Coverage for src/wiktextract/page.py: 88%
307 statements
« prev ^ index » next coverage.py v7.16.2, created at 2026-10-07 00:55 +0000
1# Code for parsing information from a single Wiktionary page.
2#
3# Copyright (c) 2018-2022 Tatu Ylonen. See file LICENSE and https://ylonen.org
5import re
6from collections import defaultdict
7from copy import copy
8from typing import Any, Callable, Optional, Union
10from mediawiki_langcodes import name_to_code
11from wikitextprocessor.core import (
12 NamespaceDataEntry,
13 PostTemplateFnCallable,
14 TemplateArgs,
15 TemplateFnCallable,
16)
17from wikitextprocessor.node_expand import NodeHandlerFnCallable
18from wikitextprocessor.parser import GeneralNode, NodeKind, WikiNode
20from .clean import clean_value, remove_invisible_markup
21from .datautils import data_append, data_extend
22from .import_utils import import_extractor_module
23from .wxr_context import WiktextractContext
25# NodeKind values for subtitles
26LEVEL_KINDS = {
27 NodeKind.LEVEL2,
28 NodeKind.LEVEL3,
29 NodeKind.LEVEL4,
30 NodeKind.LEVEL5,
31 NodeKind.LEVEL6,
32}
35def parse_page(
36 wxr: WiktextractContext, page_title: str, page_text: str
37) -> list[dict[str, Any]]:
38 """Parses the text of a Wiktionary page and returns a list of
39 dictionaries, one for each word/part-of-speech defined on the page
40 for the languages specified by ``capture_language_codes`` (None means
41 all available languages). ``word`` is page title, and ``text`` is
42 page text in Wikimedia format. Other arguments indicate what is
43 captured."""
44 page_extractor_mod = import_extractor_module(wxr.wtp.lang_code, "page")
45 page_data = page_extractor_mod.parse_page(wxr, page_title, page_text)
46 if wxr.config.extract_thesaurus_pages:
47 inject_linkages(wxr, page_data)
48 if wxr.config.dump_file_lang_code == "en":
49 process_categories(wxr, page_data)
50 remove_duplicate_data(page_data)
51 return page_data
54def is_panel_template(wxr: WiktextractContext, template_name: str) -> bool:
55 """Checks if `Template_name` is a known panel template name (i.e., one that
56 produces an infobox in Wiktionary, but this also recognizes certain other
57 templates that we do not wish to expand)."""
58 page_extractor_mod = import_extractor_module(wxr.wtp.lang_code, "page")
59 if (
60 hasattr(page_extractor_mod, "PANEL_TEMPLATES")
61 and template_name in page_extractor_mod.PANEL_TEMPLATES
62 ):
63 return True
64 if hasattr(
65 page_extractor_mod, "PANEL_PREFIXES"
66 ) and template_name.startswith(tuple(page_extractor_mod.PANEL_PREFIXES)):
67 return True
68 return False
71def recursively_extract(
72 contents: Union[WikiNode, str, list[Union[str, WikiNode]]],
73 fn: Callable[[Union[WikiNode, list[WikiNode]]], bool],
74) -> tuple[list[Union[str, WikiNode]], list[Union[str, WikiNode]]]:
75 """Recursively extracts elements from contents for which ``fn`` returns
76 True. This returns two lists, the extracted elements and the remaining
77 content (with the extracted elements removed at each level). Only
78 WikiNode objects can be extracted."""
79 # If contents is a list, process each element separately
80 extracted = []
81 new_contents = []
82 if isinstance(contents, (list, tuple)):
83 for x in contents:
84 e1, c1 = recursively_extract(x, fn)
85 extracted.extend(e1)
86 new_contents.extend(c1)
87 return extracted, new_contents
88 # If content is not WikiNode, just return it as new contents.
89 if not isinstance(contents, WikiNode):
90 return [], [contents]
91 # Check if this content should be extracted
92 if fn(contents):
93 return [contents], []
94 # Otherwise content is WikiNode, and we must recurse into it.
