Testing the ability to include libraries with the bot

lib/bs4/builder/__init__.py

@@ -0,0 +1,316 @@
+from collections import defaultdict
+import itertools
+import sys
+from bs4.element import (
+    CharsetMetaAttributeValue,
+    ContentMetaAttributeValue,
+    whitespace_re
+    )
+
+__all__ = [
+    'HTMLTreeBuilder',
+    'SAXTreeBuilder',
+    'TreeBuilder',
+    'TreeBuilderRegistry',
+    ]
+
+# Some useful features for a TreeBuilder to have.
+FAST = 'fast'
+PERMISSIVE = 'permissive'
+STRICT = 'strict'
+XML = 'xml'
+HTML = 'html'
+HTML_5 = 'html5'
+
+
+class TreeBuilderRegistry(object):
+
+    def __init__(self):
+        self.builders_for_feature = defaultdict(list)
+        self.builders = []
+
+    def register(self, treebuilder_class):
+        """Register a treebuilder based on its advertised features."""
+        for feature in treebuilder_class.features:
+            self.builders_for_feature[feature].insert(0, treebuilder_class)
+        self.builders.insert(0, treebuilder_class)
+
+    def lookup(self, *features):
+        if len(self.builders) == 0:
+            # There are no builders at all.
+            return None
+
+        if len(features) == 0:
+            # They didn't ask for any features. Give them the most
+            # recently registered builder.
+            return self.builders[0]
+
+        # Go down the list of features in order, and eliminate any builders
+        # that don't match every feature.
+        features = list(features)
+        features.reverse()
+        candidates = None
+        candidate_set = None
+        while len(features) > 0:
+            feature = features.pop()
+            we_have_the_feature = self.builders_for_feature.get(feature, [])
+            if len(we_have_the_feature) > 0:
+                if candidates is None:
+                    candidates = we_have_the_feature
+                    candidate_set = set(candidates)
+                else:
+                    # Eliminate any candidates that don't have this feature.
+                    candidate_set = candidate_set.intersection(
+                        set(we_have_the_feature))
+
+        # The only valid candidates are the ones in candidate_set.
+        # Go through the original list of candidates and pick the first one
+        # that's in candidate_set.
+        if candidate_set is None:
+            return None
+        for candidate in candidates:
+            if candidate in candidate_set:
+                return candidate
+        return None
+
+# The BeautifulSoup class will take feature lists from developers and use them
+# to look up builders in this registry.
+builder_registry = TreeBuilderRegistry()
+
+class TreeBuilder(object):
+    """Turn a document into a Beautiful Soup object tree."""
+
+    features = []
+
+    is_xml = False
+    preserve_whitespace_tags = set()
+    empty_element_tags = None # A tag will be considered an empty-element
+                              # tag when and only when it has no contents.
+
+    # A value for these tag/attribute combinations is a space- or
+    # comma-separated list of CDATA, rather than a single CDATA.
+    cdata_list_attributes = {}
+
+
+    def __init__(self):
+        self.soup = None
+
+    def reset(self):
+        pass
+
+    def can_be_empty_element(self, tag_name):
+        """Might a tag with this name be an empty-element tag?
+
+        The final markup may or may not actually present this tag as
+        self-closing.
+
+        For instance: an HTMLBuilder does not consider a <p> tag to be
+        an empty-element tag (it's not in
+        HTMLBuilder.empty_element_tags). This means an empty <p> tag
+        will be presented as "<p></p>", not "<p />".
+
+        The default implementation has no opinion about which tags are
+        empty-element tags, so a tag will be presented as an
+        empty-element tag if and only if it has no contents.
+        "<foo></foo>" will become "<foo />", and "<foo>bar</foo>" will
+        be left alone.
+        """
+        if self.empty_element_tags is None:
+            return True
+        return tag_name in self.empty_element_tags
+
+    def feed(self, markup):
+        raise NotImplementedError()
+
+    def prepare_markup(self, markup, user_specified_encoding=None,
+                       document_declared_encoding=None):
+        return markup, None, None, False
+
+    def test_fragment_to_document(self, fragment):
+        """Wrap an HTML fragment to make it look like a document.
+
+        Different parsers do this differently. For instance, lxml
+        introduces an empty <head> tag, and html5lib
+        doesn't. Abstracting this away lets us write simple tests
+        which run HTML fragments through the parser and compare the
+        results against other HTML fragments.
