BeautifulSoup Parser

BeautifulSoup is a Python package for working with real-world and broken HTML, just like lxml.html. As of version 4.x, it can use different HTML parsers, each of which has its advantages and disadvantages (see the link).

lxml can make use of BeautifulSoup as a parser backend, just like BeautifulSoup can employ lxml as a parser. When using BeautifulSoup from lxml, however, the default is to use Python's integrated HTML parser in the html.parser module. In order to make use of the HTML5 parser of html5lib instead, it is better to go directly through the html5parser module in lxml.html.

A very nice feature of BeautifulSoup is its excellent support for encoding detection which can provide better results for real-world HTML pages that do not (correctly) declare their encoding.

lxml interfaces with BeautifulSoup through the lxml.html.soupparser module. It provides three main functions: fromstring() and parse() to parse a string or file using BeautifulSoup into an lxml.html document, and convert_tree() to convert an existing BeautifulSoup tree into a list of top-level Elements.


Parsing with the soupparser

The functions fromstring() and parse() behave as known from lxml. The first returns a root Element, the latter returns an ElementTree.

There is also a legacy module called lxml.html.ElementSoup, which mimics the interface provided by Fredrik Lundh's ElementSoup module. Note that the soupparser module was added in lxml 2.0.3. Previous versions of lxml 2.0.x only have the ElementSoup module.

Here is a document full of tag soup, similar to, but not quite like, HTML:

>>> tag_soup = '''
... <meta/><head><title>Hello</head><body onload=crash()>Hi all<p>'''

All you need to do is pass it to the fromstring() function:

>>> from lxml.html.soupparser import fromstring
>>> root = fromstring(tag_soup)

To see what we have here, you can serialise it:

>>> from lxml.etree import tostring
>>> print(tostring(root, pretty_print=True).strip())
  <body onload="crash()">Hi all<p/></body>

Not quite what you'd expect from an HTML page, but, well, it was broken already, right? The parser did its best, and so now it's a tree.

To control how Element objects are created during the conversion of the tree, you can pass a makeelement factory function to parse() and fromstring(). By default, this is based on the HTML parser defined in lxml.html.

For a quick comparison, libxml2 2.9.1 parses the same tag soup as follows. The only difference is that libxml2 tries harder to adhere to the structure of an HTML document and moves misplaced tags where they (likely) belong. Note, however, that the result can vary between parser versions.

  <body onload="crash()">Hi all<p/></body>

Entity handling

By default, the BeautifulSoup parser also replaces the entities it finds by their character equivalent.

>>> tag_soup = '<body>&copy;&euro;&#45;&#245;&#445;<p>'
>>> body = fromstring(tag_soup).find('.//body')
>>> body.text

If you want them back on the way out, you can just serialise with the default encoding, which is 'US-ASCII'.

>>> tostring(body)

>>> tostring(body, method="html")

Any other encoding will output the respective byte sequences.

>>> tostring(body, encoding="utf-8")

>>> tostring(body, method="html", encoding="utf-8")

>>> tostring(body, encoding='unicode')

>>> tostring(body, method="html", encoding='unicode')

Using soupparser as a fallback

The downside of using this parser is that it is much slower than the C implemented HTML parser of libxml2 that lxml uses. So if performance matters, you might want to consider using soupparser only as a fallback for certain cases.

One common problem of lxml's parser is that it might not get the encoding right in cases where the document contains a <meta> tag at the wrong place. In this case, you can exploit the fact that lxml serialises much faster than most other HTML libraries for Python. Just serialise the document to unicode and if that gives you an exception, re-parse it with BeautifulSoup to see if that works better.

>>> tag_soup = '''\
... <meta http-equiv="Content-Type"
...       content="text/html;charset=utf-8" />
... <html>
...   <head>
...     <title>Hello W\xc3\xb6rld!</title>
...   </head>
...   <body>Hi all</body>
... </html>'''

>>> import lxml.html
>>> import lxml.html.soupparser

>>> root = lxml.html.fromstring(tag_soup)
>>> try:
...     ignore = tostring(root, encoding='unicode')
... except UnicodeDecodeError:
...     root = lxml.html.soupparser.fromstring(tag_soup)

Using only the encoding detection

Even if you prefer lxml's fast HTML parser, you can still benefit from BeautifulSoup's support for encoding detection in the UnicodeDammit class. Once it succeeds in decoding the data, you can simply pass the resulting Unicode string into lxml's parser.

>>> try:
...    from bs4 import UnicodeDammit             # BeautifulSoup 4
...    def decode_html(html_string):
...        converted = UnicodeDammit(html_string)
...        if not converted.unicode_markup:
...            raise UnicodeDecodeError(
...                "Failed to detect encoding, tried [%s]",
...                ', '.join(converted.tried_encodings))
...        # print converted.original_encoding
...        return converted.unicode_markup
... except ImportError:
...    from BeautifulSoup import UnicodeDammit   # BeautifulSoup 3
...    def decode_html(html_string):
...        converted = UnicodeDammit(html_string, isHTML=True)
...        if not converted.unicode:
...            raise UnicodeDecodeError(
...                "Failed to detect encoding, tried [%s]",
...                ', '.join(converted.triedEncodings))
...        # print converted.originalEncoding
...        return converted.unicode

>>> root = lxml.html.fromstring(decode_html(tag_soup))