95 kind = contents.kind
96 new_node = copy(contents)
97 new_node.children = []
98 new_node.sarg = ""
99 new_node.largs = []
100 new_node.attrs = {}
101 new_contents.append(new_node)
102 if kind in LEVEL_KINDS or kind == NodeKind.LINK:
103 # Process args and children
104 new_args = []
105 for arg in contents.largs:
106 e1, c1 = recursively_extract(arg, fn)
107 new_args.append(c1)
108 extracted.extend(e1)
109 new_node.largs = new_args
110 e1, c1 = recursively_extract(contents.children, fn)
111 extracted.extend(e1)
112 new_node.children = c1
113 elif kind in {
114 NodeKind.ITALIC,
115 NodeKind.BOLD,
116 NodeKind.TABLE,
117 NodeKind.TABLE_CAPTION,
118 NodeKind.TABLE_ROW,
119 NodeKind.TABLE_HEADER_CELL,
120 NodeKind.TABLE_CELL,
121 NodeKind.PRE,
122 NodeKind.PREFORMATTED,
123 }:
124 # Process only children
125 e1, c1 = recursively_extract(contents.children, fn)
126 extracted.extend(e1)
127 new_node.children = c1
128 elif kind in (NodeKind.HLINE,): 128 ↛ 130line 128 didn't jump to line 130 because the condition on line 128 was never true
129 # No arguments or children
130 pass
131 elif kind in (NodeKind.LIST, NodeKind.LIST_ITEM):
132 # Keep args as-is, process children
133 new_node.sarg = contents.sarg
134 e1, c1 = recursively_extract(contents.children, fn)
135 extracted.extend(e1)
136 new_node.children = c1
137 elif kind in {
138 NodeKind.TEMPLATE,
139 NodeKind.TEMPLATE_ARG,
140 NodeKind.PARSER_FN,
141 NodeKind.URL,
142 }:
143 # Process only args
144 new_args = []
145 for arg in contents.largs:
146 e1, c1 = recursively_extract(arg, fn)
147 new_args.append(c1)
148 extracted.extend(e1)
149 new_node.largs = new_args
150 elif kind == NodeKind.HTML: 150 ↛ 158line 150 didn't jump to line 158 because the condition on line 150 was always true
151 # Keep attrs and args as-is, process children
152 new_node.attrs = contents.attrs
153 new_node.sarg = contents.sarg
154 e1, c1 = recursively_extract(contents.children, fn)
155 extracted.extend(e1)
156 new_node.children = c1
157 else:
158 raise RuntimeError(f"recursively_extract: unhandled kind {kind}")
159 return extracted, new_contents
162def inject_linkages(wxr: WiktextractContext, page_data: list[dict]) -> None:
163 # Inject linkages from thesaurus entries
164 from .thesaurus import search_thesaurus
166 local_thesaurus_ns = wxr.wtp.NAMESPACE_DATA.get("Thesaurus", {}).get("name") # type: ignore[call-overload]
167 for data in page_data:
168 if "pos" not in data: 168 ↛ 169line 168 didn't jump to line 169 because the condition on line 168 was never true
169 continue
170 word = data["word"]
171 lang_code = data["lang_code"]
172 pos = data["pos"]
173 for term in search_thesaurus(
174 wxr.thesaurus_db_conn, # type:ignore[arg-type]
175 word,
176 lang_code,
177 pos, # type: ignore[arg-type]
178 ):
179 for dt in data.get(term.linkage, ()):
180 if dt.get("word") == term.term and ( 180 ↛ 183line 180 didn't jump to line 183 because the condition on line 180 was never true
181 not term.sense or dt.get("sense") == term.sense
182 ):
183 break
184 else:
185 dt = {
186 "word": term.term,
187 "source": f"{local_thesaurus_ns}:{word}",
188 }
189 if len(term.sense) > 0: 189 ↛ 190line 189 didn't jump to line 190 because the condition on line 189 was never true
190 dt["sense"] = term.sense
191 if len(term.tags) > 0: 191 ↛ 192line 191 didn't jump to line 192 because the condition on line 191 was never true
192 dt["tags"] = term.tags
193 if len(term.raw_tags) > 0: 193 ↛ 194line 193 didn't jump to line 194 because the condition on line 193 was never true
194 dt["raw_tags"] = term.raw_tags
195 if len(term.topics) > 0: 195 ↛ 196line 195 didn't jump to line 196 because the condition on line 195 was never true
196 dt["topics"] = term.topics
197 if len(term.roman) > 0: 197 ↛ 198line 197 didn't jump to line 198 because the condition on line 197 was never true
198 dt["roman"] = term.roman
199 data_append(data, term.linkage, dt)
202def process_categories(
203 wxr: WiktextractContext, page_data: list[dict[str, Any]]
204) -> None:
205 # Categories are not otherwise disambiguated, but if there is only
206 # one sense and only one data in ret for the same language, move
207 # categories to the only sense. Note that categories are commonly
208 # specified for the page, and thus if we have multiple data in
209 # ret, we don't know which one they belong to (not even which
210 # language necessarily?).