+
+        This method should not be used outside of tests.
+        """
+        return fragment
+
+    def set_up_substitutions(self, tag):
+        return False
+
+    def _replace_cdata_list_attribute_values(self, tag_name, attrs):
+        """Replaces class="foo bar" with class=["foo", "bar"]
+
+        Modifies its input in place.
+        """
+        if self.cdata_list_attributes:
+            universal = self.cdata_list_attributes.get('*', [])
+            tag_specific = self.cdata_list_attributes.get(
+                tag_name.lower(), [])
+            for cdata_list_attr in itertools.chain(universal, tag_specific):
+                if cdata_list_attr in dict(attrs):
+                    # Basically, we have a "class" attribute whose
+                    # value is a whitespace-separated list of CSS
+                    # classes. Split it into a list.
+                    value = attrs[cdata_list_attr]
+                    if isinstance(value, basestring):
+                        values = whitespace_re.split(value)
+                    else:
+                        # html5lib sometimes calls setAttributes twice
+                        # for the same tag when rearranging the parse
+                        # tree. On the second call the attribute value
+                        # here is already a list.  If this happens,
+                        # leave the value alone rather than trying to
+                        # split it again.
+                        values = value
+                    attrs[cdata_list_attr] = values
+        return attrs
+
+class SAXTreeBuilder(TreeBuilder):
+    """A Beautiful Soup treebuilder that listens for SAX events."""
+
+    def feed(self, markup):
+        raise NotImplementedError()
+
+    def close(self):
+        pass
+
+    def startElement(self, name, attrs):
+        attrs = dict((key[1], value) for key, value in list(attrs.items()))
+        #print "Start %s, %r" % (name, attrs)
+        self.soup.handle_starttag(name, attrs)
+
+    def endElement(self, name):
+        #print "End %s" % name
+        self.soup.handle_endtag(name)
+
+    def startElementNS(self, nsTuple, nodeName, attrs):
+        # Throw away (ns, nodeName) for now.
+        self.startElement(nodeName, attrs)
+
+    def endElementNS(self, nsTuple, nodeName):
+        # Throw away (ns, nodeName) for now.
+        self.endElement(nodeName)
+        #handler.endElementNS((ns, node.nodeName), node.nodeName)
+
+    def startPrefixMapping(self, prefix, nodeValue):
+        # Ignore the prefix for now.
+        pass
+
+    def endPrefixMapping(self, prefix):
+        # Ignore the prefix for now.
+        # handler.endPrefixMapping(prefix)
+        pass
+
+    def characters(self, content):
+        self.soup.handle_data(content)
+
+    def startDocument(self):
+        pass
+
+    def endDocument(self):
+        pass
+
+
+class HTMLTreeBuilder(TreeBuilder):
+    """This TreeBuilder knows facts about HTML.
+
+    Such as which tags are empty-element tags.
+    """
+
+    preserve_whitespace_tags = set(['pre', 'textarea'])
+    empty_element_tags = set(['br' , 'hr', 'input', 'img', 'meta',
+                              'spacer', 'link', 'frame', 'base'])
+
+    # The HTML standard defines these attributes as containing a
+    # space-separated list of values, not a single value. That is,
+    # class="foo bar" means that the 'class' attribute has two values,
+    # 'foo' and 'bar', not the single value 'foo bar'.  When we
+    # encounter one of these attributes, we will parse its value into
+    # a list of values if possible. Upon output, the list will be
+    # converted back into a string.
+    cdata_list_attributes = {
+        "*" : ['class', 'accesskey', 'dropzone'],
+        "a" : ['rel', 'rev'],
+        "link" :  ['rel', 'rev'],
+        "td" : ["headers"],
+        "th" : ["headers"],
+        "td" : ["headers"],
+        "form" : ["accept-charset"],
+        "object" : ["archive"],
+
+        # These are HTML5 specific, as are *.accesskey and *.dropzone above.
+        "area" : ["rel"],
+        "icon" : ["sizes"],
+        "iframe" : ["sandbox"],
+        "output" : ["for"],
+        }
+
+    def set_up_substitutions(self, tag):
+        # We are only interested in <meta> tags
+        if tag.name != 'meta':
+            return False
+
+        http_equiv = tag.get('http-equiv')
+        content = tag.get('content')
+        charset = tag.get('charset')
+
+        # We are interested in <meta> tags that say what encoding the
+        # document was originally in. This means HTML 5-style <meta>
+        # tags that provide the "charset" attribute. It also means
+        # HTML 4-style <meta> tags that provide the "content"
+        # attribute and have "http-equiv" set to "content-type".
