f542d10148aeb79f3ee48240d5e4f7040dbc0..4e28d3658db30644d70f3b7cb91f28dd16561513 100644 (file)
@@ -7,7 +7,7 @@
 """
 Datastructures for a tokenized query.
 """
-from typing import List, Tuple, Optional, NamedTuple
+from typing import List, Tuple, Optional, NamedTuple, Iterator
 from abc import ABC, abstractmethod
 import dataclasses
 import enum
@@ -124,6 +124,13 @@ class TokenList:
     tokens: List[Token]
 
 
+    def add_penalty(self, penalty: float) -> None:
+        """ Add the given penalty to all tokens in the list.
+        """
+        for token in self.tokens:
+            token.penalty += penalty
+
+
 @dataclasses.dataclass
 class QueryNode:
     """ A node of the querry representing a break between terms.
@@ -226,6 +233,14 @@ class QueryStruct:
                           for i in range(trange.start, trange.end)]
 
 
+    def iter_token_lists(self) -> Iterator[Tuple[int, QueryNode, TokenList]]:
+        """ Iterator over all token lists in the query.
+        """
+        for i, node in enumerate(self.nodes):
+            for tlist in node.starting:
+                yield i, node, tlist
+
+
     def find_lookup_word_by_id(self, token: int) -> str:
         """ Find the first token with the given token ID and return
             its lookup word. Returns 'None' if no such token exists.
diff --git a/test/python/api/search/test_icu_query_analyzer.py b/test/python/api/search/test_icu_query_analyzer.py
new file mode 100644 (file)
index 0000000..78cd2c4
--- /dev/null
@@ -0,0 +1,186 @@
+# SPDX-License-Identifier: GPL-3.0-or-later
+#
+# This file is part of Nominatim. (https://nominatim.org)
+#
+# Copyright (C) 2023 by the Nominatim developer community.
+# For a full list of authors see the git log.
+"""
+Tests for query analyzer for ICU tokenizer.
+"""
+from pathlib import Path
+
+import pytest
+import pytest_asyncio
+
+from nominatim.api import NominatimAPIAsync
+from nominatim.api.search.query import Phrase, PhraseType, TokenType, BreakType
+import nominatim.api.search.icu_tokenizer as tok
+from nominatim.api.logging import set_log_output, get_and_disable
+
+async def add_word(conn, word_id, word_token, wtype, word, info = None):
+    t = conn.t.meta.tables['word']
+    await conn.execute(t.insert(), {'word_id': word_id,
+                                    'word_token': word_token,
+                                    'type': wtype,
+                                    'word': word,
+                                    'info': info})
+
+
+def make_phrase(query):
+    return [Phrase(PhraseType.NONE, s) for s in query.split(',')]
+
+@pytest_asyncio.fixture
+async def conn(table_factory):
+    """ Create an asynchronous SQLAlchemy engine for the test DB.
+    """
+    table_factory('nominatim_properties',
+                  definition='property TEXT, value TEXT',
+                  content=(('tokenizer_import_normalisation', ':: lower();'),
+                           ('tokenizer_import_transliteration', "'1' > '/1/'; 'ä' > 'ä '")))
+    table_factory('word',
+                  definition='word_id INT, word_token TEXT, type TEXT, word TEXT, info JSONB')
+
+    api = NominatimAPIAsync(Path('/invalid'), {})
+    async with api.begin() as conn:
+        yield conn
+    await api.close()
+
+
+@pytest.mark.asyncio
+async def test_empty_phrase(conn):
+    ana = await tok.create_query_analyzer(conn)
+
+    query = await ana.analyze_query([])
+
+    assert len(query.source) == 0
+    assert query.num_token_slots() == 0
+
+
+@pytest.mark.asyncio
+async def test_single_phrase_with_unknown_terms(conn):
+    ana = await tok.create_query_analyzer(conn)
+
+    await add_word(conn, 1, 'foo', 'w', 'FOO')
+
+    query = await ana.analyze_query(make_phrase('foo BAR'))
+
+    assert len(query.source) == 1
+    assert query.source[0].ptype == PhraseType.NONE
+    assert query.source[0].text == 'foo bar'
+
+    assert query.num_token_slots() == 2
+    assert len(query.nodes[0].starting) == 1
