Find file
Fetching contributors…
Cannot retrieve contributors at this time
1712 lines (1479 sloc) 65.3 KB
"""Beautiful Soup
Elixir and Tonic
"The Screen-Scraper's Friend"
Beautiful Soup parses a (possibly invalid) XML or HTML document into a
tree representation. It provides methods and Pythonic idioms that make
it easy to navigate, search, and modify the tree.
A well-formed XML/HTML document yields a well-formed data
structure. An ill-formed XML/HTML document yields a correspondingly
ill-formed data structure. If your document is only locally
well-formed, you can use this library to find and process the
well-formed part of it.
Beautiful Soup works with Python 2.2 and up. It has no external
dependencies, but you'll have more success at converting data to UTF-8
if you also install these three packages:
* chardet, for auto-detecting character encodings
* cjkcodecs and iconv_codec, which add more encodings to the ones supported
by stock Python.
Beautiful Soup defines classes for two main parsing strategies:
* BeautifulStoneSoup, for parsing XML, SGML, or your domain-specific
language that kind of looks like XML.
* BeautifulSoup, for parsing run-of-the-mill HTML code, be it valid
or invalid. This class has web browser-like heuristics for
obtaining a sensible parse tree in the face of common HTML errors.
Beautiful Soup also defines a class (UnicodeDammit) for autodetecting
the encoding of an HTML or XML document, and converting it to
Unicode. Much of this code is taken from Mark Pilgrim's Universal Feed Parser.
For more than you ever wanted to know about Beautiful Soup, see the
from __future__ import generators
__author__ = "Leonard Richardson ("
__version__ = "3.0.0"
__date__ = "$Date: 2004/10/18 00:14:20 $"
__copyright__ = "Copyright (c) 2004-2005 Leonard Richardson"
__license__ = "PSF"
from sgmllib import SGMLParser, SGMLParseError
import codecs
import types
import re
import sgmllib
from htmlentitydefs import name2codepoint
#This code makes Beautiful Soup able to parse XML with namespaces
sgmllib.tagfind = re.compile('[a-zA-Z][-_.:a-zA-Z0-9]*')
# First, the classes that represent markup elements.
class PageElement:
"""Contains the navigational information for some part of the page
(either a tag or a piece of text)"""
def setup(self, parent=None, previous=None):
"""Sets up the initial relations between this element and
other elements."""
self.parent = parent
self.previous = previous = None
self.previousSibling = None
self.nextSibling = None
if self.parent and self.parent.contents:
self.previousSibling = self.parent.contents[-1]
self.previousSibling.nextSibling = self
def replaceWith(self, replaceWith):
oldParent = self.parent
myIndex = self.parent.contents.index(self)
if hasattr(replaceWith, 'parent') and replaceWith.parent == self.parent:
# We're replacing this element with one of its siblings.
index = self.parent.contents.index(replaceWith)
if index and index < myIndex:
# Furthermore, it comes before this element. That
# means that when we extract it, the index of this
# element will change.
myIndex = myIndex - 1
oldParent.insert(myIndex, replaceWith)
def extract(self):
"""Destructively rips this element out of the tree."""
if self.parent:
except ValueError:
#Find the two elements that would be next to each other if
#this element (and any children) hadn't been parsed. Connect
#the two.
lastChild = self._lastRecursiveChild()
nextElement =
if self.previous: = nextElement
if nextElement:
nextElement.previous = self.previous
self.previous = None = None
self.parent = None
if self.previousSibling:
self.previousSibling.nextSibling = self.nextSibling
if self.nextSibling:
self.nextSibling.previousSibling = self.previousSibling
self.previousSibling = self.nextSibling = None
def _lastRecursiveChild(self):
"Finds the last element beneath this object to be parsed."
lastChild = self
while hasattr(lastChild, 'contents') and lastChild.contents:
lastChild = lastChild.contents[-1]
return lastChild
def insert(self, position, newChild):
if (isinstance(newChild, basestring)
or isinstance(newChild, unicode)) \
and not isinstance(newChild, NavigableString):
newChild = NavigableString(newChild)
position = min(position, len(self.contents))
if hasattr(newChild, 'parent') and newChild.parent != None:
# We're 'inserting' an element that's already one
# of this object's children.
if newChild.parent == self:
index = self.find(newChild)
if index and index < position:
# Furthermore we're moving it further down the
# list of this object's children. That means that
# when we extract this element, our target index
# will jump down one.
position = position - 1
newChild.parent = self
previousChild = None
if position == 0:
newChild.previousSibling = None
newChild.previous = self
previousChild = self.contents[position-1]
newChild.previousSibling = previousChild
newChild.previousSibling.nextSibling = newChild
newChild.previous = previousChild._lastRecursiveChild()
if newChild.previous: = newChild
newChildsLastElement = newChild._lastRecursiveChild()
if position >= len(self.contents):
newChild.nextSibling = None
parent = self
parentsNextSibling = None
while not parentsNextSibling:
parentsNextSibling = parent.nextSibling
parent = parent.parent
if not parent: # This is the last element in the document.
if parentsNextSibling: = parentsNextSibling
else: = None
nextChild = self.contents[position]
newChild.nextSibling = nextChild
if newChild.nextSibling:
newChild.nextSibling.previousSibling = newChild = nextChild
if = newChildsLastElement
self.contents.insert(position, newChild)
def findNext(self, name=None, attrs={}, text=None, **kwargs):
"""Returns the first item that matches the given criteria and
appears after this Tag in the document."""
return self._findOne(self.findAllNext, name, attrs, text, **kwargs)
def findAllNext(self, name=None, attrs={}, text=None, limit=None,
"""Returns all items that match the given criteria and appear
before after Tag in the document."""
return self._findAll(name, attrs, text, limit, self.nextGenerator)
def findNextSibling(self, name=None, attrs={}, text=None, **kwargs):
"""Returns the closest sibling to this Tag that matches the
given criteria and appears after this Tag in the document."""
return self._findOne(self.findNextSiblings, name, attrs, text,
def findNextSiblings(self, name=None, attrs={}, text=None, limit=None,
"""Returns the siblings of this Tag that match the given
criteria and appear after this Tag in the document."""