+        #
+        # In both cases we will replace the value of the appropriate
+        # attribute with a standin object that can take on any
+        # encoding.
+        meta_encoding = None
+        if charset is not None:
+            # HTML 5 style:
+            # <meta charset="utf8">
+            meta_encoding = charset
+            tag['charset'] = CharsetMetaAttributeValue(charset)
+
+        elif (content is not None and http_equiv is not None
+              and http_equiv.lower() == 'content-type'):
+            # HTML 4 style:
+            # <meta http-equiv="content-type" content="text/html; charset=utf8">
+            tag['content'] = ContentMetaAttributeValue(content)
+
+        return (meta_encoding is not None)
+
+def register_treebuilders_from(module):
+    """Copy TreeBuilders from the given module into this module."""
+    # I'm fairly sure this is not the best way to do this.
+    this_module = sys.modules['bs4.builder']
+    for name in module.__all__:
+        obj = getattr(module, name)
+
+        if issubclass(obj, TreeBuilder):
+            setattr(this_module, name, obj)
+            this_module.__all__.append(name)
+            # Register the builder while we're at it.
+            this_module.builder_registry.register(obj)
+
+# Builders are registered in reverse order of priority, so that custom
+# builder registrations will take precedence. In general, we want lxml
+# to take precedence over html5lib, because it's faster. And we only
+# want to use HTMLParser as a last result.
+from . import _htmlparser
+register_treebuilders_from(_htmlparser)
+try:
+    from . import _html5lib
+    register_treebuilders_from(_html5lib)
+except ImportError:
+    # They don't have html5lib installed.
+    pass
+try:
+    from . import _lxml
+    register_treebuilders_from(_lxml)
+except ImportError:
+    # They don't have lxml installed.
+    pass

lib/bs4/builder/_html5lib.py

@@ -0,0 +1,222 @@
+__all__ = [
+    'HTML5TreeBuilder',
+    ]
+
+import warnings
+from bs4.builder import (
+    PERMISSIVE,
+    HTML,
+    HTML_5,
+    HTMLTreeBuilder,
+    )
+from bs4.element import NamespacedAttribute
+import html5lib
+from html5lib.constants import namespaces
+from bs4.element import (
+    Comment,
+    Doctype,
+    NavigableString,
+    Tag,
+    )
+
+class HTML5TreeBuilder(HTMLTreeBuilder):
+    """Use html5lib to build a tree."""
+
+    features = ['html5lib', PERMISSIVE, HTML_5, HTML]
+
+    def prepare_markup(self, markup, user_specified_encoding):
+        # Store the user-specified encoding for use later on.
+        self.user_specified_encoding = user_specified_encoding
+        return markup, None, None, False
+
+    # These methods are defined by Beautiful Soup.
+    def feed(self, markup):
+        if self.soup.parse_only is not None:
+            warnings.warn("You provided a value for parse_only, but the html5lib tree builder doesn't support parse_only. The entire document will be parsed.")
+        parser = html5lib.HTMLParser(tree=self.create_treebuilder)
+        doc = parser.parse(markup, encoding=self.user_specified_encoding)
+
+        # Set the character encoding detected by the tokenizer.
+        if isinstance(markup, unicode):
+            # We need to special-case this because html5lib sets
+            # charEncoding to UTF-8 if it gets Unicode input.
+            doc.original_encoding = None
+        else:
+            doc.original_encoding = parser.tokenizer.stream.charEncoding[0]
+
+    def create_treebuilder(self, namespaceHTMLElements):
+        self.underlying_builder = TreeBuilderForHtml5lib(
+            self.soup, namespaceHTMLElements)
+        return self.underlying_builder
+
+    def test_fragment_to_document(self, fragment):
+        """See `TreeBuilder`."""
