Dive Into Python

Example 8.2. dialect.py

import re
from BaseHTMLProcessor import BaseHTMLProcessor

class Dialectizer(BaseHTMLProcessor):
    subs = ()

    def reset(self):
        # extend (called from __init__ in ancestor)
        # Reset all data attributes
        self.verbatim = 0
        BaseHTMLProcessor.reset(self)

    def start_pre(self, attrs):            
        # called for every <pre> tag in HTML source
        # Increment verbatim mode count, then handle tag like normal
        self.verbatim += 1                 
        self.unknown_starttag("pre", attrs)

    def end_pre(self):   
        # called for every </pre> tag in HTML source
        # Decrement verbatim mode count
        self.unknown_endtag("pre")         
        self.verbatim -= 1                 

    def handle_data(self, text):    
        # override
        # called for every block of text in HTML source
        # If in verbatim mode, save text unaltered;
        # otherwise process the text with a series of substitutions
        self.pieces.append(self.verbatim and text or self.process(text))

    def process(self, text):
        # called from handle_data
        # Process text block by performing series of regular expression
        # substitutions (actual substitions are defined in descendant)
        for fromPattern, toPattern in self.subs:
            text = re.sub(fromPattern, toPattern, text)
        return text

class ChefDialectizer(Dialectizer):
    """convert HTML to Swedish Chef-speak
    
    based on the classic chef.x, copyright (c) 1992, 1993 John Hagerman
    """
    subs = ((r'a([nu])', r'u\1'),
            (r'A([nu])', r'U\1'),
            (r'a\B', r'e'),
            (r'A\B', r'E'),
            (r'en\b', r'ee'),
            (r'\Bew', r'oo'),
            (r'\Be\b', r'e-a'),
            (r'\be', r'i'),
            (r'\bE', r'I'),
            (r'\Bf', r'ff'),
            (r'\Bir', r'ur'),
            (r'(\w*?)i(\w*?)$', r'\1ee\2'),
            (r'\bow', r'oo'),
            (r'\bo', r'oo'),
            (r'\bO', r'Oo'),
            (r'the', r'zee'),
            (r'The', r'Zee'),
            (r'th\b', r't'),
            (r'\Btion', r'shun'),
            (r'\Bu', r'oo'),
            (r'\BU', r'Oo'),
            (r'v', r'f'),
            (r'V', r'F'),
            (r'w', r'w'),
            (r'W', r'W'),
            (r'([a-z])[.]', r'\1.  Bork Bork Bork!'))

class FuddDialectizer(Dialectizer):
    """convert HTML to Elmer Fudd-speak"""
    subs = ((r'[rl]', r'w'),
            (r'qu', r'qw'),
            (r'th\b', r'f'),
            (r'th', r'd'),
            (r'n[.]', r'n, uh-hah-hah-hah.'))

class OldeDialectizer(Dialectizer):
    """convert HTML to mock Middle English"""
    subs = ((r'i([bcdfghjklmnpqrstvwxyz])e\b', r'y\1'),
            (r'i([bcdfghjklmnpqrstvwxyz])e', r'y\1\1e'),
            (r'ick\b', r'yk'),
            (r'ia([bcdfghjklmnpqrstvwxyz])', r'e\1e'),
            (r'e[ea]([bcdfghjklmnpqrstvwxyz])', r'e\1e'),
            (r'([bcdfghjklmnpqrstvwxyz])y', r'\1ee'),
            (r'([bcdfghjklmnpqrstvwxyz])er', r'\1re'),
            (r'([aeiou])re\b', r'\1r'),
            (r'ia([bcdfghjklmnpqrstvwxyz])', r'i\1e'),
            (r'tion\b', r'cioun'),
            (r'ion\b', r'ioun'),
            (r'aid', r'ayde'),
            (r'ai', r'ey'),
            (r'ay\b', r'y'),
            (r'ay', r'ey'),
            (r'ant', r'aunt'),
            (r'ea', r'ee'),
            (r'oa', r'oo'),
            (r'ue', r'e'),
            (r'oe', r'o'),
            (r'ou', r'ow'),
            (r'ow', r'ou'),
            (r'\bhe', r'hi'),
            (r've\b', r'veth'),
            (r'se\b', r'e'),
            (r"'s\b", r'es'),
            (r'ic\b', r'ick'),
            (r'ics\b', r'icc'),
            (r'ical\b', r'ick'),
            (r'tle\b', r'til'),
            (r'll\b', r'l'),
            (r'ould\b', r'olde'),
            (r'own\b', r'oune'),
            (r'un\b', r'onne'),
            (r'rry\b', r'rye'),
            (r'est\b', r'este'),
            (r'pt\b', r'pte'),
            (r'th\b', r'the'),
            (r'ch\b', r'che'),
            (r'ss\b', r'sse'),
            (r'([wybdp])\b', r'\1e'),
            (r'([rnt])\b', r'\1\1e'),
            (r'from', r'fro'),
            (r'when', r'whan'))

def translate(url, dialectName="chef"):
    """fetch URL and translate using dialect
    
    dialect in ("chef", "fudd", "olde")"""
    import urllib    
    sock = urllib.urlopen(url)         
    htmlSource = sock.read()           
    sock.close()     
    parserName = "%sDialectizer" % dialectName.capitalize()
    parserClass = globals()[parserName]  
    parser = parserClass()               
    parser.feed(htmlSource)
    parser.close()         
    return parser.output() 

def test(url):
    """test all dialects against URL"""
    for dialect in ("chef", "fudd", "olde"):
        outfile = "%s.html" % dialect
        fsock = open(outfile, "wb")
        fsock.write(translate(url, dialect))
        fsock.close()
        import webbrowser
        webbrowser.open_new(outfile)

if __name__ == "__main__":
    test("http://diveintopython3.org/odbchelper_list.html")

Example 8.3. Output of dialect.py

Running this script will translate Section 3.2, “Introducing Lists” into mock Swedish Chef-speak (from The Muppets), mock Elmer Fudd-speak (from Bugs Bunny cartoons), and mock Middle English (loosely based on Chaucer's The Canterbury Tales). If you look at the HTML source of the output pages, you'll see that all the HTML tags and attributes are untouched, but the text between the tags has been “translated” into the mock language. If you look closer, you'll see that, in fact, only the titles and paragraphs were translated; the code listings and screen examples were left untouched.

<div class="abstract">
<p>Lists awe <span class="application">Pydon</span>'s wowkhowse datatype.
If youw onwy expewience wif wists is awways in
<span class="application">Visuaw Basic</span> ow (God fowbid) de datastowe
in <span class="application">Powewbuiwdew</span>, bwace youwsewf fow
<span class="application">Pydon</span> wists.</p>
</div>

8.2. Introducing sgmllib.py

HTML processing is broken into three steps: breaking down the HTML into its constituent pieces, fiddling with the pieces, and reconstructing the pieces into HTML again. The first step is done by sgmllib.py, a part of the standard Python library.

The key to understanding this chapter is to realize that HTML is not just text, it is structured text. The structure is derived from the more-or-less-hierarchical sequence of start tags and end tags. Usually you don't work with HTML this way; you work with it textually in a text editor, or visually in a web browser or web authoring tool. sgmllib.py presents HTML structurally.

sgmllib.py contains one important class: SGMLParser. SGMLParser parses HTML into useful pieces, like start tags and end tags. As soon as it succeeds in breaking down some data into a useful piece, it calls a method on itself based on what it found. In order to use the parser, you subclass the SGMLParser class and override these methods. This is what I meant when I said that it presents HTML structurally: the structure of the HTML determines the sequence of method calls and the arguments passed to each method.

SGMLParser parses HTML into 8 kinds of data, and calls a separate method for each of them:

Start tag

An HTML tag that starts a block, like <html>, <head>, <body>, or <pre>, or a standalone tag like <br> or <img>. When it finds a start tag tagname, SGMLParser will look for a method called start_tagname or do_tagname. For instance, when it finds a <pre> tag, it will look for a start_pre or do_pre method. If found, SGMLParser calls this method with a list of the tag's attributes; otherwise, it calls unknown_starttag with the tag name and list of attributes.

End tag

An HTML tag that ends a block, like </html>, </head>, </body>, or </pre>. When it finds an end tag, SGMLParser will look for a method called end_tagname. If found, SGMLParser calls this method, otherwise it calls unknown_endtag with the tag name.

Character reference

An escaped character referenced by its decimal or hexadecimal equivalent, like  . When found, SGMLParser calls handle_charref with the text of the decimal or hexadecimal character equivalent.

Entity reference

An HTML entity, like ©. When found, SGMLParser calls handle_entityref with the name of the HTML entity.

Comment

An HTML comment, enclosed in <!-- ... -->. When found, SGMLParser calls handle_comment with the body of the comment.

Processing instruction

An HTML processing instruction, enclosed in <? ... >. When found, SGMLParser calls handle_pi with the body of the processing instruction.

Declaration

An HTML declaration, such as a DOCTYPE, enclosed in <! ... >. When found, SGMLParser calls handle_decl with the body of the declaration.

Text data

A block of text. Anything that doesn't fit into the other 7 categories. When found, SGMLParser calls handle_data with the text.

	Python 2.0 had a bug where SGMLParser would not recognize declarations at all (handle_decl would never be called), which meant that DOCTYPEs were silently ignored. This is fixed in Python 2.1.

sgmllib.py comes with a test suite to illustrate this. You can run sgmllib.py, passing the name of an HTML file on the command line, and it will print out the tags and other elements as it parses them. It does this by subclassing the SGMLParser class and defining unknown_starttag, unknown_endtag, handle_data and other methods which simply print their arguments.

	In the ActivePython IDE on Windows, you can specify command line arguments in the “Run script” dialog. Separate multiple arguments with spaces.

Example 8.4. Sample test of sgmllib.py

Here is a snippet from the table of contents of the HTML version of this book. Of course your paths may vary. (If you haven't downloaded the HTML version of the book, you can do so at http://diveintopython3.org/.

c:\python23\lib> type "c:\downloads\diveintopython3\html\toc\index.html"

<!DOCTYPE html
  PUBLIC "-//W3C//DTD HTML 4.01//EN" "http://www.w3.org/TR/html4/strict.dtd">
<html>
   <head>
      <meta http-equiv="Content-Type" content="text/html; charset=ISO-8859-1">
   
      <title>Dive Into Python</title>
      <link rel="stylesheet" href="diveintopython3.css" type="text/css">

... rest of file omitted for brevity ...

Running this through the test suite of sgmllib.py yields this output:

c:\python23\lib> python sgmllib.py "c:\downloads\diveintopython3\html\toc\index.html"
data: '\n\n'
start tag: <html >
data: '\n   '
start tag: <head>
data: '\n      '
start tag: <meta http-equiv="Content-Type" content="text/html; charset=ISO-8859-1" >
data: '\n   \n      '
start tag: <title>
data: 'Dive Into Python'
end tag: </title>
data: '\n      '
start tag: <link rel="stylesheet" href="diveintopython3.css" type="text/css" >
data: '\n      '

... rest of output omitted for brevity ...

Here's the roadmap for the rest of the chapter:

• Subclass SGMLParser to create classes that extract interesting data out of HTML documents.
• Subclass SGMLParser to create BaseHTMLProcessor, which overrides all 8 handler methods and uses them to reconstruct the original HTML from the pieces.
• Subclass BaseHTMLProcessor to create Dialectizer, which adds some methods to process specific HTML tags specially, and overrides the handle_data method to provide a framework for processing the text blocks between the HTML tags.
• Subclass Dialectizer to create classes that define text processing rules used by Dialectizer.handle_data.
• Write a test suite that grabs a real web page from http://diveintopython3.org/ and processes it.

Along the way, you'll also learn about locals, globals, and dictionary-based string formatting.

8.3. Extracting data from HTML documents

To extract data from HTML documents, subclass the SGMLParser class and define methods for each tag or entity you want to capture.

The first step to extracting data from an HTML document is getting some HTML. If you have some HTML lying around on your hard drive, you can use file functions to read it, but the real fun begins when you get HTML from live web pages.

Example 8.5. Introducing urllib

>>> import urllib   
>>> sock = urllib.urlopen("http://diveintopython3.org/") 
>>> htmlSource = sock.read()          
>>> sock.close()    
>>> print htmlSource
<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd"><html><head>
      <meta http-equiv='Content-Type' content='text/html; charset=ISO-8859-1'>
   <title>Dive Into Python</title>
<link rel='stylesheet' href='diveintopython3.css' type='text/css'>
<link rev='made' href='mailto:mark@diveintopython3.org'>
<meta name='keywords' content='Python, Dive Into Python, tutorial, object-oriented, programming, documentation, book, free'>
<meta name='description' content='a free Python tutorial for experienced programmers'>
</head>
<body bgcolor='white' text='black' link='#0000FF' vlink='#840084' alink='#0000FF'>
<table cellpadding='0' cellspacing='0' border='0' width='100%'>
<tr><td class='header' width='1%' valign='top'>diveintopython3.org</td>
<td width='99%' align='right'><hr size='1' noshade></td></tr>
<tr><td class='tagline' colspan='2'>Python&nbsp;for&nbsp;experienced&nbsp;programmers</td></tr>

[...snip...]

	The urllib module is part of the standard Python library. It contains functions for getting information about and actually retrieving data from Internet-based URLs (mainly web pages).
	The simplest use of urllib is to retrieve the entire text of a web page using the urlopen function. Opening a URL is similar to opening a file. The return value of urlopen is a file-like object, which has some of the same methods as a file object.
	The simplest thing to do with the file-like object returned by urlopen is read, which reads the entire HTML of the web page into a single string. The object also supports readlines, which reads the text line by line into a list.
	When you're done with the object, make sure to close it, just like a normal file object.
	You now have the complete HTML of the home page of http://diveintopython3.org/ in a string, and you're ready to parse it.

Example 8.6. Introducing urllister.py

If you have not already done so, you can download this and other examples used in this book.

from sgmllib import SGMLParser

class URLLister(SGMLParser):
    def reset(self):            
        SGMLParser.reset(self)
        self.urls = []

    def start_a(self, attrs):   
        href = [v for k, v in attrs if k=='href']  
        if href:
            self.urls.extend(href)

	reset is called by the __init__ method of SGMLParser, and it can also be called manually once an instance of the parser has been created. So if you need to do any initialization, do it in reset, not in __init__, so that it will be re-initialized properly when someone re-uses a parser instance.
	start_a is called by SGMLParser whenever it finds an <a> tag. The tag may contain an href attribute, and/or other attributes, like name or title. The attrs parameter is a list of tuples, [(attribute, value), (attribute, value), ...]. Or it may be just an <a>, a valid (if useless) HTML tag, in which case attrs would be an empty list.
	You can find out whether this <a> tag has an href attribute with a simple multi-variable list comprehension.
	String comparisons like k=='href' are always case-sensitive, but that's safe in this case, because SGMLParser converts attribute names to lowercase while building attrs.

Example 8.7. Using urllister.py

>>> import urllib, urllister
>>> usock = urllib.urlopen("http://diveintopython3.org/")
>>> parser = urllister.URLLister()
>>> parser.feed(usock.read())         
>>> usock.close()   
>>> parser.close()  
>>> for url in parser.urls: print url 
toc/index.html
#download
#languages
toc/index.html
appendix/history.html
download/diveintopython3-html-5.0.zip
download/diveintopython3-pdf-5.0.zip
download/diveintopython3-word-5.0.zip
download/diveintopython3-text-5.0.zip
download/diveintopython3-html-flat-5.0.zip
download/diveintopython3-xml-5.0.zip
download/diveintopython3-common-5.0.zip


... rest of output omitted for brevity ...

	Call the feed method, defined in SGMLParser, to get HTML into the parser.[1] It takes a string, which is what usock.read() returns.
	Like files, you should close your URL objects as soon as you're done with them.
	You should close your parser object, too, but for a different reason. You've read all the data and fed it to the parser, but the feed method isn't guaranteed to have actually processed all the HTML you give it; it may buffer it, waiting for more. Be sure to call close to flush the buffer and force everything to be fully parsed.
	Once the parser is closed, the parsing is complete, and parser.urls contains a list of all the linked URLs in the HTML document. (Your output may look different, if the download links have been updated by the time you read this.)

8.4. Introducing BaseHTMLProcessor.py

SGMLParser doesn't produce anything by itself. It parses and parses and parses, and it calls a method for each interesting thing it finds, but the methods don't do anything. SGMLParser is an HTML consumer: it takes HTML and breaks it down into small, structured pieces. As you saw in the previous section, you can subclass SGMLParser to define classes that catch specific tags and produce useful things, like a list of all the links on a web page. Now you'll take this one step further by defining a class that catches everything SGMLParser throws at it and reconstructs the complete HTML document. In technical terms, this class will be an HTML producer.

BaseHTMLProcessor subclasses SGMLParser and provides all 8 essential handler methods: unknown_starttag, unknown_endtag, handle_charref, handle_entityref, handle_comment, handle_pi, handle_decl, and handle_data.

Example 8.8. Introducing BaseHTMLProcessor

class BaseHTMLProcessor(SGMLParser):
    def reset(self):      
        self.pieces = []
        SGMLParser.reset(self)

    def unknown_starttag(self, tag, attrs): 
        strattrs = "".join([' %s="%s"' % (key, value) for key, value in attrs])
        self.pieces.append("<%(tag)s%(strattrs)s>" % locals())

    def unknown_endtag(self, tag):          
        self.pieces.append("</%(tag)s>" % locals())

    def handle_charref(self, ref):          
        self.pieces.append("&#%(ref)s;" % locals())

    def handle_entityref(self, ref):        
        self.pieces.append("&%(ref)s" % locals())
        if htmlentitydefs.entitydefs.has_key(ref):
            self.pieces.append(";")

    def handle_data(self, text):            
        self.pieces.append(text)

    def handle_comment(self, text):         
        self.pieces.append("<!--%(text)s-->" % locals())

    def handle_pi(self, text):              
        self.pieces.append("<?%(text)s>" % locals())

    def handle_decl(self, text):
        self.pieces.append("<!%(text)s>" % locals())

	reset, called by SGMLParser.__init__, initializes self.pieces as an empty list before calling the ancestor method. self.pieces is a data attribute which will hold the pieces of the HTML document you're constructing. Each handler method will reconstruct the HTML that SGMLParser parsed, and each method will append that string to self.pieces. Note that self.pieces is a list. You might be tempted to define it as a string and just keep appending each piece to it. That would work, but Python is much more efficient at dealing with lists.[2]
	Since BaseHTMLProcessor does not define any methods for specific tags (like the start_a method in URLLister), SGMLParser will call unknown_starttag for every start tag. This method takes the tag (tag) and the list of attribute name/value pairs (attrs), reconstructs the original HTML, and appends it to self.pieces. The string formatting here is a little strange; you'll untangle that (and also the odd-looking locals function) later in this chapter.
	Reconstructing end tags is much simpler; just take the tag name and wrap it in the </...> brackets.
	When SGMLParser finds a character reference, it calls handle_charref with the bare reference. If the HTML document contains the reference &#160;, ref will be 160. Reconstructing the original complete character reference just involves wrapping ref in &#...; characters.
	Entity references are similar to character references, but without the hash mark. Reconstructing the original entity reference requires wrapping ref in &...; characters. (Actually, as an erudite reader pointed out to me, it's slightly more complicated than this. Only certain standard HTML entites end in a semicolon; other similar-looking entities do not. Luckily for us, the set of standard HTML entities is defined in a dictionary in a Python module called htmlentitydefs. Hence the extra if statement.)
	Blocks of text are simply appended to self.pieces unaltered.
	HTML comments are wrapped in <!--...--> characters.
	Processing instructions are wrapped in <?...> characters.

The HTML specification requires that all non-HTML (like client-side JavaScript) must be enclosed in HTML comments, but not all web pages do this properly (and all modern web browsers are forgiving if they don't). BaseHTMLProcessor is not forgiving; if script is improperly embedded, it will be parsed as if it were HTML. For instance, if the script contains less-than and equals signs, SGMLParser may incorrectly think that it has found tags and attributes. SGMLParser always converts tags and attribute names to lowercase, which may break the script, and BaseHTMLProcessor always encloses attribute values in double quotes (even if the original HTML document used single quotes or no quotes), which will certainly break the script. Always protect your client-side script within HTML comments.

Example 8.9. BaseHTMLProcessor output

    def output(self):               
        """Return processed HTML as a single string"""
        return "".join(self.pieces) 

	This is the one method in BaseHTMLProcessor that is never called by the ancestor SGMLParser. Since the other handler methods store their reconstructed HTML in self.pieces, this function is needed to join all those pieces into one string. As noted before, Python is great at lists and mediocre at strings, so you only create the complete string when somebody explicitly asks for it.
	If you prefer, you could use the join method of the string module instead: string.join(self.pieces, "")

Further reading

• W3C discusses character and entity references.
• Python Library Reference confirms your suspicions that the htmlentitydefs module is exactly what it sounds like.

8.5. locals and globals

Let's digress from HTML processing for a minute and talk about how Python handles variables. Python has two built-in functions, locals and globals, which provide dictionary-based access to local and global variables.

Remember locals? You first saw it here:

    def unknown_starttag(self, tag, attrs):
        strattrs = "".join([' %s="%s"' % (key, value) for key, value in attrs])
        self.pieces.append("<%(tag)s%(strattrs)s>" % locals())

No, wait, you can't learn about locals yet. First, you need to learn about namespaces. This is dry stuff, but it's important, so pay attention.

Python uses what are called namespaces to keep track of variables. A namespace is just like a dictionary where the keys are names of variables and the dictionary values are the values of those variables. In fact, you can access a namespace as a Python dictionary, as you'll see in a minute.

At any particular point in a Python program, there are several namespaces available. Each function has its own namespace, called the local namespace, which keeps track of the function's variables, including function arguments and locally defined variables. Each module has its own namespace, called the global namespace, which keeps track of the module's variables, including functions, classes, any other imported modules, and module-level variables and constants. And there is the built-in namespace, accessible from any module, which holds built-in functions and exceptions.

When a line of code asks for the value of a variable x, Python will search for that variable in all the available namespaces, in order:

1. local namespace - specific to the current function or class method. If the function defines a local variable x, or has an argument x, Python will use this and stop searching.
2. global namespace - specific to the current module. If the module has defined a variable, function, or class called x, Python will use that and stop searching.
3. built-in namespace - global to all modules. As a last resort, Python will assume that x is the name of built-in function or variable.

If Python doesn't find x in any of these namespaces, it gives up and raises a NameError with the message There is no variable named 'x', which you saw back in Example 3.18, “Referencing an Unbound Variable”, but you didn't appreciate how much work Python was doing before giving you that error.

Python 2.2 introduced a subtle but important change that affects the namespace search order: nested scopes. In versions of Python prior to 2.2, when you reference a variable within a nested function or lambda function, Python will search for that variable in the current (nested or lambda) function's namespace, then in the module's namespace. Python 2.2 will search for the variable in the current (nested or lambda) function's namespace, then in the parent function's namespace, then in the module's namespace. Python 2.1 can work either way; by default, it works like Python 2.0, but you can add the following line of code at the top of your module to make your module work like Python 2.2: from __future__ import nested_scopes

Are you confused yet? Don't despair! This is really cool, I promise. Like many things in Python, namespaces are directly accessible at run-time. How? Well, the local namespace is accessible via the built-in locals function, and the global (module level) namespace is accessible via the built-in globals function.

Example 8.10. Introducing locals

>>> def foo(arg): 
...     x = 1
...     print locals()
...     
>>> foo(7)        
{'arg': 7, 'x': 1}
>>> foo('bar')    
{'arg': 'bar', 'x': 1}

	The function foo has two variables in its local namespace: arg, whose value is passed in to the function, and x, which is defined within the function.
	locals returns a dictionary of name/value pairs. The keys of this dictionary are the names of the variables as strings; the values of the dictionary are the actual values of the variables. So calling foo with 7 prints the dictionary containing the function's two local variables: arg (7) and x (1).
	Remember, Python has dynamic typing, so you could just as easily pass a string in for arg; the function (and the call to locals) would still work just as well. locals works with all variables of all datatypes.

What locals does for the local (function) namespace, globals does for the global (module) namespace. globals is more exciting, though, because a module's namespace is more exciting.^[3] Not only does the module's namespace include module-level variables and constants, it includes all the functions and classes defined in the module. Plus, it includes anything that was imported into the module.

Remember the difference between from module import and import module? With import module, the module itself is imported, but it retains its own namespace, which is why you need to use the module name to access any of its functions or attributes: module.function. But with from module import, you're actually importing specific functions and attributes from another module into your own namespace, which is why you access them directly without referencing the original module they came from. With the globals function, you can actually see this happen.

Example 8.11. Introducing globals

Look at the following block of code at the bottom of BaseHTMLProcessor.py:

if __name__ == "__main__":
    for k, v in globals().items():             
        print k, "=", v

Just so you don't get intimidated, remember that you've seen all this before. The globals function returns a dictionary, and you're iterating through the dictionary using the items method and multi-variable assignment. The only thing new here is the globals function.

Now running the script from the command line gives this output (note that your output may be slightly different, depending on your platform and where you installed Python):

c:\docbook\dip\py> python BaseHTMLProcessor.py

SGMLParser = sgmllib.SGMLParser                
htmlentitydefs = <module 'htmlentitydefs' from 'C:\Python23\lib\htmlentitydefs.py'> 
BaseHTMLProcessor = __main__.BaseHTMLProcessor 
__name__ = __main__          
... rest of output omitted for brevity...

	SGMLParser was imported from sgmllib, using from module import. That means that it was imported directly into the module's namespace, and here it is.
	Contrast this with htmlentitydefs, which was imported using import. That means that the htmlentitydefs module itself is in the namespace, but the entitydefs variable defined within htmlentitydefs is not.
	This module only defines one class, BaseHTMLProcessor, and here it is. Note that the value here is the class itself, not a specific instance of the class.
	Remember the if __name__ trick? When running a module (as opposed to importing it from another module), the built-in __name__ attribute is a special value, __main__. Since you ran this module as a script from the command line, __name__ is __main__, which is why the little test code to print the globals got executed.

	Using the locals and globals functions, you can get the value of arbitrary variables dynamically, providing the variable name as a string. This mirrors the functionality of the getattr function, which allows you to access arbitrary functions dynamically by providing the function name as a string.

There is one other important difference between the locals and globals functions, which you should learn now before it bites you. It will bite you anyway, but at least then you'll remember learning it.