return self._findAll(name, attrs, text, limit,
self.nextSiblingGenerator, **kwargs)
def findPrevious(self, name=None, attrs={}, text=None, **kwargs):
"""Returns the first item that matches the given criteria and
appears before this Tag in the document."""
return self._findOne(self.findAllPrevious, name, attrs, text, **kwargs)
def findAllPrevious(self, name=None, attrs={}, text=None, limit=None,
"""Returns all items that match the given criteria and appear
before this Tag in the document."""
return self._findAll(name, attrs, text, limit, self.previousGenerator,
def findPreviousSibling(self, name=None, attrs={}, text=None, **kwargs):
"""Returns the closest sibling to this Tag that matches the
given criteria and appears before this Tag in the document."""
return self._findOne(self.findPreviousSiblings, name, attrs, text,
def findPreviousSiblings(self, name=None, attrs={}, text=None,
limit=None, **kwargs):
"""Returns the siblings of this Tag that match the given
criteria and appear before this Tag in the document."""
return self._findAll(name, attrs, text, limit,
self.previousSiblingGenerator, **kwargs)
def findParent(self, name=None, attrs={}, **kwargs):
"""Returns the closest parent of this Tag that matches the given
# NOTE: We can't use _findOne because findParents takes a different
# set of arguments.
r = None
l = self.findParents(name, attrs, 1)
if l:
r = l[0]
return r
def findParents(self, name=None, attrs={}, limit=None, **kwargs):
"""Returns the parents of this Tag that match the given
return self._findAll(name, attrs, None, limit, self.parentGenerator,
#These methods do the real heavy lifting.
def _findOne(self, method, name, attrs, text, **kwargs):
r = None
l = method(name, attrs, text, 1, **kwargs)
if l:
r = l[0]
return r
def _findAll(self, name, attrs, text, limit, generator, **kwargs):
"Iterates over a generator looking for things that match."
if isinstance(name, SoupStrainer):
strainer = name
# Build a SoupStrainer
strainer = SoupStrainer(name, attrs, text, **kwargs)
results = ResultSet(strainer)
g = generator()
while True:
i =
except StopIteration:
if i:
found =
if found:
if limit and len(results) >= limit:
return results
#These Generators can be used to navigate starting from both
#NavigableStrings and Tags.
def nextGenerator(self):
i = self
while i:
i =
yield i
def nextSiblingGenerator(self):
i = self
while i:
i = i.nextSibling
yield i
def previousGenerator(self):
i = self
while i:
i = i.previous
yield i
def previousSiblingGenerator(self):
i = self
while i:
i = i.previousSibling
yield i
def parentGenerator(self):
i = self
while i:
i = i.parent
yield i
# Utility methods
def substituteEncoding(self, str, encoding=None):
encoding = encoding or "utf-8"
return str.replace("%SOUP-ENCODING%", encoding)
def toEncoding(self, s, encoding=None):
"""Encodes an object to a string in some encoding, or to Unicode.
if isinstance(s, unicode):
if encoding:
s = s.encode(encoding)
elif isinstance(s, str):
if encoding:
s = s.encode(encoding)
s = unicode(s)
if encoding:
s = self.toEncoding(str(s), encoding)
s = unicode(s)
return s
class NavigableString(unicode, PageElement):
def __getattr__(self, attr):
"""text.string gives you text. This is for backwards
compatibility for Navigable*String, but for CData* it lets you
get the string without the CData wrapper."""
if attr == 'string':
return self
raise AttributeError, "'%s' object has no attribute '%s'" % (self.__class__.__name__, attr)
def __unicode__(self):
return __str__(self, None)
def __str__(self, encoding=DEFAULT_OUTPUT_ENCODING):
if encoding:
return self.encode(encoding)
return self
class CData(NavigableString):
def __str__(self, encoding=DEFAULT_OUTPUT_ENCODING):
return "<![CDATA[%s]]>" % NavigableString.__str__(self, encoding)
class ProcessingInstruction(NavigableString):
def __str__(self, encoding=DEFAULT_OUTPUT_ENCODING):
output = self
if "%SOUP-ENCODING%" in output:
output = self.substituteEncoding(output, encoding)
return "<?%s?>" % self.toEncoding(output, encoding)
class Comment(NavigableString):
def __str__(self, encoding=DEFAULT_OUTPUT_ENCODING):
return "<!--%s-->" % NavigableString.__str__(self, encoding)
class Declaration(NavigableString):
def __str__(self, encoding=DEFAULT_OUTPUT_ENCODING):
return "<!%s>" % NavigableString.__str__(self, encoding)
class Tag(PageElement):
"""Represents a found HTML tag with its attributes and contents."""
def __init__(self, parser, name, attrs=None, parent=None,
"Basic constructor."
# We don't actually store the parser object: that lets extracted
# chunks be garbage-collected
self.parserClass = parser.__class__
self.isSelfClosing = parser.isSelfClosingTag(name) = name
if attrs == None:
attrs = []
self.attrs = attrs
self.contents = []
self.setup(parent, previous)
self.hidden = False
self.containsSubstitutions = False
def get(self, key, default=None):
"""Returns the value of the 'key' attribute for the tag, or
the value given for 'default' if it doesn't have that
return self._getAttrMap().get(key, default)
def has_key(self, key):
return self._getAttrMap().has_key(key)
def __getitem__(self, key):
"""tag[key] returns the value of the 'key' attribute for the tag,
and throws an exception if it's not there."""
return self._getAttrMap()[key]
def __iter__(self):
"Iterating over a tag iterates over its contents."
return iter(self.contents)
def __len__(self):
"The length of a tag is the length of its list of contents."
return len(self.contents)
def __contains__(self, x):
return x in self.contents
def __nonzero__(self):
"A tag is non-None even if it has no contents."
return True
def __setitem__(self, key, value):
"""Setting tag[key] sets the value of the 'key' attribute for the
self.attrMap[key] = value
found = False
for i in range(0, len(self.attrs)):
if self.attrs[i][0] == key:
self.attrs[i] = (key, value)
found = True
if not found:
self.attrs.append((key, value))
self._getAttrMap()[key] = value
def __delitem__(self, key):
"Deleting tag[key] deletes all 'key' attributes for the tag."
for item in self.attrs:
if item[0] == key:
#We don't break because bad HTML can define the same
#attribute multiple times.
if self.attrMap.has_key(key):
del self.attrMap[key]
def __call__(self, *args, **kwargs):
"""Calling a tag like a function is the same as calling its
findAll() method. Eg. tag('a') returns a list of all the A tags
found within this tag."""
return apply(self.findAll, args, kwargs)
def __getattr__(self, tag):
#print "Getattr %s.%s" % (self.__class__, tag)
if len(tag) > 3 and tag.rfind('Tag') == len(tag)-3:
return self.find(tag[:-3])
elif tag.find('__') != 0:
return self.find(tag)
def __eq__(self, other):
"""Returns true iff this tag has the same name, the same attributes,
and the same contents (recursively) as the given tag.