+        return u'<html><head></head><body>%s</body></html>' % fragment
+
+
+class TreeBuilderForHtml5lib(html5lib.treebuilders._base.TreeBuilder):
+
+    def __init__(self, soup, namespaceHTMLElements):
+        self.soup = soup
+        super(TreeBuilderForHtml5lib, self).__init__(namespaceHTMLElements)
+
+    def documentClass(self):
+        self.soup.reset()
+        return Element(self.soup, self.soup, None)
+
+    def insertDoctype(self, token):
+        name = token["name"]
+        publicId = token["publicId"]
+        systemId = token["systemId"]
+
+        doctype = Doctype.for_name_and_ids(name, publicId, systemId)
+        self.soup.object_was_parsed(doctype)
+
+    def elementClass(self, name, namespace):
+        tag = self.soup.new_tag(name, namespace)
+        return Element(tag, self.soup, namespace)
+
+    def commentClass(self, data):
+        return TextNode(Comment(data), self.soup)
+
+    def fragmentClass(self):
+        self.soup = BeautifulSoup("")
+        self.soup.name = "[document_fragment]"
+        return Element(self.soup, self.soup, None)
+
+    def appendChild(self, node):
+        # XXX This code is not covered by the BS4 tests.
+        self.soup.append(node.element)
+
+    def getDocument(self):
+        return self.soup
+
+    def getFragment(self):
+        return html5lib.treebuilders._base.TreeBuilder.getFragment(self).element
+
+class AttrList(object):
+    def __init__(self, element):
+        self.element = element
+        self.attrs = dict(self.element.attrs)
+    def __iter__(self):
+        return list(self.attrs.items()).__iter__()
+    def __setitem__(self, name, value):
+        "set attr", name, value
+        self.element[name] = value
+    def items(self):
+        return list(self.attrs.items())
+    def keys(self):
+        return list(self.attrs.keys())
+    def __len__(self):
+        return len(self.attrs)
+    def __getitem__(self, name):
+        return self.attrs[name]
+    def __contains__(self, name):
+        return name in list(self.attrs.keys())
+
+
+class Element(html5lib.treebuilders._base.Node):
+    def __init__(self, element, soup, namespace):
+        html5lib.treebuilders._base.Node.__init__(self, element.name)
+        self.element = element
+        self.soup = soup
+        self.namespace = namespace
+
+    def appendChild(self, node):
+        if (node.element.__class__ == NavigableString and self.element.contents
+            and self.element.contents[-1].__class__ == NavigableString):
+            # Concatenate new text onto old text node
+            # XXX This has O(n^2) performance, for input like
+            # "a</a>a</a>a</a>..."
+            old_element = self.element.contents[-1]
+            new_element = self.soup.new_string(old_element + node.element)
+            old_element.replace_with(new_element)
+        else:
+            self.element.append(node.element)
+            node.parent = self
+
+    def getAttributes(self):
+        return AttrList(self.element)
+
+    def setAttributes(self, attributes):
+        if attributes is not None and len(attributes) > 0:
+
+            converted_attributes = []
+            for name, value in list(attributes.items()):
+                if isinstance(name, tuple):
+                    new_name = NamespacedAttribute(*name)
+                    del attributes[name]
+                    attributes[new_name] = value
+
+            self.soup.builder._replace_cdata_list_attribute_values(
+                self.name, attributes)
+            for name, value in attributes.items():
+                self.element[name] = value
+
+            # The attributes may contain variables that need substitution.
+            # Call set_up_substitutions manually.
+            #
+            # The Tag constructor called this method when the Tag was created,
+            # but we just set/changed the attributes, so call it again.
+            self.soup.builder.set_up_substitutions(self.element)
+    attributes = property(getAttributes, setAttributes)
+
+    def insertText(self, data, insertBefore=None):
+        text = TextNode(self.soup.new_string(data), self.soup)
+        if insertBefore:
+            self.insertBefore(text, insertBefore)
+        else:
+            self.appendChild(text)
+
+    def insertBefore(self, node, refNode):
+        index = self.element.index(refNode.element)
+        if (node.element.__class__ == NavigableString and self.element.contents
+            and self.element.contents[index-1].__class__ == NavigableString):
+            # (See comments in appendChild)
+            old_node = self.element.contents[index-1]
+            new_str = self.soup.new_string(old_node + node.element)
+            old_node.replace_with(new_str)
+        else:
+            self.element.insert(index, node.element)
+            node.parent = self
+
+    def removeChild(self, node):
+        node.element.extract()
+
+    def reparentChildren(self, newParent):
+        while self.element.contents:
+            child = self.element.contents[0]
+            child.extract()
+            if isinstance(child, Tag):
+                newParent.appendChild(
+                    Element(child, self.soup, namespaces["html"]))
+            else:
+                newParent.appendChild(
+                    TextNode(child, self.soup))
+
+    def cloneNode(self):
+        tag = self.soup.new_tag(self.element.name, self.namespace)
+        node = Element(tag, self.soup, self.namespace)
+        for key,value in self.attributes:
+            node.attributes[key] = value
+        return node
+
+    def hasContent(self):
+        return self.element.contents
+
+    def getNameTuple(self):
+        if self.namespace == None:
+            return namespaces["html"], self.name
+        else:
+            return self.namespace, self.name
+
+    nameTuple = property(getNameTuple)
+
+class TextNode(Element):
+    def __init__(self, element, soup):
+        html5lib.treebuilders._base.Node.__init__(self, None)
+        self.element = element
+        self.soup = soup
+
+    def cloneNode(self):
+        raise NotImplementedError

lib/bs4/builder/_htmlparser.py

@@ -0,0 +1,244 @@
+"""Use the HTMLParser library to parse HTML files that aren't too bad."""