211 # XXX can Category links be specified globally (i.e., in a different
212 # language?)
213 by_lang = defaultdict(list)
214 for data in page_data:
215 by_lang[data["lang"]].append(data)
216 for la, lst in by_lang.items():
217 if len(lst) > 1:
218 # Propagate categories from the last entry for the language to
219 # its other entries. It is common for them to only be specified
220 # in the last part-of-speech.
221 last = lst[-1]
222 for field in ("categories",):
223 if field not in last:
224 continue
225 vals = last[field]
226 for data in lst[:-1]:
227 assert data is not last
228 assert data.get(field) is not vals
229 if data.get("alt_of") or data.get("form_of"): 229 ↛ 230line 229 didn't jump to line 230 because the condition on line 229 was never true
230 continue # Don't add to alt-of/form-of entries
231 data_extend(data, field, vals)
232 continue
233 if len(lst) != 1: 233 ↛ 234line 233 didn't jump to line 234 because the condition on line 233 was never true
234 continue
235 data = lst[0]
236 senses = data.get("senses") or []
237 if len(senses) != 1:
238 continue
239 # Only one sense for this language. Move categories and certain other
240 # data to sense.
241 for field in ("categories", "topics", "wikidata", "wikipedia"):
242 if field in data:
243 v = data[field]
244 del data[field]
245 data_extend(senses[0], field, v)
247 # If the last part-of-speech of the last language (i.e., last item in "ret")
248 # has categories or topics not bound to a sense, propagate those
249 # categories and topics to all datas on "ret". It is common for categories
250 # to be specified at the end of an article. Apparently these can also
251 # apply to different languages.
252 if len(page_data) > 1:
253 last = page_data[-1]
254 for field in ("categories",):
255 if field not in last:
256 continue
257 lst = last[field]
258 for data in page_data[:-1]:
259 if data.get("form_of") or data.get("alt_of"): 259 ↛ 260line 259 didn't jump to line 260 because the condition on line 259 was never true
260 continue # Don't add to form_of or alt_of entries
261 data_extend(data, field, lst)
263 # Remove category links that start with a language name from entries for
264 # different languages
265 rhymes_ns_prefix = (
266 wxr.wtp.NAMESPACE_DATA.get("Rhymes", {}).get("name", "") + ":" # type: ignore[call-overload]
267 )
268 for data in page_data:
269 lang_code = data.get("lang_code")
270 cats = data.get("categories", [])
271 new_cats = []
272 for cat in cats:
273 no_prefix_cat = cat.removeprefix(rhymes_ns_prefix)
274 cat_lang = no_prefix_cat.split(maxsplit=1)[0].split(
275 "/", maxsplit=1
276 )[0]
277 cat_lang_code = name_to_code(cat_lang, "en")
278 if (
279 cat_lang_code != ""
280 and cat_lang_code != lang_code
281 and not (lang_code == "mul" and cat_lang_code == "en")
282 ):
283 continue
284 new_cats.append(cat)
285 if len(new_cats) == 0:
286 if "categories" in data:
287 del data["categories"]
288 else:
289 data["categories"] = new_cats
292def remove_duplicate_data(page_data: dict) -> None:
293 # Remove duplicates from tags, categories, etc.