NOTE: right now this will return false if two tags have the
same attributes in a different order. Should this be fixed?"""
if not hasattr(other, 'name') or not hasattr(other, 'attrs') or not hasattr(other, 'contents') or != or self.attrs != other.attrs or len(self) != len(other):
return False
for i in range(0, len(self.contents)):
if self.contents[i] != other.contents[i]:
return False
return True
def __ne__(self, other):
"""Returns true iff this tag is not identical to the other tag,
as defined in __eq__."""
return not self == other
def __repr__(self, encoding=DEFAULT_OUTPUT_ENCODING):
"""Renders this tag as a string."""
return self.__str__(encoding)
def __unicode__(self):
return self.__str__(None)
def __str__(self, encoding=DEFAULT_OUTPUT_ENCODING,
prettyPrint=False, indentLevel=0):
"""Returns a string or Unicode representation of this tag and
its contents. To get Unicode, pass None for encoding.
NOTE: since Python's HTML parser consumes whitespace, this
method is not certain to reproduce the whitespace present in
the original string."""
encodedName = self.toEncoding(, encoding)
attrs = []
if self.attrs:
for key, val in self.attrs:
fmt = '%s="%s"'
if isString(val):
if self.containsSubstitutions and '%SOUP-ENCODING%' in val:
val = self.substituteEncoding(val, encoding)
if '"' in val:
fmt = "%s='%s'"
# This can't happen naturally, but it can happen
# if you modify an attribute value and print it out.
if "'" in val:
val = val.replace("'", "&squot;")
attrs.append(fmt % (self.toEncoding(key, encoding),
self.toEncoding(val, encoding)))
close = ''
closeTag = ''
if self.isSelfClosing:
close = ' /'
closeTag = '</%s>' % encodedName
indentTag, indentContents = 0, 0
if prettyPrint:
indentTag = indentLevel
space = (' ' * (indentTag-1))
indentContents = indentTag + 1
contents = self.renderContents(encoding, prettyPrint, indentContents)
if self.hidden:
s = contents
s = []
attributeString = ''
if attrs:
attributeString = ' ' + ' '.join(attrs)
if prettyPrint:
s.append('<%s%s%s>' % (encodedName, attributeString, close))
if prettyPrint:
if prettyPrint and contents and contents[-1] != "\n":
if prettyPrint and closeTag:
if prettyPrint and closeTag and self.nextSibling:
s = ''.join(s)
return s
def prettify(self, encoding=DEFAULT_OUTPUT_ENCODING):
return self.__str__(encoding, True)
def renderContents(self, encoding=DEFAULT_OUTPUT_ENCODING,
prettyPrint=False, indentLevel=0):
"""Renders the contents of this tag as a string in the given
encoding. If encoding is None, returns a Unicode string.."""
for c in self:
text = None
if isinstance(c, NavigableString):
text = c.__str__(encoding)
elif isinstance(c, Tag):
s.append(c.__str__(encoding, prettyPrint, indentLevel))
if text and prettyPrint:
text = text.strip()
if text:
if prettyPrint:
s.append(" " * (indentLevel-1))
if prettyPrint:
return ''.join(s)
#Soup methods
def find(self, name=None, attrs={}, recursive=True, text=None,
"""Return only the first child of this
Tag matching the given criteria."""
r = None
l = self.findAll(name, attrs, recursive, text, 1)
if l:
r = l[0]
return r
findChild = find
def findAll(self, name=None, attrs={}, recursive=True, text=None,
limit=None, **kwargs):
"""Extracts a list of Tag objects that match the given
criteria. You can specify the name of the Tag and any
attributes you want the Tag to have.
The value of a key-value pair in the 'attrs' map can be a
string, a list of strings, a regular expression object, or a
callable that takes a string and returns whether or not the
string matches for some custom definition of 'matches'. The
same is true of the tag name."""
generator = self.recursiveChildGenerator
if not recursive:
generator = self.childGenerator
return self._findAll(name, attrs, text, limit, generator, **kwargs)
findAllChildren = findAll
#Utility methods
def append(self, tag):
"""Appends the given tag to the contents of this tag."""
#Private methods
def _getAttrMap(self):
"""Initializes a map representation of this tag's attributes,
if not already initialized."""
if not getattr(self, 'attrMap'):
self.attrMap = {}
for (key, value) in self.attrs:
self.attrMap[key] = value
return self.attrMap
#Generator methods
def childGenerator(self):
for i in range(0, len(self.contents)):
yield self.contents[i]
raise StopIteration
def recursiveChildGenerator(self):
stack = [(self, 0)]
while stack:
tag, start = stack.pop()
if isinstance(tag, Tag):
for i in range(start, len(tag.contents)):
a = tag.contents[i]
yield a
if isinstance(a, Tag) and tag.contents:
if i < len(tag.contents) - 1:
stack.append((tag, i+1))
stack.append((a, 0))
raise StopIteration
# Next, a couple classes to represent queries and their results.
class SoupStrainer:
"""Encapsulates a number of ways of matching a markup element (tag or
def __init__(self, name=None, attrs={}, text=None, **kwargs):
self.text = text
def __str__(self):
if self.text:
return self.text
return "%s|%s" % (, self.attrs)
def searchTag(self, markupName=None, markupAttrs={}):
found = None
markup = None
if isinstance(markupName, Tag):
markup = markupName
markupAttrs = markup
callFunctionWithTagData = callable( \
and not isinstance(markupName, Tag)
if (not \
or callFunctionWithTagData \
or (markup and self._matches(markup, \
or (not markup and self._matches(markupName,
if callFunctionWithTagData:
match =, markupAttrs)
match = True
markupAttrMap = None
for attr, matchAgainst in self.attrs.items():
if not markupAttrMap:
if hasattr(markupAttrs, 'get'):
markupAttrMap = markupAttrs
markupAttrMap = {}
for k,v in markupAttrs:
markupAttrMap[k] = v
attrValue = markupAttrMap.get(attr)
if not self._matches(attrValue, matchAgainst):
match = False
if match:
if markup:
found = markup
found = markupName
return found
def search(self, markup):
#print 'looking for %s in %s' % (self, markup)
found = None
# If given a list of items, scan it for a text element that
# matches.
if isList(markup) and not isinstance(markup, Tag):
for element in markup:
