1 # SPDX-License-Identifier: GPL-3.0-or-later
3 # This file is part of Nominatim. (https://nominatim.org)
5 # Copyright (C) 2023 by the Nominatim developer community.
6 # For a full list of authors see the git log.
8 Datastructures for a tokenized query.
10 from typing import List, Tuple, Optional, NamedTuple, Iterator
11 from abc import ABC, abstractmethod
15 class BreakType(enum.Enum):
16 """ Type of break between tokens.
19 """ Begin of the query. """
21 """ End of the query. """
23 """ Break between two phrases. """
25 """ Break between words. """
27 """ Break inside a word, for example a hyphen or apostrophe. """
29 """ Break created as a result of tokenization.
30 This may happen in languages without spaces between words.
34 class TokenType(enum.Enum):
38 """ Full name of a place. """
40 """ Word term without breaks, does not necessarily represent a full name. """
41 HOUSENUMBER = enum.auto()
42 """ Housenumber term. """
43 POSTCODE = enum.auto()
44 """ Postal code term. """
46 """ Country name or reference. """
47 QUALIFIER = enum.auto()
48 """ Special term used together with name (e.g. _Hotel_ Bellevue). """
49 CATEGORY = enum.auto()
50 """ Special term used as searchable object(e.g. supermarket in ...). """
53 class PhraseType(enum.Enum):
54 """ Designation of a phrase.
57 """ No specific designation (i.e. source is free-form query). """
59 """ Contains name or type of a POI. """
61 """ Contains a street name optionally with a housenumber. """
63 """ Contains the postal city. """
65 """ Contains the equivalent of a county. """
67 """ Contains a state or province. """
68 POSTCODE = enum.auto()
69 """ Contains a postal code. """
71 """ Contains the country name or code. """
73 def compatible_with(self, ttype: TokenType) -> bool:
74 """ Check if the given token type can be used with the phrase type.
76 if self == PhraseType.NONE:
78 if self == PhraseType.AMENITY:
79 return ttype in (TokenType.WORD, TokenType.PARTIAL,
80 TokenType.QUALIFIER, TokenType.CATEGORY)
81 if self == PhraseType.STREET:
82 return ttype in (TokenType.WORD, TokenType.PARTIAL, TokenType.HOUSENUMBER)
83 if self == PhraseType.POSTCODE:
84 return ttype == TokenType.POSTCODE
85 if self == PhraseType.COUNTRY:
86 return ttype == TokenType.COUNTRY
88 return ttype in (TokenType.WORD, TokenType.PARTIAL)
91 @dataclasses.dataclass
93 """ Base type for tokens.
94 Specific query analyzers must implement the concrete token class.
105 def get_category(self) -> Tuple[str, str]:
106 """ Return the category restriction for qualifier terms and
111 class TokenRange(NamedTuple):
112 """ Indexes of query nodes over which a token spans.
117 def replace_start(self, new_start: int) -> 'TokenRange':
118 """ Return a new token range with the new start.
120 return TokenRange(new_start, self.end)
123 def replace_end(self, new_end: int) -> 'TokenRange':
124 """ Return a new token range with the new end.
126 return TokenRange(self.start, new_end)
129 def split(self, index: int) -> Tuple['TokenRange', 'TokenRange']:
130 """ Split the span into two spans at the given index.
131 The index must be within the span.
133 return self.replace_end(index), self.replace_start(index)
136 @dataclasses.dataclass
138 """ List of all tokens of a given type going from one breakpoint to another.
145 def add_penalty(self, penalty: float) -> None:
146 """ Add the given penalty to all tokens in the list.
148 for token in self.tokens:
149 token.penalty += penalty
152 @dataclasses.dataclass
154 """ A node of the querry representing a break between terms.
158 starting: List[TokenList] = dataclasses.field(default_factory=list)
160 def has_tokens(self, end: int, *ttypes: TokenType) -> bool:
161 """ Check if there are tokens of the given types ending at the
164 return any(tl.end == end and tl.ttype in ttypes for tl in self.starting)
167 def get_tokens(self, end: int, ttype: TokenType) -> Optional[List[Token]]:
168 """ Get the list of tokens of the given type starting at this node
169 and ending at the node 'end'. Returns 'None' if no such
172 for tlist in self.starting:
173 if tlist.end == end and tlist.ttype == ttype:
178 @dataclasses.dataclass
180 """ A normalized query part. Phrases may be typed which means that
181 they then represent a specific part of the address.
188 """ A tokenized search query together with the normalized source
189 from which the tokens have been parsed.
191 The query contains a list of nodes that represent the breaks
192 between words. Tokens span between nodes, which don't necessarily
193 need to be direct neighbours. Thus the query is represented as a
194 directed acyclic graph.
196 When created, a query contains a single node: the start of the
197 query. Further nodes can be added by appending to 'nodes'.
200 def __init__(self, source: List[Phrase]) -> None:
202 self.nodes: List[QueryNode] = \
203 [QueryNode(BreakType.START, source[0].ptype if source else PhraseType.NONE)]
206 def num_token_slots(self) -> int:
207 """ Return the length of the query in vertice steps.
209 return len(self.nodes) - 1
212 def add_node(self, btype: BreakType, ptype: PhraseType) -> None:
213 """ Append a new break node with the given break type.
214 The phrase type denotes the type for any tokens starting
217 self.nodes.append(QueryNode(btype, ptype))
220 def add_token(self, trange: TokenRange, ttype: TokenType, token: Token) -> None:
221 """ Add a token to the query. 'start' and 'end' are the indexes of the
222 nodes from which to which the token spans. The indexes must exist
223 and are expected to be in the same phrase.
224 'ttype' denotes the type of the token and 'token' the token to
227 If the token type is not compatible with the phrase it should
228 be added to, then the token is silently dropped.
230 snode = self.nodes[trange.start]
231 if snode.ptype.compatible_with(ttype):
232 tlist = snode.get_tokens(trange.end, ttype)
234 snode.starting.append(TokenList(trange.end, ttype, [token]))
239 def get_tokens(self, trange: TokenRange, ttype: TokenType) -> List[Token]:
240 """ Get the list of tokens of a given type, spanning the given
241 nodes. The nodes must exist. If no tokens exist, an
242 empty list is returned.
244 return self.nodes[trange.start].get_tokens(trange.end, ttype) or []
247 def get_partials_list(self, trange: TokenRange) -> List[Token]:
248 """ Create a list of partial tokens between the given nodes.
249 The list is composed of the first token of type PARTIAL
250 going to the subsequent node. Such PARTIAL tokens are
253 return [next(iter(self.get_tokens(TokenRange(i, i+1), TokenType.PARTIAL)))
254 for i in range(trange.start, trange.end)]
257 def iter_token_lists(self) -> Iterator[Tuple[int, QueryNode, TokenList]]:
258 """ Iterator over all token lists in the query.
260 for i, node in enumerate(self.nodes):
261 for tlist in node.starting:
265 def find_lookup_word_by_id(self, token: int) -> str:
266 """ Find the first token with the given token ID and return
267 its lookup word. Returns 'None' if no such token exists.
268 The function is very slow and must only be used for
271 for node in self.nodes:
272 for tlist in node.starting:
273 for t in tlist.tokens:
275 return f"[{tlist.ttype.name[0]}]{t.lookup_word}"