Beautiful Soup Documentation

"The Fish-Footman began by producing from under his arm a great letter, nearly as large as himself."

Beautiful Soup is a Python library for pulling data out of HTML and XML files. It works with your favorite parser to provide idiomatic ways of navigating, searching, and modifying the parse tree. It commonly saves programmers hours or days of work.

These instructions illustrate all major features of Beautiful Soup 4, with examples. I show you what the library is good for, how it works, how to use it, how to make it do what you want, and what to do when it violates your expectations.

This document covers Beautiful Soup version 4.11.2. The examples in this documentation were written for Python 3.8.

You might be looking for the documentation for Beautiful Soup 3. If so, you should know that Beautiful Soup 3 is no longer being developed and that all support for it was dropped on December 31, 2020. If you want to learn about the differences between Beautiful Soup 3 and Beautiful Soup 4, see Porting code to BS4.

This documentation has been translated into other languages by Beautiful Soup users:

Getting help

If you have questions about Beautiful Soup, or run into problems, send mail to the discussion group. If your problem involves parsing an HTML document, be sure to mention what the diagnose() function says about that document.

Quick Start

Here’s an HTML document I’ll be using as an example throughout this document. It’s part of a story from Alice in Wonderland:

html_doc = """<html><head><title>The Dormouse's story</title></head>
<body>
<p class="title"><b>The Dormouse's story</b></p>

<p class="story">Once upon a time there were three little sisters; and their names were
<a href="http://example.com/elsie" class="sister" id="link1">Elsie</a>,
<a href="http://example.com/lacie" class="sister" id="link2">Lacie</a> and
<a href="http://example.com/tillie" class="sister" id="link3">Tillie</a>;
and they lived at the bottom of a well.</p>

<p class="story">...</p>
"""

Running the “three sisters” document through Beautiful Soup gives us a BeautifulSoup object, which represents the document as a nested data structure:

from bs4 import BeautifulSoup
soup = BeautifulSoup(html_doc, 'html.parser')

print(soup.prettify())
# <html>
#  <head>
#   <title>
#    The Dormouse's story
#   </title>
#  </head>
#  <body>
#   <p class="title">
#    <b>
#     The Dormouse's story
#    </b>
#   </p>
#   <p class="story">
#    Once upon a time there were three little sisters; and their names were
#    <a class="sister" href="http://example.com/elsie" id="link1">
#     Elsie
#    </a>
#    ,
#    <a class="sister" href="http://example.com/lacie" id="link2">
#     Lacie
#    </a>
#    and
#    <a class="sister" href="http://example.com/tillie" id="link3">
#     Tillie
#    </a>
#    ; and they lived at the bottom of a well.
#   </p>
#   <p class="story">
#    ...
#   </p>
#  </body>
# </html>

Here are some simple ways to navigate that data structure:

soup.title
# <title>The Dormouse's story</title>

soup.title.name
# u'title'

soup.title.string
# u'The Dormouse's story'

soup.title.parent.name
# u'head'

soup.p
# <p class="title"><b>The Dormouse's story</b></p>

soup.p['class']
# u'title'

soup.a
# <a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>

soup.find_all('a')
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

soup.find(id="link3")
# <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>

One common task is extracting all the URLs found within a page’s <a> tags:

for link in soup.find_all('a'):
    print(link.get('href'))
# http://example.com/elsie
# http://example.com/lacie
# http://example.com/tillie

Another common task is extracting all the text from a page:

print(soup.get_text())
# The Dormouse's story
#
# The Dormouse's story
#
# Once upon a time there were three little sisters; and their names were
# Elsie,
# Lacie and
# Tillie;
# and they lived at the bottom of a well.
#
# ...

Does this look like what you need? If so, read on.

Installing Beautiful Soup

If you’re using a recent version of Debian or Ubuntu Linux, you can install Beautiful Soup with the system package manager:

$ apt-get install python3-bs4

Beautiful Soup 4 is published through PyPi, so if you can’t install it with the system packager, you can install it with easy_install or pip. The package name is beautifulsoup4. Make sure you use the right version of pip or easy_install for your Python version (these may be named pip3 and easy_install3 respectively).

$ easy_install beautifulsoup4

$ pip install beautifulsoup4

(The BeautifulSoup package is not what you want. That’s the previous major release, Beautiful Soup 3. Lots of software uses BS3, so it’s still available, but if you’re writing new code you should install beautifulsoup4.)

If you don’t have easy_install or pip installed, you can download the Beautiful Soup 4 source tarball and install it with setup.py.

$ python setup.py install

If all else fails, the license for Beautiful Soup allows you to package the entire library with your application. You can download the tarball, copy its bs4 directory into your application’s codebase, and use Beautiful Soup without installing it at all.

I use Python 3.8 to develop Beautiful Soup, but it should work with other recent versions.

Installing a parser

Beautiful Soup supports the HTML parser included in Python’s standard library, but it also supports a number of third-party Python parsers. One is the lxml parser. Depending on your setup, you might install lxml with one of these commands:

$ apt-get install python-lxml

$ easy_install lxml

$ pip install lxml

Another alternative is the pure-Python html5lib parser, which parses HTML the way a web browser does. Depending on your setup, you might install html5lib with one of these commands:

$ apt-get install python-html5lib

$ easy_install html5lib

$ pip install html5lib

This table summarizes the advantages and disadvantages of each parser library:

Parser

Typical usage

Advantages

Disadvantages

Python’s html.parser

BeautifulSoup(markup, "html.parser")

  • Batteries included

  • Decent speed

  • Lenient (As of Python 3.2)

  • Not as fast as lxml, less lenient than html5lib.

lxml’s HTML parser

BeautifulSoup(markup, "lxml")

  • Very fast

  • Lenient

  • External C dependency

lxml’s XML parser

BeautifulSoup(markup, "lxml-xml") BeautifulSoup(markup, "xml")

  • Very fast

  • The only currently supported XML parser

  • External C dependency

html5lib

BeautifulSoup(markup, "html5lib")

  • Extremely lenient

  • Parses pages the same way a web browser does

  • Creates valid HTML5

  • Very slow

  • External Python dependency

If you can, I recommend you install and use lxml for speed. If you’re using a very old version of Python – earlier than 3.2.2 – it’s essential that you install lxml or html5lib. Python’s built-in HTML parser is just not very good in those old versions.

Note that if a document is invalid, different parsers will generate different Beautiful Soup trees for it. See Differences between parsers for details.

Making the soup

To parse a document, pass it into the BeautifulSoup constructor. You can pass in a string or an open filehandle:

from bs4 import BeautifulSoup

with open("index.html") as fp:
    soup = BeautifulSoup(fp, 'html.parser')

soup = BeautifulSoup("<html>a web page</html>", 'html.parser')

First, the document is converted to Unicode, and HTML entities are converted to Unicode characters:

print(BeautifulSoup("<html><head></head><body>Sacr&eacute; bleu!</body></html>", "html.parser"))
# <html><head></head><body>Sacré bleu!</body></html>

Beautiful Soup then parses the document using the best available parser. It will use an HTML parser unless you specifically tell it to use an XML parser. (See Parsing XML.)

Kinds of objects

Beautiful Soup transforms a complex HTML document into a complex tree of Python objects. But you’ll only ever have to deal with about four kinds of objects: Tag, NavigableString, BeautifulSoup, and Comment.

Tag

A Tag object corresponds to an XML or HTML tag in the original document:

soup = BeautifulSoup('<b class="boldest">Extremely bold</b>', 'html.parser')
tag = soup.b
type(tag)
# <class 'bs4.element.Tag'>

Tags have a lot of attributes and methods, and I’ll cover most of them in Navigating the tree and Searching the tree. For now, the most important features of a tag are its name and attributes.

Name

Every tag has a name, accessible as .name:

tag.name
# 'b'

If you change a tag’s name, the change will be reflected in any HTML markup generated by Beautiful Soup:

tag.name = "blockquote"
tag
# <blockquote class="boldest">Extremely bold</blockquote>

Attributes

A tag may have any number of attributes. The tag <b id="boldest"> has an attribute “id” whose value is “boldest”. You can access a tag’s attributes by treating the tag like a dictionary:

tag = BeautifulSoup('<b id="boldest">bold</b>', 'html.parser').b
tag['id']
# 'boldest'

You can access that dictionary directly as .attrs:

tag.attrs
# {'id': 'boldest'}

You can add, remove, and modify a tag’s attributes. Again, this is done by treating the tag as a dictionary:

tag['id'] = 'verybold'
tag['another-attribute'] = 1
tag
# <b another-attribute="1" id="verybold"></b>

del tag['id']
del tag['another-attribute']
tag
# <b>bold</b>

tag['id']
# KeyError: 'id'
tag.get('id')
# None

Multi-valued attributes

HTML 4 defines a few attributes that can have multiple values. HTML 5 removes a couple of them, but defines a few more. The most common multi-valued attribute is class (that is, a tag can have more than one CSS class). Others include rel, rev, accept-charset, headers, and accesskey. By default, Beautiful Soup parses the value(s) of a multi-valued attribute into a list:

css_soup = BeautifulSoup('<p class="body"></p>', 'html.parser')
css_soup.p['class']
# ['body']

css_soup = BeautifulSoup('<p class="body strikeout"></p>', 'html.parser')
css_soup.p['class']
# ['body', 'strikeout']

If an attribute looks like it has more than one value, but it’s not a multi-valued attribute as defined by any version of the HTML standard, Beautiful Soup will leave the attribute alone:

id_soup = BeautifulSoup('<p id="my id"></p>', 'html.parser')
id_soup.p['id']
# 'my id'

When you turn a tag back into a string, multiple attribute values are consolidated:

rel_soup = BeautifulSoup('<p>Back to the <a rel="index first">homepage</a></p>', 'html.parser')
rel_soup.a['rel']
# ['index', 'first']
rel_soup.a['rel'] = ['index', 'contents']
print(rel_soup.p)
# <p>Back to the <a rel="index contents">homepage</a></p>

You can force all attributes to be parsed as strings by passing multi_valued_attributes=None as a keyword argument into the BeautifulSoup constructor:

no_list_soup = BeautifulSoup('<p class="body strikeout"></p>', 'html.parser', multi_valued_attributes=None)
no_list_soup.p['class']
# 'body strikeout'

You can use get_attribute_list to get a value that’s always a list, whether or not it’s a multi-valued atribute:

id_soup.p.get_attribute_list('id')
# ["my id"]

If you parse a document as XML, there are no multi-valued attributes:

xml_soup = BeautifulSoup('<p class="body strikeout"></p>', 'xml')
xml_soup.p['class']
# 'body strikeout'

Again, you can configure this using the multi_valued_attributes argument:

class_is_multi= { '*' : 'class'}
xml_soup = BeautifulSoup('<p class="body strikeout"></p>', 'xml', multi_valued_attributes=class_is_multi)
xml_soup.p['class']
# ['body', 'strikeout']

You probably won’t need to do this, but if you do, use the defaults as a guide. They implement the rules described in the HTML specification:

from bs4.builder import builder_registry
builder_registry.lookup('html').DEFAULT_CDATA_LIST_ATTRIBUTES

BeautifulSoup

The BeautifulSoup object represents the parsed document as a whole. For most purposes, you can treat it as a Tag object. This means it supports most of the methods described in Navigating the tree and Searching the tree.