if isinstance(element, NavigableString) \
found = element
# If it's a Tag, make sure its name or attributes match.
# Don't bother with Tags if we're searching for text.
elif isinstance(markup, Tag):
if not self.text:
found = self.searchTag(markup)
# If it's text, make sure the text matches.
elif isinstance(markup, NavigableString) or \
if self._matches(markup, self.text):
found = markup
raise Exception, "I don't know how to match against a %s" \
% markup.__class__
return found
def _matches(self, markup, matchAgainst):
#print "Matching %s against %s" % (markup, matchAgainst)
result = False
if matchAgainst == True and type(matchAgainst) == types.BooleanType:
result = markup != None
elif callable(matchAgainst):
result = matchAgainst(markup)
#Custom match methods take the tag as an argument, but all
#other ways of matching match the tag name as a string.
if isinstance(markup, Tag):
markup =
if markup and not isString(markup):
markup = unicode(markup)
#Now we know that chunk is either a string, or None.
if hasattr(matchAgainst, 'match'):
# It's a regexp object.
result = markup and
elif isList(matchAgainst):
result = markup in matchAgainst
elif hasattr(matchAgainst, 'items'):
result = markup.has_key(matchAgainst)
elif matchAgainst and isString(markup):
if isinstance(markup, unicode):
matchAgainst = unicode(matchAgainst)
matchAgainst = str(matchAgainst)
if not result:
result = matchAgainst == markup
return result
class ResultSet(list):
"""A ResultSet is just a list that keeps track of the SoupStrainer
that created it."""
def __init__(self, source):
self.source = source
# Now, some helper functions.
def isList(l):
"""Convenience method that works with all 2.x versions of Python
to determine whether or not something is listlike."""
return hasattr(l, '__iter__') \
or (type(l) in (types.ListType, types.TupleType))
def isString(s):
"""Convenience method that works with all 2.x versions of Python
to determine whether or not something is stringlike."""
return isinstance(s, unicode) or isintance(s, basestring)
except NameError:
return isinstance(s, str)
def buildTagMap(default, *args):
"""Turns a list of maps, lists, or scalars into a single map.
NESTING_RESET_TAGS maps out of lists and partial maps."""
built = {}
for portion in args:
if hasattr(portion, 'items'):
#It's a map. Merge it.
for k,v in portion.items():
built[k] = v
elif isList(portion):
#It's a list. Map each item to the default.
for k in portion:
built[k] = default
#It's a scalar. Map it to the default.
built[portion] = default
return built
# Now, the parser classes.
class BeautifulStoneSoup(Tag, SGMLParser):
"""This class contains the basic parser and search code. It defines
a parser that knows nothing about tag behavior except for the
You can't close a tag without closing all the tags it encloses.
That is, "<foo><bar></foo>" actually means
[Another possible explanation is "<foo><bar /></foo>", but since
this class defines no SELF_CLOSING_TAGS, it will never use that
This class is useful for parsing XML or made-up markup languages,
or when BeautifulSoup makes an assumption counter to what you were
for i in ["quot", "apos", "amp", "lt", "gt"]:
MARKUP_MASSAGE = [(re.compile('(<[^<>]*)/>'),
lambda x: + ' />'),
lambda x: '<!' + + '>')
ROOT_TAG_NAME = u'[document]'
def __init__(self, markup="", parseOnlyThese=None, fromEncoding=None,
markupMassage=True, smartQuotesTo=XML_ENTITIES,
convertEntities=None, selfClosingTags=None):
"""The Soup object is initialized as the 'root tag', and the
provided markup (which can be a string or a file-like object)
is fed into the underlying parser.
sgmllib will process most bad HTML, and the BeautifulSoup
class has some tricks for dealing with some HTML that kills
sgmllib, but Beautiful Soup can nonetheless choke or lose data
if your data uses self-closing tags or declarations
By default, Beautiful Soup uses regexes to sanitize input,
avoiding the vast majority of these problems. If the problems
don't apply to you, pass in False for markupMassage, and
you'll get better performance.
The default parser massage techniques fix the two most common
instances of invalid HTML that choke sgmllib:
<br/> (No space between name of closing tag and tag close)
<! --Comment--> (Extraneous whitespace in declaration)
You can pass in a custom list of (RE object, replace method)
tuples to get Beautiful Soup to scrub your input the way you
self.parseOnlyThese = parseOnlyThese
self.fromEncoding = fromEncoding
self.smartQuotesTo = smartQuotesTo
self.convertEntities = convertEntities
if self.convertEntities:
# It doesn't make sense to convert encoded characters to
# entities even while you're converting entities to Unicode.
# Just convert it all to Unicode.
self.smartQuotesTo = None
self.instanceSelfClosingTags = buildTagMap(None, selfClosingTags)
if hasattr(markup, 'read'): # It's a file-type object.
markup =
self.markup = markup
self.markupMassage = markupMassage
except StopParsing:
self.markup = None # The markup can now be GCed
def _feed(self, inDocumentEncoding=None):
# Convert the document to Unicode.
markup = self.markup
if isinstance(markup, unicode):
if not hasattr(self, 'originalEncoding'):
self.originalEncoding = None
dammit = UnicodeDammit\
(markup, [self.fromEncoding, inDocumentEncoding],
markup = dammit.unicode
self.originalEncoding = dammit.originalEncoding
if markup:
if self.markupMassage:
if not isList(self.markupMassage):
self.markupMassage = self.MARKUP_MASSAGE
for fix, m in self.markupMassage:
markup = fix.sub(m, markup)
SGMLParser.feed(self, markup)
# Close out any unfinished strings and close all the open tags.
while != self.ROOT_TAG_NAME:
def __getattr__(self, methodName):
"""This method routes method call requests to either the SGMLParser
superclass or the Tag superclass, depending on the method name."""
