Shortcuts on this page
r m x toggle line displays
j k next/prev highlighted chunk
0 (zero) top of page
1 (one) first highlighted chunk
1# encoding: utf-8
2"""Use the HTMLParser library to parse HTML files that aren't too bad."""
3from __future__ import annotations
5# Use of this source code is governed by the MIT license.
6__license__ = "MIT"
8__all__ = [
9 "HTMLParserTreeBuilder",
10]
12from html.parser import HTMLParser
13import re
15from typing import (
16 Any,
17 Callable,
18 cast,
19 Dict,
20 Iterable,
21 List,
22 Optional,
23 TYPE_CHECKING,
24 Tuple,
25 Type,
26 Union,
27)
29from bs4.element import (
30 AttributeDict,
31 CData,
32 Comment,
33 Declaration,
34 Doctype,
35 ProcessingInstruction,
36)
37from bs4.dammit import EntitySubstitution, UnicodeDammit
39from bs4.builder import (
40 DetectsXMLParsedAsHTML,
41 HTML,
42 HTMLTreeBuilder,
43 STRICT,
44)
46from bs4.exceptions import ParserRejectedMarkup
48if TYPE_CHECKING:
49 from bs4 import BeautifulSoup
50 from bs4.element import NavigableString
51 from bs4._typing import (
52 _Encoding,
53 _Encodings,
54 _RawMarkup,
55 )
57HTMLPARSER = "html.parser"
59_DuplicateAttributeHandler = Callable[[Dict[str, str], str, str], None]
62class BeautifulSoupHTMLParser(HTMLParser, DetectsXMLParsedAsHTML):
63 #: Constant to handle duplicate attributes by ignoring later values
64 #: and keeping the earlier ones.
65 REPLACE: str = "replace"
67 #: Constant to handle duplicate attributes by replacing earlier values
68 #: with later ones.
69 IGNORE: str = "ignore"
71 """A subclass of the Python standard library's HTMLParser class, which
72 listens for HTMLParser events and translates them into calls
73 to Beautiful Soup's tree construction API.
75 :param on_duplicate_attribute: A strategy for what to do if a
76 tag includes the same attribute more than once. Accepted
77 values are: REPLACE (replace earlier values with later
78 ones, the default), IGNORE (keep the earliest value
79 encountered), or a callable. A callable must take three
80 arguments: the dictionary of attributes already processed,
81 the name of the duplicate attribute, and the most recent value
82 encountered.
83 """
85 def __init__(
86 self,
87 soup: BeautifulSoup,
88 *args: Any,
89 on_duplicate_attribute: Union[str, _DuplicateAttributeHandler] = REPLACE,
90 **kwargs: Any,
91 ):
92 self.soup = soup
93 self.on_duplicate_attribute = on_duplicate_attribute
94 self.attribute_dict_class = soup.builder.attribute_dict_class
95 HTMLParser.__init__(self, *args, **kwargs)
97 # Keep a list of empty-element tags that were encountered
98 # without an explicit closing tag. If we encounter a closing tag
99 # of this type, we'll associate it with one of those entries.
100 #
101 # This isn't a stack because we don't care about the
102 # order. It's a list of closing tags we've already handled and
103 # will ignore, assuming they ever show up.
104 self.already_closed_empty_element = []
106 self._initialize_xml_detector()
108 on_duplicate_attribute: Union[str, _DuplicateAttributeHandler]
109 already_closed_empty_element: List[str]
110 soup: BeautifulSoup
112 def error(self, message: str) -> None:
113 # NOTE: This method is required so long as Python 3.9 is
114 # supported. The corresponding code is removed from HTMLParser
115 # in 3.5, but not removed from ParserBase until 3.10.
116 # https://github.com/python/cpython/issues/76025
117 #
118 # The original implementation turned the error into a warning,
119 # but in every case I discovered, this made HTMLParser
120 # immediately crash with an error message that was less
121 # helpful than the warning. The new implementation makes it
122 # more clear that html.parser just can't parse this
123 # markup. The 3.10 implementation does the same, though it
124 # raises AssertionError rather than calling a method. (We
125 # catch this error and wrap it in a ParserRejectedMarkup.)
126 raise ParserRejectedMarkup(message)
128 def handle_startendtag(
129 self, tag: str, attrs: List[Tuple[str, Optional[str]]]
130 ) -> None:
131 """Handle an incoming empty-element tag.
133 html.parser only calls this method when the markup looks like
134 <tag/>.
135 """
136 # `handle_empty_element` tells handle_starttag not to close the tag
137 # just because its name matches a known empty-element tag. We
138 # know that this is an empty-element tag, and we want to call
139 # handle_endtag ourselves.