Example 8.12. locals is read-only, globals is not

def foo(arg):
    x = 1
    print locals()    
    locals()["x"] = 2 
    print "x=",x      

z = 7
print "z=",z
foo(3)
globals()["z"] = 8    
print "z=",z          

	Since foo is called with 3, this will print {'arg': 3, 'x': 1}. This should not be a surprise.
	locals is a function that returns a dictionary, and here you are setting a value in that dictionary. You might think that this would change the value of the local variable x to 2, but it doesn't. locals does not actually return the local namespace, it returns a copy. So changing it does nothing to the value of the variables in the local namespace.
	This prints x= 1, not x= 2.
	After being burned by locals, you might think that this wouldn't change the value of z, but it does. Due to internal differences in how Python is implemented (which I'd rather not go into, since I don't fully understand them myself), globals returns the actual global namespace, not a copy: the exact opposite behavior of locals. So any changes to the dictionary returned by globals directly affect your global variables.
	This prints z= 8, not z= 7.

8.6. Dictionary-based string formatting

Why did you learn about locals and globals? So you can learn about dictionary-based string formatting. As you recall, regular string formatting provides an easy way to insert values into strings. Values are listed in a tuple and inserted in order into the string in place of each formatting marker. While this is efficient, it is not always the easiest code to read, especially when multiple values are being inserted. You can't simply scan through the string in one pass and understand what the result will be; you're constantly switching between reading the string and reading the tuple of values.

There is an alternative form of string formatting that uses dictionaries instead of tuples of values.

Example 8.13. Introducing dictionary-based string formatting

>>> params = {"server":"mpilgrim", "database":"master", "uid":"sa", "pwd":"secret"}
>>> "%(pwd)s" % params
'secret'
>>> "%(pwd)s is not a good password for %(uid)s" % params 
'secret is not a good password for sa'
>>> "%(database)s of mind, %(database)s of body" % params 
'master of mind, master of body'

	Instead of a tuple of explicit values, this form of string formatting uses a dictionary, params. And instead of a simple %s marker in the string, the marker contains a name in parentheses. This name is used as a key in the params dictionary and subsitutes the corresponding value, secret, in place of the %(pwd)s marker.
	Dictionary-based string formatting works with any number of named keys. Each key must exist in the given dictionary, or the formatting will fail with a KeyError.
	You can even specify the same key twice; each occurrence will be replaced with the same value.

So why would you use dictionary-based string formatting? Well, it does seem like overkill to set up a dictionary of keys and values simply to do string formatting in the next line; it's really most useful when you happen to have a dictionary of meaningful keys and values already. Like locals.

Example 8.14. Dictionary-based string formatting in BaseHTMLProcessor.py

    def handle_comment(self, text):        
        self.pieces.append("<!--%(text)s-->" % locals()) 

Using the built-in locals function is the most common use of dictionary-based string formatting. It means that you can use the names of local variables within your string (in this case, text, which was passed to the class method as an argument) and each named variable will be replaced by its value. If text is 'Begin page footer', the string formatting "<!--%(text)s-->" % locals() will resolve to the string '<!--Begin page footer-->'.

Example 8.15. More dictionary-based string formatting

    def unknown_starttag(self, tag, attrs):
        strattrs = "".join([' %s="%s"' % (key, value) for key, value in attrs]) 
        self.pieces.append("<%(tag)s%(strattrs)s>" % locals())    

When this method is called, attrs is a list of key/value tuples, just like the items of a dictionary, which means you can use multi-variable assignment to iterate through it. This should be a familiar pattern by now, but there's a lot going on here, so let's break it down: Suppose attrs is [('href', 'index.html'), ('title', 'Go to home page')]. In the first round of the list comprehension, key will get 'href', and value will get 'index.html'. The string formatting ' %s="%s"' % (key, value) will resolve to ' href="index.html"'. This string becomes the first element of the list comprehension's return value. In the second round, key will get 'title', and value will get 'Go to home page'. The string formatting will resolve to ' title="Go to home page"'. The list comprehension returns a list of these two resolved strings, and strattrs will join both elements of this list together to form ' href="index.html" title="Go to home page"'.

Now, using dictionary-based string formatting, you insert the value of tag and strattrs into a string. So if tag is 'a', the final result would be '<a href="index.html" title="Go to home page">', and that is what gets appended to self.pieces.

	Using dictionary-based string formatting with locals is a convenient way of making complex string formatting expressions more readable, but it comes with a price. There is a slight performance hit in making the call to locals, since locals builds a copy of the local namespace.

8.7. Quoting attribute values

A common question on comp.lang.python is “I have a bunch of HTML documents with unquoted attribute values, and I want to properly quote them all. How can I do this?”^[4] (This is generally precipitated by a project manager who has found the HTML-is-a-standard religion joining a large project and proclaiming that all pages must validate against an HTML validator. Unquoted attribute values are a common violation of the HTML standard.) Whatever the reason, unquoted attribute values are easy to fix by feeding HTML through BaseHTMLProcessor.

BaseHTMLProcessor consumes HTML (since it's descended from SGMLParser) and produces equivalent HTML, but the HTML output is not identical to the input. Tags and attribute names will end up in lowercase, even if they started in uppercase or mixed case, and attribute values will be enclosed in double quotes, even if they started in single quotes or with no quotes at all. It is this last side effect that you can take advantage of.

Example 8.16. Quoting attribute values

>>> htmlSource = """        
...     <html>
...     <head>
...     <title>Test page</title>
...     </head>
...     <body>
...     <ul>
...     <li><a href=index.html>Home</a></li>
...     <li><a href=toc.html>Table of contents</a></li>
...     <li><a href=history.html>Revision history</a></li>
...     </body>
...     </html>
...     """
>>> from BaseHTMLProcessor import BaseHTMLProcessor
>>> parser = BaseHTMLProcessor()
>>> parser.feed(htmlSource) 
>>> print parser.output()   
<html>
<head>
<title>Test page</title>
</head>
<body>
<ul>
<li><a href="index.html">Home</a></li>
<li><a href="toc.html">Table of contents</a></li>
<li><a href="history.html">Revision history</a></li>
</body>
</html>

	Note that the attribute values of the href attributes in the <a> tags are not properly quoted. (Also note that you're using triple quotes for something other than a docstring. And directly in the IDE, no less. They're very useful.)
	Feed the parser.
	Using the output function defined in BaseHTMLProcessor, you get the output as a single string, complete with quoted attribute values. While this may seem anti-climactic, think about how much has actually happened here: SGMLParser parsed the entire HTML document, breaking it down into tags, refs, data, and so forth; BaseHTMLProcessor used those elements to reconstruct pieces of HTML (which are still stored in parser.pieces, if you want to see them); finally, you called parser.output, which joined all the pieces of HTML into one string.

8.8. Introducing dialect.py

Dialectizer is a simple (and silly) descendant of BaseHTMLProcessor. It runs blocks of text through a series of substitutions, but it makes sure that anything within a <pre>...</pre> block passes through unaltered.

To handle the <pre> blocks, you define two methods in Dialectizer: start_pre and end_pre.

Example 8.17. Handling specific tags

    def start_pre(self, attrs):             
        self.verbatim += 1
        self.unknown_starttag("pre", attrs) 

    def end_pre(self):    
        self.unknown_endtag("pre")          
        self.verbatim -= 1

	start_pre is called every time SGMLParser finds a <pre> tag in the HTML source. (In a minute, you'll see exactly how this happens.) The method takes a single parameter, attrs, which contains the attributes of the tag (if any). attrs is a list of key/value tuples, just like unknown_starttag takes.
	In the reset method, you initialize a data attribute that serves as a counter for <pre> tags. Every time you hit a <pre> tag, you increment the counter; every time you hit a </pre> tag, you'll decrement the counter. (You could just use this as a flag and set it to 1 and reset it to 0, but it's just as easy to do it this way, and this handles the odd (but possible) case of nested <pre> tags.) In a minute, you'll see how this counter is put to good use.
	That's it, that's the only special processing you do for <pre> tags. Now you pass the list of attributes along to unknown_starttag so it can do the default processing.
	end_pre is called every time SGMLParser finds a </pre> tag. Since end tags can not contain attributes, the method takes no parameters.
	First, you want to do the default processing, just like any other end tag.
	Second, you decrement your counter to signal that this <pre> block has been closed.

At this point, it's worth digging a little further into SGMLParser. I've claimed repeatedly (and you've taken it on faith so far) that SGMLParser looks for and calls specific methods for each tag, if they exist. For instance, you just saw the definition of start_pre and end_pre to handle <pre> and </pre>. But how does this happen? Well, it's not magic, it's just good Python coding.

Example 8.18. SGMLParser

    def finish_starttag(self, tag, attrs):               
        try:        
            method = getattr(self, 'start_' + tag)       
        except AttributeError:         
            try:    
                method = getattr(self, 'do_' + tag)      
            except AttributeError:    
                self.unknown_starttag(tag, attrs)        
                return -1             
            else:   
                self.handle_starttag(tag, method, attrs) 
                return 0              
        else:       
            self.stack.append(tag)    
            self.handle_starttag(tag, method, attrs)    
            return 1 

    def handle_starttag(self, tag, method, attrs):      
        method(attrs)

	At this point, SGMLParser has already found a start tag and parsed the attribute list. The only thing left to do is figure out whether there is a specific handler method for this tag, or whether you should fall back on the default method (unknown_starttag).
	The “magic” of SGMLParser is nothing more than your old friend, getattr. What you may not have realized before is that getattr will find methods defined in descendants of an object as well as the object itself. Here the object is self, the current instance. So if tag is 'pre', this call to getattr will look for a start_pre method on the current instance, which is an instance of the Dialectizer class.
	getattr raises an AttributeError if the method it's looking for doesn't exist in the object (or any of its descendants), but that's okay, because you wrapped the call to getattr inside a try...except block and explicitly caught the AttributeError.
	Since you didn't find a start_xxx method, you'll also look for a do_xxx method before giving up. This alternate naming scheme is generally used for standalone tags, like <br>, which have no corresponding end tag. But you can use either naming scheme; as you can see, SGMLParser tries both for every tag. (You shouldn't define both a start_xxx and do_xxx handler method for the same tag, though; only the start_xxx method will get called.)
	Another AttributeError, which means that the call to getattr failed with do_xxx. Since you found neither a start_xxx nor a do_xxx method for this tag, you catch the exception and fall back on the default method, unknown_starttag.
	Remember, try...except blocks can have an else clause, which is called if no exception is raised during the try...except block. Logically, that means that you did find a do_xxx method for this tag, so you're going to call it.
	By the way, don't worry about these different return values; in theory they mean something, but they're never actually used. Don't worry about the self.stack.append(tag) either; SGMLParser keeps track internally of whether your start tags are balanced by appropriate end tags, but it doesn't do anything with this information either. In theory, you could use this module to validate that your tags were fully balanced, but it's probably not worth it, and it's beyond the scope of this chapter. You have better things to worry about right now.
	start_xxx and do_xxx methods are not called directly; the tag, method, and attributes are passed to this function, handle_starttag, so that descendants can override it and change the way all start tags are dispatched. You don't need that level of control, so you just let this method do its thing, which is to call the method (start_xxx or do_xxx) with the list of attributes. Remember, method is a function, returned from getattr, and functions are objects. (I know you're getting tired of hearing it, and I promise I'll stop saying it as soon as I run out of ways to use it to my advantage.) Here, the function object is passed into this dispatch method as an argument, and this method turns around and calls the function. At this point, you don't need to know what the function is, what it's named, or where it's defined; the only thing you need to know about the function is that it is called with one argument, attrs.

Now back to our regularly scheduled program: Dialectizer. When you left, you were in the process of defining specific handler methods for <pre> and </pre> tags. There's only one thing left to do, and that is to process text blocks with the pre-defined substitutions. For that, you need to override the handle_data method.

Example 8.19. Overriding the handle_data method

    def handle_data(self, text):     
        self.pieces.append(self.verbatim and text or self.process(text)) 

handle_data is called with only one argument, the text to process.

In the ancestor BaseHTMLProcessor, the handle_data method simply appended the text to the output buffer, self.pieces. Here the logic is only slightly more complicated. If you're in the middle of a <pre>...</pre> block, self.verbatim will be some value greater than 0, and you want to put the text in the output buffer unaltered. Otherwise, you will call a separate method to process the substitutions, then put the result of that into the output buffer. In Python, this is a one-liner, using the and-or trick.

You're close to completely understanding Dialectizer. The only missing link is the nature of the text substitutions themselves. If you know any Perl, you know that when complex text substitutions are required, the only real solution is regular expressions. The classes later in dialect.py define a series of regular expressions that operate on the text between the HTML tags. But you just had a whole chapter on regular expressions. You don't really want to slog through regular expressions again, do you? God knows I don't. I think you've learned enough for one chapter.

8.9. Putting it all together

It's time to put everything you've learned so far to good use. I hope you were paying attention.

Example 8.20. The translate function, part 1

def translate(url, dialectName="chef"): 
    import urllib     
    sock = urllib.urlopen(url)          
    htmlSource = sock.read()           
    sock.close()     

	The translate function has an optional argument dialectName, which is a string that specifies the dialect you'll be using. You'll see how this is used in a minute.
	Hey, wait a minute, there's an import statement in this function! That's perfectly legal in Python. You're used to seeing import statements at the top of a program, which means that the imported module is available anywhere in the program. But you can also import modules within a function, which means that the imported module is only available within the function. If you have a module that is only ever used in one function, this is an easy way to make your code more modular. (When you find that your weekend hack has turned into an 800-line work of art and decide to split it up into a dozen reusable modules, you'll appreciate this.)
	Now you get the source of the given URL.

Example 8.21. The translate function, part 2: curiouser and curiouser

    parserName = "%sDialectizer" % dialectName.capitalize() 
    parserClass = globals()[parserName]   
    parser = parserClass()                

	capitalize is a string method you haven't seen before; it simply capitalizes the first letter of a string and forces everything else to lowercase. Combined with some string formatting, you've taken the name of a dialect and transformed it into the name of the corresponding Dialectizer class. If dialectName is the string 'chef', parserName will be the string 'ChefDialectizer'.
	You have the name of a class as a string (parserName), and you have the global namespace as a dictionary (globals()). Combined, you can get a reference to the class which the string names. (Remember, classes are objects, and they can be assigned to variables just like any other object.) If parserName is the string 'ChefDialectizer', parserClass will be the class ChefDialectizer.
	Finally, you have a class object (parserClass), and you want an instance of the class. Well, you already know how to do that: call the class like a function. The fact that the class is being stored in a local variable makes absolutely no difference; you just call the local variable like a function, and out pops an instance of the class. If parserClass is the class ChefDialectizer, parser will be an instance of the class ChefDialectizer.

Why bother? After all, there are only 3 Dialectizer classes; why not just use a case statement? (Well, there's no case statement in Python, but why not just use a series of if statements?) One reason: extensibility. The translate function has absolutely no idea how many Dialectizer classes you've defined. Imagine if you defined a new FooDialectizer tomorrow; translate would work by passing 'foo' as the dialectName.

Even better, imagine putting FooDialectizer in a separate module, and importing it with from module import. You've already seen that this includes it in globals(), so translate would still work without modification, even though FooDialectizer was in a separate file.

Now imagine that the name of the dialect is coming from somewhere outside the program, maybe from a database or from a user-inputted value on a form. You can use any number of server-side Python scripting architectures to dynamically generate web pages; this function could take a URL and a dialect name (both strings) in the query string of a web page request, and output the “translated” web page.

Finally, imagine a Dialectizer framework with a plug-in architecture. You could put each Dialectizer class in a separate file, leaving only the translate function in dialect.py. Assuming a consistent naming scheme, the translate function could dynamic import the appropiate class from the appropriate file, given nothing but the dialect name. (You haven't seen dynamic importing yet, but I promise to cover it in a later chapter.) To add a new dialect, you would simply add an appropriately-named file in the plug-ins directory (like foodialect.py which contains the FooDialectizer class). Calling the translate function with the dialect name 'foo' would find the module foodialect.py, import the class FooDialectizer, and away you go.

Example 8.22. The translate function, part 3

    parser.feed(htmlSource) 
    parser.close()          
    return parser.output()  

	After all that imagining, this is going to seem pretty boring, but the feed function is what does the entire transformation. You had the entire HTML source in a single string, so you only had to call feed once. However, you can call feed as often as you want, and the parser will just keep parsing. So if you were worried about memory usage (or you knew you were going to be dealing with very large HTML pages), you could set this up in a loop, where you read a few bytes of HTML and fed it to the parser. The result would be the same.
	Because feed maintains an internal buffer, you should always call the parser's close method when you're done (even if you fed it all at once, like you did). Otherwise you may find that your output is missing the last few bytes.
	Remember, output is the function you defined on BaseHTMLProcessor that joins all the pieces of output you've buffered and returns them in a single string.

And just like that, you've “translated” a web page, given nothing but a URL and the name of a dialect.

Further reading

• You thought I was kidding about the server-side scripting idea. So did I, until I found this web-based dialectizer. Unfortunately, source code does not appear to be available.

8.10. Summary

Python provides you with a powerful tool, sgmllib.py, to manipulate HTML by turning its structure into an object model. You can use this tool in many different ways.

• parsing the HTML looking for something specific
• aggregating the results, like the URL lister
• altering the structure along the way, like the attribute quoter
• transforming the HTML into something else by manipulating the text while leaving the tags alone, like the Dialectizer

Along with these examples, you should be comfortable doing all of the following things:

• Using locals() and globals() to access namespaces
• Formatting strings using dictionary-based substitutions

---

^[1] The technical term for a parser like SGMLParser is a consumer: it consumes HTML and breaks it down. Presumably, the name feed was chosen to fit into the whole “consumer” motif. Personally, it makes me think of an exhibit in the zoo where there's just a dark cage with no trees or plants or evidence of life of any kind, but if you stand perfectly still and look really closely you can make out two beady eyes staring back at you from the far left corner, but you convince yourself that that's just your mind playing tricks on you, and the only way you can tell that the whole thing isn't just an empty cage is a small innocuous sign on the railing that reads, “Do not feed the parser.” But maybe that's just me. In any event, it's an interesting mental image.

^[2] The reason Python is better at lists than strings is that lists are mutable but strings are immutable. This means that appending to a list just adds the element and updates the index. Since strings can not be changed after they are created, code like s = s + newpiece will create an entirely new string out of the concatenation of the original and the new piece, then throw away the original string. This involves a lot of expensive memory management, and the amount of effort involved increases as the string gets longer, so doing s = s + newpiece in a loop is deadly. In technical terms, appending n items to a list is O(n), while appending n items to a string is O(n2).

^[3] I don't get out much.

^[4] All right, it's not that common a question. It's not up there with “What editor should I use to write Python code?” (answer: Emacs) or “Is Python better or worse than Perl?” (answer: “Perl is worse than Python because people wanted it worse.” -Larry Wall, 10/14/1998) But questions about HTML processing pop up in one form or another about once a month, and among those questions, this is a popular one.

Chapter 9. XML Processing

9.1. Diving in

These next two chapters are about XML processing in Python. It would be helpful if you already knew what an XML document looks like, that it's made up of structured tags to form a hierarchy of elements, and so on. If this doesn't make sense to you, there are many XML tutorials that can explain the basics.

If you're not particularly interested in XML, you should still read these chapters, which cover important topics like Python packages, Unicode, command line arguments, and how to use getattr for method dispatching.

Being a philosophy major is not required, although if you have ever had the misfortune of being subjected to the writings of Immanuel Kant, you will appreciate the example program a lot more than if you majored in something useful, like computer science.

There are two basic ways to work with XML. One is called SAX (“Simple API for XML”), and it works by reading the XML a little bit at a time and calling a method for each element it finds. (If you read Chapter 8, HTML Processing, this should sound familiar, because that's how the sgmllib module works.) The other is called DOM (“Document Object Model”), and it works by reading in the entire XML document at once and creating an internal representation of it using native Python classes linked in a tree structure. Python has standard modules for both kinds of parsing, but this chapter will only deal with using the DOM.

The following is a complete Python program which generates pseudo-random output based on a context-free grammar defined in an XML format. Don't worry yet if you don't understand what that means; you'll examine both the program's input and its output in more depth throughout these next two chapters.

Example 9.1. kgp.py

If you have not already done so, you can download this and other examples used in this book.

"""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.pygenerates 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://diveintopython3.org/kant.xml")
        - a filename of a local XML file ("~/diveintopython3/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:])

Example 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)) 

Run the program kgp.py by itself, and it will parse the default XML-based grammar, in kant.xml, and print several paragraphs worth of philosophy in the style of Immanuel Kant.

Example 9.3. Sample output of kgp.py

[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...]

This is, of course, complete gibberish. Well, not complete gibberish. It is syntactically and grammatically correct (although very verbose -- Kant wasn't what you would call a get-to-the-point kind of guy). Some of it may actually be true (or at least the sort of thing that Kant would have agreed with), some of it is blatantly false, and most of it is simply incoherent. But all of it is in the style of Immanuel Kant.

Let me repeat that this is much, much funnier if you are now or have ever been a philosophy major.

The interesting thing about this program is that there is nothing Kant-specific about it. All the content in the previous example was derived from the grammar file, kant.xml. If you tell the program to use a different grammar file (which you can specify on the command line), the output will be completely different.

Example 9.4. Simpler output from kgp.py

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

You will take a closer look at the structure of the grammar file later in this chapter. For now, all you need to know is that the grammar file defines the structure of the output, and the kgp.py program reads through the grammar and makes random decisions about which words to plug in where.

9.2. Packages

Actually parsing an XML document is very simple: one line of code. However, before you get to that line of code, you need to take a short detour to talk about packages.

Example 9.5. Loading an XML document (a sneak peek)

>>> from xml.dom import minidom 
>>> xmldoc = minidom.parse('~/diveintopython3/common/py/kgp/binary.xml')

This is a syntax you haven't seen before. It looks almost like the from module import you know and love, but the "." gives it away as something above and beyond a simple import. In fact, xml is what is known as a package, dom is a nested package within xml, and minidom is a module within xml.dom.

That sounds complicated, but it's really not. Looking at the actual implementation may help. Packages are little more than directories of modules; nested packages are subdirectories. The modules within a package (or a nested package) are still just .py files, like always, except that they're in a subdirectory instead of the main lib/ directory of your Python installation.

Example 9.6. File layout of a package

Python21/           root Python installation (home of the executable)
|
+--lib/             library directory (home of the standard library modules)
   |
   +-- xml/         xml package (really just a directory with other stuff in it)
       |
       +--sax/      xml.sax package (again, just a directory)
       |
       +--dom/      xml.dom package (contains minidom.py)
       |
       +--parsers/  xml.parsers package (used internally)

So when you say from xml.dom import minidom, Python figures out that that means “look in the xml directory for a dom directory, and look in that for the minidom module, and import it as minidom”. But Python is even smarter than that; not only can you import entire modules contained within a package, you can selectively import specific classes or functions from a module contained within a package. You can also import the package itself as a module. The syntax is all the same; Python figures out what you mean based on the file layout of the package, and automatically does the right thing.

Example 9.7. Packages are modules, too

>>> from xml.dom import minidom         
>>> minidom
<module 'xml.dom.minidom' from 'C:\Python21\lib\xml\dom\minidom.pyc'>
>>> minidom.Element
<class xml.dom.minidom.Element at 01095744>
>>> from xml.dom.minidom import Element 
>>> Element
<class xml.dom.minidom.Element at 01095744>
>>> minidom.Element
<class xml.dom.minidom.Element at 01095744>
>>> from xml import dom                 
>>> dom
<module 'xml.dom' from 'C:\Python21\lib\xml\dom\__init__.pyc'>
>>> import xml        
>>> xml
<module 'xml' from 'C:\Python21\lib\xml\__init__.pyc'>

	Here you're importing a module (minidom) from a nested package (xml.dom). The result is that minidom is imported into your namespace, and in order to reference classes within the minidom module (like Element), you need to preface them with the module name.
	Here you are importing a class (Element) from a module (minidom) from a nested package (xml.dom). The result is that Element is imported directly into your namespace. Note that this does not interfere with the previous import; the Element class can now be referenced in two ways (but it's all still the same class).
	Here you are importing the dom package (a nested package of xml) as a module in and of itself. Any level of a package can be treated as a module, as you'll see in a moment. It can even have its own attributes and methods, just the modules you've seen before.
	Here you are importing the root level xml package as a module.

So how can a package (which is just a directory on disk) be imported and treated as a module (which is always a file on disk)? The answer is the magical __init__.py file. You see, packages are not simply directories; they are directories with a specific file, __init__.py, inside. This file defines the attributes and methods of the package. For instance, xml.dom contains a Node class, which is defined in xml/dom/__init__.py. When you import a package as a module (like dom from xml), you're really importing its __init__.py file.

	A package is a directory with the special __init__.py file in it. The __init__.py file defines the attributes and methods of the package. It doesn't need to define anything; it can just be an empty file, but it has to exist. But if __init__.py doesn't exist, the directory is just a directory, not a package, and it can't be imported or contain modules or nested packages.

So why bother with packages? Well, they provide a way to logically group related modules. Instead of having an xml package with sax and dom packages inside, the authors could have chosen to put all the sax functionality in xmlsax.py and all the dom functionality in xmldom.py, or even put all of it in a single module. But that would have been unwieldy (as of this writing, the XML package has over 3000 lines of code) and difficult to manage (separate source files mean multiple people can work on different areas simultaneously).

If you ever find yourself writing a large subsystem in Python (or, more likely, when you realize that your small subsystem has grown into a large one), invest some time designing a good package architecture. It's one of the many things Python is good at, so take advantage of it.

9.3. Parsing XML

As I was saying, actually parsing an XML document is very simple: one line of code. Where you go from there is up to you.

