第 9 章 XML 处理

9.1. 概览

下面两章是关于 PythonXML 处理的。如果你已经对 XML 文档有了一个大概的了解,比如它是由结构化标记构成的,这些标记形成了层次模型的元素,等等这些知识都是有帮助的。如果你不明白这些,这里有很多 XML 教程能够解释这些基础知识。

如果你对 XML 不是很感兴趣,你还是应该读一下这些章节,它们涵盖了不少重要的主题,比如 Python 包、Unicode、命令行参数以及如何使用 getattr 进行方法分发。

如果你在大学里主修哲学 (而不是像计算机科学这样的实用专业),并且曾不幸地被伊曼努尔·康德的著作折磨地够呛,那么你会非常欣赏本章的样例程序。(这当然不意味着你必须修过哲学。)

处理 XML 有两种基本的方式。一种叫做 SAX (“Simple API for XML”),它的工作方式是,一次读出一点 XML 内容,然后对发现的每一个元素调用一个方法。(如果你读了 第 8 章 HTML 处理,这应该听起来很熟悉,因为这是 sgmllib 工作的方式。) 另一种方式叫做 DOM (“Document Object Model”),它的工作方式是,一次性读入整个 XML 文档,然后使用 Python 类创建一个内部表示形式 (以树结构进行连接)。Python 拥有这两种解析方式的标准模块,但是本章只涉及 DOM

下面是一个完整的 Python 程序,它根据 XML 格式定义的上下文无关语法生成伪随机输出。如果你不明白是什么意思,不用担心,下面两章中将会深入检视这个程序的输入和输出。

例 9.1. kgp.py

如果您还没有下载本书附带的样例程序, 可以 下载本程序和其他样例程序

"""Kant Generator for Python

Generates mock philosophy based on a context-free grammar

Usage: python kgp.py [options] [source]

Options:
  -g ..., --grammar=...   use specified grammar file or URL
  -h, --help              show this help
  -d                      show debugging information while parsing

Examples:
  kgp.py                  generates several paragraphs of Kantian philosophy
  kgp.py -g husserl.xml   generates several paragraphs of Husserl
  kpg.py "<xref id='paragraph'/>"  generates a paragraph of Kant
  kgp.py template.xml     reads from template.xml to decide what to generate
"""
from xml.dom import minidom
import random
import toolbox
import sys
import getopt

_debug = 0

class NoSourceError(Exception): pass

class KantGenerator:
    """generates mock philosophy based on a context-free grammar"""

    def __init__(self, grammar, source=None):
        self.loadGrammar(grammar)
        self.loadSource(source and source or self.getDefaultSource())
        self.refresh()

    def _load(self, source):
        """load XML input source, return parsed XML document

        - a URL of a remote XML file ("http://diveintopython.org/kant.xml")
        - a filename of a local XML file ("~/diveintopython/common/py/kant.xml")
        - standard input ("-")
        - the actual XML document, as a string
        """
        sock = toolbox.openAnything(source)
        xmldoc = minidom.parse(sock).documentElement
        sock.close()
        return xmldoc

    def loadGrammar(self, grammar):                         
        """load context-free grammar"""                     
        self.grammar = self._load(grammar)                  
        self.refs = {}                                      
        for ref in self.grammar.getElementsByTagName("ref"):
            self.refs[ref.attributes["id"].value] = ref     

    def loadSource(self, source):
        """load source"""
        self.source = self._load(source)

    def getDefaultSource(self):
        """guess default source of the current grammar
        
        The default source will be one of the <ref>s that is not
        cross-referenced.  This sounds complicated but it's not.
        Example: The default source for kant.xml is
        "<xref id='section'/>", because 'section' is the one <ref>
        that is not <xref>'d anywhere in the grammar.
        In most grammars, the default source will produce the
        longest (and most interesting) output.
        """
        xrefs = {}
        for xref in self.grammar.getElementsByTagName("xref"):
            xrefs[xref.attributes["id"].value] = 1
        xrefs = xrefs.keys()
        standaloneXrefs = [e for e in self.refs.keys() if e not in xrefs]
        if not standaloneXrefs:
            raise NoSourceError, "can't guess source, and no source specified"
        return '<xref id="%s"/>' % random.choice(standaloneXrefs)
        
    def reset(self):
        """reset parser"""
        self.pieces = []
        self.capitalizeNextWord = 0

    def refresh(self):
        """reset output buffer, re-parse entire source file, and return output
        