You can also pass a BeautifulSoup object into one of the methods defined in Modifying the tree, just as you would a Tag. This lets you do things like combine two parsed documents:

doc = BeautifulSoup("<document><content/>INSERT FOOTER HERE</document", "xml")
footer = BeautifulSoup("<footer>Here's the footer</footer>", "xml")
doc.find(text="INSERT FOOTER HERE").replace_with(footer)
# 'INSERT FOOTER HERE'
print(doc)
# <?xml version="1.0" encoding="utf-8"?>
# <document><content/><footer>Here's the footer</footer></document>

Since the BeautifulSoup object doesn’t correspond to an actual HTML or XML tag, it has no name and no attributes. But sometimes it’s useful to look at its .name, so it’s been given the special .name “[document]”:

soup.name
# '[document]'

Comments and other special strings

Tag, NavigableString, and BeautifulSoup cover almost everything you’ll see in an HTML or XML file, but there are a few leftover bits. The main one you’ll probably encounter is the comment:

markup = "<b><!--Hey, buddy. Want to buy a used parser?--></b>"
soup = BeautifulSoup(markup, 'html.parser')
comment = soup.b.string
type(comment)
# <class 'bs4.element.Comment'>

The Comment object is just a special type of NavigableString:

comment
# 'Hey, buddy. Want to buy a used parser'

But when it appears as part of an HTML document, a Comment is displayed with special formatting:

print(soup.b.prettify())
# <b>
#  <!--Hey, buddy. Want to buy a used parser?-->
# </b>

Beautiful Soup also defines classes called Stylesheet, Script, and TemplateString, for embedded CSS stylesheets (any strings found inside a <style> tag), embedded Javascript (any strings found in a <script> tag), and HTML templates (any strings inside a <template> tag). These classes work exactly the same way as NavigableString; their only purpose is to make it easier to pick out the main body of the page, by ignoring strings that represent something else. (These classes are new in Beautiful Soup 4.9.0, and the html5lib parser doesn’t use them.)

Beautiful Soup defines classes for anything else that might show up in an XML document: CData, ProcessingInstruction, Declaration, and Doctype. Like Comment, these classes are subclasses of NavigableString that add something extra to the string. Here’s an example that replaces the comment with a CDATA block:

from bs4 import CData
cdata = CData("A CDATA block")
comment.replace_with(cdata)

print(soup.b.prettify())
# <b>
#  <![CDATA[A CDATA block]]>
# </b>

Searching the tree

Beautiful Soup defines a lot of methods for searching the parse tree, but they’re all very similar. I’m going to spend a lot of time explaining the two most popular methods: find() and find_all(). The other methods take almost exactly the same arguments, so I’ll just cover them briefly.

Once again, I’ll be using the “three sisters” document as an example:

html_doc = """
<html><head><title>The Dormouse's story</title></head>
<body>
<p class="title"><b>The Dormouse's story</b></p>

<p class="story">Once upon a time there were three little sisters; and their names were
<a href="http://example.com/elsie" class="sister" id="link1">Elsie</a>,
<a href="http://example.com/lacie" class="sister" id="link2">Lacie</a> and
<a href="http://example.com/tillie" class="sister" id="link3">Tillie</a>;
and they lived at the bottom of a well.</p>

<p class="story">...</p>
"""

from bs4 import BeautifulSoup
soup = BeautifulSoup(html_doc, 'html.parser')

By passing in a filter to an argument like find_all(), you can zoom in on the parts of the document you’re interested in.

Kinds of filters

Before talking in detail about find_all() and similar methods, I want to show examples of different filters you can pass into these methods. These filters show up again and again, throughout the search API. You can use them to filter based on a tag’s name, on its attributes, on the text of a string, or on some combination of these.

A string

The simplest filter is a string. Pass a string to a search method and Beautiful Soup will perform a match against that exact string. This code finds all the <b> tags in the document:

soup.find_all('b')
# [<b>The Dormouse's story</b>]

If you pass in a byte string, Beautiful Soup will assume the string is encoded as UTF-8. You can avoid this by passing in a Unicode string instead.

A regular expression

If you pass in a regular expression object, Beautiful Soup will filter against that regular expression using its search() method. This code finds all the tags whose names start with the letter “b”; in this case, the <body> tag and the <b> tag:

import re
for tag in soup.find_all(re.compile("^b")):
    print(tag.name)
# body
# b

This code finds all the tags whose names contain the letter ‘t’:

for tag in soup.find_all(re.compile("t")):
    print(tag.name)
# html
# title

A list

If you pass in a list, Beautiful Soup will allow a string match against any item in that list. This code finds all the <a> tags and all the <b> tags:

soup.find_all(["a", "b"])
# [<b>The Dormouse's story</b>,
#  <a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

True

The value True matches everything it can. This code finds all the tags in the document, but none of the text strings:

for tag in soup.find_all(True):
    print(tag.name)
# html
# head
# title
# body
# p
# b
# p
# a
# a
# a
# p

A function

If none of the other matches work for you, define a function that takes an element as its only argument. The function should return True if the argument matches, and False otherwise.

Here’s a function that returns True if a tag defines the “class” attribute but doesn’t define the “id” attribute:

def has_class_but_no_id(tag):
    return tag.has_attr('class') and not tag.has_attr('id')

Pass this function into find_all() and you’ll pick up all the <p> tags:

soup.find_all(has_class_but_no_id)
# [<p class="title"><b>The Dormouse's story</b></p>,
#  <p class="story">Once upon a time there were…bottom of a well.</p>,
#  <p class="story">...</p>]

This function only picks up the <p> tags. It doesn’t pick up the <a> tags, because those tags define both “class” and “id”. It doesn’t pick up tags like <html> and <title>, because those tags don’t define “class”.

If you pass in a function to filter on a specific attribute like href, the argument passed into the function will be the attribute value, not the whole tag. Here’s a function that finds all a tags whose href attribute does not match a regular expression:

import re
def not_lacie(href):
    return href and not re.compile("lacie").search(href)

soup.find_all(href=not_lacie)
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

The function can be as complicated as you need it to be. Here’s a function that returns True if a tag is surrounded by string objects:

from bs4 import NavigableString
def surrounded_by_strings(tag):
    return (isinstance(tag.next_element, NavigableString)
            and isinstance(tag.previous_element, NavigableString))

for tag in soup.find_all(surrounded_by_strings):
    print(tag.name)
# body
# p
# a
# a
# a
# p

Now we’re ready to look at the search methods in detail.

find_all()

Method signature: find_all(name, attrs, recursive, string, limit, **kwargs)

The find_all() method looks through a tag’s descendants and retrieves all descendants that match your filters. I gave several examples in Kinds of filters, but here are a few more:

soup.find_all("title")
# [<title>The Dormouse's story</title>]

soup.find_all("p", "title")
# [<p class="title"><b>The Dormouse's story</b></p>]

soup.find_all("a")
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

soup.find_all(id="link2")
# [<a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>]

import re
soup.find(string=re.compile("sisters"))
# 'Once upon a time there were three little sisters; and their names were\n'

Some of these should look familiar, but others are new. What does it mean to pass in a value for string, or id? Why does find_all("p", "title") find a <p> tag with the CSS class “title”? Let’s look at the arguments to find_all().

The name argument

Pass in a value for name and you’ll tell Beautiful Soup to only consider tags with certain names. Text strings will be ignored, as will tags whose names that don’t match.

This is the simplest usage:

soup.find_all("title")
# [<title>The Dormouse's story</title>]

Recall from Kinds of filters that the value to name can be a string, a regular expression, a list, a function, or the value True.

The keyword arguments

Any argument that’s not recognized will be turned into a filter on one of a tag’s attributes. If you pass in a value for an argument called id, Beautiful Soup will filter against each tag’s ‘id’ attribute:

soup.find_all(id='link2')
# [<a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>]

If you pass in a value for href, Beautiful Soup will filter against each tag’s ‘href’ attribute:

soup.find_all(href=re.compile("elsie"))
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>]

You can filter an attribute based on a string, a regular expression, a list, a function, or the value True.

This code finds all tags whose id attribute has a value, regardless of what the value is:

soup.find_all(id=True)
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

You can filter multiple attributes at once by passing in more than one keyword argument:

soup.find_all(href=re.compile("elsie"), id='link1')
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>]

Some attributes, like the data-* attributes in HTML 5, have names that can’t be used as the names of keyword arguments:

data_soup = BeautifulSoup('<div data-foo="value">foo!</div>', 'html.parser')
data_soup.find_all(data-foo="value")
# SyntaxError: keyword can't be an expression

You can use these attributes in searches by putting them into a dictionary and passing the dictionary into find_all() as the attrs argument:

data_soup.find_all(attrs={"data-foo": "value"})
# [<div data-foo="value">foo!</div>]

You can’t use a keyword argument to search for HTML’s ‘name’ element, because Beautiful Soup uses the name argument to contain the name of the tag itself. Instead, you can give a value to ‘name’ in the attrs argument:

name_soup = BeautifulSoup('<input name="email"/>', 'html.parser')
name_soup.find_all(name="email")
# []
name_soup.find_all(attrs={"name": "email"})
# [<input name="email"/>]

Searching by CSS class

It’s very useful to search for a tag that has a certain CSS class, but the name of the CSS attribute, “class”, is a reserved word in Python. Using class as a keyword argument will give you a syntax error. As of Beautiful Soup 4.1.2, you can search by CSS class using the keyword argument class_:

soup.find_all("a", class_="sister")
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

As with any keyword argument, you can pass class_ a string, a regular expression, a function, or True:

soup.find_all(class_=re.compile("itl"))
# [<p class="title"><b>The Dormouse's story</b></p>]

def has_six_characters(css_class):
    return css_class is not None and len(css_class) == 6

soup.find_all(class_=has_six_characters)
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