Example 9.8. Loading an XML document (for real this time)

>>> from xml.dom import minidom      
>>> xmldoc = minidom.parse('~/diveintopython3/common/py/kgp/binary.xml')  
>>> xmldoc         
<xml.dom.minidom.Document instance at 010BE87C>
>>> print xmldoc.toxml()             
<?xml version="1.0" ?>
<grammar>
<ref id="bit">
  <p>0</p>
  <p>1</p>
</ref>
<ref id="byte">
  <p><xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/>\
<xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/></p>
</ref>
</grammar>

	As you saw in the previous section, this imports the minidom module from the xml.dom package.
	Here is the one line of code that does all the work: minidom.parse takes one argument and returns a parsed representation of the XML document. The argument can be many things; in this case, it's simply a filename of an XML document on my local disk. (To follow along, you'll need to change the path to point to your downloaded examples directory.) But you can also pass a file object, or even a file-like object. You'll take advantage of this flexibility later in this chapter.
	The object returned from minidom.parse is a Document object, a descendant of the Node class. This Document object is the root level of a complex tree-like structure of interlocking Python objects that completely represent the XML document you passed to minidom.parse.
	toxml is a method of the Node class (and is therefore available on the Document object you got from minidom.parse). toxml prints out the XML that this Node represents. For the Document node, this prints out the entire XML document.

Now that you have an XML document in memory, you can start traversing through it.

Example 9.9. Getting child nodes

>>> xmldoc.childNodes    
[<DOM Element: grammar at 17538908>]
>>> xmldoc.childNodes[0] 
<DOM Element: grammar at 17538908>
>>> xmldoc.firstChild    
<DOM Element: grammar at 17538908>

	Every Node has a childNodes attribute, which is a list of the Node objects. A Document always has only one child node, the root element of the XML document (in this case, the grammar element).
	To get the first (and in this case, the only) child node, just use regular list syntax. Remember, there is nothing special going on here; this is just a regular Python list of regular Python objects.
	Since getting the first child node of a node is a useful and common activity, the Node class has a firstChild attribute, which is synonymous with childNodes[0]. (There is also a lastChild attribute, which is synonymous with childNodes[-1].)

Example 9.10. toxml works on any node

>>> grammarNode = xmldoc.firstChild
>>> print grammarNode.toxml() 
<grammar>
<ref id="bit">
  <p>0</p>
  <p>1</p>
</ref>
<ref id="byte">
  <p><xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/>\
<xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/></p>
</ref>
</grammar>

Since the toxml method is defined in the Node class, it is available on any XML node, not just the Document element.

Example 9.11. Child nodes can be text

>>> grammarNode.childNodes
[<DOM Text node "\n">, <DOM Element: ref at 17533332>, \
<DOM Text node "\n">, <DOM Element: ref at 17549660>, <DOM Text node "\n">]
>>> print grammarNode.firstChild.toxml()    



>>> print grammarNode.childNodes[1].toxml() 
<ref id="bit">
  <p>0</p>
  <p>1</p>
</ref>
>>> print grammarNode.childNodes[3].toxml() 
<ref id="byte">
  <p><xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/>\
<xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/></p>
</ref>
>>> print grammarNode.lastChild.toxml()     

	Looking at the XML in binary.xml, you might think that the grammar has only two child nodes, the two ref elements. But you're missing something: the carriage returns! After the '<grammar>' and before the first '<ref>' is a carriage return, and this text counts as a child node of the grammar element. Similarly, there is a carriage return after each '</ref>'; these also count as child nodes. So grammar.childNodes is actually a list of 5 objects: 3 Text objects and 2 Element objects.
	The first child is a Text object representing the carriage return after the '<grammar>' tag and before the first '<ref>' tag.
	The second child is an Element object representing the first ref element.
	The fourth child is an Element object representing the second ref element.
	The last child is a Text object representing the carriage return after the '</ref>' end tag and before the '</grammar>' end tag.

Example 9.12. Drilling down all the way to text

>>> grammarNode
<DOM Element: grammar at 19167148>
>>> refNode = grammarNode.childNodes[1] 
>>> refNode
<DOM Element: ref at 17987740>
>>> refNode.childNodes
[<DOM Text node "\n">, <DOM Text node "  ">, <DOM Element: p at 19315844>, \
<DOM Text node "\n">, <DOM Text node "  ">, \
<DOM Element: p at 19462036>, <DOM Text node "\n">]
>>> pNode = refNode.childNodes[2]
>>> pNode
<DOM Element: p at 19315844>
>>> print pNode.toxml()                 
<p>0</p>
>>> pNode.firstChild  
<DOM Text node "0">
>>> pNode.firstChild.data               
u'0'

	As you saw in the previous example, the first ref element is grammarNode.childNodes[1], since childNodes[0] is a Text node for the carriage return.
	The ref element has its own set of child nodes, one for the carriage return, a separate one for the spaces, one for the p element, and so forth.
	You can even use the toxml method here, deeply nested within the document.
	The p element has only one child node (you can't tell that from this example, but look at pNode.childNodes if you don't believe me), and it is a Text node for the single character '0'.
	The .data attribute of a Text node gives you the actual string that the text node represents. But what is that 'u' in front of the string? The answer to that deserves its own section.

9.4. Unicode

Unicode is a system to represent characters from all the world's different languages. When Python parses an XML document, all data is stored in memory as unicode.

You'll get to all that in a minute, but first, some background.

Historical note. Before unicode, there were separate character encoding systems for each language, each using the same numbers (0-255) to represent that language's characters. Some languages (like Russian) have multiple conflicting standards about how to represent the same characters; other languages (like Japanese) have so many characters that they require multiple-byte character sets. Exchanging documents between systems was difficult because there was no way for a computer to tell for certain which character encoding scheme the document author had used; the computer only saw numbers, and the numbers could mean different things. Then think about trying to store these documents in the same place (like in the same database table); you would need to store the character encoding alongside each piece of text, and make sure to pass it around whenever you passed the text around. Then think about multilingual documents, with characters from multiple languages in the same document. (They typically used escape codes to switch modes; poof, you're in Russian koi8-r mode, so character 241 means this; poof, now you're in Mac Greek mode, so character 241 means something else. And so on.) These are the problems which unicode was designed to solve.

To solve these problems, unicode represents each character as a 2-byte number, from 0 to 65535.^[5] Each 2-byte number represents a unique character used in at least one of the world's languages. (Characters that are used in multiple languages have the same numeric code.) There is exactly 1 number per character, and exactly 1 character per number. Unicode data is never ambiguous.

Of course, there is still the matter of all these legacy encoding systems. 7-bit ASCII, for instance, which stores English characters as numbers ranging from 0 to 127. (65 is capital “A”, 97 is lowercase “a”, and so forth.) English has a very simple alphabet, so it can be completely expressed in 7-bit ASCII. Western European languages like French, Spanish, and German all use an encoding system called ISO-8859-1 (also called “latin-1”), which uses the 7-bit ASCII characters for the numbers 0 through 127, but then extends into the 128-255 range for characters like n-with-a-tilde-over-it (241), and u-with-two-dots-over-it (252). And unicode uses the same characters as 7-bit ASCII for 0 through 127, and the same characters as ISO-8859-1 for 128 through 255, and then extends from there into characters for other languages with the remaining numbers, 256 through 65535.

When dealing with unicode data, you may at some point need to convert the data back into one of these other legacy encoding systems. For instance, to integrate with some other computer system which expects its data in a specific 1-byte encoding scheme, or to print it to a non-unicode-aware terminal or printer. Or to store it in an XML document which explicitly specifies the encoding scheme.

And on that note, let's get back to Python.

Python has had unicode support throughout the language since version 2.0. The XML package uses unicode to store all parsed XML data, but you can use unicode anywhere.

Example 9.13. Introducing unicode

>>> s = u'Dive in'            
>>> s
u'Dive in'
>>> print s 
Dive in

To create a unicode string instead of a regular ASCII string, add the letter “u” before the string. Note that this particular string doesn't have any non-ASCII characters. That's fine; unicode is a superset of ASCII (a very large superset at that), so any regular ASCII string can also be stored as unicode.

When printing a string, Python will attempt to convert it to your default encoding, which is usually ASCII. (More on this in a minute.) Since this unicode string is made up of characters that are also ASCII characters, printing it has the same result as printing a normal ASCII string; the conversion is seamless, and if you didn't know that s was a unicode string, you'd never notice the difference.

Example 9.14. Storing non-ASCII characters

>>> s = u'La Pe\xf1a'         
>>> print s 
Traceback (innermost last):
  File "<interactive input>", line 1, in ?
UnicodeError: ASCII encoding error: ordinal not in range(128)
>>> print s.encode('latin-1') 
La Peña

	The real advantage of unicode, of course, is its ability to store non-ASCII characters, like the Spanish “ñ” (n with a tilde over it). The unicode character code for the tilde-n is 0xf1 in hexadecimal (241 in decimal), which you can type like this: \xf1.
	Remember I said that the print function attempts to convert a unicode string to ASCII so it can print it? Well, that's not going to work here, because your unicode string contains non-ASCII characters, so Python raises a UnicodeError error.
	Here's where the conversion-from-unicode-to-other-encoding-schemes comes in. s is a unicode string, but print can only print a regular string. To solve this problem, you call the encode method, available on every unicode string, to convert the unicode string to a regular string in the given encoding scheme, which you pass as a parameter. In this case, you're using latin-1 (also known as iso-8859-1), which includes the tilde-n (whereas the default ASCII encoding scheme did not, since it only includes characters numbered 0 through 127).

Remember I said Python usually converted unicode to ASCII whenever it needed to make a regular string out of a unicode string? Well, this default encoding scheme is an option which you can customize.

Example 9.15. sitecustomize.py

# sitecustomize.py 
# this file can be anywhere in your Python path,
# but it usually goes in ${pythondir}/lib/site-packages/
import sys
sys.setdefaultencoding('iso-8859-1') 

	sitecustomize.py is a special script; Python will try to import it on startup, so any code in it will be run automatically. As the comment mentions, it can go anywhere (as long as import can find it), but it usually goes in the site-packages directory within your Python lib directory.
	setdefaultencoding function sets, well, the default encoding. This is the encoding scheme that Python will try to use whenever it needs to auto-coerce a unicode string into a regular string.

Example 9.16. Effects of setting the default encoding

>>> import sys
>>> sys.getdefaultencoding() 
'iso-8859-1'
>>> s = u'La Pe\xf1a'
>>> print s
La Peña

This example assumes that you have made the changes listed in the previous example to your sitecustomize.py file, and restarted Python. If your default encoding still says 'ascii', you didn't set up your sitecustomize.py properly, or you didn't restart Python. The default encoding can only be changed during Python startup; you can't change it later. (Due to some wacky programming tricks that I won't get into right now, you can't even call sys.setdefaultencoding after Python has started up. Dig into site.py and search for “setdefaultencoding” to find out how.)

Now that the default encoding scheme includes all the characters you use in your string, Python has no problem auto-coercing the string and printing it.

Example 9.17. Specifying encoding in .py files

If you are going to be storing non-ASCII strings within your Python code, you'll need to specify the encoding of each individual .py file by putting an encoding declaration at the top of each file. This declaration defines the .py file to be UTF-8:

#!/usr/bin/env python
# -*- coding: UTF-8 -*-

Now, what about XML? Well, every XML document is in a specific encoding. Again, ISO-8859-1 is a popular encoding for data in Western European languages. KOI8-R is popular for Russian texts. The encoding, if specified, is in the header of the XML document.

Example 9.18. russiansample.xml

<?xml version="1.0" encoding="koi8-r"?>       
<preface>
<title>Предисловие</title>  
</preface>

	This is a sample extract from a real Russian XML document; it's part of a Russian translation of this very book. Note the encoding, koi8-r, specified in the header.
	These are Cyrillic characters which, as far as I know, spell the Russian word for “Preface”. If you open this file in a regular text editor, the characters will most likely like gibberish, because they're encoded using the koi8-r encoding scheme, but they're being displayed in iso-8859-1.

Example 9.19. Parsing russiansample.xml

>>> from xml.dom import minidom
>>> xmldoc = minidom.parse('russiansample.xml') 
>>> title = xmldoc.getElementsByTagName('title')[0].firstChild.data
>>> title   
u'\u041f\u0440\u0435\u0434\u0438\u0441\u043b\u043e\u0432\u0438\u0435'
>>> print title               
Traceback (innermost last):
  File "<interactive input>", line 1, in ?
UnicodeError: ASCII encoding error: ordinal not in range(128)
>>> convertedtitle = title.encode('koi8-r')     
>>> convertedtitle
'\xf0\xd2\xc5\xc4\xc9\xd3\xcc\xcf\xd7\xc9\xc5'
>>> print convertedtitle      
Предисловие

	I'm assuming here that you saved the previous example as russiansample.xml in the current directory. I am also, for the sake of completeness, assuming that you've changed your default encoding back to 'ascii' by removing your sitecustomize.py file, or at least commenting out the setdefaultencoding line.
	Note that the text data of the title tag (now in the title variable, thanks to that long concatenation of Python functions which I hastily skipped over and, annoyingly, won't explain until the next section) -- the text data inside the XML document's title element is stored in unicode.
	Printing the title is not possible, because this unicode string contains non-ASCII characters, so Python can't convert it to ASCII because that doesn't make sense.
	You can, however, explicitly convert it to koi8-r, in which case you get a (regular, not unicode) string of single-byte characters (f0, d2, c5, and so forth) that are the koi8-r-encoded versions of the characters in the original unicode string.
	Printing the koi8-r-encoded string will probably show gibberish on your screen, because your Python IDE is interpreting those characters as iso-8859-1, not koi8-r. But at least they do print. (And, if you look carefully, it's the same gibberish that you saw when you opened the original XML document in a non-unicode-aware text editor. Python converted it from koi8-r into unicode when it parsed the XML document, and you've just converted it back.)

To sum up, unicode itself is a bit intimidating if you've never seen it before, but unicode data is really very easy to handle in Python. If your XML documents are all 7-bit ASCII (like the examples in this chapter), you will literally never think about unicode. Python will convert the ASCII data in the XML documents into unicode while parsing, and auto-coerce it back to ASCII whenever necessary, and you'll never even notice. But if you need to deal with that in other languages, Python is ready.

Further reading

• Unicode.org is the home page of the unicode standard, including a brief technical introduction.
• Unicode Tutorial has some more examples of how to use Python's unicode functions, including how to force Python to coerce unicode into ASCII even when it doesn't really want to.
• PEP 263 goes into more detail about how and when to define a character encoding in your .py files.

9.5. Searching for elements

Traversing XML documents by stepping through each node can be tedious. If you're looking for something in particular, buried deep within your XML document, there is a shortcut you can use to find it quickly: getElementsByTagName.

For this section, you'll be using the binary.xml grammar file, which looks like this:

Example 9.20. binary.xml

<?xml version="1.0"?>
<!DOCTYPE grammar PUBLIC "-//diveintopython3.org//DTD Kant Generator Pro v1.0//EN" "kgp.dtd">
<grammar>
<ref id="bit">
  <p>0</p>
  <p>1</p>
</ref>
<ref id="byte">
  <p><xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/>\
<xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/></p>
</ref>
</grammar>

It has two refs, 'bit' and 'byte'. A bit is either a '0' or '1', and a byte is 8 bits.

Example 9.21. Introducing getElementsByTagName

>>> from xml.dom import minidom
>>> xmldoc = minidom.parse('binary.xml')
>>> reflist = xmldoc.getElementsByTagName('ref') 
>>> reflist
[<DOM Element: ref at 136138108>, <DOM Element: ref at 136144292>]
>>> print reflist[0].toxml()
<ref id="bit">
  <p>0</p>
  <p>1</p>
</ref>
>>> print reflist[1].toxml()
<ref id="byte">
  <p><xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/>\
<xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/></p>
</ref>

getElementsByTagName takes one argument, the name of the element you wish to find. It returns a list of Element objects, corresponding to the XML elements that have that name. In this case, you find two ref elements.

Example 9.22. Every element is searchable

>>> firstref = reflist[0]    
>>> print firstref.toxml()
<ref id="bit">
  <p>0</p>
  <p>1</p>
</ref>
>>> plist = firstref.getElementsByTagName("p") 
>>> plist
[<DOM Element: p at 136140116>, <DOM Element: p at 136142172>]
>>> print plist[0].toxml()   
<p>0</p>
>>> print plist[1].toxml()
<p>1</p>

	Continuing from the previous example, the first object in your reflist is the 'bit' ref element.
	You can use the same getElementsByTagName method on this Element to find all the <p> elements within the 'bit' ref element.
	Just as before, the getElementsByTagName method returns a list of all the elements it found. In this case, you have two, one for each bit.

Example 9.23. Searching is actually recursive

>>> plist = xmldoc.getElementsByTagName("p") 
>>> plist
[<DOM Element: p at 136140116>, <DOM Element: p at 136142172>, <DOM Element: p at 136146124>]
>>> plist[0].toxml()       
'<p>0</p>'
>>> plist[1].toxml()
'<p>1</p>'
>>> plist[2].toxml()       
'<p><xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/>\
<xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/></p>'

	Note carefully the difference between this and the previous example. Previously, you were searching for p elements within firstref, but here you are searching for p elements within xmldoc, the root-level object that represents the entire XML document. This does find the p elements nested within the ref elements within the root grammar element.
	The first two p elements are within the first ref (the 'bit' ref).
	The last p element is the one within the second ref (the 'byte' ref).

9.6. Accessing element attributes

XML elements can have one or more attributes, and it is incredibly simple to access them once you have parsed an XML document.

For this section, you'll be using the binary.xml grammar file that you saw in the previous section.

This section may be a little confusing, because of some overlapping terminology. Elements in an XML document have attributes, and Python objects also have attributes. When you parse an XML document, you get a bunch of Python objects that represent all the pieces of the XML document, and some of these Python objects represent attributes of the XML elements. But the (Python) objects that represent the (XML) attributes also have (Python) attributes, which are used to access various parts of the (XML) attribute that the object represents. I told you it was confusing. I am open to suggestions on how to distinguish these more clearly.

Example 9.24. Accessing element attributes

>>> xmldoc = minidom.parse('binary.xml')
>>> reflist = xmldoc.getElementsByTagName('ref')
>>> bitref = reflist[0]
>>> print bitref.toxml()
<ref id="bit">
  <p>0</p>
  <p>1</p>
</ref>
>>> bitref.attributes          
<xml.dom.minidom.NamedNodeMap instance at 0x81e0c9c>
>>> bitref.attributes.keys()    
[u'id']
>>> bitref.attributes.values() 
[<xml.dom.minidom.Attr instance at 0x81d5044>]
>>> bitref.attributes["id"]    
<xml.dom.minidom.Attr instance at 0x81d5044>

	Each Element object has an attribute called attributes, which is a NamedNodeMap object. This sounds scary, but it's not, because a NamedNodeMap is an object that acts like a dictionary, so you already know how to use it.
	Treating the NamedNodeMap as a dictionary, you can get a list of the names of the attributes of this element by using attributes.keys(). This element has only one attribute, 'id'.
	Attribute names, like all other text in an XML document, are stored in unicode.
	Again treating the NamedNodeMap as a dictionary, you can get a list of the values of the attributes by using attributes.values(). The values are themselves objects, of type Attr. You'll see how to get useful information out of this object in the next example.
	Still treating the NamedNodeMap as a dictionary, you can access an individual attribute by name, using normal dictionary syntax. (Readers who have been paying extra-close attention will already know how the NamedNodeMap class accomplishes this neat trick: by defining a __getitem__ special method. Other readers can take comfort in the fact that they don't need to understand how it works in order to use it effectively.)

Example 9.25. Accessing individual attributes

>>> a = bitref.attributes["id"]
>>> a
<xml.dom.minidom.Attr instance at 0x81d5044>
>>> a.name  
u'id'
>>> a.value 
u'bit'

	The Attr object completely represents a single XML attribute of a single XML element. The name of the attribute (the same name as you used to find this object in the bitref.attributes NamedNodeMap pseudo-dictionary) is stored in a.name.
	The actual text value of this XML attribute is stored in a.value.

Like a dictionary, attributes of an XML element have no ordering. Attributes may happen to be listed in a certain order in the original XML document, and the Attr objects may happen to be listed in a certain order when the XML document is parsed into Python objects, but these orders are arbitrary and should carry no special meaning. You should always access individual attributes by name, like the keys of a dictionary.

9.7. Segue

OK, that's it for the hard-core XML stuff. The next chapter will continue to use these same example programs, but focus on other aspects that make the program more flexible: using streams for input processing, using getattr for method dispatching, and using command-line flags to allow users to reconfigure the program without changing the code.

Before moving on to the next chapter, you should be comfortable doing all of these things:

• Parsing XML documents using minidom, searching through the parsed document, and accessing arbitrary element attributes and element children
• Organizing complex libraries into packages
• Converting unicode strings to different character encodings

---

^[5] This, sadly, is still an oversimplification. Unicode now has been extended to handle ancient Chinese, Korean, and Japanese texts, which had so many different characters that the 2-byte unicode system could not represent them all. But Python doesn't currently support that out of the box, and I don't know if there is a project afoot to add it. You've reached the limits of my expertise, sorry.

Chapter 10. Scripts and Streams

10.1. Abstracting input sources

One of Python's greatest strengths is its dynamic binding, and one powerful use of dynamic binding is the file-like object.

Many functions which require an input source could simply take a filename, go open the file for reading, read it, and close it when they're done. But they don't. Instead, they take a file-like object.

In the simplest case, a file-like object is any object with a read method with an optional size parameter, which returns a string. When called with no size parameter, it reads everything there is to read from the input source and returns all the data as a single string. When called with a size parameter, it reads that much from the input source and returns that much data; when called again, it picks up where it left off and returns the next chunk of data.

This is how reading from real files works; the difference is that you're not limiting yourself to real files. The input source could be anything: a file on disk, a web page, even a hard-coded string. As long as you pass a file-like object to the function, and the function simply calls the object's read method, the function can handle any kind of input source without specific code to handle each kind.

In case you were wondering how this relates to XML processing, minidom.parse is one such function which can take a file-like object.

Example 10.1. Parsing XML from a file

>>> from xml.dom import minidom
>>> fsock = open('binary.xml')    
>>> xmldoc = minidom.parse(fsock) 
>>> fsock.close()                 
>>> print xmldoc.toxml()          
<?xml version="1.0" ?>
<grammar>
<ref id="bit">
  <p>0</p>
  <p>1</p>
</ref>
<ref id="byte">
  <p><xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/>\
<xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/></p>
</ref>
</grammar>

	First, you open the file on disk. This gives you a file object.
	You pass the file object to minidom.parse, which calls the read method of fsock and reads the XML document from the file on disk.
	Be sure to call the close method of the file object after you're done with it. minidom.parse will not do this for you.
	Calling the toxml() method on the returned XML document prints out the entire thing.

Well, that all seems like a colossal waste of time. After all, you've already seen that minidom.parse can simply take the filename and do all the opening and closing nonsense automatically. And it's true that if you know you're just going to be parsing a local file, you can pass the filename and minidom.parse is smart enough to Do The Right Thing™. But notice how similar -- and easy -- it is to parse an XML document straight from the Internet.

Example 10.2. Parsing XML from a URL

>>> import urllib
>>> usock = urllib.urlopen('http://slashdot.org/slashdot.rdf') 
>>> xmldoc = minidom.parse(usock)            
>>> usock.close()          
>>> print xmldoc.toxml()   
<?xml version="1.0" ?>
<rdf:RDF xmlns="http://my.netscape.com/rdf/simple/0.9/"
 xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">

<channel>
<title>Slashdot</title>
<link>http://slashdot.org/</link>
<description>News for nerds, stuff that matters</description>
</channel>

<image>
<title>Slashdot</title>
<url>http://images.slashdot.org/topics/topicslashdot.gif</url>
<link>http://slashdot.org/</link>
</image>

<item>
<title>To HDTV or Not to HDTV?</title>
<link>http://slashdot.org/article.pl?sid=01/12/28/0421241</link>
</item>

[...snip...]

	As you saw in a previous chapter, urlopen takes a web page URL and returns a file-like object. Most importantly, this object has a read method which returns the HTML source of the web page.
	Now you pass the file-like object to minidom.parse, which obediently calls the read method of the object and parses the XML data that the read method returns. The fact that this XML data is now coming straight from a web page is completely irrelevant. minidom.parse doesn't know about web pages, and it doesn't care about web pages; it just knows about file-like objects.
	As soon as you're done with it, be sure to close the file-like object that urlopen gives you.
	By the way, this URL is real, and it really is XML. It's an XML representation of the current headlines on Slashdot, a technical news and gossip site.

Example 10.3. Parsing XML from a string (the easy but inflexible way)

>>> contents = "<grammar><ref id='bit'><p>0</p><p>1</p></ref></grammar>"
>>> xmldoc = minidom.parseString(contents) 
>>> print xmldoc.toxml()
<?xml version="1.0" ?>
<grammar><ref id="bit"><p>0</p><p>1</p></ref></grammar>

minidom has a method, parseString, which takes an entire XML document as a string and parses it. You can use this instead of minidom.parse if you know you already have your entire XML document in a string.

OK, so you can use the minidom.parse function for parsing both local files and remote URLs, but for parsing strings, you use... a different function. That means that if you want to be able to take input from a file, a URL, or a string, you'll need special logic to check whether it's a string, and call the parseString function instead. How unsatisfying.

If there were a way to turn a string into a file-like object, then you could simply pass this object to minidom.parse. And in fact, there is a module specifically designed for doing just that: StringIO.

Example 10.4. Introducing StringIO

>>> contents = "<grammar><ref id='bit'><p>0</p><p>1</p></ref></grammar>"
>>> import StringIO
>>> ssock = StringIO.StringIO(contents)   
>>> ssock.read()        
"<grammar><ref id='bit'><p>0</p><p>1</p></ref></grammar>"
>>> ssock.read()        
''
>>> ssock.seek(0)       
>>> ssock.read(15)      
'<grammar><ref i'
>>> ssock.read(15)
"d='bit'><p>0</p"
>>> ssock.read()
'><p>1</p></ref></grammar>'
>>> ssock.close()       

	The StringIO module contains a single class, also called StringIO, which allows you to turn a string into a file-like object. The StringIO class takes the string as a parameter when creating an instance.
	Now you have a file-like object, and you can do all sorts of file-like things with it. Like read, which returns the original string.
	Calling read again returns an empty string. This is how real file objects work too; once you read the entire file, you can't read any more without explicitly seeking to the beginning of the file. The StringIO object works the same way.
	You can explicitly seek to the beginning of the string, just like seeking through a file, by using the seek method of the StringIO object.
	You can also read the string in chunks, by passing a size parameter to the read method.
	At any time, read will return the rest of the string that you haven't read yet. All of this is exactly how file objects work; hence the term file-like object.