        Since parsing involves a good deal of randomness, this is an
        easy way to get new output without having to reload a grammar file
        each time.
        """
        self.reset()
        self.parse(self.source)
        return self.output()

    def output(self):
        """output generated text"""
        return "".join(self.pieces)

    def randomChildElement(self, node):
        """choose a random child element of a node
        
        This is a utility method used by do_xref and do_choice.
        """
        choices = [e for e in node.childNodes
                   if e.nodeType == e.ELEMENT_NODE]
        chosen = random.choice(choices)            
        if _debug:                                 
            sys.stderr.write('%s available choices: %s\n' % \
                (len(choices), [e.toxml() for e in choices]))
            sys.stderr.write('Chosen: %s\n' % chosen.toxml())
        return chosen                              

    def parse(self, node):         
        """parse a single XML node
        
        A parsed XML document (from minidom.parse) is a tree of nodes
        of various types.  Each node is represented by an instance of the
        corresponding Python class (Element for a tag, Text for
        text data, Document for the top-level document).  The following
        statement constructs the name of a class method based on the type
        of node we're parsing ("parse_Element" for an Element node,
        "parse_Text" for a Text node, etc.) and then calls the method.
        """
        parseMethod = getattr(self, "parse_%s" % node.__class__.__name__)
        parseMethod(node)

    def parse_Document(self, node):
        """parse the document node
        
        The document node by itself isn't interesting (to us), but
        its only child, node.documentElement, is: it's the root node
        of the grammar.
        """
        self.parse(node.documentElement)

    def parse_Text(self, node):    
        """parse a text node
        
        The text of a text node is usually added to the output buffer
        verbatim.  The one exception is that <p class='sentence'> sets
        a flag to capitalize the first letter of the next word.  If
        that flag is set, we capitalize the text and reset the flag.
        """
        text = node.data
        if self.capitalizeNextWord:
            self.pieces.append(text[0].upper())
            self.pieces.append(text[1:])
            self.capitalizeNextWord = 0
        else:
            self.pieces.append(text)

    def parse_Element(self, node): 
        """parse an element
        
        An XML element corresponds to an actual tag in the source:
        <xref id='...'>, <p chance='...'>, <choice>, etc.
        Each element type is handled in its own method.  Like we did in
        parse(), we construct a method name based on the name of the
        element ("do_xref" for an <xref> tag, etc.) and
        call the method.
        """
        handlerMethod = getattr(self, "do_%s" % node.tagName)
        handlerMethod(node)

    def parse_Comment(self, node):
        """parse a comment
        
        The grammar can contain XML comments, but we ignore them
        """
        pass
    
    def do_xref(self, node):
        """handle <xref id='...'> tag
        
        An <xref id='...'> tag is a cross-reference to a <ref id='...'>
        tag.  <xref id='sentence'/> evaluates to a randomly chosen child of
        <ref id='sentence'>.
        """
        id = node.attributes["id"].value
        self.parse(self.randomChildElement(self.refs[id]))

    def do_p(self, node):
        """handle <p> tag
        
        The <p> tag is the core of the grammar.  It can contain almost
        anything: freeform text, <choice> tags, <xref> tags, even other
        <p> tags.  If a "class='sentence'" attribute is found, a flag
        is set and the next word will be capitalized.  If a "chance='X'"
        attribute is found, there is an X% chance that the tag will be
        evaluated (and therefore a (100-X)% chance that it will be
        completely ignored)
        """
        keys = node.attributes.keys()
        if "class" in keys:
            if node.attributes["class"].value == "sentence":
                self.capitalizeNextWord = 1
        if "chance" in keys:
            chance = int(node.attributes["chance"].value)
            doit = (chance > random.randrange(100))
        else:
            doit = 1
        if doit:
            for child in node.childNodes: self.parse(child)

    def do_choice(self, node):
        """handle <choice> tag
        
        A <choice> tag contains one or more <p> tags.  One <p> tag
        is chosen at random and evaluated; the rest are ignored.
        """
        self.parse(self.randomChildElement(node))

def usage():
    print __doc__

def main(argv):                         
    grammar = "kant.xml"                
    try:                                
        opts, args = getopt.getopt(argv, "hg:d", ["help", "grammar="])
    except getopt.GetoptError:          
        usage()                         
        sys.exit(2)                     
    for opt, arg in opts:               
        if opt in ("-h", "--help"):     
            usage()                     
            sys.exit()                  
        elif opt == '-d':               
            global _debug               
            _debug = 1                  
        elif opt in ("-g", "--grammar"):
            grammar = arg               
    
    source = "".join(args)              

    k = KantGenerator(grammar, source)
    print k.output()

if __name__ == "__main__":
    main(sys.argv[1:])

例 9.2. toolbox.py

"""Miscellaneous utility functions"""

def openAnything(source):            
    """URI, filename, or string --> stream

    This function lets you define parsers that take any input source
    (URL, pathname to local or network file, or actual data as a string)
    and deal with it in a uniform manner.  Returned object is guaranteed
    to have all the basic stdio read methods (read, readline, readlines).
    Just .close() the object when you're done with it.
    