Remember that a single tag can have multiple values for its “class” attribute. When you search for a tag that matches a certain CSS class, you’re matching against any of its CSS classes:

css_soup = BeautifulSoup('<p class="body strikeout"></p>', 'html.parser')
css_soup.find_all("p", class_="strikeout")
# [<p class="body strikeout"></p>]

css_soup.find_all("p", class_="body")
# [<p class="body strikeout"></p>]

You can also search for the exact string value of the class attribute:

css_soup.find_all("p", class_="body strikeout")
# [<p class="body strikeout"></p>]

But searching for variants of the string value won’t work:

css_soup.find_all("p", class_="strikeout body")
# []

If you want to search for tags that match two or more CSS classes, you should use a CSS selector:

css_soup.select("p.strikeout.body")
# [<p class="body strikeout"></p>]

In older versions of Beautiful Soup, which don’t have the class_ shortcut, you can use the attrs trick mentioned above. Create a dictionary whose value for “class” is the string (or regular expression, or whatever) you want to search for:

soup.find_all("a", attrs={"class": "sister"})
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

The string argument

With string you can search for strings instead of tags. As with name and the keyword arguments, you can pass in a string, a regular expression, a list, a function, or the value True. Here are some examples:

soup.find_all(string="Elsie")
# ['Elsie']

soup.find_all(string=["Tillie", "Elsie", "Lacie"])
# ['Elsie', 'Lacie', 'Tillie']

soup.find_all(string=re.compile("Dormouse"))
# ["The Dormouse's story", "The Dormouse's story"]

def is_the_only_string_within_a_tag(s):
    """Return True if this string is the only child of its parent tag."""
    return (s == s.parent.string)

soup.find_all(string=is_the_only_string_within_a_tag)
# ["The Dormouse's story", "The Dormouse's story", 'Elsie', 'Lacie', 'Tillie', '...']

Although string is for finding strings, you can combine it with arguments that find tags: Beautiful Soup will find all tags whose .string matches your value for string. This code finds the <a> tags whose .string is “Elsie”:

soup.find_all("a", string="Elsie")
# [<a href="http://example.com/elsie" class="sister" id="link1">Elsie</a>]

The string argument is new in Beautiful Soup 4.4.0. In earlier versions it was called text:

soup.find_all("a", text="Elsie")
# [<a href="http://example.com/elsie" class="sister" id="link1">Elsie</a>]

The limit argument

find_all() returns all the tags and strings that match your filters. This can take a while if the document is large. If you don’t need all the results, you can pass in a number for limit. This works just like the LIMIT keyword in SQL. It tells Beautiful Soup to stop gathering results after it’s found a certain number.

There are three links in the “three sisters” document, but this code only finds the first two:

soup.find_all("a", limit=2)
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>]

The recursive argument

If you call mytag.find_all(), Beautiful Soup will examine all the descendants of mytag: its children, its children’s children, and so on. If you only want Beautiful Soup to consider direct children, you can pass in recursive=False. See the difference here:

soup.html.find_all("title")
# [<title>The Dormouse's story</title>]

soup.html.find_all("title", recursive=False)
# []

Here’s that part of the document:

<html>
 <head>
  <title>
   The Dormouse's story
  </title>
 </head>
...

The <title> tag is beneath the <html> tag, but it’s not directly beneath the <html> tag: the <head> tag is in the way. Beautiful Soup finds the <title> tag when it’s allowed to look at all descendants of the <html> tag, but when recursive=False restricts it to the <html> tag’s immediate children, it finds nothing.

Beautiful Soup offers a lot of tree-searching methods (covered below), and they mostly take the same arguments as find_all(): name, attrs, string, limit, and the keyword arguments. But the recursive argument is different: find_all() and find() are the only methods that support it. Passing recursive=False into a method like find_parents() wouldn’t be very useful.

Calling a tag is like calling find_all()

Because find_all() is the most popular method in the Beautiful Soup search API, you can use a shortcut for it. If you treat the BeautifulSoup object or a Tag object as though it were a function, then it’s the same as calling find_all() on that object. These two lines of code are equivalent:

soup.find_all("a")
soup("a")

These two lines are also equivalent:

soup.title.find_all(string=True)
soup.title(string=True)

find()

Method signature: find(name, attrs, recursive, string, **kwargs)

The find_all() method scans the entire document looking for results, but sometimes you only want to find one result. If you know a document only has one <body> tag, it’s a waste of time to scan the entire document looking for more. Rather than passing in limit=1 every time you call find_all, you can use the find() method. These two lines of code are nearly equivalent:

soup.find_all('title', limit=1)
# [<title>The Dormouse's story</title>]

soup.find('title')
# <title>The Dormouse's story</title>

The only difference is that find_all() returns a list containing the single result, and find() just returns the result.

If find_all() can’t find anything, it returns an empty list. If find() can’t find anything, it returns None:

print(soup.find("nosuchtag"))
# None

Remember the soup.head.title trick from Navigating using tag names? That trick works by repeatedly calling find():

soup.head.title
# <title>The Dormouse's story</title>

soup.find("head").find("title")
# <title>The Dormouse's story</title>

find_parents() and find_parent()

Method signature: find_parents(name, attrs, string, limit, **kwargs)

Method signature: find_parent(name, attrs, string, **kwargs)

I spent a lot of time above covering find_all() and find(). The Beautiful Soup API defines ten other methods for searching the tree, but don’t be afraid. Five of these methods are basically the same as find_all(), and the other five are basically the same as find(). The only differences are in what parts of the tree they search.

First let’s consider find_parents() and find_parent(). Remember that find_all() and find() work their way down the tree, looking at tag’s descendants. These methods do the opposite: they work their way up the tree, looking at a tag’s (or a string’s) parents. Let’s try them out, starting from a string buried deep in the “three daughters” document:

a_string = soup.find(string="Lacie")
a_string
# 'Lacie'

a_string.find_parents("a")
# [<a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>]

a_string.find_parent("p")
# <p class="story">Once upon a time there were three little sisters; and their names were
#  <a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a> and
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>;
#  and they lived at the bottom of a well.</p>

a_string.find_parents("p", class_="title")
# []

One of the three <a> tags is the direct parent of the string in question, so our search finds it. One of the three <p> tags is an indirect parent of the string, and our search finds that as well. There’s a <p> tag with the CSS class “title” somewhere in the document, but it’s not one of this string’s parents, so we can’t find it with find_parents().

You may have made the connection between find_parent() and find_parents(), and the .parent and .parents attributes mentioned earlier. The connection is very strong. These search methods actually use .parents to iterate over all the parents, and check each one against the provided filter to see if it matches.

find_next_siblings() and find_next_sibling()

Method signature: find_next_siblings(name, attrs, string, limit, **kwargs)

Method signature: find_next_sibling(name, attrs, string, **kwargs)

These methods use .next_siblings to iterate over the rest of an element’s siblings in the tree. The find_next_siblings() method returns all the siblings that match, and find_next_sibling() only returns the first one:

first_link = soup.a
first_link
# <a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>

first_link.find_next_siblings("a")
# [<a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

first_story_paragraph = soup.find("p", "story")
first_story_paragraph.find_next_sibling("p")
# <p class="story">...</p>

find_previous_siblings() and find_previous_sibling()

Method signature: find_previous_siblings(name, attrs, string, limit, **kwargs)

Method signature: find_previous_sibling(name, attrs, string, **kwargs)

These methods use .previous_siblings to iterate over an element’s siblings that precede it in the tree. The find_previous_siblings() method returns all the siblings that match, and find_previous_sibling() only returns the first one:

last_link = soup.find("a", id="link3")
last_link
# <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>

last_link.find_previous_siblings("a")
# [<a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>]

first_story_paragraph = soup.find("p", "story")
first_story_paragraph.find_previous_sibling("p")
# <p class="title"><b>The Dormouse's story</b></p>

find_all_next() and find_next()

Method signature: find_all_next(name, attrs, string, limit, **kwargs)

Method signature: find_next(name, attrs, string, **kwargs)

These methods use .next_elements to iterate over whatever tags and strings that come after it in the document. The find_all_next() method returns all matches, and find_next() only returns the first match:

first_link = soup.a
first_link
# <a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>

first_link.find_all_next(string=True)
# ['Elsie', ',\n', 'Lacie', ' and\n', 'Tillie',
#  ';\nand they lived at the bottom of a well.', '\n', '...', '\n']

first_link.find_next("p")
# <p class="story">...</p>

In the first example, the string “Elsie” showed up, even though it was contained within the <a> tag we started from. In the second example, the last <p> tag in the document showed up, even though it’s not in the same part of the tree as the <a> tag we started from. For these methods, all that matters is that an element match the filter, and show up later in the document than the starting element.

find_all_previous() and find_previous()

Method signature: find_all_previous(name, attrs, string, limit, **kwargs)

Method signature: find_previous(name, attrs, string, **kwargs)

These methods use .previous_elements to iterate over the tags and strings that came before it in the document. The find_all_previous() method returns all matches, and find_previous() only returns the first match:

first_link = soup.a
first_link
# <a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>

first_link.find_all_previous("p")
# [<p class="story">Once upon a time there were three little sisters; ...</p>,
#  <p class="title"><b>The Dormouse's story</b></p>]

first_link.find_previous("title")
# <title>The Dormouse's story</title>

The call to find_all_previous("p") found the first paragraph in the document (the one with class=”title”), but it also finds the second paragraph, the <p> tag that contains the <a> tag we started with. This shouldn’t be too surprising: we’re looking at all the tags that show up earlier in the document than the one we started with. A <p> tag that contains an <a> tag must have shown up before the <a> tag it contains.

CSS selectors

BeautifulSoup has a .select() method which uses the SoupSieve package to run a CSS selector against a parsed document and return all the matching elements. Tag has a similar method which runs a CSS selector against the contents of a single tag.

(The SoupSieve integration was added in Beautiful Soup 4.7.0. Earlier versions also have the .select() method, but only the most commonly-used CSS selectors are supported. If you installed Beautiful Soup through pip, SoupSieve was installed at the same time, so you don’t have to do anything extra.)