Example 10.5. Parsing XML from a string (the file-like object way)

>>> contents = "<grammar><ref id='bit'><p>0</p><p>1</p></ref></grammar>"
>>> ssock = StringIO.StringIO(contents)
>>> xmldoc = minidom.parse(ssock) 
>>> ssock.close()
>>> print xmldoc.toxml()
<?xml version="1.0" ?>
<grammar><ref id="bit"><p>0</p><p>1</p></ref></grammar>

Now you can pass the file-like object (really a StringIO) to minidom.parse, which will call the object's read method and happily parse away, never knowing that its input came from a hard-coded string.

So now you know how to use a single function, minidom.parse, to parse an XML document stored on a web page, in a local file, or in a hard-coded string. For a web page, you use urlopen to get a file-like object; for a local file, you use open; and for a string, you use StringIO. Now let's take it one step further and generalize these differences as well.

Example 10.6. openAnything

def openAnything(source):
    # 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))  

	The openAnything function takes a single parameter, source, and returns a file-like object. source is a string of some sort; it can either be a URL (like 'http://slashdot.org/slashdot.rdf'), a full or partial pathname to a local file (like 'binary.xml'), or a string that contains actual XML data to be parsed.
	First, you see if source is a URL. You do this through brute force: you try to open it as a URL and silently ignore errors caused by trying to open something which is not a URL. This is actually elegant in the sense that, if urllib ever supports new types of URLs in the future, you will also support them without recoding. If urllib is able to open source, then the return kicks you out of the function immediately and the following try statements never execute.
	On the other hand, if urllib yelled at you and told you that source wasn't a valid URL, you assume it's a path to a file on disk and try to open it. Again, you don't do anything fancy to check whether source is a valid filename or not (the rules for valid filenames vary wildly between different platforms anyway, so you'd probably get them wrong anyway). Instead, you just blindly open the file, and silently trap any errors.
	By this point, you need to assume that source is a string that has hard-coded data in it (since nothing else worked), so you use StringIO to create a file-like object out of it and return that. (In fact, since you're using the str function, source doesn't even need to be a string; it could be any object, and you'll use its string representation, as defined by its __str__ special method.)

Now you can use this openAnything function in conjunction with minidom.parse to make a function that takes a source that refers to an XML document somehow (either as a URL, or a local filename, or a hard-coded XML document in a string) and parses it.

Example 10.7. Using openAnything in kgp.py

class KantGenerator:
    def _load(self, source):
        sock = toolbox.openAnything(source)
        xmldoc = minidom.parse(sock).documentElement
        sock.close()
        return xmldoc

10.2. Standard input, output, and error

UNIX users are already familiar with the concept of standard input, standard output, and standard error. This section is for the rest of you.

Standard output and standard error (commonly abbreviated stdout and stderr) are pipes that are built into every UNIX system. When you print something, it goes to the stdout pipe; when your program crashes and prints out debugging information (like a traceback in Python), it goes to the stderr pipe. Both of these pipes are ordinarily just connected to the terminal window where you are working, so when a program prints, you see the output, and when a program crashes, you see the debugging information. (If you're working on a system with a window-based Python IDE, stdout and stderr default to your “Interactive Window”.)

Example 10.8. Introducing stdout and stderr

>>> for i in range(3):
...     print 'Dive in'             
Dive in
Dive in
Dive in
>>> import sys
>>> for i in range(3):
...     sys.stdout.write('Dive in') 
Dive inDive inDive in
>>> for i in range(3):
...     sys.stderr.write('Dive in') 
Dive inDive inDive in

	As you saw in Example 6.9, “Simple Counters”, you can use Python's built-in range function to build simple counter loops that repeat something a set number of times.
	stdout is a file-like object; calling its write function will print out whatever string you give it. In fact, this is what the print function really does; it adds a carriage return to the end of the string you're printing, and calls sys.stdout.write.
	In the simplest case, stdout and stderr send their output to the same place: the Python IDE (if you're in one), or the terminal (if you're running Python from the command line). Like stdout, stderr does not add carriage returns for you; if you want them, add them yourself.

stdout and stderr are both file-like objects, like the ones you discussed in Section 10.1, “Abstracting input sources”, but they are both write-only. They have no read method, only write. Still, they are file-like objects, and you can assign any other file- or file-like object to them to redirect their output.

Example 10.9. Redirecting output

[you@localhost kgp]$ python stdout.py
Dive in
[you@localhost kgp]$ cat out.log
This message will be logged instead of displayed

(On Windows, you can use type instead of cat to display the contents of a file.)

If you have not already done so, you can download this and other examples used in this book.

#stdout.py
import sys

print 'Dive in'      
saveout = sys.stdout 
fsock = open('out.log', 'w')           
sys.stdout = fsock   
print 'This message will be logged instead of displayed' 
sys.stdout = saveout 
fsock.close()        

	This will print to the IDE “Interactive Window” (or the terminal, if running the script from the command line).
	Always save stdout before redirecting it, so you can set it back to normal later.
	Open a file for writing. If the file doesn't exist, it will be created. If the file does exist, it will be overwritten.
	Redirect all further output to the new file you just opened.
	This will be “printed” to the log file only; it will not be visible in the IDE window or on the screen.
	Set stdout back to the way it was before you mucked with it.
	Close the log file.

Redirecting stderr works exactly the same way, using sys.stderr instead of sys.stdout.

Example 10.10. Redirecting error information

[you@localhost kgp]$ python stderr.py
[you@localhost kgp]$ cat error.log
Traceback (most recent line last):
  File "stderr.py", line 5, in ?
    raise Exception, 'this error will be logged'
Exception: this error will be logged

If you have not already done so, you can download this and other examples used in this book.

#stderr.py
import sys

fsock = open('error.log', 'w')               
sys.stderr = fsock         
raise Exception, 'this error will be logged'  

	Open the log file where you want to store debugging information.
	Redirect standard error by assigning the file object of the newly-opened log file to stderr.
	Raise an exception. Note from the screen output that this does not print anything on screen. All the normal traceback information has been written to error.log.
	Also note that you're not explicitly closing your log file, nor are you setting stderr back to its original value. This is fine, since once the program crashes (because of the exception), Python will clean up and close the file for us, and it doesn't make any difference that stderr is never restored, since, as I mentioned, the program crashes and Python ends. Restoring the original is more important for stdout, if you expect to go do other stuff within the same script afterwards.

Since it is so common to write error messages to standard error, there is a shorthand syntax that can be used instead of going through the hassle of redirecting it outright.

Example 10.11. Printing to stderr

>>> print 'entering function'
entering function
>>> import sys
>>> print >> sys.stderr, 'entering function' 
entering function

This shorthand syntax of the print statement can be used to write to any open file, or file-like object. In this case, you can redirect a single print statement to stderr without affecting subsequent print statements.

Standard input, on the other hand, is a read-only file object, and it represents the data flowing into the program from some previous program. This will likely not make much sense to classic Mac OS users, or even Windows users unless you were ever fluent on the MS-DOS command line. The way it works is that you can construct a chain of commands in a single line, so that one program's output becomes the input for the next program in the chain. The first program simply outputs to standard output (without doing any special redirecting itself, just doing normal print statements or whatever), and the next program reads from standard input, and the operating system takes care of connecting one program's output to the next program's input.

Example 10.12. Chaining commands

[you@localhost kgp]$ python kgp.py -g binary.xml         
01100111
[you@localhost kgp]$ cat binary.xml    
<?xml version="1.0"?>
<!DOCTYPE grammar PUBLIC "-//diveintopython3.org//DTD Kant Generator Pro v1.0//EN" "kgp.dtd">
<grammar>
<ref id="bit">
  <p>0</p>
  <p>1</p>
</ref>
<ref id="byte">
  <p><xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/>\
<xref id="bit"/><xref id="bit"/><xref id="bit"/><xref id="bit"/></p>
</ref>
</grammar>
[you@localhost kgp]$ cat binary.xml | python kgp.py -g -  
10110001

	As you saw in Section 9.1, “Diving in”, this will print a string of eight random bits, 0 or 1.
	This simply prints out the entire contents of binary.xml. (Windows users should use type instead of cat.)
	This prints the contents of binary.xml, but the “|” character, called the “pipe” character, means that the contents will not be printed to the screen. Instead, they will become the standard input of the next command, which in this case calls your Python script.
	Instead of specifying a module (like binary.xml), you specify “-”, which causes your script to load the grammar from standard input instead of from a specific file on disk. (More on how this happens in the next example.) So the effect is the same as the first syntax, where you specified the grammar filename directly, but think of the expansion possibilities here. Instead of simply doing cat binary.xml, you could run a script that dynamically generates the grammar, then you can pipe it into your script. It could come from anywhere: a database, or some grammar-generating meta-script, or whatever. The point is that you don't need to change your kgp.py script at all to incorporate any of this functionality. All you need to do is be able to take grammar files from standard input, and you can separate all the other logic into another program.

So how does the script “know” to read from standard input when the grammar file is “-”? It's not magic; it's just code.

Example 10.13. Reading from standard input in kgp.py

def openAnything(source):
    if source == "-":    
        import sys
        return sys.stdin

    # try to open with urllib (if source is http, ftp, or file URL)
    import urllib
    try:

[... snip ...]

This is the openAnything function from toolbox.py, which you previously examined in Section 10.1, “Abstracting input sources”. All you've done is add three lines of code at the beginning of the function to check if the source is “-”; if so, you return sys.stdin. Really, that's it! Remember, stdin is a file-like object with a read method, so the rest of the code (in kgp.py, where you call openAnything) doesn't change a bit.

10.3. Caching node lookups

kgp.py employs several tricks which may or may not be useful to you in your XML processing. The first one takes advantage of the consistent structure of the input documents to build a cache of nodes.

A grammar file defines a series of ref elements. Each ref contains one or more p elements, which can contain a lot of different things, including xrefs. Whenever you encounter an xref, you look for a corresponding ref element with the same id attribute, and choose one of the ref element's children and parse it. (You'll see how this random choice is made in the next section.)

This is how you build up the grammar: define ref elements for the smallest pieces, then define ref elements which "include" the first ref elements by using xref, and so forth. Then you parse the "largest" reference and follow each xref, and eventually output real text. The text you output depends on the (random) decisions you make each time you fill in an xref, so the output is different each time.

This is all very flexible, but there is one downside: performance. When you find an xref and need to find the corresponding ref element, you have a problem. The xref has an id attribute, and you want to find the ref element that has that same id attribute, but there is no easy way to do that. The slow way to do it would be to get the entire list of ref elements each time, then manually loop through and look at each id attribute. The fast way is to do that once and build a cache, in the form of a dictionary.

Example 10.14. loadGrammar

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

	Start by creating an empty dictionary, self.refs.
	As you saw in Section 9.5, “Searching for elements”, getElementsByTagName returns a list of all the elements of a particular name. You easily can get a list of all the ref elements, then simply loop through that list.
	As you saw in Section 9.6, “Accessing element attributes”, you can access individual attributes of an element by name, using standard dictionary syntax. So the keys of the self.refs dictionary will be the values of the id attribute of each ref element.
	The values of the self.refs dictionary will be the ref elements themselves. As you saw in Section 9.3, “Parsing XML”, each element, each node, each comment, each piece of text in a parsed XML document is an object.

Once you build this cache, whenever you come across an xref and need to find the ref element with the same id attribute, you can simply look it up in self.refs.

Example 10.15. Using the ref element cache

    def do_xref(self, node):
        id = node.attributes["id"].value
        self.parse(self.randomChildElement(self.refs[id]))

You'll explore the randomChildElement function in the next section.

10.4. Finding direct children of a node

Another useful techique when parsing XML documents is finding all the direct child elements of a particular element. For instance, in the grammar files, a ref element can have several p elements, each of which can contain many things, including other p elements. You want to find just the p elements that are children of the ref, not p elements that are children of other p elements.

You might think you could simply use getElementsByTagName for this, but you can't. getElementsByTagName searches recursively and returns a single list for all the elements it finds. Since p elements can contain other p elements, you can't use getElementsByTagName, because it would return nested p elements that you don't want. To find only direct child elements, you'll need to do it yourself.

Example 10.16. Finding direct child elements

    def randomChildElement(self, node):
        choices = [e for e in node.childNodes
 if e.nodeType == e.ELEMENT_NODE]   
        chosen = random.choice(choices)             
        return chosen            

	As you saw in Example 9.9, “Getting child nodes”, the childNodes attribute returns a list of all the child nodes of an element.
	However, as you saw in Example 9.11, “Child nodes can be text”, the list returned by childNodes contains all different types of nodes, including text nodes. That's not what you're looking for here. You only want the children that are elements.
	Each node has a nodeType attribute, which can be ELEMENT_NODE, TEXT_NODE, COMMENT_NODE, or any number of other values. The complete list of possible values is in the __init__.py file in the xml.dom package. (See Section 9.2, “Packages” for more on packages.) But you're just interested in nodes that are elements, so you can filter the list to only include those nodes whose nodeType is ELEMENT_NODE.
	Once you have a list of actual elements, choosing a random one is easy. Python comes with a module called random which includes several useful functions. The random.choice function takes a list of any number of items and returns a random item. For example, if the ref elements contains several p elements, then choices would be a list of p elements, and chosen would end up being assigned exactly one of them, selected at random.

10.5. Creating separate handlers by node type

The third useful XML processing tip involves separating your code into logical functions, based on node types and element names. Parsed XML documents are made up of various types of nodes, each represented by a Python object. The root level of the document itself is represented by a Document object. The Document then contains one or more Element objects (for actual XML tags), each of which may contain other Element objects, Text objects (for bits of text), or Comment objects (for embedded comments). Python makes it easy to write a dispatcher to separate the logic for each node type.

Example 10.17. Class names of parsed XML objects

>>> from xml.dom import minidom
>>> xmldoc = minidom.parse('kant.xml') 
>>> xmldoc
<xml.dom.minidom.Document instance at 0x01359DE8>
>>> xmldoc.__class__ 
<class xml.dom.minidom.Document at 0x01105D40>
>>> xmldoc.__class__.__name__          
'Document'

	Assume for a moment that kant.xml is in the current directory.
	As you saw in Section 9.2, “Packages”, the object returned by parsing an XML document is a Document object, as defined in the minidom.py in the xml.dom package. As you saw in Section 5.4, “Instantiating Classes”, __class__ is built-in attribute of every Python object.
	Furthermore, __name__ is a built-in attribute of every Python class, and it is a string. This string is not mysterious; it's the same as the class name you type when you define a class yourself. (See Section 5.3, “Defining Classes”.)

Fine, so now you can get the class name of any particular XML node (since each XML node is represented as a Python object). How can you use this to your advantage to separate the logic of parsing each node type? The answer is getattr, which you first saw in Section 4.4, “Getting Object References With getattr”.

Example 10.18. parse, a generic XML node dispatcher

    def parse(self, node):          
        parseMethod = getattr(self, "parse_%s" % node.__class__.__name__)  
        parseMethod(node) 

	First off, notice that you're constructing a larger string based on the class name of the node you were passed (in the node argument). So if you're passed a Document node, you're constructing the string 'parse_Document', and so forth.
	Now you can treat that string as a function name, and get a reference to the function itself using getattr
	Finally, you can call that function and pass the node itself as an argument. The next example shows the definitions of each of these functions.

Example 10.19. Functions called by the parse dispatcher

    def parse_Document(self, node): 
        self.parse(node.documentElement)

    def parse_Text(self, node):    
        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_Comment(self, node): 
        pass

    def parse_Element(self, node): 
        handlerMethod = getattr(self, "do_%s" % node.tagName)
        handlerMethod(node)

	parse_Document is only ever called once, since there is only one Document node in an XML document, and only one Document object in the parsed XML representation. It simply turns around and parses the root element of the grammar file.
	parse_Text is called on nodes that represent bits of text. The function itself does some special processing to handle automatic capitalization of the first word of a sentence, but otherwise simply appends the represented text to a list.
	parse_Comment is just a pass, since you don't care about embedded comments in the grammar files. Note, however, that you still need to define the function and explicitly make it do nothing. If the function did not exist, the generic parse function would fail as soon as it stumbled on a comment, because it would try to find the non-existent parse_Comment function. Defining a separate function for every node type, even ones you don't use, allows the generic parse function to stay simple and dumb.
	The parse_Element method is actually itself a dispatcher, based on the name of the element's tag. The basic idea is the same: take what distinguishes elements from each other (their tag names) and dispatch to a separate function for each of them. You construct a string like 'do_xref' (for an <xref> tag), find a function of that name, and call it. And so forth for each of the other tag names that might be found in the course of parsing a grammar file (<p> tags, <choice> tags).

In this example, the dispatch functions parse and parse_Element simply find other methods in the same class. If your processing is very complex (or you have many different tag names), you could break up your code into separate modules, and use dynamic importing to import each module and call whatever functions you needed. Dynamic importing will be discussed in Chapter 16, Functional Programming.

10.6. Handling command-line arguments

Python fully supports creating programs that can be run on the command line, complete with command-line arguments and either short- or long-style flags to specify various options. None of this is XML-specific, but this script makes good use of command-line processing, so it seemed like a good time to mention it.

It's difficult to talk about command-line processing without understanding how command-line arguments are exposed to your Python program, so let's write a simple program to see them.

Example 10.20. Introducing sys.argv

If you have not already done so, you can download this and other examples used in this book.

#argecho.py
import sys

for arg in sys.argv: 
    print arg

Each command-line argument passed to the program will be in sys.argv, which is just a list. Here you are printing each argument on a separate line.

Example 10.21. The contents of sys.argv

[you@localhost py]$ python argecho.py             
argecho.py
[you@localhost py]$ python argecho.py abc def     
argecho.py
abc
def
[you@localhost py]$ python argecho.py --help      
argecho.py
--help
[you@localhost py]$ python argecho.py -m kant.xml 
argecho.py
-m
kant.xml

	The first thing to know about sys.argv is that it contains the name of the script you're calling. You will actually use this knowledge to your advantage later, in Chapter 16, Functional Programming. Don't worry about it for now.
	Command-line arguments are separated by spaces, and each shows up as a separate element in the sys.argv list.
	Command-line flags, like --help, also show up as their own element in the sys.argv list.
	To make things even more interesting, some command-line flags themselves take arguments. For instance, here you have a flag (-m) which takes an argument (kant.xml). Both the flag itself and the flag's argument are simply sequential elements in the sys.argv list. No attempt is made to associate one with the other; all you get is a list.

So as you can see, you certainly have all the information passed on the command line, but then again, it doesn't look like it's going to be all that easy to actually use it. For simple programs that only take a single argument and have no flags, you can simply use sys.argv[1] to access the argument. There's no shame in this; I do it all the time. For more complex programs, you need the getopt module.

Example 10.22. Introducing getopt

def main(argv):       
    grammar = "kant.xml"                 
    try:              
        opts, args = getopt.getopt(argv, "hg:d", ["help", "grammar="]) 
    except getopt.GetoptError:           
        usage()        
        sys.exit(2)   

...

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

	First off, look at the bottom of the example and notice that you're calling the main function with sys.argv[1:]. Remember, sys.argv[0] is the name of the script that you're running; you don't care about that for command-line processing, so you chop it off and pass the rest of the list.
	This is where all the interesting processing happens. The getopt function of the getopt module takes three parameters: the argument list (which you got from sys.argv[1:]), a string containing all the possible single-character command-line flags that this program accepts, and a list of longer command-line flags that are equivalent to the single-character versions. This is quite confusing at first glance, and is explained in more detail below.
	If anything goes wrong trying to parse these command-line flags, getopt will raise an exception, which you catch. You told getopt all the flags you understand, so this probably means that the end user passed some command-line flag that you don't understand.
	As is standard practice in the UNIX world, when the script is passed flags it doesn't understand, you print out a summary of proper usage and exit gracefully. Note that I haven't shown the usage function here. You would still need to code that somewhere and have it print out the appropriate summary; it's not automatic.

So what are all those parameters you pass to the getopt function? Well, the first one is simply the raw list of command-line flags and arguments (not including the first element, the script name, which you already chopped off before calling the main function). The second is the list of short command-line flags that the script accepts.

"hg:d"

-h

print usage summary

-g ...

use specified grammar file or URL

-d

show debugging information while parsing

The first and third flags are simply standalone flags; you specify them or you don't, and they do things (print help) or change state (turn on debugging). However, the second flag (-g) must be followed by an argument, which is the name of the grammar file to read from. In fact it can be a filename or a web address, and you don't know which yet (you'll figure it out later), but you know it has to be something. So you tell getopt this by putting a colon after the g in that second parameter to the getopt function.

To further complicate things, the script accepts either short flags (like -h) or long flags (like --help), and you want them to do the same thing. This is what the third parameter to getopt is for, to specify a list of the long flags that correspond to the short flags you specified in the second parameter.

["help", "grammar="]

--help

print usage summary

--grammar ...

use specified grammar file or URL

Three things of note here:

1. All long flags are preceded by two dashes on the command line, but you don't include those dashes when calling getopt. They are understood.
2. The --grammar flag must always be followed by an additional argument, just like the -g flag. This is notated by an equals sign, "grammar=".
3. The list of long flags is shorter than the list of short flags, because the -d flag does not have a corresponding long version. This is fine; only -d will turn on debugging. But the order of short and long flags needs to be the same, so you'll need to specify all the short flags that do have corresponding long flags first, then all the rest of the short flags.

Confused yet? Let's look at the actual code and see if it makes sense in context.

Example 10.23. Handling command-line arguments in kgp.py

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()

	The grammar variable will keep track of the grammar file you're using. You initialize it here in case it's not specified on the command line (using either the -g or the --grammar flag).
	The opts variable that you get back from getopt contains a list of tuples: flag and argument. If the flag doesn't take an argument, then arg will simply be None. This makes it easier to loop through the flags.
	getopt validates that the command-line flags are acceptable, but it doesn't do any sort of conversion between short and long flags. If you specify the -h flag, opt will contain "-h"; if you specify the --help flag, opt will contain "--help". So you need to check for both.
	Remember, the -d flag didn't have a corresponding long flag, so you only need to check for the short form. If you find it, you set a global variable that you'll refer to later to print out debugging information. (I used this during the development of the script. What, you thought all these examples worked on the first try?)
	If you find a grammar file, either with a -g flag or a --grammar flag, you save the argument that followed it (stored in arg) into the grammar variable, overwriting the default that you initialized at the top of the main function.
	That's it. You've looped through and dealt with all the command-line flags. That means that anything left must be command-line arguments. These come back from the getopt function in the args variable. In this case, you're treating them as source material for the parser. If there are no command-line arguments specified, args will be an empty list, and source will end up as the empty string.

10.7. Putting it all together

You've covered a lot of ground. Let's step back and see how all the pieces fit together.

To start with, this is a script that takes its arguments on the command line, using the getopt module.

def main(argv):       
...
    try:              
        opts, args = getopt.getopt(argv, "hg:d", ["help", "grammar="])
    except getopt.GetoptError:          
...
    for opt, arg in opts:               
...

You create a new instance of the KantGenerator class, and pass it the grammar file and source that may or may not have been specified on the command line.

    k = KantGenerator(grammar, source)

The KantGenerator instance automatically loads the grammar, which is an XML file. You use your custom openAnything function to open the file (which could be stored in a local file or a remote web server), then use the built-in minidom parsing functions to parse the XML into a tree of Python objects.

    def _load(self, source):
        sock = toolbox.openAnything(source)
        xmldoc = minidom.parse(sock).documentElement
        sock.close()

Oh, and along the way, you take advantage of your knowledge of the structure of the XML document to set up a little cache of references, which are just elements in the XML document.

    def loadGrammar(self, grammar):       
        for ref in self.grammar.getElementsByTagName("ref"):
            self.refs[ref.attributes["id"].value] = ref     

If you specified some source material on the command line, you use that; otherwise you rip through the grammar looking for the "top-level" reference (that isn't referenced by anything else) and use that as a starting point.

    def getDefaultSource(self):
        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]
        return '<xref id="%s"/>' % random.choice(standaloneXrefs)

Now you rip through the source material. The source material is also XML, and you parse it one node at a time. To keep the code separated and more maintainable, you use separate handlers for each node type.

    def parse_Element(self, node): 
        handlerMethod = getattr(self, "do_%s" % node.tagName)
        handlerMethod(node)

You bounce through the grammar, parsing all the children of each p element,

    def do_p(self, node):
...
        if doit:
            for child in node.childNodes: self.parse(child)

replacing choice elements with a random child,

    def do_choice(self, node):
        self.parse(self.randomChildElement(node))

and replacing xref elements with a random child of the corresponding ref element, which you previously cached.

    def do_xref(self, node):
        id = node.attributes["id"].value
        self.parse(self.randomChildElement(self.refs[id]))

Eventually, you parse your way down to plain text,

    def parse_Text(self, node):    
        text = node.data
...
            self.pieces.append(text)

which you print out.

def main(argv):       
...
    k = KantGenerator(grammar, source)
    print k.output()

10.8. Summary

Python comes with powerful libraries for parsing and manipulating XML documents. The minidom takes an XML file and parses it into Python objects, providing for random access to arbitrary elements. Furthermore, this chapter shows how Python can be used to create a "real" standalone command-line script, complete with command-line flags, command-line arguments, error handling, even the ability to take input from the piped result of a previous program.

Before moving on to the next chapter, you should be comfortable doing all of these things:

• Chaining programs with standard input and output
• Defining dynamic dispatchers with getattr.
• Using command-line flags and validating them with getopt

Chapter 11. HTTP Web Services

11.1. Diving in

You've learned about HTML processing and XML processing, and along the way you saw how to download a web page and how to parse XML from a URL, but let's dive into the more general topic of HTTP web services.

Simply stated, HTTP web services are programmatic ways of sending and receiving data from remote servers using the operations of HTTP directly. If you want to get data from the server, use a straight HTTP GET; if you want to send new data to the server, use HTTP POST. (Some more advanced HTTP web service APIs also define ways of modifying existing data and deleting data, using HTTP PUT and HTTP DELETE.) In other words, the “verbs” built into the HTTP protocol (GET, POST, PUT, and DELETE) map directly to application-level operations for receiving, sending, modifying, and deleting data.

The main advantage of this approach is simplicity, and its simplicity has proven popular with a lot of different sites. Data -- usually XML data -- can be built and stored statically, or generated dynamically by a server-side script, and all major languages include an HTTP library for downloading it. Debugging is also easier, because you can load up the web service in any web browser and see the raw data. Modern browsers will even nicely format and pretty-print XML data for you, to allow you to quickly navigate through it.