+
+__all__ = [
+    'HTMLParserTreeBuilder',
+    ]
+
+from HTMLParser import (
+    HTMLParser,
+    HTMLParseError,
+    )
+import sys
+import warnings
+
+# Starting in Python 3.2, the HTMLParser constructor takes a 'strict'
+# argument, which we'd like to set to False. Unfortunately,
+# http://bugs.python.org/issue13273 makes strict=True a better bet
+# before Python 3.2.3.
+#
+# At the end of this file, we monkeypatch HTMLParser so that
+# strict=True works well on Python 3.2.2.
+major, minor, release = sys.version_info[:3]
+CONSTRUCTOR_TAKES_STRICT = (
+    major > 3
+    or (major == 3 and minor > 2)
+    or (major == 3 and minor == 2 and release >= 3))
+
+from bs4.element import (
+    CData,
+    Comment,
+    Declaration,
+    Doctype,
+    ProcessingInstruction,
+    )
+from bs4.dammit import EntitySubstitution, UnicodeDammit
+
+from bs4.builder import (
+    HTML,
+    HTMLTreeBuilder,
+    STRICT,
+    )
+
+
+HTMLPARSER = 'html.parser'
+
+class BeautifulSoupHTMLParser(HTMLParser):
+    def handle_starttag(self, name, attrs):
+        # XXX namespace
+        self.soup.handle_starttag(name, None, None, dict(attrs))
+
+    def handle_endtag(self, name):
+        self.soup.handle_endtag(name)
+
+    def handle_data(self, data):
+        self.soup.handle_data(data)
+
+    def handle_charref(self, name):
+        # XXX workaround for a bug in HTMLParser. Remove this once
+        # it's fixed.
+        if name.startswith('x'):
+            real_name = int(name.lstrip('x'), 16)
+        else:
+            real_name = int(name)
+
+        try:
+            data = unichr(real_name)
+        except (ValueError, OverflowError), e:
+            data = u"\N{REPLACEMENT CHARACTER}"
+
+        self.handle_data(data)
+
+    def handle_entityref(self, name):
+        character = EntitySubstitution.HTML_ENTITY_TO_CHARACTER.get(name)
+        if character is not None:
+            data = character
+        else:
+            data = "&%s;" % name
+        self.handle_data(data)
+
+    def handle_comment(self, data):
+        self.soup.endData()
+        self.soup.handle_data(data)
+        self.soup.endData(Comment)
+
+    def handle_decl(self, data):
+        self.soup.endData()
+        if data.startswith("DOCTYPE "):
+            data = data[len("DOCTYPE "):]
+        self.soup.handle_data(data)
+        self.soup.endData(Doctype)
+
+    def unknown_decl(self, data):
+        if data.upper().startswith('CDATA['):
+            cls = CData
+            data = data[len('CDATA['):]
+        else:
+            cls = Declaration
+        self.soup.endData()
+        self.soup.handle_data(data)
+        self.soup.endData(cls)
+
+    def handle_pi(self, data):
+        self.soup.endData()
+        if data.endswith("?") and data.lower().startswith("xml"):
+            # "An XHTML processing instruction using the trailing '?'
+            # will cause the '?' to be included in data." - HTMLParser
+            # docs.
+            #
+            # Strip the question mark so we don't end up with two
+            # question marks.