#print "__getattr__ called on %s.%s" % (self.__class__, methodName)
if methodName.find('start_') == 0 or methodName.find('end_') == 0 \
or methodName.find('do_') == 0:
return SGMLParser.__getattr__(self, methodName)
elif methodName.find('__') != 0:
return Tag.__getattr__(self, methodName)
raise AttributeError
def isSelfClosingTag(self, name):
"""Returns true iff the given string is the name of a
self-closing tag according to this parser."""
return self.SELF_CLOSING_TAGS.has_key(name) \
or self.instanceSelfClosingTags.has_key(name)
def reset(self):
Tag.__init__(self, self, self.ROOT_TAG_NAME)
self.hidden = 1
self.currentData = []
self.currentTag = None
self.tagStack = []
self.quoteStack = []
def popTag(self):
tag = self.tagStack.pop()
# Tags with just one string-owning child get the child as a
# 'string' property, so that soup.tag.string is shorthand for
# soup.tag.contents[0]
if len(self.currentTag.contents) == 1 and \
isinstance(self.currentTag.contents[0], NavigableString):
self.currentTag.string = self.currentTag.contents[0]
#print "Pop",
if self.tagStack:
self.currentTag = self.tagStack[-1]
return self.currentTag
def pushTag(self, tag):
#print "Push",
if self.currentTag:
self.currentTag = self.tagStack[-1]
def endData(self, containerClass=NavigableString):
if self.currentData:
currentData = ''.join(self.currentData)
if not currentData.strip():
if '\n' in currentData:
currentData = '\n'
currentData = ' '
self.currentData = []
if self.parseOnlyThese and len(self.tagStack) <= 1 and \
(not self.parseOnlyThese.text or \
o = containerClass(currentData)
o.setup(self.currentTag, self.previous)
if self.previous: = o
self.previous = o
def _popToTag(self, name, inclusivePop=True):
"""Pops the tag stack up to and including the most recent
instance of the given tag. If inclusivePop is false, pops the tag
stack up to but *not* including the most recent instqance of
the given tag."""
#print "Popping to %s" % name
if name == self.ROOT_TAG_NAME:
numPops = 0
mostRecentTag = None
for i in range(len(self.tagStack)-1, 0, -1):
if name == self.tagStack[i].name:
numPops = len(self.tagStack)-i
if not inclusivePop:
numPops = numPops - 1
for i in range(0, numPops):
mostRecentTag = self.popTag()
return mostRecentTag
def _smartPop(self, name):
"""We need to pop up to the previous tag of this type, unless
one of this tag's nesting reset triggers comes between this
tag and the previous tag of this type, OR unless this tag is a
generic nesting trigger and another generic nesting trigger
comes between this tag and the previous tag of this type.
<p>Foo<b>Bar<p> should pop to 'p', not 'b'.
<p>Foo<table>Bar<p> should pop to 'table', not 'p'.
<p>Foo<table><tr>Bar<p> should pop to 'tr', not 'p'.
<p>Foo<b>Bar<p> should pop to 'p', not 'b'.
<li><ul><li> *<li>* should pop to 'ul', not the first 'li'.
<tr><table><tr> *<tr>* should pop to 'table', not the first 'tr'
<td><tr><td> *<td>* should pop to 'tr', not the first 'td'
nestingResetTriggers = self.NESTABLE_TAGS.get(name)
isNestable = nestingResetTriggers != None
isResetNesting = self.RESET_NESTING_TAGS.has_key(name)
popTo = None
inclusive = True
for i in range(len(self.tagStack)-1, 0, -1):
p = self.tagStack[i]
if (not p or == name) and not isNestable:
#Non-nestable tags get popped to the top or to their
#last occurance.
popTo = name
if (nestingResetTriggers != None
and in nestingResetTriggers) \
or (nestingResetTriggers == None and isResetNesting
and self.RESET_NESTING_TAGS.has_key(
#If we encounter one of the nesting reset triggers
#peculiar to this tag, or we encounter another tag
#that causes nesting to reset, pop up to but not
#including that tag.
popTo =
inclusive = False
p = p.parent
if popTo:
self._popToTag(popTo, inclusive)
def unknown_starttag(self, name, attrs, selfClosing=0):
#print "Start tag %s" % name
if self.quoteStack:
#This is not a real tag.
#print "<%s> is not real!" % name
attrs = ''.join(map(lambda(x, y): ' %s="%s"' % (x, y), attrs))
self.handle_data('<%s%s>' % (name, attrs))
if not self.isSelfClosingTag(name) and not selfClosing:
if self.parseOnlyThese and len(self.tagStack) <= 1 \
and (self.parseOnlyThese.text or not self.parseOnlyThese.searchTag(name, attrs)):
tag = Tag(self, name, attrs, self.currentTag, self.previous)
if self.previous: = tag
self.previous = tag
if selfClosing or self.isSelfClosingTag(name):
if name in self.QUOTE_TAGS:
#print "Beginning quote (%s)" % name
self.literal = 1
return tag
def unknown_endtag(self, name):
#print "End tag %s" % name
if self.quoteStack and self.quoteStack[-1] != name:
#This is not a real end tag.