Examples of pure XML-over-HTTP web services:

• Amazon API allows you to retrieve product information from the Amazon.com online store.
• National Weather Service (United States) and Hong Kong Observatory (Hong Kong) offer weather alerts as a web service.
• Atom API for managing web-based content.
• Syndicated feeds from weblogs and news sites bring you up-to-the-minute news from a variety of sites.

In later chapters, you'll explore APIs which use HTTP as a transport for sending and receiving data, but don't map application semantics to the underlying HTTP semantics. (They tunnel everything over HTTP POST.) But this chapter will concentrate on using HTTP GET to get data from a remote server, and you'll explore several HTTP features you can use to get the maximum benefit out of pure HTTP web services.

Here is a more advanced version of the openanything module that you saw in the previous chapter:

Example 11.1. openanything.py

If you have not already done so, you can download this and other examples used in this book.

import urllib2, urlparse, gzip
from StringIO import StringIO

USER_AGENT = 'OpenAnything/1.0 +http://diveintopython3.org/http_web_services/'

class SmartRedirectHandler(urllib2.HTTPRedirectHandler):    
    def http_error_301(self, req, fp, code, msg, headers):  
        result = urllib2.HTTPRedirectHandler.http_error_301(
            self, req, fp, code, msg, headers)              
        result.status = code              
        return result   

    def http_error_302(self, req, fp, code, msg, headers):  
        result = urllib2.HTTPRedirectHandler.http_error_302(
            self, req, fp, code, msg, headers)              
        result.status = code              
        return result   

class DefaultErrorHandler(urllib2.HTTPDefaultErrorHandler):   
    def http_error_default(self, req, fp, code, msg, headers):
        result = urllib2.HTTPError(         
            req.get_full_url(), code, msg, headers, fp)       
        result.status = code                
        return result     

def openAnything(source, etag=None, lastmodified=None, agent=USER_AGENT):
    '''URL, 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.

    If the etag argument is supplied, it will be used as the value of an
    If-None-Match request header.

    If the lastmodified argument is supplied, it must be a formatted
    date/time string in GMT (as returned in the Last-Modified header of
    a previous request).  The formatted date/time will be used
    as the value of an If-Modified-Since request header.

    If the agent argument is supplied, it will be used as the value of a
    User-Agent request header.
    '''

    if hasattr(source, 'read'):
        return source

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

    if urlparse.urlparse(source)[0] == 'http':  
        # open URL with urllib2                 
        request = urllib2.Request(source)       
        request.add_header('User-Agent', agent) 
        if etag:              
            request.add_header('If-None-Match', etag)             
        if lastmodified:      
            request.add_header('If-Modified-Since', lastmodified) 
        request.add_header('Accept-encoding', 'gzip')             
        opener = urllib2.build_opener(SmartRedirectHandler(), DefaultErrorHandler())
        return opener.open(request)             
    
    # try to open with native open function (if source is a filename)
    try:
        return open(source)
    except (IOError, OSError):
        pass

    # treat source as string
    return StringIO(str(source))

def fetch(source, etag=None, last_modified=None, agent=USER_AGENT):  
    '''Fetch data and metadata from a URL, file, stream, or string'''
    result = {}
    f = openAnything(source, etag, last_modified, agent)             
    result['data'] = f.read()    
    if hasattr(f, 'headers'):    
        # save ETag, if the server sent one        
        result['etag'] = f.headers.get('ETag')     
        # save Last-Modified header, if the server sent one          
        result['lastmodified'] = f.headers.get('Last-Modified')      
        if f.headers.get('content-encoding', '') == 'gzip':          
            # data came back gzip-compressed, decompress it          
            result['data'] = gzip.GzipFile(fileobj=StringIO(result['data']])).read()
    if hasattr(f, 'url'):        
        result['url'] = f.url    
        result['status'] = 200   
    if hasattr(f, 'status'):     
        result['status'] = f.status                
    f.close()  
    return result                

Further reading

• Paul Prescod believes that pure HTTP web services are the future of the Internet.

11.2. How not to fetch data over HTTP

Let's say you want to download a resource over HTTP, such as a syndicated Atom feed. But you don't just want to download it once; you want to download it over and over again, every hour, to get the latest news from the site that's offering the news feed. Let's do it the quick-and-dirty way first, and then see how you can do better.

Example 11.2. Downloading a feed the quick-and-dirty way

>>> import urllib
>>> data = urllib.urlopen('http://diveintomark.org/xml/atom.xml').read()    
>>> print data
<?xml version="1.0" encoding="iso-8859-1"?>
<feed version="0.3"
  xmlns="http://purl.org/atom/ns#"
  xmlns:dc="http://purl.org/dc/elements/1.1/"
  xml:lang="en">
  <title mode="escaped">dive into mark</title>
  <link rel="alternate" type="text/html" href="http://diveintomark.org/"/>
  <-- rest of feed omitted for brevity -->

Downloading anything over HTTP is incredibly easy in Python; in fact, it's a one-liner. The urllib module has a handy urlopen function that takes the address of the page you want, and returns a file-like object that you can just read() from to get the full contents of the page. It just can't get much easier.

So what's wrong with this? Well, for a quick one-off during testing or development, there's nothing wrong with it. I do it all the time. I wanted the contents of the feed, and I got the contents of the feed. The same technique works for any web page. But once you start thinking in terms of a web service that you want to access on a regular basis -- and remember, you said you were planning on retrieving this syndicated feed once an hour -- then you're being inefficient, and you're being rude.

Let's talk about some of the basic features of HTTP.

11.3. Features of HTTP

There are five important features of HTTP which you should support.

11.3.1. User-Agent

The User-Agent is simply a way for a client to tell a server who it is when it requests a web page, a syndicated feed, or any sort of web service over HTTP. When the client requests a resource, it should always announce who it is, as specifically as possible. This allows the server-side administrator to get in touch with the client-side developer if anything is going fantastically wrong.

By default, Python sends a generic User-Agent: Python-urllib/1.15. In the next section, you'll see how to change this to something more specific.

11.3.2. Redirects

Sometimes resources move around. Web sites get reorganized, pages move to new addresses. Even web services can reorganize. A syndicated feed at http://example.com/index.xml might be moved to http://example.com/xml/atom.xml. Or an entire domain might move, as an organization expands and reorganizes; for instance, http://www.example.com/index.xml might be redirected to http://server-farm-1.example.com/index.xml.

Every time you request any kind of resource from an HTTP server, the server includes a status code in its response. Status code 200 means “everything's normal, here's the page you asked for”. Status code 404 means “page not found”. (You've probably seen 404 errors while browsing the web.)

HTTP has two different ways of signifying that a resource has moved. Status code 302 is a temporary redirect; it means “oops, that got moved over here temporarily” (and then gives the temporary address in a Location: header). Status code 301 is a permanent redirect; it means “oops, that got moved permanently” (and then gives the new address in a Location: header). If you get a 302 status code and a new address, the HTTP specification says you should use the new address to get what you asked for, but the next time you want to access the same resource, you should retry the old address. But if you get a 301 status code and a new address, you're supposed to use the new address from then on.

urllib.urlopen will automatically “follow” redirects when it receives the appropriate status code from the HTTP server, but unfortunately, it doesn't tell you when it does so. You'll end up getting data you asked for, but you'll never know that the underlying library “helpfully” followed a redirect for you. So you'll continue pounding away at the old address, and each time you'll get redirected to the new address. That's two round trips instead of one: not very efficient! Later in this chapter, you'll see how to work around this so you can deal with permanent redirects properly and efficiently.

11.3.3. Last-Modified/If-Modified-Since

Some data changes all the time. The home page of CNN.com is constantly updating every few minutes. On the other hand, the home page of Google.com only changes once every few weeks (when they put up a special holiday logo, or advertise a new service). Web services are no different; usually the server knows when the data you requested last changed, and HTTP provides a way for the server to include this last-modified date along with the data you requested.

If you ask for the same data a second time (or third, or fourth), you can tell the server the last-modified date that you got last time: you send an If-Modified-Since header with your request, with the date you got back from the server last time. If the data hasn't changed since then, the server sends back a special HTTP status code 304, which means “this data hasn't changed since the last time you asked for it”. Why is this an improvement? Because when the server sends a 304, it doesn't re-send the data. All you get is the status code. So you don't need to download the same data over and over again if it hasn't changed; the server assumes you have the data cached locally.

All modern web browsers support last-modified date checking. If you've ever visited a page, re-visited the same page a day later and found that it hadn't changed, and wondered why it loaded so quickly the second time -- this could be why. Your web browser cached the contents of the page locally the first time, and when you visited the second time, your browser automatically sent the last-modified date it got from the server the first time. The server simply says 304: Not Modified, so your browser knows to load the page from its cache. Web services can be this smart too.

Python's URL library has no built-in support for last-modified date checking, but since you can add arbitrary headers to each request and read arbitrary headers in each response, you can add support for it yourself.

11.3.4. ETag/If-None-Match

ETags are an alternate way to accomplish the same thing as the last-modified date checking: don't re-download data that hasn't changed. The way it works is, the server sends some sort of hash of the data (in an ETag header) along with the data you requested. Exactly how this hash is determined is entirely up to the server. The second time you request the same data, you include the ETag hash in an If-None-Match: header, and if the data hasn't changed, the server will send you back a 304 status code. As with the last-modified date checking, the server just sends the 304; it doesn't send you the same data a second time. By including the ETag hash in your second request, you're telling the server that there's no need to re-send the same data if it still matches this hash, since you still have the data from the last time.

Python's URL library has no built-in support for ETags, but you'll see how to add it later in this chapter.

11.3.5. Compression

The last important HTTP feature is gzip compression. When you talk about HTTP web services, you're almost always talking about moving XML back and forth over the wire. XML is text, and quite verbose text at that, and text generally compresses well. When you request a resource over HTTP, you can ask the server that, if it has any new data to send you, to please send it in compressed format. You include the Accept-encoding: gzip header in your request, and if the server supports compression, it will send you back gzip-compressed data and mark it with a Content-encoding: gzip header.

Python's URL library has no built-in support for gzip compression per se, but you can add arbitrary headers to the request. And Python comes with a separate gzip module, which has functions you can use to decompress the data yourself.

Note that our little one-line script to download a syndicated feed did not support any of these HTTP features. Let's see how you can improve it.

11.4. Debugging HTTP web services

First, let's turn on the debugging features of Python's HTTP library and see what's being sent over the wire. This will be useful throughout the chapter, as you add more and more features.

Example 11.3. Debugging HTTP

>>> import httplib
>>> httplib.HTTPConnection.debuglevel = 1             
>>> import urllib
>>> feeddata = urllib.urlopen('http://diveintomark.org/xml/atom.xml').read()
connect: (diveintomark.org, 80)     
send: '
GET /xml/atom.xml HTTP/1.0          
Host: diveintomark.org              
User-agent: Python-urllib/1.15      
'
reply: 'HTTP/1.1 200 OK\r\n'        
header: Date: Wed, 14 Apr 2004 22:27:30 GMT
header: Server: Apache/2.0.49 (Debian GNU/Linux)
header: Content-Type: application/atom+xml
header: Last-Modified: Wed, 14 Apr 2004 22:14:38 GMT  
header: ETag: "e8284-68e0-4de30f80" 
header: Accept-Ranges: bytes
header: Content-Length: 26848
header: Connection: close

	urllib relies on another standard Python library, httplib. Normally you don't need to import httplib directly (urllib does that automatically), but you will here so you can set the debugging flag on the HTTPConnection class that urllib uses internally to connect to the HTTP server. This is an incredibly useful technique. Some other Python libraries have similar debug flags, but there's no particular standard for naming them or turning them on; you need to read the documentation of each library to see if such a feature is available.
	Now that the debugging flag is set, information on the the HTTP request and response is printed out in real time. The first thing it tells you is that you're connecting to the server diveintomark.org on port 80, which is the standard port for HTTP.
	When you request the Atom feed, urllib sends three lines to the server. The first line specifies the HTTP verb you're using, and the path of the resource (minus the domain name). All the requests in this chapter will use GET, but in the next chapter on SOAP, you'll see that it uses POST for everything. The basic syntax is the same, regardless of the verb.
	The second line is the Host header, which specifies the domain name of the service you're accessing. This is important, because a single HTTP server can host multiple separate domains. My server currently hosts 12 domains; other servers can host hundreds or even thousands.
	The third line is the User-Agent header. What you see here is the generic User-Agent that the urllib library adds by default. In the next section, you'll see how to customize this to be more specific.
	The server replies with a status code and a bunch of headers (and possibly some data, which got stored in the feeddata variable). The status code here is 200, meaning “everything's normal, here's the data you requested”. The server also tells you the date it responded to your request, some information about the server itself, and the content type of the data it's giving you. Depending on your application, this might be useful, or not. It's certainly reassuring that you thought you were asking for an Atom feed, and lo and behold, you're getting an Atom feed (application/atom+xml, which is the registered content type for Atom feeds).
	The server tells you when this Atom feed was last modified (in this case, about 13 minutes ago). You can send this date back to the server the next time you request the same feed, and the server can do last-modified checking.
	The server also tells you that this Atom feed has an ETag hash of "e8284-68e0-4de30f80". The hash doesn't mean anything by itself; there's nothing you can do with it, except send it back to the server the next time you request this same feed. Then the server can use it to tell you if the data has changed or not.

11.5. Setting the User-Agent

The first step to improving your HTTP web services client is to identify yourself properly with a User-Agent. To do that, you need to move beyond the basic urllib and dive into urllib2.

Example 11.4. Introducing urllib2

>>> import httplib
>>> httplib.HTTPConnection.debuglevel = 1           
>>> import urllib2
>>> request = urllib2.Request('http://diveintomark.org/xml/atom.xml') 
>>> opener = urllib2.build_opener()                 
>>> feeddata = opener.open(request).read()          
connect: (diveintomark.org, 80)
send: '
GET /xml/atom.xml HTTP/1.0
Host: diveintomark.org
User-agent: Python-urllib/2.1
'
reply: 'HTTP/1.1 200 OK\r\n'
header: Date: Wed, 14 Apr 2004 23:23:12 GMT
header: Server: Apache/2.0.49 (Debian GNU/Linux)
header: Content-Type: application/atom+xml
header: Last-Modified: Wed, 14 Apr 2004 22:14:38 GMT
header: ETag: "e8284-68e0-4de30f80"
header: Accept-Ranges: bytes
header: Content-Length: 26848
header: Connection: close

	If you still have your Python IDE open from the previous section's example, you can skip this, but this turns on HTTP debugging so you can see what you're actually sending over the wire, and what gets sent back.
	Fetching an HTTP resource with urllib2 is a three-step process, for good reasons that will become clear shortly. The first step is to create a Request object, which takes the URL of the resource you'll eventually get around to retrieving. Note that this step doesn't actually retrieve anything yet.
	The second step is to build a URL opener. This can take any number of handlers, which control how responses are handled. But you can also build an opener without any custom handlers, which is what you're doing here. You'll see how to define and use custom handlers later in this chapter when you explore redirects.
	The final step is to tell the opener to open the URL, using the Request object you created. As you can see from all the debugging information that gets printed, this step actually retrieves the resource and stores the returned data in feeddata.

Example 11.5. Adding headers with the Request

>>> request            
<urllib2.Request instance at 0x00250AA8>
>>> request.get_full_url()
http://diveintomark.org/xml/atom.xml
>>> request.add_header('User-Agent',
...     'OpenAnything/1.0 +http://diveintopython3.org/')    
>>> feeddata = opener.open(request).read()                 
connect: (diveintomark.org, 80)
send: '
GET /xml/atom.xml HTTP/1.0
Host: diveintomark.org
User-agent: OpenAnything/1.0 +http://diveintopython3.org/   
'
reply: 'HTTP/1.1 200 OK\r\n'
header: Date: Wed, 14 Apr 2004 23:45:17 GMT
header: Server: Apache/2.0.49 (Debian GNU/Linux)
header: Content-Type: application/atom+xml
header: Last-Modified: Wed, 14 Apr 2004 22:14:38 GMT
header: ETag: "e8284-68e0-4de30f80"
header: Accept-Ranges: bytes
header: Content-Length: 26848
header: Connection: close

	You're continuing from the previous example; you've already created a Request object with the URL you want to access.
	Using the add_header method on the Request object, you can add arbitrary HTTP headers to the request. The first argument is the header, the second is the value you're providing for that header. Convention dictates that a User-Agent should be in this specific format: an application name, followed by a slash, followed by a version number. The rest is free-form, and you'll see a lot of variations in the wild, but somewhere it should include a URL of your application. The User-Agent is usually logged by the server along with other details of your request, and including a URL of your application allows server administrators looking through their access logs to contact you if something is wrong.
	The opener object you created before can be reused too, and it will retrieve the same feed again, but with your custom User-Agent header.
	And here's you sending your custom User-Agent, in place of the generic one that Python sends by default. If you look closely, you'll notice that you defined a User-Agent header, but you actually sent a User-agent header. See the difference? urllib2 changed the case so that only the first letter was capitalized. It doesn't really matter; HTTP specifies that header field names are completely case-insensitive.

11.6. Handling Last-Modified and ETag

Now that you know how to add custom HTTP headers to your web service requests, let's look at adding support for Last-Modified and ETag headers.

These examples show the output with debugging turned off. If you still have it turned on from the previous section, you can turn it off by setting httplib.HTTPConnection.debuglevel = 0. Or you can just leave debugging on, if that helps you.

Example 11.6. Testing Last-Modified

>>> import urllib2
>>> request = urllib2.Request('http://diveintomark.org/xml/atom.xml')
>>> opener = urllib2.build_opener()
>>> firstdatastream = opener.open(request)
>>> firstdatastream.headers.dict     
{'date': 'Thu, 15 Apr 2004 20:42:41 GMT', 
 'server': 'Apache/2.0.49 (Debian GNU/Linux)', 
 'content-type': 'application/atom+xml',
 'last-modified': 'Thu, 15 Apr 2004 19:45:21 GMT', 
 'etag': '"e842a-3e53-55d97640"',
 'content-length': '15955', 
 'accept-ranges': 'bytes', 
 'connection': 'close'}
>>> request.add_header('If-Modified-Since',
...     firstdatastream.headers.get('Last-Modified'))  
>>> seconddatastream = opener.open(request)            
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
  File "c:\python23\lib\urllib2.py", line 326, in open
    '_open', req)
  File "c:\python23\lib\urllib2.py", line 306, in _call_chain
    result = func(*args)
  File "c:\python23\lib\urllib2.py", line 901, in http_open
    return self.do_open(httplib.HTTP, req)
  File "c:\python23\lib\urllib2.py", line 895, in do_open
    return self.parent.error('http', req, fp, code, msg, hdrs)
  File "c:\python23\lib\urllib2.py", line 352, in error
    return self._call_chain(*args)
  File "c:\python23\lib\urllib2.py", line 306, in _call_chain
    result = func(*args)
  File "c:\python23\lib\urllib2.py", line 412, in http_error_default
    raise HTTPError(req.get_full_url(), code, msg, hdrs, fp)
urllib2.HTTPError: HTTP Error 304: Not Modified

	Remember all those HTTP headers you saw printed out when you turned on debugging? This is how you can get access to them programmatically: firstdatastream.headers is an object that acts like a dictionary and allows you to get any of the individual headers returned from the HTTP server.
	On the second request, you add the If-Modified-Since header with the last-modified date from the first request. If the data hasn't changed, the server should return a 304 status code.
	Sure enough, the data hasn't changed. You can see from the traceback that urllib2 throws a special exception, HTTPError, in response to the 304 status code. This is a little unusual, and not entirely helpful. After all, it's not an error; you specifically asked the server not to send you any data if it hadn't changed, and the data didn't change, so the server told you it wasn't sending you any data. That's not an error; that's exactly what you were hoping for.

urllib2 also raises an HTTPError exception for conditions that you would think of as errors, such as 404 (page not found). In fact, it will raise HTTPError for any status code other than 200 (OK), 301 (permanent redirect), or 302 (temporary redirect). It would be more helpful for your purposes to capture the status code and simply return it, without throwing an exception. To do that, you'll need to define a custom URL handler.

Example 11.7. Defining URL handlers

This custom URL handler is part of openanything.py.

class DefaultErrorHandler(urllib2.HTTPDefaultErrorHandler):    
    def http_error_default(self, req, fp, code, msg, headers): 
        result = urllib2.HTTPError(         
            req.get_full_url(), code, msg, headers, fp)       
        result.status = code                 
        return result     

	urllib2 is designed around URL handlers. Each handler is just a class that can define any number of methods. When something happens -- like an HTTP error, or even a 304 code -- urllib2 introspects into the list of defined handlers for a method that can handle it. You used a similar introspection in Chapter 9, XML Processing to define handlers for different node types, but urllib2 is more flexible, and introspects over as many handlers as are defined for the current request.
	urllib2 searches through the defined handlers and calls the http_error_default method when it encounters a 304 status code from the server. By defining a custom error handler, you can prevent urllib2 from raising an exception. Instead, you create the HTTPError object, but return it instead of raising it.
	This is the key part: before returning, you save the status code returned by the HTTP server. This will allow you easy access to it from the calling program.

Example 11.8. Using custom URL handlers

>>> request.headers         
{'If-modified-since': 'Thu, 15 Apr 2004 19:45:21 GMT'}
>>> import openanything
>>> opener = urllib2.build_opener(
...     openanything.DefaultErrorHandler())   
>>> seconddatastream = opener.open(request)
>>> seconddatastream.status 
304
>>> seconddatastream.read() 
''

	You're continuing the previous example, so the Request object is already set up, and you've already added the If-Modified-Since header.
	This is the key: now that you've defined your custom URL handler, you need to tell urllib2 to use it. Remember how I said that urllib2 broke up the process of accessing an HTTP resource into three steps, and for good reason? This is why building the URL opener is its own step, because you can build it with your own custom URL handlers that override urllib2's default behavior.
	Now you can quietly open the resource, and what you get back is an object that, along with the usual headers (use seconddatastream.headers.dict to acess them), also contains the HTTP status code. In this case, as you expected, the status is 304, meaning this data hasn't changed since the last time you asked for it.
	Note that when the server sends back a 304 status code, it doesn't re-send the data. That's the whole point: to save bandwidth by not re-downloading data that hasn't changed. So if you actually want that data, you'll need to cache it locally the first time you get it.

Handling ETag works much the same way, but instead of checking for Last-Modified and sending If-Modified-Since, you check for ETag and send If-None-Match. Let's start with a fresh IDE session.

Example 11.9. Supporting ETag/If-None-Match

>>> import urllib2, openanything
>>> request = urllib2.Request('http://diveintomark.org/xml/atom.xml')
>>> opener = urllib2.build_opener(
...     openanything.DefaultErrorHandler())
>>> firstdatastream = opener.open(request)
>>> firstdatastream.headers.get('ETag')        
'"e842a-3e53-55d97640"'
>>> firstdata = firstdatastream.read()
>>> print firstdata          
<?xml version="1.0" encoding="iso-8859-1"?>
<feed version="0.3"
  xmlns="http://purl.org/atom/ns#"
  xmlns:dc="http://purl.org/dc/elements/1.1/"
  xml:lang="en">
  <title mode="escaped">dive into mark</title>
  <link rel="alternate" type="text/html" href="http://diveintomark.org/"/>
  <-- rest of feed omitted for brevity -->
>>> request.add_header('If-None-Match',
...     firstdatastream.headers.get('ETag'))   
>>> seconddatastream = opener.open(request)
>>> seconddatastream.status  
304
>>> seconddatastream.read()  
''

	Using the firstdatastream.headers pseudo-dictionary, you can get the ETag returned from the server. (What happens if the server didn't send back an ETag? Then this line would return None.)
	OK, you got the data.
	Now set up the second call by setting the If-None-Match header to the ETag you got from the first call.
	The second call succeeds quietly (without throwing an exception), and once again you see that the server has sent back a 304 status code. Based on the ETag you sent the second time, it knows that the data hasn't changed.
	Regardless of whether the 304 is triggered by Last-Modified date checking or ETag hash matching, you'll never get the data along with the 304. That's the whole point.

	In these examples, the HTTP server has supported both Last-Modified and ETag headers, but not all servers do. As a web services client, you should be prepared to support both, but you must code defensively in case a server only supports one or the other, or neither.

11.7. Handling redirects

You can support permanent and temporary redirects using a different kind of custom URL handler.

First, let's see why a redirect handler is necessary in the first place.

Example 11.10. Accessing web services without a redirect handler

>>> import urllib2, httplib
>>> httplib.HTTPConnection.debuglevel = 1           
>>> request = urllib2.Request(
...     'http://diveintomark.org/redir/example301.xml') 
>>> opener = urllib2.build_opener()
>>> f = opener.open(request)
connect: (diveintomark.org, 80)
send: '
GET /redir/example301.xml HTTP/1.0
Host: diveintomark.org
User-agent: Python-urllib/2.1
'
reply: 'HTTP/1.1 301 Moved Permanently\r\n'             
header: Date: Thu, 15 Apr 2004 22:06:25 GMT
header: Server: Apache/2.0.49 (Debian GNU/Linux)
header: Location: http://diveintomark.org/xml/atom.xml  
header: Content-Length: 338
header: Connection: close
header: Content-Type: text/html; charset=iso-8859-1
connect: (diveintomark.org, 80)
send: '
GET /xml/atom.xml HTTP/1.0            
Host: diveintomark.org
User-agent: Python-urllib/2.1
'
reply: 'HTTP/1.1 200 OK\r\n'
header: Date: Thu, 15 Apr 2004 22:06:25 GMT
header: Server: Apache/2.0.49 (Debian GNU/Linux)
header: Last-Modified: Thu, 15 Apr 2004 19:45:21 GMT
header: ETag: "e842a-3e53-55d97640"
header: Accept-Ranges: bytes
header: Content-Length: 15955
header: Connection: close
header: Content-Type: application/atom+xml
>>> f.url           
'http://diveintomark.org/xml/atom.xml'
>>> f.headers.dict
{'content-length': '15955', 
'accept-ranges': 'bytes', 
'server': 'Apache/2.0.49 (Debian GNU/Linux)', 
'last-modified': 'Thu, 15 Apr 2004 19:45:21 GMT', 
'connection': 'close', 
'etag': '"e842a-3e53-55d97640"', 
'date': 'Thu, 15 Apr 2004 22:06:25 GMT', 
'content-type': 'application/atom+xml'}
>>> f.status
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
AttributeError: addinfourl instance has no attribute 'status'

	You'll be better able to see what's happening if you turn on debugging.
	This is a URL which I have set up to permanently redirect to my Atom feed at http://diveintomark.org/xml/atom.xml.
	Sure enough, when you try to download the data at that address, the server sends back a 301 status code, telling you that the resource has moved permanently.
	The server also sends back a Location: header that gives the new address of this data.
	urllib2 notices the redirect status code and automatically tries to retrieve the data at the new location specified in the Location: header.
	The object you get back from the opener contains the new permanent address and all the headers returned from the second request (retrieved from the new permanent address). But the status code is missing, so you have no way of knowing programmatically whether this redirect was temporary or permanent. And that matters very much: if it was a temporary redirect, then you should continue to ask for the data at the old location. But if it was a permanent redirect (as this was), you should ask for the data at the new location from now on.