+            data = data[:-1]
+        self.soup.handle_data(data)
+        self.soup.endData(ProcessingInstruction)
+
+
+class HTMLParserTreeBuilder(HTMLTreeBuilder):
+
+    is_xml = False
+    features = [HTML, STRICT, HTMLPARSER]
+
+    def __init__(self, *args, **kwargs):
+        if CONSTRUCTOR_TAKES_STRICT:
+            kwargs['strict'] = False
+        self.parser_args = (args, kwargs)
+
+    def prepare_markup(self, markup, user_specified_encoding=None,
+                       document_declared_encoding=None):
+        """
+        :return: A 4-tuple (markup, original encoding, encoding
+        declared within markup, whether any characters had to be
+        replaced with REPLACEMENT CHARACTER).
+        """
+        if isinstance(markup, unicode):
+            return markup, None, None, False
+
+        try_encodings = [user_specified_encoding, document_declared_encoding]
+        dammit = UnicodeDammit(markup, try_encodings, is_html=True)
+        return (dammit.markup, dammit.original_encoding,
+                dammit.declared_html_encoding,
+                dammit.contains_replacement_characters)
+
+    def feed(self, markup):
+        args, kwargs = self.parser_args
+        parser = BeautifulSoupHTMLParser(*args, **kwargs)
+        parser.soup = self.soup
+        try:
+            parser.feed(markup)
+        except HTMLParseError, e:
+            warnings.warn(RuntimeWarning(
+                "Python's built-in HTMLParser cannot parse the given document. This is not a bug in Beautiful Soup. The best solution is to install an external parser (lxml or html5lib), and use Beautiful Soup with that parser. See http://www.crummy.com/software/BeautifulSoup/bs4/doc/#installing-a-parser for help."))
+            raise e
+
+# Patch 3.2 versions of HTMLParser earlier than 3.2.3 to use some
+# 3.2.3 code. This ensures they don't treat markup like <p></p> as a
+# string.
+#
+# XXX This code can be removed once most Python 3 users are on 3.2.3.
+if major == 3 and minor == 2 and not CONSTRUCTOR_TAKES_STRICT:
+    import re
+    attrfind_tolerant = re.compile(
+        r'\s*((?<=[\'"\s])[^\s/>][^\s/=>]*)(\s*=+\s*'
+        r'(\'[^\']*\'|"[^"]*"|(?![\'"])[^>\s]*))?')
+    HTMLParserTreeBuilder.attrfind_tolerant = attrfind_tolerant
+
+    locatestarttagend = re.compile(r"""
+  <[a-zA-Z][-.a-zA-Z0-9:_]*          # tag name
+  (?:\s+                             # whitespace before attribute name
+    (?:[a-zA-Z_][-.:a-zA-Z0-9_]*     # attribute name
+      (?:\s*=\s*                     # value indicator
+        (?:'[^']*'                   # LITA-enclosed value
+          |\"[^\"]*\"                # LIT-enclosed value
+          |[^'\">\s]+                # bare value
+         )
+       )?
+     )
+   )*
+  \s*                                # trailing whitespace
+""", re.VERBOSE)
+    BeautifulSoupHTMLParser.locatestarttagend = locatestarttagend
+
+    from html.parser import tagfind, attrfind
+
+    def parse_starttag(self, i):
+        self.__starttag_text = None
+        endpos = self.check_for_whole_start_tag(i)
+        if endpos < 0:
+            return endpos
+        rawdata = self.rawdata
+        self.__starttag_text = rawdata[i:endpos]
+
+        # Now parse the data between i+1 and j into a tag and attrs
+        attrs = []
+        match = tagfind.match(rawdata, i+1)
+        assert match, 'unexpected call to parse_starttag()'
+        k = match.end()
+        self.lasttag = tag = rawdata[i+1:k].lower()
+        while k < endpos:
+            if self.strict:
+                m = attrfind.match(rawdata, k)
+            else:
+                m = attrfind_tolerant.match(rawdata, k)
+            if not m:
+                break
+            attrname, rest, attrvalue = m.group(1, 2, 3)
+            if not rest:
+                attrvalue = None
+            elif attrvalue[:1] == '\'' == attrvalue[-1:] or \
+                 attrvalue[:1] == '"' == attrvalue[-1:]:
+                attrvalue = attrvalue[1:-1]
+            if attrvalue:
+                attrvalue = self.unescape(attrvalue)
+            attrs.append((attrname.lower(), attrvalue))