The SoupSieve documentation lists all the currently supported CSS selectors, but here are some of the basics:

You can find tags:

soup.select("title")
# [<title>The Dormouse's story</title>]

soup.select("p:nth-of-type(3)")
# [<p class="story">...</p>]

Find tags beneath other tags:

soup.select("body a")
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie"  id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

soup.select("html head title")
# [<title>The Dormouse's story</title>]

Find tags directly beneath other tags:

soup.select("head > title")
# [<title>The Dormouse's story</title>]

soup.select("p > a")
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie"  id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

soup.select("p > a:nth-of-type(2)")
# [<a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>]

soup.select("p > #link1")
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>]

soup.select("body > a")
# []

Find the siblings of tags:

soup.select("#link1 ~ .sister")
# [<a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie"  id="link3">Tillie</a>]

soup.select("#link1 + .sister")
# [<a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>]

Find tags by CSS class:

soup.select(".sister")
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

soup.select("[class~=sister]")
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

Find tags by ID:

soup.select("#link1")
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>]

soup.select("a#link2")
# [<a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>]

Find tags that match any selector from a list of selectors:

soup.select("#link1,#link2")
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>]

Test for the existence of an attribute:

soup.select('a[href]')
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

Find tags by attribute value:

soup.select('a[href="http://example.com/elsie"]')
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>]

soup.select('a[href^="http://example.com/"]')
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>,
#  <a class="sister" href="http://example.com/lacie" id="link2">Lacie</a>,
#  <a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

soup.select('a[href$="tillie"]')
# [<a class="sister" href="http://example.com/tillie" id="link3">Tillie</a>]

soup.select('a[href*=".com/el"]')
# [<a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>]

There’s also a method called select_one(), which finds only the first tag that matches a selector:

soup.select_one(".sister")
# <a class="sister" href="http://example.com/elsie" id="link1">Elsie</a>

If you’ve parsed XML that defines namespaces, you can use them in CSS selectors.:

from bs4 import BeautifulSoup
xml = """<tag xmlns:ns1="http://namespace1/" xmlns:ns2="http://namespace2/">
 <ns1:child>I'm in namespace 1</ns1:child>
 <ns2:child>I'm in namespace 2</ns2:child>
</tag> """
soup = BeautifulSoup(xml, "xml")

soup.select("child")
# [<ns1:child>I'm in namespace 1</ns1:child>, <ns2:child>I'm in namespace 2</ns2:child>]

soup.select("ns1|child")
# [<ns1:child>I'm in namespace 1</ns1:child>]

When handling a CSS selector that uses namespaces, Beautiful Soup always tries to use namespace prefixes that make sense based on what it saw while parsing the document. You can always provide your own dictionary of abbreviations:

namespaces = dict(first="http://namespace1/", second="http://namespace2/")
soup.select("second|child", namespaces=namespaces)
# [<ns1:child>I'm in namespace 2</ns1:child>]

All this CSS selector stuff is a convenience for people who already know the CSS selector syntax. You can do all of this with the Beautiful Soup API. And if CSS selectors are all you need, you should parse the document with lxml: it’s a lot faster. But this lets you combine CSS selectors with the Beautiful Soup API.

Modifying the tree

Beautiful Soup’s main strength is in searching the parse tree, but you can also modify the tree and write your changes as a new HTML or XML document.

Changing tag names and attributes

I covered this earlier, in Attributes, but it bears repeating. You can rename a tag, change the values of its attributes, add new attributes, and delete attributes:

soup = BeautifulSoup('<b class="boldest">Extremely bold</b>', 'html.parser')
tag = soup.b

tag.name = "blockquote"
tag['class'] = 'verybold'
tag['id'] = 1
tag
# <blockquote class="verybold" id="1">Extremely bold</blockquote>

del tag['class']
del tag['id']
tag
# <blockquote>Extremely bold</blockquote>

Modifying .string

If you set a tag’s .string attribute to a new string, the tag’s contents are replaced with that string:

markup = '<a href="http://example.com/">I linked to <i>example.com</i></a>'
soup = BeautifulSoup(markup, 'html.parser')

tag = soup.a
tag.string = "New link text."
tag
# <a href="http://example.com/">New link text.</a>

Be careful: if the tag contained other tags, they and all their contents will be destroyed.

append()

You can add to a tag’s contents with Tag.append(). It works just like calling .append() on a Python list:

soup = BeautifulSoup("<a>Foo</a>", 'html.parser')
soup.a.append("Bar")

soup
# <a>FooBar</a>
soup.a.contents
# ['Foo', 'Bar']

extend()

Starting in Beautiful Soup 4.7.0, Tag also supports a method called .extend(), which adds every element of a list to a Tag, in order:

soup = BeautifulSoup("<a>Soup</a>", 'html.parser')
soup.a.extend(["'s", " ", "on"])

soup
# <a>Soup's on</a>
soup.a.contents
# ['Soup', ''s', ' ', 'on']

insert()

Tag.insert() is just like Tag.append(), except the new element doesn’t necessarily go at the end of its parent’s .contents. It’ll be inserted at whatever numeric position you say. It works just like .insert() on a Python list:

markup = '<a href="http://example.com/">I linked to <i>example.com</i></a>'
soup = BeautifulSoup(markup, 'html.parser')
tag = soup.a

tag.insert(1, "but did not endorse ")
tag
# <a href="http://example.com/">I linked to but did not endorse <i>example.com</i></a>
tag.contents
# ['I linked to ', 'but did not endorse', <i>example.com</i>]

insert_before() and insert_after()

The insert_before() method inserts tags or strings immediately before something else in the parse tree:

soup = BeautifulSoup("<b>leave</b>", 'html.parser')
tag = soup.new_tag("i")
tag.string = "Don't"
soup.b.string.insert_before(tag)
soup.b
# <b><i>Don't</i>leave</b>

The insert_after() method inserts tags or strings immediately following something else in the parse tree:

div = soup.new_tag('div')
div.string = 'ever'
soup.b.i.insert_after(" you ", div)
soup.b
# <b><i>Don't</i> you <div>ever</div> leave</b>
soup.b.contents
# [<i>Don't</i>, ' you', <div>ever</div>, 'leave']

clear()

Tag.clear() removes the contents of a tag:

markup = '<a href="http://example.com/">I linked to <i>example.com</i></a>'
soup = BeautifulSoup(markup, 'html.parser')
tag = soup.a

tag.clear()
tag
# <a href="http://example.com/"></a>

extract()

PageElement.extract() removes a tag or string from the tree. It returns the tag or string that was extracted:

markup = '<a href="http://example.com/">I linked to <i>example.com</i></a>'
soup = BeautifulSoup(markup, 'html.parser')
a_tag = soup.a

i_tag = soup.i.extract()

a_tag
# <a href="http://example.com/">I linked to</a>

i_tag
# <i>example.com</i>

print(i_tag.parent)
# None

At this point you effectively have two parse trees: one rooted at the BeautifulSoup object you used to parse the document, and one rooted at the tag that was extracted. You can go on to call extract on a child of the element you extracted:

my_string = i_tag.string.extract()
my_string
# 'example.com'

print(my_string.parent)
# None
i_tag
# <i></i>

decompose()

Tag.decompose() removes a tag from the tree, then completely destroys it and its contents:

markup = '<a href="http://example.com/">I linked to <i>example.com</i></a>'
soup = BeautifulSoup(markup, 'html.parser')
a_tag = soup.a
i_tag = soup.i

i_tag.decompose()
a_tag
# <a href="http://example.com/">I linked to</a>

The behavior of a decomposed Tag or NavigableString is not defined and you should not use it for anything. If you’re not sure whether something has been decomposed, you can check its .decomposed property (new in Beautiful Soup 4.9.0):

i_tag.decomposed
# True

a_tag.decomposed
# False

replace_with()

PageElement.replace_with() removes a tag or string from the tree, and replaces it with one or more tags or strings of your choice:

markup = '<a href="http://example.com/">I linked to <i>example.com</i></a>'
soup = BeautifulSoup(markup, 'html.parser')
a_tag = soup.a

new_tag = soup.new_tag("b")
new_tag.string = "example.com"
a_tag.i.replace_with(new_tag)

a_tag
# <a href="http://example.com/">I linked to <b>example.com</b></a>

bold_tag = soup.new_tag("b")
bold_tag.string = "example"
i_tag = soup.new_tag("i")
i_tag.string = "net"
a_tag.b.replace_with(bold_tag, ".", i_tag)

a_tag
# <a href="http://example.com/">I linked to <b>example</b>.<i>net</i></a>

replace_with() returns the tag or string that got replaced, so that you can examine it or add it back to another part of the tree.

The ability to pass multiple arguments into replace_with() is new in Beautiful Soup 4.10.0.

wrap()

PageElement.wrap() wraps an element in the tag you specify. It returns the new wrapper:

soup = BeautifulSoup("<p>I wish I was bold.</p>", 'html.parser')
soup.p.string.wrap(soup.new_tag("b"))
# <b>I wish I was bold.</b>

soup.p.wrap(soup.new_tag("div"))
# <div><p><b>I wish I was bold.</b></p></div>

This method is new in Beautiful Soup 4.0.5.

unwrap()

Tag.unwrap() is the opposite of wrap(). It replaces a tag with whatever’s inside that tag. It’s good for stripping out markup:

markup = '<a href="http://example.com/">I linked to <i>example.com</i></a>'
soup = BeautifulSoup(markup, 'html.parser')
a_tag = soup.a

a_tag.i.unwrap()
a_tag
# <a href="http://example.com/">I linked to example.com</a>

Like replace_with(), unwrap() returns the tag that was replaced.

smooth()

After calling a bunch of methods that modify the parse tree, you may end up with two or more NavigableString objects next to each other. Beautiful Soup doesn’t have any problems with this, but since it can’t happen in a freshly parsed document, you might not expect behavior like the following:

soup = BeautifulSoup("<p>A one</p>", 'html.parser')
soup.p.append(", a two")

soup.p.contents
# ['A one', ', a two']

print(soup.p.encode())
# b'<p>A one, a two</p>'

print(soup.p.prettify())
# <p>
#  A one
#  , a two
# </p>

You can call Tag.smooth() to clean up the parse tree by consolidating adjacent strings:

soup.smooth()

soup.p.contents
# ['A one, a two']

print(soup.p.prettify())
# <p>
#  A one, a two
# </p>

This method is new in Beautiful Soup 4.8.0.