140 self.handle_starttag(tag, attrs, handle_empty_element=False)
141 self.handle_endtag(tag)
143 def handle_starttag(
144 self,
145 tag: str,
146 attrs: List[Tuple[str, Optional[str]]],
147 handle_empty_element: bool = True,
148 ) -> None:
149 """Handle an opening tag, e.g. '<tag>'
151 :param handle_empty_element: True if this tag is known to be
152 an empty-element tag (i.e. there is not expected to be any
153 closing tag).
154 """
155 # TODO: handle namespaces here?
156 attr_dict: AttributeDict = self.attribute_dict_class()
157 for key, value in attrs:
158 # Change None attribute values to the empty string
159 # for consistency with the other tree builders.
160 if value is None:
161 value = ""
162 if key in attr_dict:
163 # A single attribute shows up multiple times in this
164 # tag. How to handle it depends on the
165 # on_duplicate_attribute setting.
166 on_dupe = self.on_duplicate_attribute
167 if on_dupe == self.IGNORE:
168 pass
169 elif on_dupe in (None, self.REPLACE):
170 attr_dict[key] = value
171 else:
172 on_dupe = cast(_DuplicateAttributeHandler, on_dupe)
173 on_dupe(attr_dict, key, value)
174 else:
175 attr_dict[key] = value
176 # print("START", tag)
177 sourceline: Optional[int]
178 sourcepos: Optional[int]
179 if self.soup.builder.store_line_numbers:
180 sourceline, sourcepos = self.getpos()
181 else:
182 sourceline = sourcepos = None
183 tagObj = self.soup.handle_starttag(
184 tag, None, None, attr_dict, sourceline=sourceline, sourcepos=sourcepos
185 )
186 if tagObj is not None and tagObj.is_empty_element and handle_empty_element:
187 # Unlike other parsers, html.parser doesn't send separate end tag
188 # events for empty-element tags. (It's handled in
189 # handle_startendtag, but only if the original markup looked like
190 # <tag/>.)
191 #
192 # So we need to call handle_endtag() ourselves. Since we
193 # know the start event is identical to the end event, we
194 # don't want handle_endtag() to cross off any previous end
195 # events for tags of this name.
196 self.handle_endtag(tag, check_already_closed=False)
198 # But we might encounter an explicit closing tag for this tag
199 # later on. If so, we want to ignore it.
200 self.already_closed_empty_element.append(tag)
202 if self._root_tag_name is None:
203 self._root_tag_encountered(tag)
205 def handle_endtag(self, tag: str, check_already_closed: bool = True) -> None:
206 """Handle a closing tag, e.g. '</tag>'
208 :param tag: A tag name.
209 :param check_already_closed: True if this tag is expected to
210 be the closing portion of an empty-element tag,
211 e.g. '<tag></tag>'.
212 """
213 # print("END", tag)
214 if check_already_closed and tag in self.already_closed_empty_element:
215 # This is a redundant end tag for an empty-element tag.
216 # We've already called handle_endtag() for it, so just
217 # check it off the list.
218 # print("ALREADY CLOSED", tag)
219 self.already_closed_empty_element.remove(tag)
220 else:
221 self.soup.handle_endtag(tag)
223 def handle_data(self, data: str) -> None:
224 """Handle some textual data that shows up between tags."""
225 self.soup.handle_data(data)
227 _DECIMAL_REFERENCE_WITH_FOLLOWING_DATA = re.compile("^([0-9]+)(.*)")
228 _HEX_REFERENCE_WITH_FOLLOWING_DATA = re.compile("^([0-9a-f]+)(.*)")
230 @classmethod
231 def _dereference_numeric_character_reference(cls, name:str) -> Tuple[str, bool, str]:
232 """Convert a numeric character reference into an actual character.
234 :param name: The number of the character reference, as
235 obtained by html.parser
237 :return: A 3-tuple (dereferenced, replacement_added,
238 extra_data). `dereferenced` is the dereferenced character
239 reference, or the empty string if there was no
240 reference. `replacement_added` is True if the reference
241 could only be dereferenced by replacing content with U+FFFD
242 REPLACEMENT CHARACTER. `extra_data` is a portion of data
243 following the character reference, which was deemed to be
244 normal data and not part of the reference at all.
245 """
246 dereferenced:str = ""
247 replacement_added:bool = False
248 extra_data:str = ""
250 base:int = 10
251 reg = cls._DECIMAL_REFERENCE_WITH_FOLLOWING_DATA
252 if name.startswith("x") or name.startswith("X"):
253 # Hex reference
254 name = name[1:]
255 base = 16
256 reg = cls._HEX_REFERENCE_WITH_FOLLOWING_DATA
258 real_name:Optional[int] = None
259 try:
260 real_name = int(name, base)
261 except ValueError:
262 # This is either bad data that starts with what looks like
263 # a numeric character reference, or a real numeric
264 # reference that wasn't terminated by a semicolon.