#print "</%s> is not real!" % name
self.handle_data('</%s>' % name)
if self.quoteStack and self.quoteStack[-1] == name:
self.literal = (len(self.quoteStack) > 0)
def handle_data(self, data):
def _toStringSubclass(self, text, subclass):
"""Adds a certain piece of text to the tree as a NavigableString
def handle_pi(self, text):
"""Handle a processing instruction as a ProcessingInstruction
object, possibly one with a %SOUP-ENCODING% slot into which an
encoding will be plugged later."""
if text[:3] == "xml":
text = "xml version='1.0' encoding='%SOUP-ENCODING%'"
self._toStringSubclass(text, ProcessingInstruction)
def handle_comment(self, text):
"Handle comments as Comment objects."
self._toStringSubclass(text, Comment)
def handle_charref(self, ref):
"Handle character references as data."
if self.convertEntities in [self.HTML_ENTITIES,
data = unichr(int(ref))
data = '&#%s;' % ref
def handle_entityref(self, ref):
"""Handle entity references as data, possibly converting known
HTML entity references to the corresponding Unicode
data = None
if self.convertEntities == self.HTML_ENTITIES or \
(self.convertEntities == self.XML_ENTITIES and \
data = unichr(name2codepoint[ref])
except KeyError:
if not data:
data = '&%s;' % ref
def handle_decl(self, data):
"Handle DOCTYPEs and the like as Declaration objects."
self._toStringSubclass(data, Declaration)
def parse_declaration(self, i):
"""Treat a bogus SGML declaration as raw data. Treat a CDATA
declaration as a CData object."""
j = None
if self.rawdata[i:i+9] == '<![CDATA[':
k = self.rawdata.find(']]>', i)
if k == -1:
k = len(self.rawdata)
data = self.rawdata[i+9:k]
j = k+3
self._toStringSubclass(data, CData)
j = SGMLParser.parse_declaration(self, i)
except SGMLParseError:
toHandle = self.rawdata[i:]
j = i + len(toHandle)
return j
class BeautifulSoup(BeautifulStoneSoup):
"""This parser knows the following facts about HTML:
* Some tags have no closing tag and should be interpreted as being
closed as soon as they are encountered.
* The text inside some tags (ie. 'script') may contain tags which
are not really part of the document and which should be parsed
as text, not tags. If you want to parse the text as tags, you can
always fetch it and parse it explicitly.
* Tag nesting rules:
Most tags can't be nested at all. For instance, the occurance of
a <p> tag should implicitly close the previous <p> tag.
should be transformed into:
Some tags can be nested arbitrarily. For instance, the occurance
of a <blockquote> tag should _not_ implicitly close the previous
<blockquote> tag.
Alice said: <blockquote>Bob said: <blockquote>Blah
should NOT be transformed into:
Alice said: <blockquote>Bob said: </blockquote><blockquote>Blah
Some tags can be nested, but the nesting is reset by the
interposition of other tags. For instance, a <tr> tag should
implicitly close the previous <tr> tag within the same <table>,
but not close a <tr> tag in another table.
should be transformed into:
should NOT be transformed into
Differing assumptions about tag nesting rules are a major source
of problems with the BeautifulSoup class. If BeautifulSoup is not
treating as nestable a tag your page author treats as nestable,
try ICantBelieveItsBeautifulSoup, MinimalSoup, or
BeautifulStoneSoup before writing your own subclass."""
def __init__(self, *args, **kwargs):
if not kwargs.has_key('smartQuotesTo'):
kwargs['smartQuotesTo'] = self.HTML_ENTITIES
BeautifulStoneSoup.__init__(self, *args, **kwargs)
SELF_CLOSING_TAGS = buildTagMap(None,
['br' , 'hr', 'input', 'img', 'meta',
'spacer', 'link', 'frame', 'base'])
QUOTE_TAGS = {'script': None}
#According to the HTML standard, each of these inline tags can
#contain another tag of the same type. Furthermore, it's common
#to actually use these tags this way.
NESTABLE_INLINE_TAGS = ['span', 'font', 'q', 'object', 'bdo', 'sub', 'sup',
#According to the HTML standard, these block tags can contain
#another tag of the same type. Furthermore, it's common
#to actually use these tags this way.
NESTABLE_BLOCK_TAGS = ['blockquote', 'div', 'fieldset', 'ins', 'del']
#Lists can contain other lists, but there are restrictions.
NESTABLE_LIST_TAGS = { 'ol' : [],
'ul' : [],
'li' : ['ul', 'ol'],
'dl' : [],
'dd' : ['dl'],
'dt' : ['dl'] }
#Tables can contain other tables, but there are restrictions.
NESTABLE_TABLE_TAGS = {'table' : [],
'tr' : ['table', 'tbody', 'tfoot', 'thead'],
'td' : ['tr'],
'th' : ['tr'],
'thead' : ['table'],
'tbody' : ['table'],
'tfoot' : ['table'],
NON_NESTABLE_BLOCK_TAGS = ['address', 'form', 'p', 'pre']
#If one of these tags is encountered, all tags up to the next tag of
#this type are popped.
# Used to detect the charset in a META tag; see start_meta
CHARSET_RE = re.compile("((^|;)\s*charset=)([^;]*)")
def start_meta(self, attrs):
"""Beautiful Soup can detect a charset included in a META tag,
try to convert the document to that charset, and re-parse the
document from the beginning."""
httpEquiv = None
contentType = None
contentTypeIndex = None
tagNeedsEncodingSubstitution = False
for i in range(0, len(attrs)):
key, value = attrs[i]
key = key.lower()
if key == 'http-equiv':
httpEquiv = value
elif key == 'content':
contentType = value
contentTypeIndex = i
if httpEquiv and contentType: # It's an interesting meta tag.
match =
if match:
if getattr(self, 'declaredHTMLEncoding') or \
(self.originalEncoding == self.fromEncoding):
# This is our second pass through the document, or
# else an encoding was specified explicitly and it
# worked. Rewrite the meta tag.
newAttr = self.CHARSET_RE.sub\
(lambda(match) +
"%SOUP-ENCODING%", value)
attrs[contentTypeIndex] = (attrs[contentTypeIndex][0],
tagNeedsEncodingSubstitution = True
# This is our first pass through the document.
# Go through it again with the new information.
newCharset =
if newCharset and newCharset != self.originalEncoding:
self.declaredHTMLEncoding = newCharset
raise StopParsing
tag = self.unknown_starttag("meta", attrs)
if tagNeedsEncodingSubstitution:
tag.containsSubstitutions = True
class StopParsing(Exception):
class ICantBelieveItsBeautifulSoup(BeautifulSoup):
"""The BeautifulSoup class is oriented towards skipping over
common HTML errors like unclosed tags. However, sometimes it makes
errors of its own. For instance, consider this fragment:
This is perfectly valid (if bizarre) HTML. However, the
BeautifulSoup class will implicitly close the first b tag when it
encounters the second 'b'. It will think the author wrote
"<b>Foo<b>Bar", and didn't close the first 'b' tag, because
there's no real-world reason to bold something that's already
bold. When it encounters '</b></b>' it will close two more 'b'
tags, for a grand total of three tags closed instead of two. This
can throw off the rest of your document structure. The same is
true of a number of other tags, listed below.