Output

Pretty-printing

The prettify() method will turn a Beautiful Soup parse tree into a nicely formatted Unicode string, with a separate line for each tag and each string:

markup = '<html><head><body><a href="http://example.com/">I linked to <i>example.com</i></a>'
soup = BeautifulSoup(markup, 'html.parser')
soup.prettify()
# '<html>\n <head>\n </head>\n <body>\n  <a href="http://example.com/">\n...'

print(soup.prettify())
# <html>
#  <head>
#  </head>
#  <body>
#   <a href="http://example.com/">
#    I linked to
#    <i>
#     example.com
#    </i>
#   </a>
#  </body>
# </html>

You can call prettify() on the top-level BeautifulSoup object, or on any of its Tag objects:

print(soup.a.prettify())
# <a href="http://example.com/">
#  I linked to
#  <i>
#   example.com
#  </i>
# </a>

Since it adds whitespace (in the form of newlines), prettify() changes the meaning of an HTML document and should not be used to reformat one. The goal of prettify() is to help you visually understand the structure of the documents you work with.

Non-pretty printing

If you just want a string, with no fancy formatting, you can call str() on a BeautifulSoup object, or on a Tag within it:

str(soup)
# '<html><head></head><body><a href="http://example.com/">I linked to <i>example.com</i></a></body></html>'

str(soup.a)
# '<a href="http://example.com/">I linked to <i>example.com</i></a>'

The str() function returns a string encoded in UTF-8. See Encodings for other options.

You can also call encode() to get a bytestring, and decode() to get Unicode.

Output formatters

If you give Beautiful Soup a document that contains HTML entities like “&lquot;”, they’ll be converted to Unicode characters:

soup = BeautifulSoup("&ldquo;Dammit!&rdquo; he said.", 'html.parser')
str(soup)
# '“Dammit!” he said.'

If you then convert the document to a bytestring, the Unicode characters will be encoded as UTF-8. You won’t get the HTML entities back:

soup.encode("utf8")
# b'\xe2\x80\x9cDammit!\xe2\x80\x9d he said.'

By default, the only characters that are escaped upon output are bare ampersands and angle brackets. These get turned into “&amp;”, “&lt;”, and “&gt;”, so that Beautiful Soup doesn’t inadvertently generate invalid HTML or XML:

soup = BeautifulSoup("<p>The law firm of Dewey, Cheatem, & Howe</p>", 'html.parser')
soup.p
# <p>The law firm of Dewey, Cheatem, &amp; Howe</p>

soup = BeautifulSoup('<a href="http://example.com/?foo=val1&bar=val2">A link</a>', 'html.parser')
soup.a
# <a href="http://example.com/?foo=val1&amp;bar=val2">A link</a>

You can change this behavior by providing a value for the formatter argument to prettify(), encode(), or decode(). Beautiful Soup recognizes five possible values for formatter.

The default is formatter="minimal". Strings will only be processed enough to ensure that Beautiful Soup generates valid HTML/XML:

french = "<p>Il a dit &lt;&lt;Sacr&eacute; bleu!&gt;&gt;</p>"
soup = BeautifulSoup(french, 'html.parser')
print(soup.prettify(formatter="minimal"))
# <p>
#  Il a dit &lt;&lt;Sacré bleu!&gt;&gt;
# </p>

If you pass in formatter="html", Beautiful Soup will convert Unicode characters to HTML entities whenever possible:

print(soup.prettify(formatter="html"))
# <p>
#  Il a dit &lt;&lt;Sacr&eacute; bleu!&gt;&gt;
# </p>

If you pass in formatter="html5", it’s similar to formatter="html", but Beautiful Soup will omit the closing slash in HTML void tags like “br”:

br = BeautifulSoup("<br>", 'html.parser').br

print(br.encode(formatter="html"))
# b'<br/>'

print(br.encode(formatter="html5"))
# b'<br>'

In addition, any attributes whose values are the empty string will become HTML-style boolean attributes:

option = BeautifulSoup('<option selected=""></option>').option
print(option.encode(formatter="html"))
# b'<option selected=""></option>'

print(option.encode(formatter="html5"))
# b'<option selected></option>'

(This behavior is new as of Beautiful Soup 4.10.0.)

If you pass in formatter=None, Beautiful Soup will not modify strings at all on output. This is the fastest option, but it may lead to Beautiful Soup generating invalid HTML/XML, as in these examples:

print(soup.prettify(formatter=None))
# <p>
#  Il a dit <<Sacré bleu!>>
# </p>

link_soup = BeautifulSoup('<a href="http://example.com/?foo=val1&bar=val2">A link</a>', 'html.parser')
print(link_soup.a.encode(formatter=None))
# b'<a href="http://example.com/?foo=val1&bar=val2">A link</a>'

If you need more sophisticated control over your output, you can use Beautiful Soup’s Formatter class. Here’s a formatter that converts strings to uppercase, whether they occur in a text node or in an attribute value:

from bs4.formatter import HTMLFormatter
def uppercase(str):
    return str.upper()

formatter = HTMLFormatter(uppercase)

print(soup.prettify(formatter=formatter))
# <p>
#  IL A DIT <<SACRÉ BLEU!>>
# </p>

print(link_soup.a.prettify(formatter=formatter))
# <a href="HTTP://EXAMPLE.COM/?FOO=VAL1&BAR=VAL2">
#  A LINK
# </a>

Here’s a formatter that increases the indentation when pretty-printing:

formatter = HTMLFormatter(indent=8)
print(link_soup.a.prettify(formatter=formatter))
# <a href="http://example.com/?foo=val1&bar=val2">
#         A link
# </a>

Subclassing HTMLFormatter or XMLFormatter will give you even more control over the output. For example, Beautiful Soup sorts the attributes in every tag by default:

attr_soup = BeautifulSoup(b'<p z="1" m="2" a="3"></p>', 'html.parser')
print(attr_soup.p.encode())
# <p a="3" m="2" z="1"></p>

To turn this off, you can subclass the Formatter.attributes() method, which controls which attributes are output and in what order. This implementation also filters out the attribute called “m” whenever it appears:

class UnsortedAttributes(HTMLFormatter):
    def attributes(self, tag):
        for k, v in tag.attrs.items():
            if k == 'm':
                continue
            yield k, v

print(attr_soup.p.encode(formatter=UnsortedAttributes()))
# <p z="1" a="3"></p>

One last caveat: if you create a CData object, the text inside that object is always presented exactly as it appears, with no formatting. Beautiful Soup will call your entity substitution function, just in case you’ve written a custom function that counts all the strings in the document or something, but it will ignore the return value:

from bs4.element import CData
soup = BeautifulSoup("<a></a>", 'html.parser')
soup.a.string = CData("one < three")
print(soup.a.prettify(formatter="html"))
# <a>
#  <![CDATA[one < three]]>
# </a>

get_text()

If you only want the human-readable text inside a document or tag, you can use the get_text() method. It returns all the text in a document or beneath a tag, as a single Unicode string:

markup = '<a href="http://example.com/">\nI linked to <i>example.com</i>\n</a>'
soup = BeautifulSoup(markup, 'html.parser')

soup.get_text()
'\nI linked to example.com\n'
soup.i.get_text()
'example.com'

You can specify a string to be used to join the bits of text together:

# soup.get_text("|")
'\nI linked to |example.com|\n'

You can tell Beautiful Soup to strip whitespace from the beginning and end of each bit of text:

# soup.get_text("|", strip=True)
'I linked to|example.com'

But at that point you might want to use the .stripped_strings generator instead, and process the text yourself:

[text for text in soup.stripped_strings]
# ['I linked to', 'example.com']

As of Beautiful Soup version 4.9.0, when lxml or html.parser are in use, the contents of <script>, <style>, and <template> tags are generally not considered to be ‘text’, since those tags are not part of the human-visible content of the page.

As of Beautiful Soup version 4.10.0, you can call get_text(), .strings, or .stripped_strings on a NavigableString object. It will either return the object itself, or nothing, so the only reason to do this is when you’re iterating over a mixed list.

Specifying the parser to use

If you just need to parse some HTML, you can dump the markup into the BeautifulSoup constructor, and it’ll probably be fine. Beautiful Soup will pick a parser for you and parse the data. But there are a few additional arguments you can pass in to the constructor to change which parser is used.

The first argument to the BeautifulSoup constructor is a string or an open filehandle–the markup you want parsed. The second argument is how you’d like the markup parsed.

If you don’t specify anything, you’ll get the best HTML parser that’s installed. Beautiful Soup ranks lxml’s parser as being the best, then html5lib’s, then Python’s built-in parser. You can override this by specifying one of the following:

  • What type of markup you want to parse. Currently supported are “html”, “xml”, and “html5”.

  • The name of the parser library you want to use. Currently supported options are “lxml”, “html5lib”, and “html.parser” (Python’s built-in HTML parser).

The section Installing a parser contrasts the supported parsers.

If you don’t have an appropriate parser installed, Beautiful Soup will ignore your request and pick a different parser. Right now, the only supported XML parser is lxml. If you don’t have lxml installed, asking for an XML parser won’t give you one, and asking for “lxml” won’t work either.

Differences between parsers

Beautiful Soup presents the same interface to a number of different parsers, but each parser is different. Different parsers will create different parse trees from the same document. The biggest differences are between the HTML parsers and the XML parsers. Here’s a short document, parsed as HTML using the parser that comes with Python:

BeautifulSoup("<a><b/></a>", "html.parser")
# <a><b></b></a>

Since a standalone <b/> tag is not valid HTML, html.parser turns it into a <b></b> tag pair.

Here’s the same document parsed as XML (running this requires that you have lxml installed). Note that the standalone <b/> tag is left alone, and that the document is given an XML declaration instead of being put into an <html> tag.:

print(BeautifulSoup("<a><b/></a>", "xml"))
# <?xml version="1.0" encoding="utf-8"?>
# <a><b/></a>

There are also differences between HTML parsers. If you give Beautiful Soup a perfectly-formed HTML document, these differences won’t matter. One parser will be faster than another, but they’ll all give you a data structure that looks exactly like the original HTML document.

But if the document is not perfectly-formed, different parsers will give different results. Here’s a short, invalid document parsed using lxml’s HTML parser. Note that the <a> tag gets wrapped in <body> and <html> tags, and the dangling </p> tag is simply ignored:

BeautifulSoup("<a></p>", "lxml")
# <html><body><a></a></body></html>

Here’s the same document parsed using html5lib:

BeautifulSoup("<a></p>", "html5lib")
# <html><head></head><body><a><p></p></a></body></html>

Instead of ignoring the dangling </p> tag, html5lib pairs it with an opening <p> tag. html5lib also adds an empty <head> tag; lxml didn’t bother.

Here’s the same document parsed with Python’s built-in HTML parser:

BeautifulSoup("<a></p>", "html.parser")
# <a></a>

Like lxml, this parser ignores the closing </p> tag. Unlike html5lib or lxml, this parser makes no attempt to create a well-formed HTML document by adding <html> or <body> tags.