+    assert not query.nodes[1].starting
+
+
+@pytest.mark.asyncio
+async def test_multiple_phrases(conn):
+    ana = await tok.create_query_analyzer(conn)
+
+    await add_word(conn, 1, 'one', 'w', 'one')
+    await add_word(conn, 2, 'two', 'w', 'two')
+    await add_word(conn, 100, 'one two', 'W', 'one two')
+    await add_word(conn, 3, 'three', 'w', 'three')
+
+    query = await ana.analyze_query(make_phrase('one two,three'))
+
+    assert len(query.source) == 2
+
+
+@pytest.mark.asyncio
+async def test_splitting_in_transliteration(conn):
+    ana = await tok.create_query_analyzer(conn)
+
+    await add_word(conn, 1, 'mä', 'W', 'ma')
+    await add_word(conn, 2, 'fo', 'W', 'fo')
+
+    query = await ana.analyze_query(make_phrase('mäfo'))
+
+    assert query.num_token_slots() == 2
+    assert query.nodes[0].starting
+    assert query.nodes[1].starting
+    assert query.nodes[1].btype == BreakType.TOKEN
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize('term,order', [('23456', ['POSTCODE', 'HOUSENUMBER', 'WORD', 'PARTIAL']),
+                                        ('3', ['HOUSENUMBER', 'POSTCODE', 'WORD', 'PARTIAL'])
+                                       ])
+async def test_penalty_postcodes_and_housenumbers(conn, term, order):
+    ana = await tok.create_query_analyzer(conn)
+
+    await add_word(conn, 1, term, 'P', None)
+    await add_word(conn, 2, term, 'H', term)
+    await add_word(conn, 3, term, 'w', term)
+    await add_word(conn, 4, term, 'W', term)
+
+    query = await ana.analyze_query(make_phrase(term))
+
+    assert query.num_token_slots() == 1
+
+    torder = [(tl.tokens[0].penalty, tl.ttype) for tl in query.nodes[0].starting]
+    torder.sort()
+
+    assert [t[1] for t in torder] == [TokenType[o] for o in order]
+
+@pytest.mark.asyncio
+async def test_category_words_only_at_beginning(conn):
+    ana = await tok.create_query_analyzer(conn)
+
+    await add_word(conn, 1, 'foo', 'S', 'FOO', {'op': 'in'})
+    await add_word(conn, 2, 'bar', 'w', 'BAR')
+
+    query = await ana.analyze_query(make_phrase('foo BAR foo'))
+
+    assert query.num_token_slots() == 3
+    assert len(query.nodes[0].starting) == 1
+    assert query.nodes[0].starting[0].ttype == TokenType.CATEGORY
+    assert not query.nodes[2].starting
+
+
+@pytest.mark.asyncio
+async def test_qualifier_words(conn):
+    ana = await tok.create_query_analyzer(conn)
+
+    await add_word(conn, 1, 'foo', 'S', None, {'op': '-'})
+    await add_word(conn, 2, 'bar', 'w', None)
+
+    query = await ana.analyze_query(make_phrase('foo BAR foo BAR foo'))
+
+    assert query.num_token_slots() == 5
+    assert set(t.ttype for t in query.nodes[0].starting) == {TokenType.CATEGORY, TokenType.QUALIFIER}
+    assert set(t.ttype for t in query.nodes[2].starting) == {TokenType.QUALIFIER}
+    assert set(t.ttype for t in query.nodes[4].starting) == {TokenType.CATEGORY, TokenType.QUALIFIER}
+
+
+@pytest.mark.asyncio
+async def test_add_unknown_housenumbers(conn):
+    ana = await tok.create_query_analyzer(conn)
+
+    await add_word(conn, 1, '23', 'H', '23')
+
+    query = await ana.analyze_query(make_phrase('466 23 99834 34a'))
+
+    assert query.num_token_slots() == 4
+    assert query.nodes[0].starting[0].ttype == TokenType.HOUSENUMBER
+    assert len(query.nodes[0].starting[0].tokens) == 1
+    assert query.nodes[0].starting[0].tokens[0].token == 0
+    assert query.nodes[1].starting[0].ttype == TokenType.HOUSENUMBER
+    assert len(query.nodes[1].starting[0].tokens) == 1
+    assert query.nodes[1].starting[0].tokens[0].token == 1
+    assert not query.nodes[2].starting
+    assert not query.nodes[3].starting
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize('logtype', ['text', 'html'])
+async def test_log_output(conn, logtype):
+
+    ana = await tok.create_query_analyzer(conn)
+
+    await add_word(conn, 1, 'foo', 'w', 'FOO')
+
+    set_log_output(logtype)
+    await ana.analyze_query(make_phrase('foo'))
+
+    assert get_and_disable()