294 for data in page_data:
295 for field in ("categories", "topics", "tags", "wikidata", "wikipedia"):
296 if field in data:
297 data[field] = sorted(set(data[field]))
298 for sense in data.get("senses", ()):
299 if field in sense:
300 sense[field] = sorted(set(sense[field]))
302 # If raw_glosses is identical to glosses, remove it
303 # If "empty-gloss" in tags and there are glosses, remove the tag
304 for data in page_data:
305 for s in data.get("senses", []):
306 rglosses = s.get("raw_glosses", ())
307 if not rglosses:
308 continue
309 sglosses = s.get("glosses", ())
310 if sglosses: 310 ↛ 314line 310 didn't jump to line 314 because the condition on line 310 was always true
311 tags = s.get("tags", ())
312 while "empty-gloss" in s.get("tags", ()): 312 ↛ 313line 312 didn't jump to line 313 because the condition on line 312 was never true
313 tags.remove("empty-gloss")
314 if len(rglosses) != len(sglosses):
315 continue
316 same = True
317 for rg, sg in zip(rglosses, sglosses):
318 if rg != sg:
319 same = False
320 break
321 if same:
322 del s["raw_glosses"]
325def clean_node(
326 wxr: WiktextractContext,
327 sense_data: Optional[Any],
328 wikinode: GeneralNode,
329 template_fn: Optional[TemplateFnCallable] = None,
330 post_template_fn: Optional[PostTemplateFnCallable] = None,
331 node_handler_fn: Optional[NodeHandlerFnCallable] = None,
332 collect_links: bool = False,
333 remove_anchors_from_links: bool = False,
334 no_strip=False,
335 no_html_strip=False,
336 link_collector: list[tuple[str, str]] | None = None,
337) -> str:
338 """
339 Expands node or nodes to text, cleaning up HTML tags and duplicate spaces.
341 If `sense_data` is a dictionary, expanded category links will be added to
342 it under the `categories` key. And if `collect_link` is `True`, expanded
343 links will be added to the `links` key. An optional `link_collector` also
344 receives expanded links, independently of `sense_data`. This lets callers
345 retain links alongside other text fields without changing category handling.
346 """
348 # print("CLEAN_NODE:", repr(value))
349 def clean_template_fn(name: str, ht: TemplateArgs) -> Optional[str]:
350 if template_fn is not None:
351 return template_fn(name, ht)
352 if is_panel_template(wxr, name):
353 return ""
354 return None
356 def clean_node_handler_fn_default(
357 node: WikiNode,
358 ) -> Optional[list[Union[str, WikiNode]]]:
359 assert isinstance(node, WikiNode)
360 kind = node.kind
361 if kind in {
362 NodeKind.TABLE_CELL,
363 NodeKind.TABLE_HEADER_CELL,
364 }:
365 return node.children
366 return None
368 if node_handler_fn is not None:
369 # override clean_node_handler_fn, the def above can't be accessed
370 clean_node_handler_fn = node_handler_fn
371 else:
372 clean_node_handler_fn = clean_node_handler_fn_default
374 # print("clean_node: value={!r}".format(value))
375 v = wxr.wtp.node_to_html(
376 wikinode,
377 node_handler_fn=clean_node_handler_fn,
378 template_fn=template_fn,
379 post_template_fn=post_template_fn,
380 )
381 # print("##########")
382 # print(f"{wikinode=}")
383 # print("clean_node: v={!r}".format(v))
385 # Capture categories if sense_data has been given. We also track
386 # Lua execution errors here.