+            k = m.end()
+
+        end = rawdata[k:endpos].strip()
+        if end not in (">", "/>"):
+            lineno, offset = self.getpos()
+            if "\n" in self.__starttag_text:
+                lineno = lineno + self.__starttag_text.count("\n")
+                offset = len(self.__starttag_text) \
+                         - self.__starttag_text.rfind("\n")
+            else:
+                offset = offset + len(self.__starttag_text)
+            if self.strict:
+                self.error("junk characters in start tag: %r"
+                           % (rawdata[k:endpos][:20],))
+            self.handle_data(rawdata[i:endpos])
+            return endpos
+        if end.endswith('/>'):
+            # XHTML-style empty tag: <span attr="value" />
+            self.handle_startendtag(tag, attrs)
+        else:
+            self.handle_starttag(tag, attrs)
+            if tag in self.CDATA_CONTENT_ELEMENTS:
+                self.set_cdata_mode(tag)
+        return endpos
+
+    def set_cdata_mode(self, elem):
+        self.cdata_elem = elem.lower()
+        self.interesting = re.compile(r'</\s*%s\s*>' % self.cdata_elem, re.I)
+
+    BeautifulSoupHTMLParser.parse_starttag = parse_starttag
+    BeautifulSoupHTMLParser.set_cdata_mode = set_cdata_mode
+
+    CONSTRUCTOR_TAKES_STRICT = True

lib/bs4/builder/_lxml.py

@@ -0,0 +1,199 @@
+__all__ = [
+    'LXMLTreeBuilderForXML',
+    'LXMLTreeBuilder',
+    ]
+
+from StringIO import StringIO
+import collections
+from lxml import etree
+from bs4.element import Comment, Doctype, NamespacedAttribute
+from bs4.builder import (
+    FAST,
+    HTML,
+    HTMLTreeBuilder,
+    PERMISSIVE,
+    TreeBuilder,
+    XML)
+from bs4.dammit import UnicodeDammit
+
+LXML = 'lxml'
+
+class LXMLTreeBuilderForXML(TreeBuilder):
+    DEFAULT_PARSER_CLASS = etree.XMLParser
+
+    is_xml = True
+
+    # Well, it's permissive by XML parser standards.
+    features = [LXML, XML, FAST, PERMISSIVE]
+
+    CHUNK_SIZE = 512
+
+    @property
+    def default_parser(self):
+        # This can either return a parser object or a class, which
+        # will be instantiated with default arguments.
+        return etree.XMLParser(target=self, strip_cdata=False, recover=True)
+
+    def __init__(self, parser=None, empty_element_tags=None):
+        if empty_element_tags is not None:
+            self.empty_element_tags = set(empty_element_tags)
+        if parser is None:
+            # Use the default parser.
+            parser = self.default_parser
+        if isinstance(parser, collections.Callable):
+            # Instantiate the parser with default arguments
+            parser = parser(target=self, strip_cdata=False)
+        self.parser = parser
+        self.soup = None
+        self.nsmaps = None
+
+    def _getNsTag(self, tag):
+        # Split the namespace URL out of a fully-qualified lxml tag
+        # name. Copied from lxml's src/lxml/sax.py.
+        if tag[0] == '{':
+            return tuple(tag[1:].split('}', 1))
+        else:
+            return (None, tag)
+
+    def prepare_markup(self, markup, user_specified_encoding=None,
+                       document_declared_encoding=None):
+        """
+        :return: A 3-tuple (markup, original encoding, encoding
+        declared within markup).
+        """
+        if isinstance(markup, unicode):
+            return markup, None, None, False
+
+        try_encodings = [user_specified_encoding, document_declared_encoding]
+        dammit = UnicodeDammit(markup, try_encodings, is_html=True)
+        return (dammit.markup, dammit.original_encoding,
+                dammit.declared_html_encoding,
+                dammit.contains_replacement_characters)
+
+    def feed(self, markup):
+        if isinstance(markup, basestring):
+            markup = StringIO(markup)
+        # Call feed() at least once, even if the markup is empty,
+        # or the parser won't be initialized.
+        data = markup.read(self.CHUNK_SIZE)
+        self.parser.feed(data)
+        while data != '':
+            # Now call feed() on the rest of the data, chunk by chunk.