Since the document “<a></p>” is invalid, none of these techniques is the ‘correct’ way to handle it. The html5lib parser uses techniques that are part of the HTML5 standard, so it has the best claim on being the ‘correct’ way, but all three techniques are legitimate.

Differences between parsers can affect your script. If you’re planning on distributing your script to other people, or running it on multiple machines, you should specify a parser in the BeautifulSoup constructor. That will reduce the chances that your users parse a document differently from the way you parse it.

Encodings

Any HTML or XML document is written in a specific encoding like ASCII or UTF-8. But when you load that document into Beautiful Soup, you’ll discover it’s been converted to Unicode:

markup = "<h1>Sacr\xc3\xa9 bleu!</h1>"
soup = BeautifulSoup(markup, 'html.parser')
soup.h1
# <h1>Sacré bleu!</h1>
soup.h1.string
# 'Sacr\xe9 bleu!'

It’s not magic. (That sure would be nice.) Beautiful Soup uses a sub-library called Unicode, Dammit to detect a document’s encoding and convert it to Unicode. The autodetected encoding is available as the .original_encoding attribute of the BeautifulSoup object:

soup.original_encoding
'utf-8'

Unicode, Dammit guesses correctly most of the time, but sometimes it makes mistakes. Sometimes it guesses correctly, but only after a byte-by-byte search of the document that takes a very long time. If you happen to know a document’s encoding ahead of time, you can avoid mistakes and delays by passing it to the BeautifulSoup constructor as from_encoding.

Here’s a document written in ISO-8859-8. The document is so short that Unicode, Dammit can’t get a lock on it, and misidentifies it as ISO-8859-7:

markup = b"<h1>\xed\xe5\xec\xf9</h1>"
soup = BeautifulSoup(markup, 'html.parser')
print(soup.h1)
# <h1>νεμω</h1>
print(soup.original_encoding)
# iso-8859-7

We can fix this by passing in the correct from_encoding:

soup = BeautifulSoup(markup, 'html.parser', from_encoding="iso-8859-8")
print(soup.h1)
# <h1>םולש</h1>
print(soup.original_encoding)
# iso8859-8

If you don’t know what the correct encoding is, but you know that Unicode, Dammit is guessing wrong, you can pass the wrong guesses in as exclude_encodings:

soup = BeautifulSoup(markup, 'html.parser', exclude_encodings=["iso-8859-7"])
print(soup.h1)
# <h1>םולש</h1>
print(soup.original_encoding)
# WINDOWS-1255

Windows-1255 isn’t 100% correct, but that encoding is a compatible superset of ISO-8859-8, so it’s close enough. (exclude_encodings is a new feature in Beautiful Soup 4.4.0.)

In rare cases (usually when a UTF-8 document contains text written in a completely different encoding), the only way to get Unicode may be to replace some characters with the special Unicode character “REPLACEMENT CHARACTER” (U+FFFD, �). If Unicode, Dammit needs to do this, it will set the .contains_replacement_characters attribute to True on the UnicodeDammit or BeautifulSoup object. This lets you know that the Unicode representation is not an exact representation of the original–some data was lost. If a document contains �, but .contains_replacement_characters is False, you’ll know that the � was there originally (as it is in this paragraph) and doesn’t stand in for missing data.

Output encoding

When you write out a document from Beautiful Soup, you get a UTF-8 document, even if the document wasn’t in UTF-8 to begin with. Here’s a document written in the Latin-1 encoding:

markup = b'''
 <html>
  <head>
   <meta content="text/html; charset=ISO-Latin-1" http-equiv="Content-type" />
  </head>
  <body>
   <p>Sacr\xe9 bleu!</p>
  </body>
 </html>
'''

soup = BeautifulSoup(markup, 'html.parser')
print(soup.prettify())
# <html>
#  <head>
#   <meta content="text/html; charset=utf-8" http-equiv="Content-type" />
#  </head>
#  <body>
#   <p>
#    Sacré bleu!
#   </p>
#  </body>
# </html>

Note that the <meta> tag has been rewritten to reflect the fact that the document is now in UTF-8.

If you don’t want UTF-8, you can pass an encoding into prettify():

print(soup.prettify("latin-1"))
# <html>
#  <head>
#   <meta content="text/html; charset=latin-1" http-equiv="Content-type" />
# ...

You can also call encode() on the BeautifulSoup object, or any element in the soup, just as if it were a Python string:

soup.p.encode("latin-1")
# b'<p>Sacr\xe9 bleu!</p>'

soup.p.encode("utf-8")
# b'<p>Sacr\xc3\xa9 bleu!</p>'

Any characters that can’t be represented in your chosen encoding will be converted into numeric XML entity references. Here’s a document that includes the Unicode character SNOWMAN:

markup = u"<b>\N{SNOWMAN}</b>"
snowman_soup = BeautifulSoup(markup, 'html.parser')
tag = snowman_soup.b

The SNOWMAN character can be part of a UTF-8 document (it looks like ☃), but there’s no representation for that character in ISO-Latin-1 or ASCII, so it’s converted into “&#9731” for those encodings:

print(tag.encode("utf-8"))
# b'<b>\xe2\x98\x83</b>'

print(tag.encode("latin-1"))
# b'<b>&#9731;</b>'

print(tag.encode("ascii"))
# b'<b>&#9731;</b>'

Unicode, Dammit

You can use Unicode, Dammit without using Beautiful Soup. It’s useful whenever you have data in an unknown encoding and you just want it to become Unicode:

from bs4 import UnicodeDammit
dammit = UnicodeDammit("Sacr\xc3\xa9 bleu!")
print(dammit.unicode_markup)
# Sacré bleu!
dammit.original_encoding
# 'utf-8'

Unicode, Dammit’s guesses will get a lot more accurate if you install one of these Python libraries: charset-normalizer, chardet, or cchardet. The more data you give Unicode, Dammit, the more accurately it will guess. If you have your own suspicions as to what the encoding might be, you can pass them in as a list:

dammit = UnicodeDammit("Sacr\xe9 bleu!", ["latin-1", "iso-8859-1"])
print(dammit.unicode_markup)
# Sacré bleu!
dammit.original_encoding
# 'latin-1'

Unicode, Dammit has two special features that Beautiful Soup doesn’t use.

Smart quotes

You can use Unicode, Dammit to convert Microsoft smart quotes to HTML or XML entities:

markup = b"<p>I just \x93love\x94 Microsoft Word\x92s smart quotes</p>"

UnicodeDammit(markup, ["windows-1252"], smart_quotes_to="html").unicode_markup
# '<p>I just &ldquo;love&rdquo; Microsoft Word&rsquo;s smart quotes</p>'

UnicodeDammit(markup, ["windows-1252"], smart_quotes_to="xml").unicode_markup
# '<p>I just &#x201C;love&#x201D; Microsoft Word&#x2019;s smart quotes</p>'

You can also convert Microsoft smart quotes to ASCII quotes:

UnicodeDammit(markup, ["windows-1252"], smart_quotes_to="ascii").unicode_markup
# '<p>I just "love" Microsoft Word\'s smart quotes</p>'

Hopefully you’ll find this feature useful, but Beautiful Soup doesn’t use it. Beautiful Soup prefers the default behavior, which is to convert Microsoft smart quotes to Unicode characters along with everything else:

UnicodeDammit(markup, ["windows-1252"]).unicode_markup
# '<p>I just “love” Microsoft Word’s smart quotes</p>'

Inconsistent encodings

Sometimes a document is mostly in UTF-8, but contains Windows-1252 characters such as (again) Microsoft smart quotes. This can happen when a website includes data from multiple sources. You can use UnicodeDammit.detwingle() to turn such a document into pure UTF-8. Here’s a simple example:

snowmen = (u"\N{SNOWMAN}" * 3)
quote = (u"\N{LEFT DOUBLE QUOTATION MARK}I like snowmen!\N{RIGHT DOUBLE QUOTATION MARK}")
doc = snowmen.encode("utf8") + quote.encode("windows_1252")

This document is a mess. The snowmen are in UTF-8 and the quotes are in Windows-1252. You can display the snowmen or the quotes, but not both:

print(doc)
# ☃☃☃�I like snowmen!�

print(doc.decode("windows-1252"))
# ☃☃☃“I like snowmen!”

Decoding the document as UTF-8 raises a UnicodeDecodeError, and decoding it as Windows-1252 gives you gibberish. Fortunately, UnicodeDammit.detwingle() will convert the string to pure UTF-8, allowing you to decode it to Unicode and display the snowmen and quote marks simultaneously:

new_doc = UnicodeDammit.detwingle(doc)
print(new_doc.decode("utf8"))
# ☃☃☃“I like snowmen!”

UnicodeDammit.detwingle() only knows how to handle Windows-1252 embedded in UTF-8 (or vice versa, I suppose), but this is the most common case.

Note that you must know to call UnicodeDammit.detwingle() on your data before passing it into BeautifulSoup or the UnicodeDammit constructor. Beautiful Soup assumes that a document has a single encoding, whatever it might be. If you pass it a document that contains both UTF-8 and Windows-1252, it’s likely to think the whole document is Windows-1252, and the document will come out looking like ☃☃☃“I like snowmen!”.

UnicodeDammit.detwingle() is new in Beautiful Soup 4.1.0.

Line numbers

The html.parser and html5lib parsers can keep track of where in the original document each Tag was found. You can access this information as Tag.sourceline (line number) and Tag.sourcepos (position of the start tag within a line):

markup = "<p\n>Paragraph 1</p>\n    <p>Paragraph 2</p>"
soup = BeautifulSoup(markup, 'html.parser')
for tag in soup.find_all('p'):
    print(repr((tag.sourceline, tag.sourcepos, tag.string)))
# (1, 0, 'Paragraph 1')
# (3, 4, 'Paragraph 2')

Note that the two parsers mean slightly different things by sourceline and sourcepos. For html.parser, these numbers represent the position of the initial less-than sign. For html5lib, these numbers represent the position of the final greater-than sign:

soup = BeautifulSoup(markup, 'html5lib')
for tag in soup.find_all('p'):
    print(repr((tag.sourceline, tag.sourcepos, tag.string)))
# (2, 0, 'Paragraph 1')
# (3, 6, 'Paragraph 2')

You can shut off this feature by passing store_line_numbers=False` into the ``BeautifulSoup constructor:

markup = "<p\n>Paragraph 1</p>\n    <p>Paragraph 2</p>"
soup = BeautifulSoup(markup, 'html.parser', store_line_numbers=False)
print(soup.p.sourceline)
# None

This feature is new in 4.8.1, and the parsers based on lxml don’t support it.