This is suboptimal, but easy to fix. urllib2 doesn't behave exactly as you want it to when it encounters a 301 or 302, so let's override its behavior. How? With a custom URL handler, just like you did to handle 304 codes.

Example 11.11. Defining the redirect handler

This class is defined in openanything.py.

class SmartRedirectHandler(urllib2.HTTPRedirectHandler):     
    def http_error_301(self, req, fp, code, msg, headers):  
        result = urllib2.HTTPRedirectHandler.http_error_301( 
            self, req, fp, code, msg, headers)              
        result.status = code               
        return result   

    def http_error_302(self, req, fp, code, msg, headers):   
        result = urllib2.HTTPRedirectHandler.http_error_302(
            self, req, fp, code, msg, headers)              
        result.status = code              
        return result   

	Redirect behavior is defined in urllib2 in a class called HTTPRedirectHandler. You don't want to completely override the behavior, you just want to extend it a little, so you'll subclass HTTPRedirectHandler so you can call the ancestor class to do all the hard work.
	When it encounters a 301 status code from the server, urllib2 will search through its handlers and call the http_error_301 method. The first thing ours does is just call the http_error_301 method in the ancestor, which handles the grunt work of looking for the Location: header and following the redirect to the new address.
	Here's the key: before you return, you store the status code (301), so that the calling program can access it later.
	Temporary redirects (status code 302) work the same way: override the http_error_302 method, call the ancestor, and save the status code before returning.

So what has this bought us? You can now build a URL opener with the custom redirect handler, and it will still automatically follow redirects, but now it will also expose the redirect status code.

Example 11.12. Using the redirect handler to detect permanent redirects

>>> request = urllib2.Request('http://diveintomark.org/redir/example301.xml')
>>> import openanything, httplib
>>> httplib.HTTPConnection.debuglevel = 1
>>> opener = urllib2.build_opener(
...     openanything.SmartRedirectHandler())           
>>> f = opener.open(request)
connect: (diveintomark.org, 80)
send: 'GET /redir/example301.xml HTTP/1.0
Host: diveintomark.org
User-agent: Python-urllib/2.1
'
reply: 'HTTP/1.1 301 Moved Permanently\r\n'            
header: Date: Thu, 15 Apr 2004 22:13:21 GMT
header: Server: Apache/2.0.49 (Debian GNU/Linux)
header: Location: http://diveintomark.org/xml/atom.xml
header: Content-Length: 338
header: Connection: close
header: Content-Type: text/html; charset=iso-8859-1
connect: (diveintomark.org, 80)
send: '
GET /xml/atom.xml HTTP/1.0
Host: diveintomark.org
User-agent: Python-urllib/2.1
'
reply: 'HTTP/1.1 200 OK\r\n'
header: Date: Thu, 15 Apr 2004 22:13:21 GMT
header: Server: Apache/2.0.49 (Debian GNU/Linux)
header: Last-Modified: Thu, 15 Apr 2004 19:45:21 GMT
header: ETag: "e842a-3e53-55d97640"
header: Accept-Ranges: bytes
header: Content-Length: 15955
header: Connection: close
header: Content-Type: application/atom+xml

>>> f.status       
301
>>> f.url
'http://diveintomark.org/xml/atom.xml'

	First, build a URL opener with the redirect handler you just defined.
	You sent off a request, and you got a 301 status code in response. At this point, the http_error_301 method gets called. You call the ancestor method, which follows the redirect and sends a request at the new location (http://diveintomark.org/xml/atom.xml).
	This is the payoff: now, not only do you have access to the new URL, but you have access to the redirect status code, so you can tell that this was a permanent redirect. The next time you request this data, you should request it from the new location (http://diveintomark.org/xml/atom.xml, as specified in f.url). If you had stored the location in a configuration file or a database, you need to update that so you don't keep pounding the server with requests at the old address. It's time to update your address book.

The same redirect handler can also tell you that you shouldn't update your address book.

Example 11.13. Using the redirect handler to detect temporary redirects

>>> request = urllib2.Request(
...     'http://diveintomark.org/redir/example302.xml')   
>>> f = opener.open(request)
connect: (diveintomark.org, 80)
send: '
GET /redir/example302.xml HTTP/1.0
Host: diveintomark.org
User-agent: Python-urllib/2.1
'
reply: 'HTTP/1.1 302 Found\r\n'         
header: Date: Thu, 15 Apr 2004 22:18:21 GMT
header: Server: Apache/2.0.49 (Debian GNU/Linux)
header: Location: http://diveintomark.org/xml/atom.xml
header: Content-Length: 314
header: Connection: close
header: Content-Type: text/html; charset=iso-8859-1
connect: (diveintomark.org, 80)
send: '
GET /xml/atom.xml HTTP/1.0              
Host: diveintomark.org
User-agent: Python-urllib/2.1
'
reply: 'HTTP/1.1 200 OK\r\n'
header: Date: Thu, 15 Apr 2004 22:18:21 GMT
header: Server: Apache/2.0.49 (Debian GNU/Linux)
header: Last-Modified: Thu, 15 Apr 2004 19:45:21 GMT
header: ETag: "e842a-3e53-55d97640"
header: Accept-Ranges: bytes
header: Content-Length: 15955
header: Connection: close
header: Content-Type: application/atom+xml
>>> f.status          
302
>>> f.url
http://diveintomark.org/xml/atom.xml

	This is a sample URL I've set up that is configured to tell clients to temporarily redirect to http://diveintomark.org/xml/atom.xml.
	The server sends back a 302 status code, indicating a temporary redirect. The temporary new location of the data is given in the Location: header.
	urllib2 calls your http_error_302 method, which calls the ancestor method of the same name in urllib2.HTTPRedirectHandler, which follows the redirect to the new location. Then your http_error_302 method stores the status code (302) so the calling application can get it later.
	And here you are, having successfully followed the redirect to http://diveintomark.org/xml/atom.xml. f.status tells you that this was a temporary redirect, which means that you should continue to request data from the original address (http://diveintomark.org/redir/example302.xml). Maybe it will redirect next time too, but maybe not. Maybe it will redirect to a different address. It's not for you to say. The server said this redirect was only temporary, so you should respect that. And now you're exposing enough information that the calling application can respect that.

11.8. Handling compressed data

The last important HTTP feature you want to support is compression. Many web services have the ability to send data compressed, which can cut down the amount of data sent over the wire by 60% or more. This is especially true of XML web services, since XML data compresses very well.

Servers won't give you compressed data unless you tell them you can handle it.

Example 11.14. Telling the server you would like compressed data

>>> import urllib2, httplib
>>> httplib.HTTPConnection.debuglevel = 1
>>> request = urllib2.Request('http://diveintomark.org/xml/atom.xml')
>>> request.add_header('Accept-encoding', 'gzip')        
>>> opener = urllib2.build_opener()
>>> f = opener.open(request)
connect: (diveintomark.org, 80)
send: '
GET /xml/atom.xml HTTP/1.0
Host: diveintomark.org
User-agent: Python-urllib/2.1
Accept-encoding: gzip
'
reply: 'HTTP/1.1 200 OK\r\n'
header: Date: Thu, 15 Apr 2004 22:24:39 GMT
header: Server: Apache/2.0.49 (Debian GNU/Linux)
header: Last-Modified: Thu, 15 Apr 2004 19:45:21 GMT
header: ETag: "e842a-3e53-55d97640"
header: Accept-Ranges: bytes
header: Vary: Accept-Encoding
header: Content-Encoding: gzip         
header: Content-Length: 6289           
header: Connection: close
header: Content-Type: application/atom+xml

	This is the key: once you've created your Request object, add an Accept-encoding header to tell the server you can accept gzip-encoded data. gzip is the name of the compression algorithm you're using. In theory there could be other compression algorithms, but gzip is the compression algorithm used by 99% of web servers.
	There's your header going across the wire.
	And here's what the server sends back: the Content-Encoding: gzip header means that the data you're about to receive has been gzip-compressed.
	The Content-Length header is the length of the compressed data, not the uncompressed data. As you'll see in a minute, the actual length of the uncompressed data was 15955, so gzip compression cut your bandwidth by over 60%!

Example 11.15. Decompressing the data

>>> compresseddata = f.read()            
>>> len(compresseddata)
6289
>>> import StringIO
>>> compressedstream = StringIO.StringIO(compresseddata)   
>>> import gzip
>>> gzipper = gzip.GzipFile(fileobj=compressedstream)      
>>> data = gzipper.read()                
>>> print data         
<?xml version="1.0" encoding="iso-8859-1"?>
<feed version="0.3"
  xmlns="http://purl.org/atom/ns#"
  xmlns:dc="http://purl.org/dc/elements/1.1/"
  xml:lang="en">
  <title mode="escaped">dive into mark</title>
  <link rel="alternate" type="text/html" href="http://diveintomark.org/"/>
  <-- rest of feed omitted for brevity -->
>>> len(data)
15955

	Continuing from the previous example, f is the file-like object returned from the URL opener. Using its read() method would ordinarily get you the uncompressed data, but since this data has been gzip-compressed, this is just the first step towards getting the data you really want.
	OK, this step is a little bit of messy workaround. Python has a gzip module, which reads (and actually writes) gzip-compressed files on disk. But you don't have a file on disk, you have a gzip-compressed buffer in memory, and you don't want to write out a temporary file just so you can uncompress it. So what you're going to do is create a file-like object out of the in-memory data (compresseddata), using the StringIO module. You first saw the StringIO module in the previous chapter, but now you've found another use for it.
	Now you can create an instance of GzipFile, and tell it that its “file” is the file-like object compressedstream.
	This is the line that does all the actual work: “reading” from GzipFile will decompress the data. Strange? Yes, but it makes sense in a twisted kind of way. gzipper is a file-like object which represents a gzip-compressed file. That “file” is not a real file on disk, though; gzipper is really just “reading” from the file-like object you created with StringIO to wrap the compressed data, which is only in memory in the variable compresseddata. And where did that compressed data come from? You originally downloaded it from a remote HTTP server by “reading” from the file-like object you built with urllib2.build_opener. And amazingly, this all just works. Every step in the chain has no idea that the previous step is faking it.
	Look ma, real data. (15955 bytes of it, in fact.)

“But wait!” I hear you cry. “This could be even easier!” I know what you're thinking. You're thinking that opener.open returns a file-like object, so why not cut out the StringIO middleman and just pass f directly to GzipFile? OK, maybe you weren't thinking that, but don't worry about it, because it doesn't work.

Example 11.16. Decompressing the data directly from the server

>>> f = opener.open(request)
>>> f.headers.get('Content-Encoding')         
'gzip'
>>> data = gzip.GzipFile(fileobj=f).read()    
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
  File "c:\python23\lib\gzip.py", line 217, in read
    self._read(readsize)
  File "c:\python23\lib\gzip.py", line 252, in _read
    pos = self.fileobj.tell()   # Save current position
AttributeError: addinfourl instance has no attribute 'tell'

	Continuing from the previous example, you already have a Request object set up with an Accept-encoding: gzip header.
	Simply opening the request will get you the headers (though not download any data yet). As you can see from the returned Content-Encoding header, this data has been sent gzip-compressed.
	Since opener.open returns a file-like object, and you know from the headers that when you read it, you're going to get gzip-compressed data, why not simply pass that file-like object directly to GzipFile? As you “read” from the GzipFile instance, it will “read” compressed data from the remote HTTP server and decompress it on the fly. It's a good idea, but unfortunately it doesn't work. Because of the way gzip compression works, GzipFile needs to save its position and move forwards and backwards through the compressed file. This doesn't work when the “file” is a stream of bytes coming from a remote server; all you can do with it is retrieve bytes one at a time, not move back and forth through the data stream. So the inelegant hack of using StringIO is the best solution: download the compressed data, create a file-like object out of it with StringIO, and then decompress the data from that.

11.9. Putting it all together

You've seen all the pieces for building an intelligent HTTP web services client. Now let's see how they all fit together.

Example 11.17. The openanything function

This function is defined in openanything.py.

def openAnything(source, etag=None, lastmodified=None, agent=USER_AGENT):
    # non-HTTP code omitted for brevity
    if urlparse.urlparse(source)[0] == 'http':   
        # open URL with urllib2                 
        request = urllib2.Request(source)       
        request.add_header('User-Agent', agent)  
        if etag:              
            request.add_header('If-None-Match', etag)              
        if lastmodified:      
            request.add_header('If-Modified-Since', lastmodified)  
        request.add_header('Accept-encoding', 'gzip')              
        opener = urllib2.build_opener(SmartRedirectHandler(), DefaultErrorHandler()) 
        return opener.open(request)              

	urlparse is a handy utility module for, you guessed it, parsing URLs. It's primary function, also called urlparse, takes a URL and splits it into a tuple of (scheme, domain, path, params, query string parameters, and fragment identifier). Of these, the only thing you care about is the scheme, to make sure that you're dealing with an HTTP URL (which urllib2 can handle).
	You identify yourself to the HTTP server with the User-Agent passed in by the calling function. If no User-Agent was specified, you use a default one defined earlier in the openanything.py module. You never use the default one defined by urllib2.
	If an ETag hash was given, send it in the If-None-Match header.
	If a last-modified date was given, send it in the If-Modified-Since header.
	Tell the server you would like compressed data if possible.
	Build a URL opener that uses both of the custom URL handlers: SmartRedirectHandler for handling 301 and 302 redirects, and DefaultErrorHandler for handling 304, 404, and other error conditions gracefully.
	That's it! Open the URL and return a file-like object to the caller.

Example 11.18. The fetch function

This function is defined in openanything.py.

def fetch(source, etag=None, last_modified=None, agent=USER_AGENT):  
    '''Fetch data and metadata from a URL, file, stream, or string'''
    result = {}
    f = openAnything(source, etag, last_modified, agent)              
    result['data'] = f.read()     
    if hasattr(f, 'headers'):    
        # save ETag, if the server sent one        
        result['etag'] = f.headers.get('ETag')      
        # save Last-Modified header, if the server sent one          
        result['lastmodified'] = f.headers.get('Last-Modified')       
        if f.headers.get('content-encoding', '') == 'gzip':           
            # data came back gzip-compressed, decompress it          
            result['data'] = gzip.GzipFile(fileobj=StringIO(result['data']])).read()
    if hasattr(f, 'url'):         
        result['url'] = f.url    
        result['status'] = 200   
    if hasattr(f, 'status'):      
        result['status'] = f.status                
    f.close()  
    return result                

	First, you call the openAnything function with a URL, ETag hash, Last-Modified date, and User-Agent.
	Read the actual data returned from the server. This may be compressed; if so, you'll decompress it later.
	Save the ETag hash returned from the server, so the calling application can pass it back to you next time, and you can pass it on to openAnything, which can stick it in the If-None-Match header and send it to the remote server.
	Save the Last-Modified date too.
	If the server says that it sent compressed data, decompress it.
	If you got a URL back from the server, save it, and assume that the status code is 200 until you find out otherwise.
	If one of the custom URL handlers captured a status code, then save that too.

Example 11.19. Using openanything.py

>>> import openanything
>>> useragent = 'MyHTTPWebServicesApp/1.0'
>>> url = 'http://diveintopython3.org/redir/example301.xml'
>>> params = openanything.fetch(url, agent=useragent)              
>>> params   
{'url': 'http://diveintomark.org/xml/atom.xml', 
'lastmodified': 'Thu, 15 Apr 2004 19:45:21 GMT', 
'etag': '"e842a-3e53-55d97640"', 
'status': 301,
'data': '<?xml version="1.0" encoding="iso-8859-1"?>
<feed version="0.3"
<-- rest of data omitted for brevity -->'}
>>> if params['status'] == 301:
...     url = params['url']
>>> newparams = openanything.fetch(
...     url, params['etag'], params['lastmodified'], useragent)    
>>> newparams
{'url': 'http://diveintomark.org/xml/atom.xml', 
'lastmodified': None, 
'etag': '"e842a-3e53-55d97640"', 
'status': 304,
'data': ''}  

	The very first time you fetch a resource, you don't have an ETag hash or Last-Modified date, so you'll leave those out. (They're optional parameters.)
	What you get back is a dictionary of several useful headers, the HTTP status code, and the actual data returned from the server. openanything handles the gzip compression internally; you don't care about that at this level.
	If you ever get a 301 status code, that's a permanent redirect, and you need to update your URL to the new address.
	The second time you fetch the same resource, you have all sorts of information to pass back: a (possibly updated) URL, the ETag from the last time, the Last-Modified date from the last time, and of course your User-Agent.
	What you get back is again a dictionary, but the data hasn't changed, so all you got was a 304 status code and no data.

11.10. Summary

The openanything.py and its functions should now make perfect sense.

There are 5 important features of HTTP web services that every client should support:

• Identifying your application by setting a proper User-Agent.
• Handling permanent redirects properly.
• Supporting Last-Modified date checking to avoid re-downloading data that hasn't changed.
• Supporting ETag hashes to avoid re-downloading data that hasn't changed.
• Supporting gzip compression to reduce bandwidth even when data has changed.

Chapter 12. SOAP Web Services

Chapter 11 focused on document-oriented web services over HTTP. The “input parameter” was the URL, and the “return value” was an actual XML document which it was your responsibility to parse.

This chapter will focus on SOAP web services, which take a more structured approach. Rather than dealing with HTTP requests and XML documents directly, SOAP allows you to simulate calling functions that return native data types. As you will see, the illusion is almost perfect; you can “call” a function through a SOAP library, with the standard Python calling syntax, and the function appears to return Python objects and values. But under the covers, the SOAP library has actually performed a complex transaction involving multiple XML documents and a remote server.

SOAP is a complex specification, and it is somewhat misleading to say that SOAP is all about calling remote functions. Some people would pipe up to add that SOAP allows for one-way asynchronous message passing, and document-oriented web services. And those people would be correct; SOAP can be used that way, and in many different ways. But this chapter will focus on so-called “RPC-style” SOAP -- calling a remote function and getting results back.

12.1. Diving In

You use Google, right? It's a popular search engine. Have you ever wished you could programmatically access Google search results? Now you can. Here is a program to search Google from Python.

Example 12.1. search.py

from SOAPpy import WSDL

# you'll need to configure these two values;
# see http://www.google.com/apis/
WSDLFILE = '/path/to/copy/of/GoogleSearch.wsdl'
APIKEY = 'YOUR_GOOGLE_API_KEY'

_server = WSDL.Proxy(WSDLFILE)
def search(q):
    """Search Google and return list of {title, link, description}"""
    results = _server.doGoogleSearch(
        APIKEY, q, 0, 10, False, "", False, "", "utf-8", "utf-8")
    return [{"title": r.title.encode("utf-8"),
             "link": r.URL.encode("utf-8"),
             "description": r.snippet.encode("utf-8")}
            for r in results.resultElements]

if __name__ == '__main__':
    import sys
    for r in search(sys.argv[1])[:5]:
        print r['title']
        print r['link']
        print r['description']
        print

You can import this as a module and use it from a larger program, or you can run the script from the command line. On the command line, you give the search query as a command-line argument, and it prints out the URL, title, and description of the top five Google search results.

Here is the sample output for a search for the word “python”.

Example 12.2. Sample Usage of search.py

C:\diveintopython3\common\py> python search.py "python"
<b>Python</b> Programming Language
http://www.python.org/
Home page for <b>Python</b>, an interpreted, interactive, object-oriented,
extensible<br> programming language. <b>...</b> <b>Python</b>
is OSI Certified Open Source: OSI Certified.

<b>Python</b> Documentation Index
http://www.python.org/doc/
 <b>...</b> New-style classes (aka descrintro). Regular expressions. Database
API. Email Us.<br> docs@<b>python</b>.org. (c) 2004. <b>Python</b>
Software Foundation. <b>Python</b> Documentation. <b>...</b>

Download <b>Python</b> Software
http://www.python.org/download/
Download Standard <b>Python</b> Software. <b>Python</b> 2.3.3 is the
current production<br> version of <b>Python</b>. <b>...</b>
<b>Python</b> is OSI Certified Open Source:

Pythonline
http://www.pythonline.com/


Dive Into <b>Python</b>
http://diveintopython3.org/
Dive Into <b>Python</b>. <b>Python</b> from novice to pro. Find:
<b>...</b> It is also available in multiple<br> languages. Read
Dive Into <b>Python</b>. This book is still being written. <b>...</b>

Further Reading on SOAP

• http://www.xmethods.net/ is a repository of public access SOAP web services.
• The SOAP specification is surprisingly readable, if you like that sort of thing.

12.2. Installing the SOAP Libraries

Unlike the other code in this book, this chapter relies on libraries that do not come pre-installed with Python.

Before you can dive into SOAP web services, you'll need to install three libraries: PyXML, fpconst, and SOAPpy.

12.2.1. Installing PyXML

The first library you need is PyXML, an advanced set of XML libraries that provide more functionality than the built-in XML libraries we studied in Chapter 9.

Procedure 12.1.

Here is the procedure for installing PyXML:

1. Go to http://pyxml.sourceforge.net/, click Downloads, and download the latest version for your operating system.

2. If you are using Windows, there are several choices. Make sure to download the version of PyXML that matches the version of Python you are using.

3. Double-click the installer. If you download PyXML 0.8.3 for Windows and Python 2.3, the installer program will be PyXML-0.8.3.win32-py2.3.exe.

4. Step through the installer program.

5. After the installation is complete, close the installer. There will not be any visible indication of success (no programs installed on the Start Menu or shortcuts installed on the desktop). PyXML is simply a collection of XML libraries used by other programs.

To verify that you installed PyXML correctly, run your Python IDE and check the version of the XML libraries you have installed, as shown here.

Example 12.3. Verifying PyXML Installation

>>> import xml
>>> xml.__version__
'0.8.3'

This version number should match the version number of the PyXML installer program you downloaded and ran.

12.2.2. Installing fpconst

The second library you need is fpconst, a set of constants and functions for working with IEEE754 double-precision special values. This provides support for the special values Not-a-Number (NaN), Positive Infinity (Inf), and Negative Infinity (-Inf), which are part of the SOAP datatype specification.

Procedure 12.2.

Here is the procedure for installing fpconst:

1. Download the latest version of fpconst from http://www.analytics.washington.edu/statcomp/projects/rzope/fpconst/.

2. There are two downloads available, one in .tar.gz format, the other in .zip format. If you are using Windows, download the .zip file; otherwise, download the .tar.gz file.

3. Decompress the downloaded file. On Windows XP, you can right-click on the file and choose Extract All; on earlier versions of Windows, you will need a third-party program such as WinZip. On Mac OS X, you can double-click the compressed file to decompress it with Stuffit Expander.

4. Open a command prompt and navigate to the directory where you decompressed the fpconst files.

5. Type python setup.py install to run the installation program.

To verify that you installed fpconst correctly, run your Python IDE and check the version number.

Example 12.4. Verifying fpconst Installation

>>> import fpconst
>>> fpconst.__version__
'0.6.0'

This version number should match the version number of the fpconst archive you downloaded and installed.

12.2.3. Installing SOAPpy

The third and final requirement is the SOAP library itself: SOAPpy.

Procedure 12.3.

Here is the procedure for installing SOAPpy:

1. Go to http://pywebsvcs.sourceforge.net/ and select Latest Official Release under the SOAPpy section.

2. There are two downloads available. If you are using Windows, download the .zip file; otherwise, download the .tar.gz file.

3. Decompress the downloaded file, just as you did with fpconst.

4. Open a command prompt and navigate to the directory where you decompressed the SOAPpy files.

5. Type python setup.py install to run the installation program.

To verify that you installed SOAPpy correctly, run your Python IDE and check the version number.

Example 12.5. Verifying SOAPpy Installation

>>> import SOAPpy
>>> SOAPpy.__version__
'0.11.4'

This version number should match the version number of the SOAPpy archive you downloaded and installed.

12.3. First Steps with SOAP

The heart of SOAP is the ability to call remote functions. There are a number of public access SOAP servers that provide simple functions for demonstration purposes.

The most popular public access SOAP server is http://www.xmethods.net/. This example uses a demonstration function that takes a United States zip code and returns the current temperature in that region.

Example 12.6. Getting the Current Temperature

>>> from SOAPpy import SOAPProxy            
>>> url = 'http://services.xmethods.net:80/soap/servlet/rpcrouter'
>>> namespace = 'urn:xmethods-Temperature'  
>>> server = SOAPProxy(url, namespace)      
>>> server.getTemp('27502')                 
80.0

	You access the remote SOAP server through a proxy class, SOAPProxy. The proxy handles all the internals of SOAP for you, including creating the XML request document out of the function name and argument list, sending the request over HTTP to the remote SOAP server, parsing the XML response document, and creating native Python values to return. You'll see what these XML documents look like in the next section.
	Every SOAP service has a URL which handles all the requests. The same URL is used for all function calls. This particular service only has a single function, but later in this chapter you'll see examples of the Google API, which has several functions. The service URL is shared by all functions.Each SOAP service also has a namespace, which is defined by the server and is completely arbitrary. It's simply part of the configuration required to call SOAP methods. It allows the server to share a single service URL and route requests between several unrelated services. It's like dividing Python modules into packages.
	You're creating the SOAPProxy with the service URL and the service namespace. This doesn't make any connection to the SOAP server; it simply creates a local Python object.
	Now with everything configured properly, you can actually call remote SOAP methods as if they were local functions. You pass arguments just like a normal function, and you get a return value just like a normal function. But under the covers, there's a heck of a lot going on.

Let's peek under those covers.

12.4. Debugging SOAP Web Services

The SOAP libraries provide an easy way to see what's going on behind the scenes.