+            data = markup.read(self.CHUNK_SIZE)
+            if data != '':
+                self.parser.feed(data)
+        self.parser.close()
+
+    def close(self):
+        self.nsmaps = None
+
+    def start(self, name, attrs, nsmap={}):
+        # Make sure attrs is a mutable dict--lxml may send an immutable dictproxy.
+        attrs = dict(attrs)
+
+        nsprefix = None
+        # Invert each namespace map as it comes in.
+        if len(nsmap) == 0 and self.nsmaps != None:
+            # There are no new namespaces for this tag, but namespaces
+            # are in play, so we need a separate tag stack to know
+            # when they end.
+            self.nsmaps.append(None)
+        elif len(nsmap) > 0:
+            # A new namespace mapping has come into play.
+            if self.nsmaps is None:
+                self.nsmaps = []
+            inverted_nsmap = dict((value, key) for key, value in nsmap.items())
+            self.nsmaps.append(inverted_nsmap)
+            # Also treat the namespace mapping as a set of attributes on the
+            # tag, so we can recreate it later.
+            attrs = attrs.copy()
+            for prefix, namespace in nsmap.items():
+                attribute = NamespacedAttribute(
+                    "xmlns", prefix, "http://www.w3.org/2000/xmlns/")
+                attrs[attribute] = namespace
+
+        if self.nsmaps is not None and len(self.nsmaps) > 0:
+            # Namespaces are in play. Find any attributes that came in
+            # from lxml with namespaces attached to their names, and
+            # turn then into NamespacedAttribute objects.
+            new_attrs = {}
+            for attr, value in attrs.items():
+                namespace, attr = self._getNsTag(attr)
+                if namespace is None:
+                    new_attrs[attr] = value
+                else:
+                    nsprefix = self._prefix_for_namespace(namespace)
+                    attr = NamespacedAttribute(nsprefix, attr, namespace)
+                    new_attrs[attr] = value
+            attrs = new_attrs
+
+        namespace, name = self._getNsTag(name)
+        nsprefix = self._prefix_for_namespace(namespace)
+        self.soup.handle_starttag(name, namespace, nsprefix, attrs)
+
+    def _prefix_for_namespace(self, namespace):
+        """Find the currently active prefix for the given namespace."""
+        if namespace is None:
+            return None
+        for inverted_nsmap in reversed(self.nsmaps):
+            if inverted_nsmap is not None and namespace in inverted_nsmap:
+                return inverted_nsmap[namespace]
+
+    def end(self, name):
+        self.soup.endData()
+        completed_tag = self.soup.tagStack[-1]
+        namespace, name = self._getNsTag(name)
+        nsprefix = None
+        if namespace is not None:
+            for inverted_nsmap in reversed(self.nsmaps):
+                if inverted_nsmap is not None and namespace in inverted_nsmap:
+                    nsprefix = inverted_nsmap[namespace]
+                    break
+        self.soup.handle_endtag(name, nsprefix)
+        if self.nsmaps != None:
+            # This tag, or one of its parents, introduced a namespace
+            # mapping, so pop it off the stack.
+            self.nsmaps.pop()
+            if len(self.nsmaps) == 0:
+                # Namespaces are no longer in play, so don't bother keeping
+                # track of the namespace stack.
+                self.nsmaps = None
+
+    def pi(self, target, data):
+        pass
+
+    def data(self, content):
+        self.soup.handle_data(content)
+
+    def doctype(self, name, pubid, system):
+        self.soup.endData()
+        doctype = Doctype.for_name_and_ids(name, pubid, system)
+        self.soup.object_was_parsed(doctype)
+
+    def comment(self, content):
+        "Handle comments as Comment objects."
+        self.soup.endData()
+        self.soup.handle_data(content)
+        self.soup.endData(Comment)
+
+    def test_fragment_to_document(self, fragment):
+        """See `TreeBuilder`."""
+        return u'<?xml version="1.0" encoding="utf-8"?>\n%s' % fragment
+
+
+class LXMLTreeBuilder(HTMLTreeBuilder, LXMLTreeBuilderForXML):
+
+    features = [LXML, HTML, FAST, PERMISSIVE]
+    is_xml = False
+
+    @property
+    def default_parser(self):
+        return etree.HTMLParser
+
+    def feed(self, markup):
+        self.parser.feed(markup)
+        self.parser.close()
+
+    def test_fragment_to_document(self, fragment):
+        """See `TreeBuilder`."""
+        return u'<html><body>%s</body></html>' % fragment