Comparing objects for equality

Beautiful Soup says that two NavigableString or Tag objects are equal when they represent the same HTML or XML markup. In this example, the two <b> tags are treated as equal, even though they live in different parts of the object tree, because they both look like “<b>pizza</b>”:

markup = "<p>I want <b>pizza</b> and more <b>pizza</b>!</p>"
soup = BeautifulSoup(markup, 'html.parser')
first_b, second_b = soup.find_all('b')
print(first_b == second_b)
# True

print(first_b.previous_element == second_b.previous_element)
# False

If you want to see whether two variables refer to exactly the same object, use is:

print(first_b is second_b)
# False

Copying Beautiful Soup objects

You can use copy.copy() to create a copy of any Tag or NavigableString:

import copy
p_copy = copy.copy(soup.p)
print(p_copy)
# <p>I want <b>pizza</b> and more <b>pizza</b>!</p>

The copy is considered equal to the original, since it represents the same markup as the original, but it’s not the same object:

print(soup.p == p_copy)
# True

print(soup.p is p_copy)
# False

The only real difference is that the copy is completely detached from the original Beautiful Soup object tree, just as if extract() had been called on it:

print(p_copy.parent)
# None

This is because two different Tag objects can’t occupy the same space at the same time.

Advanced parser customization

Beautiful Soup offers a number of ways to customize how the parser treats incoming HTML and XML. This section covers the most commonly used customization techniques.

Parsing only part of a document

Let’s say you want to use Beautiful Soup look at a document’s <a> tags. It’s a waste of time and memory to parse the entire document and then go over it again looking for <a> tags. It would be much faster to ignore everything that wasn’t an <a> tag in the first place. The SoupStrainer class allows you to choose which parts of an incoming document are parsed. You just create a SoupStrainer and pass it in to the BeautifulSoup constructor as the parse_only argument.

(Note that this feature won’t work if you’re using the html5lib parser. If you use html5lib, the whole document will be parsed, no matter what. This is because html5lib constantly rearranges the parse tree as it works, and if some part of the document didn’t actually make it into the parse tree, it’ll crash. To avoid confusion, in the examples below I’ll be forcing Beautiful Soup to use Python’s built-in parser.)

SoupStrainer

The SoupStrainer class takes the same arguments as a typical method from Searching the tree: name, attrs, string, and **kwargs. Here are three SoupStrainer objects:

from bs4 import SoupStrainer

only_a_tags = SoupStrainer("a")

only_tags_with_id_link2 = SoupStrainer(id="link2")

def is_short_string(string):
    return string is not None and len(string) < 10

only_short_strings = SoupStrainer(string=is_short_string)

I’m going to bring back the “three sisters” document one more time, and we’ll see what the document looks like when it’s parsed with these three SoupStrainer objects:

html_doc = """<html><head><title>The Dormouse's story</title></head>
<body>
<p class="title"><b>The Dormouse's story</b></p>

<p class="story">Once upon a time there were three little sisters; and their names were
<a href="http://example.com/elsie" class="sister" id="link1">Elsie</a>,
<a href="http://example.com/lacie" class="sister" id="link2">Lacie</a> and
<a href="http://example.com/tillie" class="sister" id="link3">Tillie</a>;
and they lived at the bottom of a well.</p>

<p class="story">...</p>
"""

print(BeautifulSoup(html_doc, "html.parser", parse_only=only_a_tags).prettify())
# <a class="sister" href="http://example.com/elsie" id="link1">
#  Elsie
# </a>
# <a class="sister" href="http://example.com/lacie" id="link2">
#  Lacie
# </a>
# <a class="sister" href="http://example.com/tillie" id="link3">
#  Tillie
# </a>

print(BeautifulSoup(html_doc, "html.parser", parse_only=only_tags_with_id_link2).prettify())
# <a class="sister" href="http://example.com/lacie" id="link2">
#  Lacie
# </a>

print(BeautifulSoup(html_doc, "html.parser", parse_only=only_short_strings).prettify())
# Elsie
# ,
# Lacie
# and
# Tillie
# ...
#

You can also pass a SoupStrainer into any of the methods covered in Searching the tree. This probably isn’t terribly useful, but I thought I’d mention it:

soup = BeautifulSoup(html_doc, 'html.parser')
soup.find_all(only_short_strings)
# ['\n\n', '\n\n', 'Elsie', ',\n', 'Lacie', ' and\n', 'Tillie',
#  '\n\n', '...', '\n']

Customizing multi-valued attributes

In an HTML document, an attribute like class is given a list of values, and an attribute like id is given a single value, because the HTML specification treats those attributes differently:

markup = '<a class="cls1 cls2" id="id1 id2">'
soup = BeautifulSoup(markup, 'html.parser')
soup.a['class']
# ['cls1', 'cls2']
soup.a['id']
# 'id1 id2'

You can turn this off by passing in multi_valued_attributes=None. Than all attributes will be given a single value:

soup = BeautifulSoup(markup, 'html.parser', multi_valued_attributes=None)
soup.a['class']
# 'cls1 cls2'
soup.a['id']
# 'id1 id2'

You can customize this behavior quite a bit by passing in a dictionary for multi_valued_attributes. If you need this, look at HTMLTreeBuilder.DEFAULT_CDATA_LIST_ATTRIBUTES to see the configuration Beautiful Soup uses by default, which is based on the HTML specification.

(This is a new feature in Beautiful Soup 4.8.0.)

Handling duplicate attributes

When using the html.parser parser, you can use the on_duplicate_attribute constructor argument to customize what Beautiful Soup does when it encounters a tag that defines the same attribute more than once:

markup = '<a href="http://url1/" href="http://url2/">'

The default behavior is to use the last value found for the tag:

soup = BeautifulSoup(markup, 'html.parser')
soup.a['href']
# http://url2/

soup = BeautifulSoup(markup, 'html.parser', on_duplicate_attribute='replace')
soup.a['href']
# http://url2/

With on_duplicate_attribute='ignore' you can tell Beautiful Soup to use the first value found and ignore the rest:

soup = BeautifulSoup(markup, 'html.parser', on_duplicate_attribute='ignore')
soup.a['href']
# http://url1/

(lxml and html5lib always do it this way; their behavior can’t be configured from within Beautiful Soup.)

If you need more, you can pass in a function that’s called on each duplicate value:

def accumulate(attributes_so_far, key, value):
    if not isinstance(attributes_so_far[key], list):
        attributes_so_far[key] = [attributes_so_far[key]]
    attributes_so_far[key].append(value)

soup = BeautifulSoup(markup, 'html.parser', on_duplicate_attribute=accumulate)
soup.a['href']
# ["http://url1/", "http://url2/"]

(This is a new feature in Beautiful Soup 4.9.1.)

Instantiating custom subclasses

When a parser tells Beautiful Soup about a tag or a string, Beautiful Soup will instantiate a Tag or NavigableString object to contain that information. Instead of that default behavior, you can tell Beautiful Soup to instantiate subclasses of Tag or NavigableString, subclasses you define with custom behavior:

from bs4 import Tag, NavigableString
class MyTag(Tag):
    pass


class MyString(NavigableString):
    pass


markup = "<div>some text</div>"
soup = BeautifulSoup(markup, 'html.parser')
isinstance(soup.div, MyTag)
# False
isinstance(soup.div.string, MyString)
# False

my_classes = { Tag: MyTag, NavigableString: MyString }
soup = BeautifulSoup(markup, 'html.parser', element_classes=my_classes)
isinstance(soup.div, MyTag)
# True
isinstance(soup.div.string, MyString)
# True

This can be useful when incorporating Beautiful Soup into a test framework.

(This is a new feature in Beautiful Soup 4.8.1.)

Troubleshooting

diagnose()

If you’re having trouble understanding what Beautiful Soup does to a document, pass the document into the diagnose() function. (New in Beautiful Soup 4.2.0.) Beautiful Soup will print out a report showing you how different parsers handle the document, and tell you if you’re missing a parser that Beautiful Soup could be using:

from bs4.diagnose import diagnose
with open("bad.html") as fp:
    data = fp.read()

diagnose(data)

# Diagnostic running on Beautiful Soup 4.2.0
# Python version 2.7.3 (default, Aug  1 2012, 05:16:07)
# I noticed that html5lib is not installed. Installing it may help.
# Found lxml version 2.3.2.0
#
# Trying to parse your data with html.parser
# Here's what html.parser did with the document:
# ...

Just looking at the output of diagnose() may show you how to solve the problem. Even if not, you can paste the output of diagnose() when asking for help.

Errors when parsing a document

There are two different kinds of parse errors. There are crashes, where you feed a document to Beautiful Soup and it raises an exception, usually an HTMLParser.HTMLParseError. And there is unexpected behavior, where a Beautiful Soup parse tree looks a lot different than the document used to create it.

Almost none of these problems turn out to be problems with Beautiful Soup. This is not because Beautiful Soup is an amazingly well-written piece of software. It’s because Beautiful Soup doesn’t include any parsing code. Instead, it relies on external parsers. If one parser isn’t working on a certain document, the best solution is to try a different parser. See Installing a parser for details and a parser comparison.

The most common parse errors are HTMLParser.HTMLParseError: malformed start tag and HTMLParser.HTMLParseError: bad end tag. These are both generated by Python’s built-in HTML parser library, and the solution is to install lxml or html5lib.

The most common type of unexpected behavior is that you can’t find a tag that you know is in the document. You saw it going in, but find_all() returns [] or find() returns None. This is another common problem with Python’s built-in HTML parser, which sometimes skips tags it doesn’t understand. Again, the best solution is to install lxml or html5lib.

Version mismatch problems

  • SyntaxError: Invalid syntax (on the line ROOT_TAG_NAME = '[document]'): Caused by running an old Python 2 version of Beautiful Soup under Python 3, without converting the code.

  • ImportError: No module named HTMLParser - Caused by running an old Python 2 version of Beautiful Soup under Python 3.

  • ImportError: No module named html.parser - Caused by running the Python 3 version of Beautiful Soup under Python 2.

  • ImportError: No module named BeautifulSoup - Caused by running Beautiful Soup 3 code on a system that doesn’t have BS3 installed. Or, by writing Beautiful Soup 4 code without knowing that the package name has changed to bs4.

  • ImportError: No module named bs4 - Caused by running Beautiful Soup 4 code on a system that doesn’t have BS4 installed.