Turning on debugging is a simple matter of setting two flags in the SOAPProxy's configuration.

Example 12.7. Debugging SOAP Web Services

>>> from SOAPpy import SOAPProxy
>>> url = 'http://services.xmethods.net:80/soap/servlet/rpcrouter'
>>> n = 'urn:xmethods-Temperature'
>>> server = SOAPProxy(url, namespace=n)     
>>> server.config.dumpSOAPOut = 1            
>>> server.config.dumpSOAPIn = 1
>>> temperature = server.getTemp('27502')    
*** Outgoing SOAP ******************************************************
<?xml version="1.0" encoding="UTF-8"?>
<SOAP-ENV:Envelope SOAP-ENV:encodingStyle="http://schemas.xmlsoap.org/soap/encoding/"
  xmlns:SOAP-ENC="http://schemas.xmlsoap.org/soap/encoding/"
  xmlns:xsi="http://www.w3.org/1999/XMLSchema-instance"
  xmlns:SOAP-ENV="http://schemas.xmlsoap.org/soap/envelope/"
  xmlns:xsd="http://www.w3.org/1999/XMLSchema">
<SOAP-ENV:Body>
<ns1:getTemp xmlns:ns1="urn:xmethods-Temperature" SOAP-ENC:root="1">
<v1 xsi:type="xsd:string">27502</v1>
</ns1:getTemp>
</SOAP-ENV:Body>
</SOAP-ENV:Envelope>
************************************************************************
*** Incoming SOAP ******************************************************
<?xml version='1.0' encoding='UTF-8'?>
<SOAP-ENV:Envelope xmlns:SOAP-ENV="http://schemas.xmlsoap.org/soap/envelope/"
  xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
  xmlns:xsd="http://www.w3.org/2001/XMLSchema">
<SOAP-ENV:Body>
<ns1:getTempResponse xmlns:ns1="urn:xmethods-Temperature"
  SOAP-ENV:encodingStyle="http://schemas.xmlsoap.org/soap/encoding/">
<return xsi:type="xsd:float">80.0</return>
</ns1:getTempResponse>

</SOAP-ENV:Body>
</SOAP-ENV:Envelope>
************************************************************************

>>> temperature
80.0

	First, create the SOAPProxy like normal, with the service URL and the namespace.
	Second, turn on debugging by setting server.config.dumpSOAPIn and server.config.dumpSOAPOut.
	Third, call the remote SOAP method as usual. The SOAP library will print out both the outgoing XML request document, and the incoming XML response document. This is all the hard work that SOAPProxy is doing for you. Intimidating, isn't it? Let's break it down.

Most of the XML request document that gets sent to the server is just boilerplate. Ignore all the namespace declarations; they're going to be the same (or similar) for all SOAP calls. The heart of the “function call” is this fragment within the <Body> element:

<ns1:getTemp               
  xmlns:ns1="urn:xmethods-Temperature"       
  SOAP-ENC:root="1">
<v1 xsi:type="xsd:string">27502</v1>         
</ns1:getTemp>

	The element name is the function name, getTemp. SOAPProxy uses getattr as a dispatcher. Instead of calling separate local methods based on the method name, it actually uses the method name to construct the XML request document.
	The function's XML element is contained in a specific namespace, which is the namespace you specified when you created the SOAPProxy object. Don't worry about the SOAP-ENC:root; that's boilerplate too.
	The arguments of the function also got translated into XML. SOAPProxy introspects each argument to determine its datatype (in this case it's a string). The argument datatype goes into the xsi:type attribute, followed by the actual string value.

The XML return document is equally easy to understand, once you know what to ignore. Focus on this fragment within the <Body>:

<ns1:getTempResponse           
  xmlns:ns1="urn:xmethods-Temperature"           
  SOAP-ENV:encodingStyle="http://schemas.xmlsoap.org/soap/encoding/">
<return xsi:type="xsd:float">80.0</return>       
</ns1:getTempResponse>

	The server wraps the function return value within a <getTempResponse> element. By convention, this wrapper element is the name of the function, plus Response. But it could really be almost anything; the important thing that SOAPProxy notices is not the element name, but the namespace.
	The server returns the response in the same namespace we used in the request, the same namespace we specified when we first create the SOAPProxy. Later in this chapter we'll see what happens if you forget to specify the namespace when creating the SOAPProxy.
	The return value is specified, along with its datatype (it's a float). SOAPProxy uses this explicit datatype to create a Python object of the correct native datatype and return it.

12.5. Introducing WSDL

The SOAPProxy class proxies local method calls and transparently turns then into invocations of remote SOAP methods. As you've seen, this is a lot of work, and SOAPProxy does it quickly and transparently. What it doesn't do is provide any means of method introspection.

Consider this: the previous two sections showed an example of calling a simple remote SOAP method with one argument and one return value, both of simple data types. This required knowing, and keeping track of, the service URL, the service namespace, the function name, the number of arguments, and the datatype of each argument. If any of these is missing or wrong, the whole thing falls apart.

That shouldn't come as a big surprise. If I wanted to call a local function, I would need to know what package or module it was in (the equivalent of service URL and namespace). I would need to know the correct function name and the correct number of arguments. Python deftly handles datatyping without explicit types, but I would still need to know how many argument to pass, and how many return values to expect.

The big difference is introspection. As you saw in Chapter 4, Python excels at letting you discover things about modules and functions at runtime. You can list the available functions within a module, and with a little work, drill down to individual function declarations and arguments.

WSDL lets you do that with SOAP web services. WSDL stands for “Web Services Description Language”. Although designed to be flexible enough to describe many types of web services, it is most often used to describe SOAP web services.

A WSDL file is just that: a file. More specifically, it's an XML file. It usually lives on the same server you use to access the SOAP web services it describes, although there's nothing special about it. Later in this chapter, we'll download the WSDL file for the Google API and use it locally. That doesn't mean we're calling Google locally; the WSDL file still describes the remote functions sitting on Google's server.

A WSDL file contains a description of everything involved in calling a SOAP web service:

• The service URL and namespace
• The type of web service (probably function calls using SOAP, although as I mentioned, WSDL is flexible enough to describe a wide variety of web services)
• The list of available functions
• The arguments for each function
• The datatype of each argument
• The return values of each function, and the datatype of each return value

In other words, a WSDL file tells you everything you need to know to be able to call a SOAP web service.

12.6. Introspecting SOAP Web Services with WSDL

Like many things in the web services arena, WSDL has a long and checkered history, full of political strife and intrigue. I will skip over this history entirely, since it bores me to tears. There were other standards that tried to do similar things, but WSDL won, so let's learn how to use it.

The most fundamental thing that WSDL allows you to do is discover the available methods offered by a SOAP server.

Example 12.8. Discovering The Available Methods

>>> from SOAPpy import WSDL          
>>> wsdlFile = 'http://www.xmethods.net/sd/2001/TemperatureService.wsdl')
>>> server = WSDL.Proxy(wsdlFile)    
>>> server.methods.keys()            
[u'getTemp']

	SOAPpy includes a WSDL parser. At the time of this writing, it was labeled as being in the early stages of development, but I had no problem parsing any of the WSDL files I tried.
	To use a WSDL file, you again use a proxy class, WSDL.Proxy, which takes a single argument: the WSDL file. Note that in this case you are passing in the URL of a WSDL file stored on the remote server, but the proxy class works just as well with a local copy of the WSDL file. The act of creating the WSDL proxy will download the WSDL file and parse it, so it there are any errors in the WSDL file (or it can't be fetched due to networking problems), you'll know about it immediately.
	The WSDL proxy class exposes the available functions as a Python dictionary, server.methods. So getting the list of available methods is as simple as calling the dictionary method keys().

Okay, so you know that this SOAP server offers a single method: getTemp. But how do you call it? The WSDL proxy object can tell you that too.

Example 12.9. Discovering A Method's Arguments

>>> callInfo = server.methods['getTemp']  
>>> callInfo.inparams   
[<SOAPpy.wstools.WSDLTools.ParameterInfo instance at 0x00CF3AD0>]
>>> callInfo.inparams[0].name             
u'zipcode'
>>> callInfo.inparams[0].type             
(u'http://www.w3.org/2001/XMLSchema', u'string')

	The server.methods dictionary is filled with a SOAPpy-specific structure called CallInfo. A CallInfo object contains information about one specific function, including the function arguments.
	The function arguments are stored in callInfo.inparams, which is a Python list of ParameterInfo objects that hold information about each parameter.
	Each ParameterInfo object contains a name attribute, which is the argument name. You are not required to know the argument name to call the function through SOAP, but SOAP does support calling functions with named arguments (just like Python), and WSDL.Proxy will correctly handle mapping named arguments to the remote function if you choose to use them.
	Each parameter is also explicitly typed, using datatypes defined in XML Schema. You saw this in the wire trace in the previous section; the XML Schema namespace was part of the “boilerplate” I told you to ignore. For our purposes here, you may continue to ignore it. The zipcode parameter is a string, and if you pass in a Python string to the WSDL.Proxy object, it will map it correctly and send it to the server.

WSDL also lets you introspect into a function's return values.

Example 12.10. Discovering A Method's Return Values

>>> callInfo.outparams            
[<SOAPpy.wstools.WSDLTools.ParameterInfo instance at 0x00CF3AF8>]
>>> callInfo.outparams[0].name    
u'return'
>>> callInfo.outparams[0].type
(u'http://www.w3.org/2001/XMLSchema', u'float')

	The adjunct to callInfo.inparams for function arguments is callInfo.outparams for return value. It is also a list, because functions called through SOAP can return multiple values, just like Python functions.
	Each ParameterInfo object contains name and type. This function returns a single value, named return, which is a float.

Let's put it all together, and call a SOAP web service through a WSDL proxy.

Example 12.11. Calling A Web Service Through A WSDL Proxy

>>> from SOAPpy import WSDL
>>> wsdlFile = 'http://www.xmethods.net/sd/2001/TemperatureService.wsdl')
>>> server = WSDL.Proxy(wsdlFile)               
>>> server.getTemp('90210')   
66.0
>>> server.soapproxy.config.dumpSOAPOut = 1     
>>> server.soapproxy.config.dumpSOAPIn = 1
>>> temperature = server.getTemp('90210')
*** Outgoing SOAP ******************************************************
<?xml version="1.0" encoding="UTF-8"?>
<SOAP-ENV:Envelope SOAP-ENV:encodingStyle="http://schemas.xmlsoap.org/soap/encoding/"
  xmlns:SOAP-ENC="http://schemas.xmlsoap.org/soap/encoding/"
  xmlns:xsi="http://www.w3.org/1999/XMLSchema-instance"
  xmlns:SOAP-ENV="http://schemas.xmlsoap.org/soap/envelope/"
  xmlns:xsd="http://www.w3.org/1999/XMLSchema">
<SOAP-ENV:Body>
<ns1:getTemp xmlns:ns1="urn:xmethods-Temperature" SOAP-ENC:root="1">
<v1 xsi:type="xsd:string">90210</v1>
</ns1:getTemp>
</SOAP-ENV:Body>
</SOAP-ENV:Envelope>
************************************************************************
*** Incoming SOAP ******************************************************
<?xml version='1.0' encoding='UTF-8'?>
<SOAP-ENV:Envelope xmlns:SOAP-ENV="http://schemas.xmlsoap.org/soap/envelope/"
  xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
  xmlns:xsd="http://www.w3.org/2001/XMLSchema">
<SOAP-ENV:Body>
<ns1:getTempResponse xmlns:ns1="urn:xmethods-Temperature"
  SOAP-ENV:encodingStyle="http://schemas.xmlsoap.org/soap/encoding/">
<return xsi:type="xsd:float">66.0</return>
</ns1:getTempResponse>

</SOAP-ENV:Body>
</SOAP-ENV:Envelope>
************************************************************************

>>> temperature
66.0

	The configuration is simpler than calling the SOAP service directly, since the WSDL file contains the both service URL and namespace you need to call the service. Creating the WSDL.Proxy object downloads the WSDL file, parses it, and configures a SOAPProxy object that it uses to call the actual SOAP web service.
	Once the WSDL.Proxy object is created, you can call a function as easily as you did with the SOAPProxy object. This is not surprising; the WSDL.Proxy is just a wrapper around the SOAPProxy with some introspection methods added, so the syntax for calling functions is the same.
	You can access the WSDL.Proxy's SOAPProxy with server.soapproxy. This is useful to turning on debugging, so that when you can call functions through the WSDL proxy, its SOAPProxy will dump the outgoing and incoming XML documents that are going over the wire.

12.7. Searching Google

Let's finally turn to the sample code that you saw that the beginning of this chapter, which does something more useful and exciting than get the current temperature.

Google provides a SOAP API for programmatically accessing Google search results. To use it, you will need to sign up for Google Web Services.

Procedure 12.4. Signing Up for Google Web Services

1. Go to http://www.google.com/apis/ and create a Google account. This requires only an email address. After you sign up you will receive your Google API license key by email. You will need this key to pass as a parameter whenever you call Google's search functions.

2. Also on http://www.google.com/apis/, download the Google Web APIs developer kit. This includes some sample code in several programming languages (but not Python), and more importantly, it includes the WSDL file.

3. Decompress the developer kit file and find GoogleSearch.wsdl. Copy this file to some permanent location on your local drive. You will need it later in this chapter.

Once you have your developer key and your Google WSDL file in a known place, you can start poking around with Google Web Services.

Example 12.12. Introspecting Google Web Services

>>> from SOAPpy import WSDL
>>> server = WSDL.Proxy('/path/to/your/GoogleSearch.wsdl') 
>>> server.methods.keys()                
[u'doGoogleSearch', u'doGetCachedPage', u'doSpellingSuggestion']
>>> callInfo = server.methods['doGoogleSearch']
>>> for arg in callInfo.inparams:        
...     print arg.name.ljust(15), arg.type
key             (u'http://www.w3.org/2001/XMLSchema', u'string')
q               (u'http://www.w3.org/2001/XMLSchema', u'string')
start           (u'http://www.w3.org/2001/XMLSchema', u'int')
maxResults      (u'http://www.w3.org/2001/XMLSchema', u'int')
filter          (u'http://www.w3.org/2001/XMLSchema', u'boolean')
restrict        (u'http://www.w3.org/2001/XMLSchema', u'string')
safeSearch      (u'http://www.w3.org/2001/XMLSchema', u'boolean')
lr              (u'http://www.w3.org/2001/XMLSchema', u'string')
ie              (u'http://www.w3.org/2001/XMLSchema', u'string')
oe              (u'http://www.w3.org/2001/XMLSchema', u'string')

	Getting started with Google web services is easy: just create a WSDL.Proxy object and point it at your local copy of Google's WSDL file.
	According to the WSDL file, Google offers three functions: doGoogleSearch, doGetCachedPage, and doSpellingSuggestion. These do exactly what they sound like: perform a Google search and return the results programmatically, get access to the cached version of a page from the last time Google saw it, and offer spelling suggestions for commonly misspelled search words.
	The doGoogleSearch function takes a number of parameters of various types. Note that while the WSDL file can tell you what the arguments are called and what datatype they are, it can't tell you what they mean or how to use them. It could theoretically tell you the acceptable range of values for each parameter, if only specific values were allowed, but Google's WSDL file is not that detailed. WSDL.Proxy can't work magic; it can only give you the information provided in the WSDL file.

Here is a brief synopsis of all the parameters to the doGoogleSearch function:

• key - Your Google API key, which you received when you signed up for Google web services.
• q - The search word or phrase you're looking for. The syntax is exactly the same as Google's web form, so if you know any advanced search syntax or tricks, they all work here as well.
• start - The index of the result to start on. Like the interactive web version of Google, this function returns 10 results at a time. If you wanted to get the second “page” of results, you would set start to 10.
• maxResults - The number of results to return. Currently capped at 10, although you can specify fewer if you are only interested in a few results and want to save a little bandwidth.
• filter - If True, Google will filter out duplicate pages from the results.
• restrict - Set this to country plus a country code to get results only from a particular country. Example: countryUK to search pages in the United Kingdom. You can also specify linux, mac, or bsd to search a Google-defined set of technical sites, or unclesam to search sites about the United States government.
• safeSearch - If True, Google will filter out porn sites.
• lr (“language restrict”) - Set this to a language code to get results only in a particular language.
• ie and oe (“input encoding” and “output encoding”) - Deprecated, both must be utf-8.

Example 12.13. Searching Google

>>> from SOAPpy import WSDL
>>> server = WSDL.Proxy('/path/to/your/GoogleSearch.wsdl')
>>> key = 'YOUR_GOOGLE_API_KEY'
>>> results = server.doGoogleSearch(key, 'mark', 0, 10, False, "",
...     False, "", "utf-8", "utf-8")             
>>> len(results.resultElements)
10
>>> results.resultElements[0].URL                
'http://diveintomark.org/'
>>> results.resultElements[0].title
'dive into <b>mark</b>'

	After setting up the WSDL.Proxy object, you can call server.doGoogleSearch with all ten parameters. Remember to use your own Google API key that you received when you signed up for Google web services.
	There's a lot of information returned, but let's look at the actual search results first. They're stored in results.resultElements, and you can access them just like a normal Python list.
	Each element in the resultElements is an object that has a URL, title, snippet, and other useful attributes. At this point you can use normal Python introspection techniques like dir(results.resultElements[0]) to see the available attributes. Or you can introspect through the WSDL proxy object and look through the function's outparams. Each technique will give you the same information.

The results object contains more than the actual search results. It also contains information about the search itself, such as how long it took and how many results were found (even though only 10 were returned). The Google web interface shows this information, and you can access it programmatically too.

Example 12.14. Accessing Secondary Information From Google

>>> results.searchTime   
0.224919
>>> results.estimatedTotalResultsCount     
29800000
>>> results.directoryCategories            
[<SOAPpy.Types.structType item at 14367400>:
 {'fullViewableName':
  'Top/Arts/Literature/World_Literature/American/19th_Century/Twain,_Mark',
  'specialEncoding': ''}]
>>> results.directoryCategories[0].fullViewableName
'Top/Arts/Literature/World_Literature/American/19th_Century/Twain,_Mark'

	This search took 0.224919 seconds. That does not include the time spent sending and receiving the actual SOAP XML documents. It's just the time that Google spent processing your request once it received it.
	In total, there were approximately 30 million results. You can access them 10 at a time by changing the start parameter and calling server.doGoogleSearch again.
	For some queries, Google also returns a list of related categories in the Google Directory. You can append these URLs to http://directory.google.com/ to construct the link to the directory category page.

12.8. Troubleshooting SOAP Web Services

Of course, the world of SOAP web services is not all happiness and light. Sometimes things go wrong.

As you've seen throughout this chapter, SOAP involves several layers. There's the HTTP layer, since SOAP is sending XML documents to, and receiving XML documents from, an HTTP server. So all the debugging techniques you learned in Chapter 11, HTTP Web Services come into play here. You can import httplib and then set httplib.HTTPConnection.debuglevel = 1 to see the underlying HTTP traffic.

Beyond the underlying HTTP layer, there are a number of things that can go wrong. SOAPpy does an admirable job hiding the SOAP syntax from you, but that also means it can be difficult to determine where the problem is when things don't work.

Here are a few examples of common mistakes that I've made in using SOAP web services, and the errors they generated.

Example 12.15. Calling a Method With an Incorrectly Configured Proxy

>>> from SOAPpy import SOAPProxy
>>> url = 'http://services.xmethods.net:80/soap/servlet/rpcrouter'
>>> server = SOAPProxy(url)    
>>> server.getTemp('27502')    
<Fault SOAP-ENV:Server.BadTargetObjectURI:
Unable to determine object id from call: is the method element namespaced?>
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
  File "c:\python23\Lib\site-packages\SOAPpy\Client.py", line 453, in __call__
    return self.__r_call(*args, **kw)
  File "c:\python23\Lib\site-packages\SOAPpy\Client.py", line 475, in __r_call
    self.__hd, self.__ma)
  File "c:\python23\Lib\site-packages\SOAPpy\Client.py", line 389, in __call
    raise p
SOAPpy.Types.faultType: <Fault SOAP-ENV:Server.BadTargetObjectURI:
Unable to determine object id from call: is the method element namespaced?>

Did you spot the mistake? You're creating a SOAPProxy manually, and you've correctly specified the service URL, but you haven't specified the namespace. Since multiple services may be routed through the same service URL, the namespace is essential to determine which service you're trying to talk to, and therefore which method you're really calling.

The server responds by sending a SOAP Fault, which SOAPpy turns into a Python exception of type SOAPpy.Types.faultType. All errors returned from any SOAP server will always be SOAP Faults, so you can easily catch this exception. In this case, the human-readable part of the SOAP Fault gives a clue to the problem: the method element is not namespaced, because the original SOAPProxy object was not configured with a service namespace.

Misconfiguring the basic elements of the SOAP service is one of the problems that WSDL aims to solve. The WSDL file contains the service URL and namespace, so you can't get it wrong. Of course, there are still other things you can get wrong.

Example 12.16. Calling a Method With the Wrong Arguments

>>> wsdlFile = 'http://www.xmethods.net/sd/2001/TemperatureService.wsdl'
>>> server = WSDL.Proxy(wsdlFile)
>>> temperature = server.getTemp(27502)              
<Fault SOAP-ENV:Server: Exception while handling service request:
services.temperature.TempService.getTemp(int) -- no signature match>   
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
  File "c:\python23\Lib\site-packages\SOAPpy\Client.py", line 453, in __call__
    return self.__r_call(*args, **kw)
  File "c:\python23\Lib\site-packages\SOAPpy\Client.py", line 475, in __r_call
    self.__hd, self.__ma)
  File "c:\python23\Lib\site-packages\SOAPpy\Client.py", line 389, in __call
    raise p
SOAPpy.Types.faultType: <Fault SOAP-ENV:Server: Exception while handling service request:
services.temperature.TempService.getTemp(int) -- no signature match>

Did you spot the mistake? It's a subtle one: you're calling server.getTemp with an integer instead of a string. As you saw from introspecting the WSDL file, the getTemp() SOAP function takes a single argument, zipcode, which must be a string. WSDL.Proxy will not coerce datatypes for you; you need to pass the exact datatypes that the server expects.

Again, the server returns a SOAP Fault, and the human-readable part of the error gives a clue as to the problem: you're calling a getTemp function with an integer value, but there is no function defined with that name that takes an integer. In theory, SOAP allows you to overload functions, so you could have two functions in the same SOAP service with the same name and the same number of arguments, but the arguments were of different datatypes. This is why it's important to match the datatypes exactly, and why WSDL.Proxy doesn't coerce datatypes for you. If it did, you could end up calling a completely different function! Good luck debugging that one. It's much easier to be picky about datatypes and fail as quickly as possible if you get them wrong.

It's also possible to write Python code that expects a different number of return values than the remote function actually returns.

Example 12.17. Calling a Method and Expecting the Wrong Number of Return Values

>>> wsdlFile = 'http://www.xmethods.net/sd/2001/TemperatureService.wsdl'
>>> server = WSDL.Proxy(wsdlFile)
>>> (city, temperature) = server.getTemp(27502)  
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
TypeError: unpack non-sequence

Did you spot the mistake? server.getTemp only returns one value, a float, but you've written code that assumes you're getting two values and trying to assign them to two different variables. Note that this does not fail with a SOAP fault. As far as the remote server is concerned, nothing went wrong at all. The error only occurred after the SOAP transaction was complete, WSDL.Proxy returned a float, and your local Python interpreter tried to accomodate your request to split it into two different variables. Since the function only returned one value, you get a Python exception trying to split it, not a SOAP Fault.

What about Google's web service? The most common problem I've had with it is that I forget to set the application key properly.

Example 12.18. Calling a Method With An Application-Specific Error

>>> from SOAPpy import WSDL
>>> server = WSDL.Proxy(r'/path/to/local/GoogleSearch.wsdl')
>>> results = server.doGoogleSearch('foo', 'mark', 0, 10, False, "", 
...     False, "", "utf-8", "utf-8")
<Fault SOAP-ENV:Server:          
 Exception from service object: Invalid authorization key: foo:
 <SOAPpy.Types.structType detail at 14164616>:
 {'stackTrace':
  'com.google.soap.search.GoogleSearchFault: Invalid authorization key: foo
   at com.google.soap.search.QueryLimits.lookUpAndLoadFromINSIfNeedBe(
     QueryLimits.java:220)
   at com.google.soap.search.QueryLimits.validateKey(QueryLimits.java:127)
   at com.google.soap.search.GoogleSearchService.doPublicMethodChecks(
     GoogleSearchService.java:825)
   at com.google.soap.search.GoogleSearchService.doGoogleSearch(
     GoogleSearchService.java:121)
   at sun.reflect.GeneratedMethodAccessor13.invoke(Unknown Source)
   at sun.reflect.DelegatingMethodAccessorImpl.invoke(Unknown Source)
   at java.lang.reflect.Method.invoke(Unknown Source)
   at org.apache.soap.server.RPCRouter.invoke(RPCRouter.java:146)
   at org.apache.soap.providers.RPCJavaProvider.invoke(
     RPCJavaProvider.java:129)
   at org.apache.soap.server.http.RPCRouterServlet.doPost(
     RPCRouterServlet.java:288)
   at javax.servlet.http.HttpServlet.service(HttpServlet.java:760)
   at javax.servlet.http.HttpServlet.service(HttpServlet.java:853)
   at com.google.gse.HttpConnection.runServlet(HttpConnection.java:237)
   at com.google.gse.HttpConnection.run(HttpConnection.java:195)
   at com.google.gse.DispatchQueue$WorkerThread.run(DispatchQueue.java:201)
Caused by: com.google.soap.search.UserKeyInvalidException: Key was of wrong size.
   at com.google.soap.search.UserKey.<init>(UserKey.java:59)
   at com.google.soap.search.QueryLimits.lookUpAndLoadFromINSIfNeedBe(
     QueryLimits.java:217)
   ... 14 more
'}>
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
  File "c:\python23\Lib\site-packages\SOAPpy\Client.py", line 453, in __call__
    return self.__r_call(*args, **kw)
  File "c:\python23\Lib\site-packages\SOAPpy\Client.py", line 475, in __r_call
    self.__hd, self.__ma)
  File "c:\python23\Lib\site-packages\SOAPpy\Client.py", line 389, in __call
    raise p
SOAPpy.Types.faultType: <Fault SOAP-ENV:Server: Exception from service object:
Invalid authorization key: foo:
<SOAPpy.Types.structType detail at 14164616>:
{'stackTrace':
  'com.google.soap.search.GoogleSearchFault: Invalid authorization key: foo
   at com.google.soap.search.QueryLimits.lookUpAndLoadFromINSIfNeedBe(
     QueryLimits.java:220)
   at com.google.soap.search.QueryLimits.validateKey(QueryLimits.java:127)
   at com.google.soap.search.GoogleSearchService.doPublicMethodChecks(
     GoogleSearchService.java:825)
   at com.google.soap.search.GoogleSearchService.doGoogleSearch(
     GoogleSearchService.java:121)
   at sun.reflect.GeneratedMethodAccessor13.invoke(Unknown Source)
   at sun.reflect.DelegatingMethodAccessorImpl.invoke(Unknown Source)
   at java.lang.reflect.Method.invoke(Unknown Source)
   at org.apache.soap.server.RPCRouter.invoke(RPCRouter.java:146)
   at org.apache.soap.providers.RPCJavaProvider.invoke(
     RPCJavaProvider.java:129)
   at org.apache.soap.server.http.RPCRouterServlet.doPost(
     RPCRouterServlet.java:288)
   at javax.servlet.http.HttpServlet.service(HttpServlet.java:760)
   at javax.servlet.http.HttpServlet.service(HttpServlet.java:853)
   at com.google.gse.HttpConnection.runServlet(HttpConnection.java:237)
   at com.google.gse.HttpConnection.run(HttpConnection.java:195)
   at com.google.gse.DispatchQueue$WorkerThread.run(DispatchQueue.java:201)
Caused by: com.google.soap.search.UserKeyInvalidException: Key was of wrong size.
   at com.google.soap.search.UserKey.<init>(UserKey.java:59)
   at com.google.soap.search.QueryLimits.lookUpAndLoadFromINSIfNeedBe(
     QueryLimits.java:217)
   ... 14 more
'}>

	Can you spot the mistake? There's nothing wrong with the calling syntax, or the number of arguments, or the datatypes. The problem is application-specific: the first argument is supposed to be my application key, but foo is not a valid Google key.
	The Google server responds with a SOAP Fault and an incredibly long error message, which includes a complete Java stack trace. Remember that all SOAP errors are signified by SOAP Faults: errors in configuration, errors in function arguments, and application-specific errors like this. Buried in there somewhere is the crucial piece of information: Invalid authorization key: foo.