387 # If collect_links=True (for glosses), capture links
388 category_ns_data: NamespaceDataEntry = wxr.wtp.NAMESPACE_DATA.get(
389 "Category",
390 {}, # type: ignore[typeddict-item]
391 )
392 category_ns_names: set[str] = {category_ns_data.get("name")} | set(
393 category_ns_data.get("aliases") # type:ignore[assignment,arg-type]
394 )
395 category_ns_names |= {"Category", "category"}
396 category_names_pattern = rf"(?:{'|'.join(category_ns_names)})"
397 if sense_data is not None:
398 # Check for Lua execution error
399 if '<strong class="error">Lua execution error' in v: 399 ↛ 400line 399 didn't jump to line 400 because the condition on line 399 was never true
400 data_append(sense_data, "tags", "error-lua-exec")
401 if '<strong class="error">Lua timeout error' in v: 401 ↛ 402line 401 didn't jump to line 402 because the condition on line 401 was never true
402 data_append(sense_data, "tags", "error-lua-timeout")
403 # Capture Category tags
404 if not collect_links:
405 for m in re.finditer(
406 rf"(?is)\[\[:?\s*{category_names_pattern}\s*:([^]|]+)",
407 v,
408 ):
409 cat = clean_value(wxr, m.group(1))
410 cat = re.sub(r"\s+", " ", cat)
411 cat = cat.strip()
412 if not cat: 412 ↛ 413line 412 didn't jump to line 413 because the condition on line 412 was never true
413 continue
414 if not sense_data_has_value(sense_data, "categories", cat):
415 data_append(sense_data, "categories", cat)
416 else:
417 links, categories = extract_links_from_node(
418 wxr,
419 v,
420 category_ns_names=category_ns_names,
421 remove_anchor_tags=remove_anchors_from_links,
422 )
423 for cat in categories:
424 # do not keep duplicated category links
425 if not sense_data_has_value(sense_data, "categories", cat): 425 ↛ 423line 425 didn't jump to line 423 because the condition on line 425 was always true
426 data_append(sense_data, "categories", cat)
427 # print(f"{links=}")
428 for ltuple in links:
429 # We want to keep link data as is, even duplicated
430 data_append(sense_data, "links", ltuple)
432 if link_collector is not None:
433 # Use the same visibility rules as the returned prose, so citations,
434 # tables and side panels cannot contribute unrelated destinations.
435 link_text = remove_invisible_markup(v)
436 captured_links, _ = extract_links_from_node(
437 wxr,
438 link_text,
439 category_ns_names=category_ns_names,
440 remove_anchor_tags=remove_anchors_from_links,
441 )
442 # An empty namespace/interwiki link (e.g. [[w:|language]]) can
443 # otherwise fall back to a bare "w:" target in the string parser.
444 link_collector.extend(
445 (label, target)
446 for label, target in captured_links
447 if not (target.endswith(":") and re.fullmatch(r"[^:]+:", target))
448 )
450 v = clean_value(wxr, v, no_strip=no_strip, no_html_strip=no_html_strip)
451 # print("After clean_value:", repr(v))
453 # Strip any unhandled templates and other stuff. This is mostly intended
454 # to clean up erroneous codings in the original text.
455 # v = re.sub(r"(?s)\{\{.*", "", v)
456 # Some templates create <sup>(Category: ...)</sup>; remove
457 v = re.sub(
458 rf"(?si)\s*(?:<sup>)?\({category_names_pattern}:[^)]+\)(?:</sup>)?",
459 "",
460 v,
461 )
462 # Some templates create question mark in <sup>, e.g.,
463 # some Korean Hanja form
464 v = re.sub(r"\^\?", "", v)
465 return v
468def sense_data_has_value(
469 sense_data: dict[str, Any], name: str, value: Any
470) -> bool:
471 """
472 Return True if `sense_data` has value in the attribute `name`'s value or
473 in the value of key `name` if `sense_date` is dictionary.
474 """
475 if hasattr(sense_data, name):
476 return value in getattr(sense_data, name)
477 elif isinstance(sense_data, dict): 477 ↛ 479line 477 didn't jump to line 479 because the condition on line 477 was always true
478 return value in sense_data.get(name, ()) # type:ignore[operator]
479 return False
482def extract_links_from_node(
483 wxr: WiktextractContext,
484 nodes: WikiNode | list[WikiNode | str] | str,
485 category_ns_names: set[str] | None = None,
486 remove_anchor_tags=False,
487 expand_nodes=False,
488) -> tuple[list[tuple[str, str]], set[str]]:
489 """Find link nodes and extract them as a list of tuples. If
490 `category_ns_names` is passed, also extract category names separately and
491 return them as a list of strings."""