    Examples:
    >>> from xml.dom import minidom
    >>> sock = openAnything("http://localhost/kant.xml")
    >>> doc = minidom.parse(sock)
    >>> sock.close()
    >>> sock = openAnything("c:\\inetpub\\wwwroot\\kant.xml")
    >>> doc = minidom.parse(sock)
    >>> sock.close()
    >>> sock = openAnything("<ref id='conjunction'><text>and</text><text>or</text></ref>")
    >>> doc = minidom.parse(sock)
    >>> sock.close()
    """
    if hasattr(source, "read"):
        return source

    if source == '-':
        import sys
        return sys.stdin

    # try to open with urllib (if source is http, ftp, or file URL)
    import urllib                         
    try:                                  
        return urllib.urlopen(source)     
    except (IOError, OSError):            
        pass                              
    
    # try to open with native open function (if source is pathname)
    try:                                  
        return open(source)               
    except (IOError, OSError):            
        pass                              
    
    # treat source as string
    import StringIO                       
    return StringIO.StringIO(str(source)) 

独立运行程序 kgp.py,它会解析 kant.xml 中默认的基于 XML 的语法,并以康德的风格打印出几段有哲学价值的段落来。

例 9.3. kgp.py 的样例输出

[you@localhost kgp]$ python kgp.py
     As is shown in the writings of Hume, our a priori concepts, in
reference to ends, abstract from all content of knowledge; in the study
of space, the discipline of human reason, in accordance with the
principles of philosophy, is the clue to the discovery of the
Transcendental Deduction.  The transcendental aesthetic, in all
theoretical sciences, occupies part of the sphere of human reason
concerning the existence of our ideas in general; still, the
never-ending regress in the series of empirical conditions constitutes
the whole content for the transcendental unity of apperception.  What
we have alone been able to show is that, even as this relates to the
architectonic of human reason, the Ideal may not contradict itself, but
it is still possible that it may be in contradictions with the
employment of the pure employment of our hypothetical judgements, but
natural causes (and I assert that this is the case) prove the validity
of the discipline of pure reason.  As we have already seen, time (and
it is obvious that this is true) proves the validity of time, and the
architectonic of human reason, in the full sense of these terms,
abstracts from all content of knowledge.  I assert, in the case of the
discipline of practical reason, that the Antinomies are just as
necessary as natural causes, since knowledge of the phenomena is a
posteriori.
    The discipline of human reason, as I have elsewhere shown, is by
its very nature contradictory, but our ideas exclude the possibility of
the Antinomies.  We can deduce that, on the contrary, the pure
employment of philosophy, on the contrary, is by its very nature
contradictory, but our sense perceptions are a representation of, in
the case of space, metaphysics.  The thing in itself is a
representation of philosophy.  Applied logic is the clue to the
discovery of natural causes.  However, what we have alone been able to
show is that our ideas, in other words, should only be used as a canon
for the Ideal, because of our necessary ignorance of the conditions.

[...snip...]

这当然是胡言乱语。噢,不完全是胡言乱语。它在句法和语法上都是正确的 (尽管非常罗嗦――康德可不是你们所说的踩得到点上的那种人)。其中一些实际上是正确的 (或者至少康德可能会认同的事情),其中一些则明显是错误的,大部分只是语无伦次。但所有内容都符合康德的风格。

让我重复一遍,如果你现在或曾经主修哲学专业,这会非常、非常有趣。

有趣之处在于,这个程序中没有一点内容是属于康德的。所有的内容都来自于上下文无关语法文件 kant.xml。如果你要程序使用不同的语法文件 (可以在命令行中指定),输出信息将完全不同。

例 9.4. kgp.py 的简单输出

[you@localhost kgp]$ python kgp.py -g binary.xml
00101001
[you@localhost kgp]$ python kgp.py -g binary.xml
10110100

在本章后面的内容中,你将近距离地观察语法文件的结构。现在,你只要知道语法文件定义了输出信息的结构,而 kgp.py 程序读取语法规则并随机确定哪些单词插入哪里。