It's much more common for someone to forget to close a 'b' tag
than to actually use nested 'b' tags, and the BeautifulSoup class
handles the common case. This class handles the not-co-common
case: where you can't believe someone wrote what they did, but
it's valid HTML and BeautifulSoup screwed up by assuming it
wouldn't be."""
['em', 'big', 'i', 'small', 'tt', 'abbr', 'acronym', 'strong',
'cite', 'code', 'dfn', 'kbd', 'samp', 'strong', 'var', 'b',
NESTABLE_TAGS = buildTagMap([], BeautifulSoup.NESTABLE_TAGS,
class MinimalSoup(BeautifulSoup):
"""The MinimalSoup class is for parsing HTML that contains
pathologically bad markup. It makes no assumptions about tag
nesting, but it does know which tags are self-closing, that
<script> tags contain Javascript and should not be parsed, that
META tags may contain encoding information, and so on.
This also makes it better for subclassing than BeautifulStoneSoup
or BeautifulSoup."""
RESET_NESTING_TAGS = buildTagMap('noscript')
class BeautifulSOAP(BeautifulStoneSoup):
"""This class will push a tag with only a single string child into
the tag's parent as an attribute. The attribute's name is the tag
name, and the value is the string child. An example should give
the flavor of the change:
<foo bar="baz"><bar>baz</bar></foo>
You can then access fooTag['bar'] instead of fooTag.barTag.string.
This is, of course, useful for scraping structures that tend to
use subelements instead of attributes, such as SOAP messages. Note
that it modifies its input, so don't print the modified version
I'm not sure how many people really want to use this class; let me
know if you do. Mainly I like the name."""
def popTag(self):
if len(self.tagStack) > 1:
tag = self.tagStack[-1]
parent = self.tagStack[-2]
if (isinstance(tag, Tag) and len(tag.contents) == 1 and
isinstance(tag.contents[0], NavigableString) and
not parent.attrMap.has_key(
parent[] = tag.contents[0]
#Enterprise class names! It has come to our attention that some people
#think the names of the Beautiful Soup parser classes are too silly
#and "unprofessional" for use in enterprise screen-scraping. We feel
#your pain! For such-minded folk, the Beautiful Soup Consortium And
#All-Night Kosher Bakery recommends renaming this file to
#"" (or, in cases of extreme enterprisitude,
#"RobustParserBeanInterface.class") and using the following
#enterprise-friendly class aliases:
class RobustXMLParser(BeautifulStoneSoup):
class RobustHTMLParser(BeautifulSoup):
class RobustWackAssHTMLParser(ICantBelieveItsBeautifulSoup):
class RobustInsanelyWackAssHTMLParser(MinimalSoup):
class SimplifyingSOAPParser(BeautifulSOAP):
# Bonus library: Unicode, Dammit
# This class forces XML data into a standard format (usually to UTF-8
# or Unicode). It is heavily based on code from Mark Pilgrim's
# Universal Feed Parser. It does not rewrite the XML or HTML to
# reflect a new encoding: that happens in BeautifulStoneSoup.handle_pi
# (XML) and BeautifulSoup.start_meta (HTML).
# Autodetects character encodings.
# Download from
import chardet
# import chardet.constants
# chardet.constants._debug = 1
chardet = None
chardet = None
# cjkcodecs and iconv_codec make Python know about more character encodings.
# Both are available from
# They're built in if you use Python 2.4.
import cjkcodecs.aliases
import iconv_codec
class UnicodeDammit:
"""A class for detecting the encoding of a *ML document and
converting it to a Unicode string. If the source encoding is
windows-1252, can replace MS smart quotes with their HTML or XML
# This dictionary maps commonly seen values for "charset" in HTML
# meta tags to the corresponding Python codec names. It only covers
# values that aren't in Python's aliases and can't be determined
# by the heuristics in find_codec.
CHARSET_ALIASES = { "macintosh" : "mac-roman",
"x-sjis" : "shift-jis" }
def __init__(self, markup, overrideEncodings=[],
self.markup, documentEncoding, sniffedEncoding = \
self.smartQuotesTo = smartQuotesTo
self.triedEncodings = []
if isinstance(markup, unicode):
return markup
u = None
for proposedEncoding in overrideEncodings:
u = self._convertFrom(proposedEncoding)
if u: break
if not u:
for proposedEncoding in (documentEncoding, sniffedEncoding):
u = self._convertFrom(proposedEncoding)
if u: break
# If no luck and we have auto-detection library, try that:
if not u and chardet and not isinstance(self.markup, unicode):
u = self._convertFrom(chardet.detect(self.markup)['encoding'])
# As a last resort, try utf-8 and windows-1252:
if not u:
for proposed_encoding in ("utf-8", "windows-1252"):
u = self._convertFrom(proposed_encoding)
if u: break
self.unicode = u
if not u: self.originalEncoding = None
def _subMSChar(self, orig):
"""Changes a MS smart quote character to an XML or HTML
sub = self.MS_CHARS.get(orig)
if type(sub) == types.TupleType:
if self.smartQuotesTo == 'xml':
sub = '&#x%s;' % sub[1]
sub = '&%s;' % sub[0]
return sub
def _convertFrom(self, proposed):
proposed = self.find_codec(proposed)
if not proposed or proposed in self.triedEncodings:
return None
markup = self.markup
# Convert smart quotes to HTML if coming from an encoding
# that might have them.
if self.smartQuotesTo and proposed in("windows-1252",
markup = re.compile("([\x80-\x9f])").sub \
(lambda(x): self._subMSChar(,
# print "Trying to convert document to %s" % proposed
u = self._toUnicode(markup, proposed)
self.markup = u
self.originalEncoding = proposed
except Exception, e:
# print "That didn't work!"
# print e
return None
#print "Correct encoding: %s" % proposed
return self.markup
def _toUnicode(self, data, encoding):
'''Given a string and its encoding, decodes the string into Unicode.