265 #
266 # The fix to https://bugs.python.org/issue13633 made it
267 # our responsibility to handle the extra data.
268 #
269 # To preserve the old behavior, we extract the numeric
270 # portion of the incoming "reference" and treat that as a
271 # numeric reference. All subsequent data will be processed
272 # as string data.
273 match = reg.search(name)
274 if match is not None:
275 real_name = int(match.groups()[0], base)
276 extra_data = match.groups()[1]
278 if real_name is None:
279 dereferenced = ""
280 extra_data = name
281 else:
282 dereferenced, replacement_added = UnicodeDammit.numeric_character_reference(real_name)
283 return dereferenced, replacement_added, extra_data
285 def handle_charref(self, name: str) -> None:
286 """Handle a numeric character reference by converting it to the
287 corresponding Unicode character and treating it as textual
288 data.
290 :param name: Character number, possibly in hexadecimal.
291 """
292 dereferenced, replacement_added, extra_data = self._dereference_numeric_character_reference(name)
293 if replacement_added:
294 self.soup.contains_replacement_characters = True
295 if dereferenced is not None:
296 self.handle_data(dereferenced)
297 if extra_data is not None:
298 self.handle_data(extra_data)
300 def handle_entityref(self, name: str) -> None:
301 """Handle a named entity reference by converting it to the
302 corresponding Unicode character(s) and treating it as textual
303 data.
305 :param name: Name of the entity reference.
306 """
307 character = EntitySubstitution.HTML_ENTITY_TO_CHARACTER.get(name)
308 if character is not None:
309 data = character
310 else:
311 # If this were XML, it would be ambiguous whether "&foo"
312 # was an character entity reference with a missing
313 # semicolon or the literal string "&foo". Since this is
314 # HTML, we have a complete list of all character entity references,
315 # and this one wasn't found, so assume it's the literal string "&foo".
316 data = "&%s" % name
317 self.handle_data(data)
319 def handle_comment(self, data: str) -> None:
320 """Handle an HTML comment.
322 :param data: The text of the comment.
323 """
324 self.soup.endData()
325 self.soup.handle_data(data)
326 self.soup.endData(Comment)
328 def handle_decl(self, decl: str) -> None:
329 """Handle a DOCTYPE declaration.
331 :param data: The text of the declaration.
332 """
333 self.soup.endData()
334 decl = decl[len("DOCTYPE ") :]
335 self.soup.handle_data(decl)
336 self.soup.endData(Doctype)
338 def unknown_decl(self, data: str) -> None:
339 """Handle a declaration of unknown type -- probably a CDATA block.
341 :param data: The text of the declaration.
342 """
343 cls: Type[NavigableString]
344 if data.upper().startswith("CDATA["):
345 cls = CData
346 data = data[len("CDATA[") :]
347 else:
348 cls = Declaration
349 self.soup.endData()
350 self.soup.handle_data(data)
351 self.soup.endData(cls)
353 def handle_pi(self, data: str) -> None:
354 """Handle a processing instruction.
356 :param data: The text of the instruction.
357 """
358 self.soup.endData()
359 self.soup.handle_data(data)
360 self._document_might_be_xml(data)
361 self.soup.endData(ProcessingInstruction)
364class HTMLParserTreeBuilder(HTMLTreeBuilder):
365 """A Beautiful soup `bs4.builder.TreeBuilder` that uses the
366 :py:class:`html.parser.HTMLParser` parser, found in the Python
367 standard library.
369 """
371 is_xml: bool = False
372 picklable: bool = True
373 NAME: str = HTMLPARSER
374 features: Iterable[str] = [NAME, HTML, STRICT]
375 parser_args: Tuple[Iterable[Any], Dict[str, Any]]
377 #: The html.parser knows which line number and position in the
378 #: original file is the source of an element.
379 TRACKS_LINE_NUMBERS: bool = True
381 def __init__(
382 self,
383 parser_args: Optional[Iterable[Any]] = None,
384 parser_kwargs: Optional[Dict[str, Any]] = None,
385 **kwargs: Any,
386 ):
387 """Constructor.
389 :param parser_args: Positional arguments to pass into
390 the BeautifulSoupHTMLParser constructor, once it's
391 invoked.
392 :param parser_kwargs: Keyword arguments to pass into
393 the BeautifulSoupHTMLParser constructor, once it's
394 invoked.
395 :param kwargs: Keyword arguments for the superclass constructor.
