Parsing HTMLTopic 3, Chapter 7
Network Programming
Kansas State University at Salina
Picking information from an HTML page
A difficult problem HTML defines page layout, not content –
advantage XML Very useful because of volume of data
available If the format of the page changes, your
program is broken.
HTML Definition: Token – one piece of information
in an HTML formatted page HTML tag – usually only relates to formatting URL or image reference Textual information
Must look at several tokens to determine context of the data
Start-tag, End-tag structure leads parsing code to use finite state machines and stacks.
( <TABLE> … </TABLE> )
Tokens
{'data': [], 'type': 'StartTag', 'name': u'html'}{'data': [], 'type': 'StartTag', 'name': u'head'}{'data': u'\n ', 'type': 'SpaceCharacters'}{'data': [], 'type': 'StartTag', 'name': u'title'}{'data': u' ', 'type': 'SpaceCharacters'}{'data': u'Tim Bower', 'type': 'Characters'}{'data': u' ', 'type': 'SpaceCharacters'}{'data': [], 'type': 'EndTag', 'name': u'title'}{'data': u'\n', 'type': 'SpaceCharacters'}{'data': [], 'type': 'EndTag', 'name': u'head'}{'data': u'\n\n', 'type': 'SpaceCharacters'}{'data': [(u'bgcolor', u'lightyellow')],
'type': 'StartTag', 'name': u'body'}{'data': u' \n\n', 'type': 'SpaceCharacters'}{'data': [], 'type': 'StartTag', 'name': u'table'}{'data': u' ', 'type': 'SpaceCharacters'}{'data': [], 'type': 'StartTag', 'name': u'tbody'}{'data': [], 'type': 'StartTag', 'name': u'tr'}{'data': u'\n', 'type': 'SpaceCharacters'}{'data': [], 'type': 'StartTag', 'name': u'td'}{'data': u'\n', 'type': 'SpaceCharacters'}{'data': [], 'type': 'StartTag', 'name': u'h1'}{'data': u'Tim Bower', 'type': 'Characters'}{'data': [], 'type': 'EndTag', 'name': u'h1'}
<HTML>
<HEAD>
<TITLE> Tim Bower </TITLE>
</HEAD>
<BODY BGCOLOR="lightyellow">
<TABLE> <TR>
<TD>
<H1>Tim Bower</H1>
Two main programming strategies The call-back approach (HTMLParser shown
in text book) Define your own class that extends the
HTMLParser class Nice use of inheritance and polymorphism Pass the HTML page to the parser and it calls
functions from your class as needed to process the start-tags, data elements, end-tags and a few other miscellaneous tags.
The document tree approach Parser builds a tree (data structure object) based
on the page contents You iterate through the tree or a list of tokens
taken from the tree looking for desired data.
HTMLParserimport HTMLParser
class TitleParser(HTMLParser): def __init__(self): self.title = '' self.readingtitle = 0 HTMLParser.__init__(self)
def handle_starttag(self, tag, \ attrs):
if tag == 'title': self.readingtitle = 1
def handle_data(self, data): if self.readingtitle: self.title += data
def handle_endtag(self, tag): if tag == 'title':
print “*** %s ***” % \ self.title self.readingtitle = 0
fd = open(sys.argv[1])tp = TitleParser()tp.feed(fd.read())
Argh!, HTMLParser is fragile and hard to debug.
Traceback (most recent call last): File "C:\Users\tim\Documents\Classes\Net_Programming\Source_code\Topic 3 - Web\weatherParser.py", line 258, in <module> parser.feed(data) File "C:\Python25\lib\HTMLParser.py", line 108, in feed self.goahead(0) File "C:\Python25\lib\HTMLParser.py", line 148, in goahead k = self.parse_starttag(i) File "C:\Python25\lib\HTMLParser.py", line 226, in parse_starttag endpos = self.check_for_whole_start_tag(i) File "C:\Python25\lib\HTMLParser.py", line 301, in check_for_whole_start_tag self.error("malformed start tag") File "C:\Python25\lib\HTMLParser.py", line 115, in error raise HTMLParseError(message, self.getpos())HTMLParseError: malformed start tag, at line 120, column 477
html5lib
Found on Python package index Install setuptools then use Python to install
html5lib (see the README file). Both are on K-State Online.
Advantages: Robust, standards based parser Filtering data after the page is parsed is easier to
follow and debug than the call-back approach Disadvantage:
Documentation of API for traversing the tree
html5lib Usage Build the tree:
Loop through tokens:
p = html5lib.HTMLParser( \tree=treebuilders.getTreeBuilder("dom"))
f = open( "weather.html", "r" )dom_tree = p.parse(f)f.close()
walker = treewalkers.getTreeWalker("dom")stream = walker(dom_tree)passtags = [ u'a', u'h1', u'h2', u'h3', u'h4',u'em', \
u'strong', u'br', u'img', \u'dl', u'dt', u'dd' ]
for token in stream: # Don't show non interesting stuff if token.has_key('name'): if token['name'] in passtags: continue print token
The DOM tree alternative
The DOM tree may be used directly. Not documented with html5lib, but xml.dom
package is standard with Python. DOM trees are normally used with XML, but
html5lib can make a DOM tree from HTML. Walk through the tree by examining children
nodes of each node. With knowledge of the page structure, you may be able to go almost directly to the desired information.
See chapter 8 and DOMtry.py posted file.
html5lib tokens
Stream of tokens is a list Each token is a dictionary
token[ ‘data’ ] String (unicode encoding) Empty list List of tuples for formatting attributes
token[ ‘type’ ] – (StartTag, EndTag, Characters, SpaceCharacters)
token[ ‘name’ ] – description of start and end tags. (table, tr, td, h1, br, ul, li, … )
See example of tokens on previous slide
html5lib token parsingdoingTitle = Falsefor token in stream: if token.has_key('name'): if token['name'] in passtags: continue else: tName = token['name'] tType = token['type'] if tType == 'StartTag': if tName == u'title': title = '' doingTitle = True if tType == 'EndTag': if tName == u'title': print "*** %s ***\n" % title doingTitle = False
if tType == 'Characters': if doingTitle: title += token['data']