%encoding is a string recognized by encodings.aliases'''
# strip Byte Order Mark (if present)
if (len(data) >= 4) and (data[:2] == '\xfe\xff') \
and (data[2:4] != '\x00\x00'):
encoding = 'utf-16be'
data = data[2:]
elif (len(data) >= 4) and (data[:2] == '\xff\xfe') \
and (data[2:4] != '\x00\x00'):
encoding = 'utf-16le'
data = data[2:]
elif data[:3] == '\xef\xbb\xbf':
encoding = 'utf-8'
data = data[3:]
elif data[:4] == '\x00\x00\xfe\xff':
encoding = 'utf-32be'
data = data[4:]
elif data[:4] == '\xff\xfe\x00\x00':
encoding = 'utf-32le'
data = data[4:]
newdata = unicode(data, encoding)
return newdata
def _detectEncoding(self, xml_data):
"""Given a document, tries to detect its XML encoding."""
xml_encoding = sniffed_xml_encoding = None
if xml_data[:4] == '\x4c\x6f\xa7\x94':
xml_data = self._ebcdic_to_ascii(xml_data)
elif xml_data[:4] == '\x00\x3c\x00\x3f':
# UTF-16BE
sniffed_xml_encoding = 'utf-16be'
xml_data = unicode(xml_data, 'utf-16be').encode('utf-8')
elif (len(xml_data) >= 4) and (xml_data[:2] == '\xfe\xff') \
and (xml_data[2:4] != '\x00\x00'):
# UTF-16BE with BOM
sniffed_xml_encoding = 'utf-16be'
xml_data = unicode(xml_data[2:], 'utf-16be').encode('utf-8')
elif xml_data[:4] == '\x3c\x00\x3f\x00':
# UTF-16LE
sniffed_xml_encoding = 'utf-16le'
xml_data = unicode(xml_data, 'utf-16le').encode('utf-8')
elif (len(xml_data) >= 4) and (xml_data[:2] == '\xff\xfe') and \
(xml_data[2:4] != '\x00\x00'):
# UTF-16LE with BOM
sniffed_xml_encoding = 'utf-16le'
xml_data = unicode(xml_data[2:], 'utf-16le').encode('utf-8')
elif xml_data[:4] == '\x00\x00\x00\x3c':
# UTF-32BE
sniffed_xml_encoding = 'utf-32be'
xml_data = unicode(xml_data, 'utf-32be').encode('utf-8')
elif xml_data[:4] == '\x3c\x00\x00\x00':
# UTF-32LE
sniffed_xml_encoding = 'utf-32le'
xml_data = unicode(xml_data, 'utf-32le').encode('utf-8')
elif xml_data[:4] == '\x00\x00\xfe\xff':
# UTF-32BE with BOM
sniffed_xml_encoding = 'utf-32be'
xml_data = unicode(xml_data[4:], 'utf-32be').encode('utf-8')
elif xml_data[:4] == '\xff\xfe\x00\x00':
# UTF-32LE with BOM
sniffed_xml_encoding = 'utf-32le'
xml_data = unicode(xml_data[4:], 'utf-32le').encode('utf-8')
elif xml_data[:3] == '\xef\xbb\xbf':
# UTF-8 with BOM
sniffed_xml_encoding = 'utf-8'
xml_data = unicode(xml_data[3:], 'utf-8').encode('utf-8')
sniffed_xml_encoding = 'ascii'
xml_encoding_match = re.compile \
xml_encoding_match = None
if xml_encoding_match:
xml_encoding = xml_encoding_match.groups()[0].lower()
if sniffed_xml_encoding and \
(xml_encoding in ('iso-10646-ucs-2', 'ucs-2', 'csunicode',
'iso-10646-ucs-4', 'ucs-4', 'csucs4',
'utf-16', 'utf-32', 'utf_16', 'utf_32',
'utf16', 'u16')):
xml_encoding = sniffed_xml_encoding
return xml_data, xml_encoding, sniffed_xml_encoding
def find_codec(self, charset):
return self._codec(self.CHARSET_ALIASES.get(charset, charset)) \
or (charset and self._codec(charset.replace("-", ""))) \
or (charset and self._codec(charset.replace("-", "_"))) \
or charset
def _codec(self, charset):
if not charset: return charset
codec = None
codec = charset
except LookupError:
return codec
def _ebcdic_to_ascii(self, s):
c = self.__class__
emap = (0,1,2,3,156,9,134,127,151,141,142,11,12,13,14,15,
import string
c.EBCDIC_TO_ASCII_MAP = string.maketrans( \
''.join(map(chr, range(256))), ''.join(map(chr, emap)))
return s.translate(c.EBCDIC_TO_ASCII_MAP)
MS_CHARS = { '\x80' : ('euro', '20AC'),
'\x81' : ' ',
'\x82' : ('sbquo', '201A'),
'\x83' : ('fnof', '192'),
'\x84' : ('bdquo', '201E'),
'\x85' : ('hellip', '2026'),
'\x86' : ('dagger', '2020'),
'\x87' : ('Dagger', '2021'),
'\x88' : ('circ', '2C6'),
'\x89' : ('permil', '2030'),
'\x8A' : ('Scaron', '160'),
'\x8B' : ('lsaquo', '2039'),
'\x8C' : ('OElig', '152'),
'\x8D' : '?',
'\x8E' : ('#x17D', '17D'),
'\x8F' : '?',
'\x90' : '?',
'\x91' : ('lsquo', '2018'),
'\x92' : ('rsquo', '2019'),
'\x93' : ('ldquo', '201C'),
'\x94' : ('rdquo', '201D'),
'\x95' : ('bull', '2022'),
'\x96' : ('ndash', '2013'),
'\x97' : ('mdash', '2014'),
'\x98' : ('tilde', '2DC'),
'\x99' : ('trade', '2122'),
'\x9a' : ('scaron', '161'),
'\x9b' : ('rsaquo', '203A'),
'\x9c' : ('oelig', '153'),
'\x9d' : '?',
'\x9e' : ('#x17E', '17E'),
'\x9f' : ('Yuml', ''),}
#By default, act as an HTML pretty-printer.
if __name__ == '__main__':
import sys
soup = BeautifulStoneSoup(
print soup.prettify()