396 """
397 # Some keyword arguments will be pulled out of kwargs and placed
398 # into parser_kwargs.
399 extra_parser_kwargs = dict()
400 for arg in ("on_duplicate_attribute",):
401 if arg in kwargs:
402 value = kwargs.pop(arg)
403 extra_parser_kwargs[arg] = value
404 super(HTMLParserTreeBuilder, self).__init__(**kwargs)
405 parser_args = parser_args or []
406 parser_kwargs = parser_kwargs or {}
407 parser_kwargs.update(extra_parser_kwargs)
408 parser_kwargs["convert_charrefs"] = False
409 self.parser_args = (parser_args, parser_kwargs)
411 def prepare_markup(
412 self,
413 markup: _RawMarkup,
414 user_specified_encoding: Optional[_Encoding] = None,
415 document_declared_encoding: Optional[_Encoding] = None,
416 exclude_encodings: Optional[_Encodings] = None,
417 ) -> Iterable[Tuple[str, Optional[_Encoding], Optional[_Encoding], bool]]:
418 """Run any preliminary steps necessary to make incoming markup
419 acceptable to the parser.
421 :param markup: Some markup -- probably a bytestring.
422 :param user_specified_encoding: The user asked to try this encoding.
423 :param document_declared_encoding: The markup itself claims to be
424 in this encoding.
425 :param exclude_encodings: The user asked _not_ to try any of
426 these encodings.
428 :yield: A series of 4-tuples: (markup, encoding, declared encoding,
429 has undergone character replacement)
431 Each 4-tuple represents a strategy for parsing the document.
432 This TreeBuilder uses Unicode, Dammit to convert the markup
433 into Unicode, so the ``markup`` element of the tuple will
434 always be a string.
435 """
436 if isinstance(markup, str):
437 # Parse Unicode as-is.
438 yield (markup, None, None, False)
439 return
441 # Ask UnicodeDammit to sniff the most likely encoding.
443 known_definite_encodings: List[_Encoding] = []
444 if user_specified_encoding:
445 # This was provided by the end-user; treat it as a known
446 # definite encoding per the algorithm laid out in the
447 # HTML5 spec. (See the EncodingDetector class for
448 # details.)
449 known_definite_encodings.append(user_specified_encoding)
451 user_encodings: List[_Encoding] = []
452 if document_declared_encoding:
453 # This was found in the document; treat it as a slightly
454 # lower-priority user encoding.
455 user_encodings.append(document_declared_encoding)
457 dammit = UnicodeDammit(
458 markup,
459 known_definite_encodings=known_definite_encodings,
460 user_encodings=user_encodings,
461 is_html=True,
462 exclude_encodings=exclude_encodings,
463 )
465 if dammit.unicode_markup is None:
466 # In every case I've seen, Unicode, Dammit is able to
467 # convert the markup into Unicode, even if it needs to use
468 # REPLACEMENT CHARACTER. But there is a code path that
469 # could result in unicode_markup being None, and
470 # HTMLParser can only parse Unicode, so here we handle
471 # that code path.
472 raise ParserRejectedMarkup(
473 "Could not convert input to Unicode, and html.parser will not accept bytestrings."
474 )
475 else:
476 yield (
477 dammit.unicode_markup,
478 dammit.original_encoding,
479 dammit.declared_html_encoding,
480 dammit.contains_replacement_characters,
481 )
483 def feed(self, markup: _RawMarkup, _parser_class:type[BeautifulSoupHTMLParser] =BeautifulSoupHTMLParser) -> None:
484 """
485 :param markup: The markup to feed into the parser.
486 :param _parser_class: An HTMLParser subclass to use. This is only intended for use in unit tests.
487 """
488 args, kwargs = self.parser_args
490 # HTMLParser.feed will only handle str, but
491 # BeautifulSoup.markup is allowed to be _RawMarkup, because
492 # it's set by the yield value of
493 # TreeBuilder.prepare_markup. Fortunately,
494 # HTMLParserTreeBuilder.prepare_markup always yields a str
495 # (UnicodeDammit.unicode_markup).
496 assert isinstance(markup, str)
498 # We know BeautifulSoup calls TreeBuilder.initialize_soup
499 # before calling feed(), so we can assume self.soup
500 # is set.
501 assert self.soup is not None
502 parser = _parser_class(self.soup, *args, **kwargs)
504 try:
505 parser.feed(markup)
506 parser.close()
507 except AssertionError as e:
508 # html.parser raises AssertionError in rare cases to
509 # indicate a fatal problem with the markup, especially
510 # when there's an error in the doctype declaration.
511 raise ParserRejectedMarkup(e)
512 parser.already_closed_empty_element = []