Parsing XML

By default, Beautiful Soup parses documents as HTML. To parse a document as XML, pass in “xml” as the second argument to the BeautifulSoup constructor:

soup = BeautifulSoup(markup, "xml")

You’ll need to have lxml installed.

Other parser problems

  • If your script works on one computer but not another, or in one virtual environment but not another, or outside the virtual environment but not inside, it’s probably because the two environments have different parser libraries available. For example, you may have developed the script on a computer that has lxml installed, and then tried to run it on a computer that only has html5lib installed. See Differences between parsers for why this matters, and fix the problem by mentioning a specific parser library in the BeautifulSoup constructor.

  • Because HTML tags and attributes are case-insensitive, all three HTML parsers convert tag and attribute names to lowercase. That is, the markup <TAG></TAG> is converted to <tag></tag>. If you want to preserve mixed-case or uppercase tags and attributes, you’ll need to parse the document as XML.

Miscellaneous

  • UnicodeEncodeError: 'charmap' codec can't encode character '\xfoo' in position bar (or just about any other UnicodeEncodeError) - This problem shows up in two main situations. First, when you try to print a Unicode character that your console doesn’t know how to display. (See this page on the Python wiki for help.) Second, when you’re writing to a file and you pass in a Unicode character that’s not supported by your default encoding. In this case, the simplest solution is to explicitly encode the Unicode string into UTF-8 with u.encode("utf8").

  • KeyError: [attr] - Caused by accessing tag['attr'] when the tag in question doesn’t define the attr attribute. The most common errors are KeyError: 'href' and KeyError: 'class'. Use tag.get('attr') if you’re not sure attr is defined, just as you would with a Python dictionary.

  • AttributeError: 'ResultSet' object has no attribute 'foo' - This usually happens because you expected find_all() to return a single tag or string. But find_all() returns a _list_ of tags and strings–a ResultSet object. You need to iterate over the list and look at the .foo of each one. Or, if you really only want one result, you need to use find() instead of find_all().

  • AttributeError: 'NoneType' object has no attribute 'foo' - This usually happens because you called find() and then tried to access the .foo` attribute of the result. But in your case, find() didn’t find anything, so it returned None, instead of returning a tag or a string. You need to figure out why your find() call isn’t returning anything.

  • AttributeError: 'NavigableString' object has no attribute 'foo' - This usually happens because you’re treating a string as though it were a tag. You may be iterating over a list, expecting that it contains nothing but tags, when it actually contains both tags and strings.

Improving Performance

Beautiful Soup will never be as fast as the parsers it sits on top of. If response time is critical, if you’re paying for computer time by the hour, or if there’s any other reason why computer time is more valuable than programmer time, you should forget about Beautiful Soup and work directly atop lxml.

That said, there are things you can do to speed up Beautiful Soup. If you’re not using lxml as the underlying parser, my advice is to start. Beautiful Soup parses documents significantly faster using lxml than using html.parser or html5lib.

You can speed up encoding detection significantly by installing the cchardet library.

Parsing only part of a document won’t save you much time parsing the document, but it can save a lot of memory, and it’ll make searching the document much faster.

Translating this documentation

New translations of the Beautiful Soup documentation are greatly appreciated. Translations should be licensed under the MIT license, just like Beautiful Soup and its English documentation are.

There are two ways of getting your translation into the main code base and onto the Beautiful Soup website:

  1. Create a branch of the Beautiful Soup repository, add your translation, and propose a merge with the main branch, the same as you would do with a proposed change to the source code.

  2. Send a message to the Beautiful Soup discussion group with a link to your translation, or attach your translation to the message.

Use the Chinese or Brazilian Portuguese translations as your model. In particular, please translate the source file doc/source/index.rst, rather than the HTML version of the documentation. This makes it possible to publish the documentation in a variety of formats, not just HTML.

Beautiful Soup 3

Beautiful Soup 3 is the previous release series, and is no longer being actively developed. It’s currently packaged with all major Linux distributions:

$ apt-get install python-beautifulsoup

It’s also published through PyPi as BeautifulSoup.:

$ easy_install BeautifulSoup

$ pip install BeautifulSoup

You can also download a tarball of Beautiful Soup 3.2.0.

If you ran easy_install beautifulsoup or easy_install BeautifulSoup, but your code doesn’t work, you installed Beautiful Soup 3 by mistake. You need to run easy_install beautifulsoup4.

The documentation for Beautiful Soup 3 is archived online.

Porting code to BS4

Most code written against Beautiful Soup 3 will work against Beautiful Soup 4 with one simple change. All you should have to do is change the package name from BeautifulSoup to bs4. So this:

from BeautifulSoup import BeautifulSoup

becomes this:

from bs4 import BeautifulSoup
  • If you get the ImportError “No module named BeautifulSoup”, your problem is that you’re trying to run Beautiful Soup 3 code, but you only have Beautiful Soup 4 installed.

  • If you get the ImportError “No module named bs4”, your problem is that you’re trying to run Beautiful Soup 4 code, but you only have Beautiful Soup 3 installed.

Although BS4 is mostly backwards-compatible with BS3, most of its methods have been deprecated and given new names for PEP 8 compliance. There are numerous other renames and changes, and a few of them break backwards compatibility.

Here’s what you’ll need to know to convert your BS3 code and habits to BS4:

You need a parser

Beautiful Soup 3 used Python’s SGMLParser, a module that was deprecated and removed in Python 3.0. Beautiful Soup 4 uses html.parser by default, but you can plug in lxml or html5lib and use that instead. See Installing a parser for a comparison.

Since html.parser is not the same parser as SGMLParser, you may find that Beautiful Soup 4 gives you a different parse tree than Beautiful Soup 3 for the same markup. If you swap out html.parser for lxml or html5lib, you may find that the parse tree changes yet again. If this happens, you’ll need to update your scraping code to deal with the new tree.

Method names

  • renderContents -> encode_contents

  • replaceWith -> replace_with

  • replaceWithChildren -> unwrap

  • findAll -> find_all

  • findAllNext -> find_all_next

  • findAllPrevious -> find_all_previous

  • findNext -> find_next

  • findNextSibling -> find_next_sibling

  • findNextSiblings -> find_next_siblings

  • findParent -> find_parent

  • findParents -> find_parents

  • findPrevious -> find_previous

  • findPreviousSibling -> find_previous_sibling

  • findPreviousSiblings -> find_previous_siblings

  • getText -> get_text

  • nextSibling -> next_sibling

  • previousSibling -> previous_sibling

Some arguments to the Beautiful Soup constructor were renamed for the same reasons:

  • BeautifulSoup(parseOnlyThese=...) -> BeautifulSoup(parse_only=...)

  • BeautifulSoup(fromEncoding=...) -> BeautifulSoup(from_encoding=...)

I renamed one method for compatibility with Python 3:

  • Tag.has_key() -> Tag.has_attr()

I renamed one attribute to use more accurate terminology:

  • Tag.isSelfClosing -> Tag.is_empty_element

I renamed three attributes to avoid using words that have special meaning to Python. Unlike the others, these changes are not backwards compatible. If you used these attributes in BS3, your code will break on BS4 until you change them.

  • UnicodeDammit.unicode -> UnicodeDammit.unicode_markup

  • Tag.next -> Tag.next_element

  • Tag.previous -> Tag.previous_element

These methods are left over from the Beautiful Soup 2 API. They’ve been deprecated since 2006, and should not be used at all:

  • Tag.fetchNextSiblings

  • Tag.fetchPreviousSiblings

  • Tag.fetchPrevious

  • Tag.fetchPreviousSiblings

  • Tag.fetchParents

  • Tag.findChild

  • Tag.findChildren

Generators

I gave the generators PEP 8-compliant names, and transformed them into properties:

  • childGenerator() -> children

  • nextGenerator() -> next_elements

  • nextSiblingGenerator() -> next_siblings

  • previousGenerator() -> previous_elements

  • previousSiblingGenerator() -> previous_siblings

  • recursiveChildGenerator() -> descendants

  • parentGenerator() -> parents

So instead of this:

for parent in tag.parentGenerator():
    ...

You can write this:

for parent in tag.parents:
    ...

(But the old code will still work.)

Some of the generators used to yield None after they were done, and then stop. That was a bug. Now the generators just stop.

There are two new generators, .strings and .stripped_strings. .strings yields NavigableString objects, and .stripped_strings yields Python strings that have had whitespace stripped.

XML

There is no longer a BeautifulStoneSoup class for parsing XML. To parse XML you pass in “xml” as the second argument to the BeautifulSoup constructor. For the same reason, the BeautifulSoup constructor no longer recognizes the isHTML argument.

Beautiful Soup’s handling of empty-element XML tags has been improved. Previously when you parsed XML you had to explicitly say which tags were considered empty-element tags. The selfClosingTags argument to the constructor is no longer recognized. Instead, Beautiful Soup considers any empty tag to be an empty-element tag. If you add a child to an empty-element tag, it stops being an empty-element tag.

Entities

An incoming HTML or XML entity is always converted into the corresponding Unicode character. Beautiful Soup 3 had a number of overlapping ways of dealing with entities, which have been removed. The BeautifulSoup constructor no longer recognizes the smartQuotesTo or convertEntities arguments. (Unicode, Dammit still has smart_quotes_to, but its default is now to turn smart quotes into Unicode.) The constants HTML_ENTITIES, XML_ENTITIES, and XHTML_ENTITIES have been removed, since they configure a feature (transforming some but not all entities into Unicode characters) that no longer exists.

If you want to turn Unicode characters back into HTML entities on output, rather than turning them into UTF-8 characters, you need to use an output formatter.

Miscellaneous

Tag.string now operates recursively. If tag A contains a single tag B and nothing else, then A.string is the same as B.string. (Previously, it was None.)

Multi-valued attributes like class have lists of strings as their values, not strings. This may affect the way you search by CSS class.

Tag objects now implement the __hash__ method, such that two Tag objects are considered equal if they generate the same markup. This may change your script’s behavior if you put Tag objects into a dictionary or set.

If you pass one of the find* methods both string and a tag-specific argument like name, Beautiful Soup will search for tags that match your tag-specific criteria and whose Tag.string matches your value for string. It will not find the strings themselves. Previously, Beautiful Soup ignored the tag-specific arguments and looked for strings.

The BeautifulSoup constructor no longer recognizes the markupMassage argument. It’s now the parser’s responsibility to handle markup correctly.

The rarely-used alternate parser classes like ICantBelieveItsBeautifulSoup and BeautifulSOAP have been removed. It’s now the parser’s decision how to handle ambiguous markup.

The prettify() method now returns a Unicode string, not a bytestring.