492 ret: list[tuple[str, str]] = []
493 cat_ret: set[str] = set()
495 # Sometimes this function may receive nodes that have not been expanded,
496 # and like a head template node. Expanding the involves turning the nodes
497 # into wikitext and then parsing and expanding them, so it's expensive.
498 if expand_nodes is True:
499 nodes = wxr.wtp.parse(wxr.wtp.node_to_wikitext(nodes), expand_all=True)
500 if not isinstance(nodes, list): 500 ↛ 502line 500 didn't jump to line 502 because the condition on line 500 was always true
501 nodes = [nodes]
502 for node in nodes:
503 # print(f"{node=}")
504 if isinstance(node, str) and node.strip():
505 for m in re.finditer(
506 r"(?is)\[\[:?(\s*([^][|:]+):)?\s*([^]|]+)(\|([^]|]+))?\]\]",
507 # 1 2 3 4 5
508 node,
509 ):
510 if (
511 m.group(2)
512 and category_ns_names is not None
513 and m.group(2).strip() in category_ns_names
514 ):
515 cat = clean_value(wxr, m.group(3))
516 cat = re.sub(r"\s+", " ", cat).strip()
517 if not cat: 517 ↛ 518line 517 didn't jump to line 518 because the condition on line 517 was never true
518 continue
519 cat_ret.add(cat)
520 elif not m.group(1):
521 if m.group(5):
522 ltext = clean_value(wxr, m.group(5))
523 ltarget = clean_value(wxr, m.group(3))
524 elif not m.group(3): 524 ↛ 525line 524 didn't jump to line 525 because the condition on line 524 was never true
525 continue
526 else:
527 txt = clean_value(wxr, m.group(3))
528 ltext = txt
529 ltarget = txt
530 ltarget = re.sub(r"\s+", " ", ltarget).strip()
531 ltext = re.sub(r"\s+", " ", ltext).strip()
532 if not ltext and not ltarget: 532 ↛ 533line 532 didn't jump to line 533 because the condition on line 532 was never true
533 continue
534 if not ltext and ltarget: 534 ↛ 535line 534 didn't jump to line 535 because the condition on line 534 was never true
535 ltext = ltarget
536 ret.append((ltext, ltarget))
537 if not isinstance(node, WikiNode):
538 continue
539 for link_node in node.find_child_recursively(NodeKind.LINK):
540 if len(link_node.largs) > 0: 540 ↛ 539line 540 didn't jump to line 539 because the condition on line 540 was always true
541 ltarget = clean_node(wxr, None, link_node.largs[0]).strip()
542 if not ltarget: 542 ↛ 543line 542 didn't jump to line 543 because the condition on line 542 was never true
543 continue
544 ltext = clean_node(wxr, None, link_node)
545 ret.append((ltext, ltarget))
546 # XXX extract category links
547 if category_ns_names is not None:
548 new_ret: list[tuple[str, str]] = []
549 for ltext, ltarget in ret:
550 if ltext.strip() or not ltarget.strip():
551 new_ret.append((ltext, ltarget))
552 continue
553 m2 = re.match(r"([^:]+):.+", ltarget)
554 if m2 is not None and m2.group(1).strip() in category_ns_names: 554 ↛ 557line 554 didn't jump to line 557 because the condition on line 554 was always true
555 cat_ret.add(ltarget[ltarget.index(":") + 1 :])
556 else:
557 new_ret.append((ltext, ltarget))
558 ret = new_ret
559 if remove_anchor_tags is True:
560 new_ret = []
561 for ltext, ltarget in ret:
562 if "#" in ltarget and not ltarget.startswith("#"):
563 ltarget = ltarget[: ltarget.index("#")]
564 new_ret.append((ltext, ltarget))
565 ret = new_ret
566 return ret, cat_ret