Further Reading on Troubleshooting SOAP

• New developments for SOAPpy steps through trying to connect to another SOAP service that doesn't quite work as advertised.

12.9. Summary

SOAP web services are very complicated. The specification is very ambitious and tries to cover many different use cases for web services. This chapter has touched on some of the simpler use cases.

Before diving into the next chapter, make sure you're comfortable doing all of these things:

• Connecting to a SOAP server and calling remote methods
• Loading a WSDL file and introspecting remote methods
• Debugging SOAP calls with wire traces
• Troubleshooting common SOAP-related errors

Chapter 13. Unit Testing

13.1. Introduction to Roman numerals

In previous chapters, you “dived in” by immediately looking at code and trying to understand it as quickly as possible. Now that you have some Python under your belt, you're going to step back and look at the steps that happen before the code gets written.

In the next few chapters, you're going to write, debug, and optimize a set of utility functions to convert to and from Roman numerals. You saw the mechanics of constructing and validating Roman numerals in Section 7.3, “Case Study: Roman Numerals”, but now let's step back and consider what it would take to expand that into a two-way utility.

The rules for Roman numerals lead to a number of interesting observations:

1. There is only one correct way to represent a particular number as Roman numerals.
2. The converse is also true: if a string of characters is a valid Roman numeral, it represents only one number (i.e. it can only be read one way).
3. There is a limited range of numbers that can be expressed as Roman numerals, specifically 1 through 3999. (The Romans did have several ways of expressing larger numbers, for instance by having a bar over a numeral to represent that its normal value should be multiplied by 1000, but you're not going to deal with that. For the purposes of this chapter, let's stipulate that Roman numerals go from 1 to 3999.)
4. There is no way to represent 0 in Roman numerals. (Amazingly, the ancient Romans had no concept of 0 as a number. Numbers were for counting things you had; how can you count what you don't have?)
5. There is no way to represent negative numbers in Roman numerals.
6. There is no way to represent fractions or non-integer numbers in Roman numerals.

Given all of this, what would you expect out of a set of functions to convert to and from Roman numerals?

roman.py requirements

1. toRoman should return the Roman numeral representation for all integers 1 to 3999.
2. toRoman should fail when given an integer outside the range 1 to 3999.
3. toRoman should fail when given a non-integer number.
4. fromRoman should take a valid Roman numeral and return the number that it represents.
5. fromRoman should fail when given an invalid Roman numeral.
6. If you take a number, convert it to Roman numerals, then convert that back to a number, you should end up with the number you started with. So fromRoman(toRoman(n)) == n for all n in 1..3999.
7. toRoman should always return a Roman numeral using uppercase letters.
8. fromRoman should only accept uppercase Roman numerals (i.e. it should fail when given lowercase input).

Further reading

• This site has more on Roman numerals, including a fascinating history of how Romans and other civilizations really used them (short answer: haphazardly and inconsistently).

13.2. Diving in

Now that you've completely defined the behavior you expect from your conversion functions, you're going to do something a little unexpected: you're going to write a test suite that puts these functions through their paces and makes sure that they behave the way you want them to. You read that right: you're going to write code that tests code that you haven't written yet.

This is called unit testing, since the set of two conversion functions can be written and tested as a unit, separate from any larger program they may become part of later. Python has a framework for unit testing, the appropriately-named unittest module.

	unittest is included with Python 2.1 and later. Python 2.0 users can download it from pyunit.sourceforge.net.

Unit testing is an important part of an overall testing-centric development strategy. If you write unit tests, it is important to write them early (preferably before writing the code that they test), and to keep them updated as code and requirements change. Unit testing is not a replacement for higher-level functional or system testing, but it is important in all phases of development:

• Before writing code, it forces you to detail your requirements in a useful fashion.
• While writing code, it keeps you from over-coding. When all the test cases pass, the function is complete.
• When refactoring code, it assures you that the new version behaves the same way as the old version.
• When maintaining code, it helps you cover your ass when someone comes screaming that your latest change broke their old code. (“But sir, all the unit tests passed when I checked it in...”)
• When writing code in a team, it increases confidence that the code you're about to commit isn't going to break other peoples' code, because you can run their unittests first. (I've seen this sort of thing in code sprints. A team breaks up the assignment, everybody takes the specs for their task, writes unit tests for it, then shares their unit tests with the rest of the team. That way, nobody goes off too far into developing code that won't play well with others.)

13.3. Introducing romantest.py

This is the complete test suite for your Roman numeral conversion functions, which are yet to be written but will eventually be in roman.py. It is not immediately obvious how it all fits together; none of these classes or methods reference any of the others. There are good reasons for this, as you'll see shortly.

Example 13.1. romantest.py

If you have not already done so, you can download this and other examples used in this book.

"""Unit test for roman.py"""

import roman
import unittest

class KnownValues(unittest.TestCase):        
    knownValues = ( (1, 'I'),
  (2, 'II'),
  (3, 'III'),
  (4, 'IV'),
  (5, 'V'),
  (6, 'VI'),
  (7, 'VII'),
  (8, 'VIII'),
  (9, 'IX'),
  (10, 'X'),
  (50, 'L'),
  (100, 'C'),
  (500, 'D'),
  (1000, 'M'),
  (31, 'XXXI'),
  (148, 'CXLVIII'),
  (294, 'CCXCIV'),
  (312, 'CCCXII'),
  (421, 'CDXXI'),
  (528, 'DXXVIII'),
  (621, 'DCXXI'),
  (782, 'DCCLXXXII'),
  (870, 'DCCCLXX'),
  (941, 'CMXLI'),
  (1043, 'MXLIII'),
  (1110, 'MCX'),
  (1226, 'MCCXXVI'),
  (1301, 'MCCCI'),
  (1485, 'MCDLXXXV'),
  (1509, 'MDIX'),
  (1607, 'MDCVII'),
  (1754, 'MDCCLIV'),
  (1832, 'MDCCCXXXII'),
  (1993, 'MCMXCIII'),
  (2074, 'MMLXXIV'),
  (2152, 'MMCLII'),
  (2212, 'MMCCXII'),
  (2343, 'MMCCCXLIII'),
  (2499, 'MMCDXCIX'),
  (2574, 'MMDLXXIV'),
  (2646, 'MMDCXLVI'),
  (2723, 'MMDCCXXIII'),
  (2892, 'MMDCCCXCII'),
  (2975, 'MMCMLXXV'),
  (3051, 'MMMLI'),
  (3185, 'MMMCLXXXV'),
  (3250, 'MMMCCL'),
  (3313, 'MMMCCCXIII'),
  (3408, 'MMMCDVIII'),
  (3501, 'MMMDI'),
  (3610, 'MMMDCX'),
  (3743, 'MMMDCCXLIII'),
  (3844, 'MMMDCCCXLIV'),
  (3888, 'MMMDCCCLXXXVIII'),
  (3940, 'MMMCMXL'),
  (3999, 'MMMCMXCIX'))     

    def testToRomanKnownValues(self):        
        """toRoman should give known result with known input"""
        for integer, numeral in self.knownValues:              
            result = roman.toRoman(integer)  
            self.assertEqual(numeral, result)

    def testFromRomanKnownValues(self):        
        """fromRoman should give known result with known input"""
        for integer, numeral in self.knownValues:                
            result = roman.fromRoman(numeral)  
            self.assertEqual(integer, result)  

class ToRomanBadInput(unittest.TestCase):          
    def testTooLarge(self):      
        """toRoman should fail with large input""" 
        self.assertRaises(roman.OutOfRangeError, roman.toRoman, 4000)

    def testZero(self):          
        """toRoman should fail with 0 input"""     
        self.assertRaises(roman.OutOfRangeError, roman.toRoman, 0)   

    def testNegative(self):      
        """toRoman should fail with negative input"""                
        self.assertRaises(roman.OutOfRangeError, roman.toRoman, -1)  

    def testNonInteger(self):    
        """toRoman should fail with non-integer input"""             
        self.assertRaises(roman.NotIntegerError, roman.toRoman, 0.5) 

class FromRomanBadInput(unittest.TestCase):  
    def testTooManyRepeatedNumerals(self):   
        """fromRoman should fail with too many repeated numerals"""              
        for s in ('MMMM', 'DD', 'CCCC', 'LL', 'XXXX', 'VV', 'IIII'):             
            self.assertRaises(roman.InvalidRomanNumeralError, roman.fromRoman, s)

    def testRepeatedPairs(self):             
        """fromRoman should fail with repeated pairs of numerals"""              
        for s in ('CMCM', 'CDCD', 'XCXC', 'XLXL', 'IXIX', 'IVIV'):               
            self.assertRaises(roman.InvalidRomanNumeralError, roman.fromRoman, s)

    def testMalformedAntecedent(self):       
        """fromRoman should fail with malformed antecedents""" 
        for s in ('IIMXCC', 'VX', 'DCM', 'CMM', 'IXIV',
'MCMC', 'XCX', 'IVI', 'LM', 'LD', 'LC'):     
            self.assertRaises(roman.InvalidRomanNumeralError, roman.fromRoman, s)

class SanityCheck(unittest.TestCase):        
    def testSanity(self):  
        """fromRoman(toRoman(n))==n for all n"""
        for integer in range(1, 4000):       
            numeral = roman.toRoman(integer) 
            result = roman.fromRoman(numeral)
            self.assertEqual(integer, result)

class CaseCheck(unittest.TestCase): 
    def testToRomanCase(self):      
        """toRoman should always return uppercase"""  
        for integer in range(1, 4000):                
            numeral = roman.toRoman(integer)          
            self.assertEqual(numeral, numeral.upper())

    def testFromRomanCase(self):    
        """fromRoman should only accept uppercase input"""
        for integer in range(1, 4000):                
            numeral = roman.toRoman(integer)          
            roman.fromRoman(numeral.upper())          
            self.assertRaises(roman.InvalidRomanNumeralError,
            roman.fromRoman, numeral.lower())

if __name__ == "__main__":
    unittest.main()   

Further reading

• The PyUnit home page has an in-depth discussion of using the unittest framework, including advanced features not covered in this chapter.
• The PyUnit FAQ explains why test cases are stored separately from the code they test.
• Python Library Reference summarizes the unittest module.
• ExtremeProgramming.org discusses why you should write unit tests.
• The Portland Pattern Repository has an ongoing discussion of unit tests, including a standard definition, why you should code unit tests first, and several in-depth case studies.

13.4. Testing for success

The most fundamental part of unit testing is constructing individual test cases. A test case answers a single question about the code it is testing.

A test case should be able to...

• ...run completely by itself, without any human input. Unit testing is about automation.
• ...determine by itself whether the function it is testing has passed or failed, without a human interpreting the results.
• ...run in isolation, separate from any other test cases (even if they test the same functions). Each test case is an island.

Given that, let's build the first test case. You have the following requirement:

1. toRoman should return the Roman numeral representation for all integers 1 to 3999.

Example 13.2. testToRomanKnownValues

class KnownValues(unittest.TestCase):         
    knownValues = ( (1, 'I'),
  (2, 'II'),
  (3, 'III'),
  (4, 'IV'),
  (5, 'V'),
  (6, 'VI'),
  (7, 'VII'),
  (8, 'VIII'),
  (9, 'IX'),
  (10, 'X'),
  (50, 'L'),
  (100, 'C'),
  (500, 'D'),
  (1000, 'M'),
  (31, 'XXXI'),
  (148, 'CXLVIII'),
  (294, 'CCXCIV'),
  (312, 'CCCXII'),
  (421, 'CDXXI'),
  (528, 'DXXVIII'),
  (621, 'DCXXI'),
  (782, 'DCCLXXXII'),
  (870, 'DCCCLXX'),
  (941, 'CMXLI'),
  (1043, 'MXLIII'),
  (1110, 'MCX'),
  (1226, 'MCCXXVI'),
  (1301, 'MCCCI'),
  (1485, 'MCDLXXXV'),
  (1509, 'MDIX'),
  (1607, 'MDCVII'),
  (1754, 'MDCCLIV'),
  (1832, 'MDCCCXXXII'),
  (1993, 'MCMXCIII'),
  (2074, 'MMLXXIV'),
  (2152, 'MMCLII'),
  (2212, 'MMCCXII'),
  (2343, 'MMCCCXLIII'),
  (2499, 'MMCDXCIX'),
  (2574, 'MMDLXXIV'),
  (2646, 'MMDCXLVI'),
  (2723, 'MMDCCXXIII'),
  (2892, 'MMDCCCXCII'),
  (2975, 'MMCMLXXV'),
  (3051, 'MMMLI'),
  (3185, 'MMMCLXXXV'),
  (3250, 'MMMCCL'),
  (3313, 'MMMCCCXIII'),
  (3408, 'MMMCDVIII'),
  (3501, 'MMMDI'),
  (3610, 'MMMDCX'),
  (3743, 'MMMDCCXLIII'),
  (3844, 'MMMDCCCXLIV'),
  (3888, 'MMMDCCCLXXXVIII'),
  (3940, 'MMMCMXL'),
  (3999, 'MMMCMXCIX'))      

    def testToRomanKnownValues(self):         
        """toRoman should give known result with known input"""
        for integer, numeral in self.knownValues:              
            result = roman.toRoman(integer)    
            self.assertEqual(numeral, result) 

	To write a test case, first subclass the TestCase class of the unittest module. This class provides many useful methods which you can use in your test case to test specific conditions.
	This is a list of integer/numeral pairs that I verified manually. It includes the lowest ten numbers, the highest number, every number that translates to a single-character Roman numeral, and a random sampling of other valid numbers. The point of a unit test is not to test every possible input, but to test a representative sample.
	Every individual test is its own method, which must take no parameters and return no value. If the method exits normally without raising an exception, the test is considered passed; if the method raises an exception, the test is considered failed.
	Here you call the actual toRoman function. (Well, the function hasn't be written yet, but once it is, this is the line that will call it.) Notice that you have now defined the API for the toRoman function: it must take an integer (the number to convert) and return a string (the Roman numeral representation). If the API is different than that, this test is considered failed.
	Also notice that you are not trapping any exceptions when you call toRoman. This is intentional. toRoman shouldn't raise an exception when you call it with valid input, and these input values are all valid. If toRoman raises an exception, this test is considered failed.
	Assuming the toRoman function was defined correctly, called correctly, completed successfully, and returned a value, the last step is to check whether it returned the right value. This is a common question, and the TestCase class provides a method, assertEqual, to check whether two values are equal. If the result returned from toRoman (result) does not match the known value you were expecting (numeral), assertEqual will raise an exception and the test will fail. If the two values are equal, assertEqual will do nothing. If every value returned from toRoman matches the known value you expect, assertEqual never raises an exception, so testToRomanKnownValues eventually exits normally, which means toRoman has passed this test.

13.5. Testing for failure

It is not enough to test that functions succeed when given good input; you must also test that they fail when given bad input. And not just any sort of failure; they must fail in the way you expect.

Remember the other requirements for toRoman:

2. toRoman should fail when given an integer outside the range 1 to 3999.
3. toRoman should fail when given a non-integer number.

In Python, functions indicate failure by raising exceptions, and the unittest module provides methods for testing whether a function raises a particular exception when given bad input.

Example 13.3. Testing bad input to toRoman

class ToRomanBadInput(unittest.TestCase):          
    def testTooLarge(self):      
        """toRoman should fail with large input""" 
        self.assertRaises(roman.OutOfRangeError, roman.toRoman, 4000) 

    def testZero(self):          
        """toRoman should fail with 0 input"""     
        self.assertRaises(roman.OutOfRangeError, roman.toRoman, 0)    

    def testNegative(self):      
        """toRoman should fail with negative input"""                
        self.assertRaises(roman.OutOfRangeError, roman.toRoman, -1)  

    def testNonInteger(self):    
        """toRoman should fail with non-integer input"""             
        self.assertRaises(roman.NotIntegerError, roman.toRoman, 0.5)  

	The TestCase class of the unittest provides the assertRaises method, which takes the following arguments: the exception you're expecting, the function you're testing, and the arguments you're passing that function. (If the function you're testing takes more than one argument, pass them all to assertRaises, in order, and it will pass them right along to the function you're testing.) Pay close attention to what you're doing here: instead of calling toRoman directly and manually checking that it raises a particular exception (by wrapping it in a try...except block), assertRaises has encapsulated all of that for us. All you do is give it the exception (roman.OutOfRangeError), the function (toRoman), and toRoman's arguments (4000), and assertRaises takes care of calling toRoman and checking to make sure that it raises roman.OutOfRangeError. (Also note that you're passing the toRoman function itself as an argument; you're not calling it, and you're not passing the name of it as a string. Have I mentioned recently how handy it is that everything in Python is an object, including functions and exceptions?)
	Along with testing numbers that are too large, you need to test numbers that are too small. Remember, Roman numerals cannot express 0 or negative numbers, so you have a test case for each of those (testZero and testNegative). In testZero, you are testing that toRoman raises a roman.OutOfRangeError exception when called with 0; if it does not raise a roman.OutOfRangeError (either because it returns an actual value, or because it raises some other exception), this test is considered failed.
	Requirement #3 specifies that toRoman cannot accept a non-integer number, so here you test to make sure that toRoman raises a roman.NotIntegerError exception when called with 0.5. If toRoman does not raise a roman.NotIntegerError, this test is considered failed.

The next two requirements are similar to the first three, except they apply to fromRoman instead of toRoman:

4. fromRoman should take a valid Roman numeral and return the number that it represents.
5. fromRoman should fail when given an invalid Roman numeral.

Requirement #4 is handled in the same way as requirement #1, iterating through a sampling of known values and testing each in turn. Requirement #5 is handled in the same way as requirements #2 and #3, by testing a series of bad inputs and making sure fromRoman raises the appropriate exception.

Example 13.4. Testing bad input to fromRoman

class FromRomanBadInput(unittest.TestCase):  
    def testTooManyRepeatedNumerals(self):   
        """fromRoman should fail with too many repeated numerals"""              
        for s in ('MMMM', 'DD', 'CCCC', 'LL', 'XXXX', 'VV', 'IIII'):             
            self.assertRaises(roman.InvalidRomanNumeralError, roman.fromRoman, s) 

    def testRepeatedPairs(self):             
        """fromRoman should fail with repeated pairs of numerals"""              
        for s in ('CMCM', 'CDCD', 'XCXC', 'XLXL', 'IXIX', 'IVIV'):               
            self.assertRaises(roman.InvalidRomanNumeralError, roman.fromRoman, s)

    def testMalformedAntecedent(self):       
        """fromRoman should fail with malformed antecedents""" 
        for s in ('IIMXCC', 'VX', 'DCM', 'CMM', 'IXIV',
'MCMC', 'XCX', 'IVI', 'LM', 'LD', 'LC'):     
            self.assertRaises(roman.InvalidRomanNumeralError, roman.fromRoman, s)

Not much new to say about these; the pattern is exactly the same as the one you used to test bad input to toRoman. I will briefly note that you have another exception: roman.InvalidRomanNumeralError. That makes a total of three custom exceptions that will need to be defined in roman.py (along with roman.OutOfRangeError and roman.NotIntegerError). You'll see how to define these custom exceptions when you actually start writing roman.py, later in this chapter.

13.6. Testing for sanity

Often, you will find that a unit of code contains a set of reciprocal functions, usually in the form of conversion functions where one converts A to B and the other converts B to A. In these cases, it is useful to create a “sanity check” to make sure that you can convert A to B and back to A without losing precision, incurring rounding errors, or triggering any other sort of bug.

Consider this requirement:

6. If you take a number, convert it to Roman numerals, then convert that back to a number, you should end up with the number you started with. So fromRoman(toRoman(n)) == n for all n in 1..3999.

Example 13.5. Testing toRoman against fromRoman

class SanityCheck(unittest.TestCase):        
    def testSanity(self):  
        """fromRoman(toRoman(n))==n for all n"""
        for integer in range(1, 4000):         
            numeral = roman.toRoman(integer) 
            result = roman.fromRoman(numeral)
            self.assertEqual(integer, result) 

	You've seen the range function before, but here it is called with two arguments, which returns a list of integers starting at the first argument (1) and counting consecutively up to but not including the second argument (4000). Thus, 1..3999, which is the valid range for converting to Roman numerals.
	I just wanted to mention in passing that integer is not a keyword in Python; here it's just a variable name like any other.
	The actual testing logic here is straightforward: take a number (integer), convert it to a Roman numeral (numeral), then convert it back to a number (result) and make sure you end up with the same number you started with. If not, assertEqual will raise an exception and the test will immediately be considered failed. If all the numbers match, assertEqual will always return silently, the entire testSanity method will eventually return silently, and the test will be considered passed.

The last two requirements are different from the others because they seem both arbitrary and trivial:

7. toRoman should always return a Roman numeral using uppercase letters.
8. fromRoman should only accept uppercase Roman numerals (i.e. it should fail when given lowercase input).

In fact, they are somewhat arbitrary. You could, for instance, have stipulated that fromRoman accept lowercase and mixed case input. But they are not completely arbitrary; if toRoman is always returning uppercase output, then fromRoman must at least accept uppercase input, or the “sanity check” (requirement #6) would fail. The fact that it only accepts uppercase input is arbitrary, but as any systems integrator will tell you, case always matters, so it's worth specifying the behavior up front. And if it's worth specifying, it's worth testing.

Example 13.6. Testing for case

class CaseCheck(unittest.TestCase): 
    def testToRomanCase(self):      
        """toRoman should always return uppercase"""  
        for integer in range(1, 4000):                
            numeral = roman.toRoman(integer)          
            self.assertEqual(numeral, numeral.upper())         

    def testFromRomanCase(self):    
        """fromRoman should only accept uppercase input"""
        for integer in range(1, 4000):                
            numeral = roman.toRoman(integer)          
            roman.fromRoman(numeral.upper())  
            self.assertRaises(roman.InvalidRomanNumeralError,
            roman.fromRoman, numeral.lower())   

	The most interesting thing about this test case is all the things it doesn't test. It doesn't test that the value returned from toRoman is right or even consistent; those questions are answered by separate test cases. You have a whole test case just to test for uppercase-ness. You might be tempted to combine this with the sanity check, since both run through the entire range of values and call toRoman.[6] But that would violate one of the fundamental rules: each test case should answer only a single question. Imagine that you combined this case check with the sanity check, and then that test case failed. You would need to do further analysis to figure out which part of the test case failed to determine what the problem was. If you need to analyze the results of your unit testing just to figure out what they mean, it's a sure sign that you've mis-designed your test cases.
	There's a similar lesson to be learned here: even though “you know” that toRoman always returns uppercase, you are explicitly converting its return value to uppercase here to test that fromRoman accepts uppercase input. Why? Because the fact that toRoman always returns uppercase is an independent requirement. If you changed that requirement so that, for instance, it always returned lowercase, the testToRomanCase test case would need to change, but this test case would still work. This was another of the fundamental rules: each test case must be able to work in isolation from any of the others. Every test case is an island.
	Note that you're not assigning the return value of fromRoman to anything. This is legal syntax in Python; if a function returns a value but nobody's listening, Python just throws away the return value. In this case, that's what you want. This test case doesn't test anything about the return value; it just tests that fromRoman accepts the uppercase input without raising an exception.
	This is a complicated line, but it's very similar to what you did in the ToRomanBadInput and FromRomanBadInput tests. You are testing to make sure that calling a particular function (roman.fromRoman) with a particular value (numeral.lower(), the lowercase version of the current Roman numeral in the loop) raises a particular exception (roman.InvalidRomanNumeralError). If it does (each time through the loop), the test passes; if even one time it does something else (like raises a different exception, or returning a value without raising an exception at all), the test fails.

In the next chapter, you'll see how to write code that passes these tests.

---

^[6] “I can resist everything except temptation.” --Oscar Wilde

Chapter 14. Test-First Programming

14.1. roman.py, stage 1

Now that the unit tests are complete, it's time to start writing the code that the test cases are attempting to test. You're going to do this in stages, so you can see all the unit tests fail, then watch them pass one by one as you fill in the gaps in roman.py.

Example 14.1. roman1.py

This file is available in py/roman/stage1/ in the examples directory.

If you have not already done so, you can download this and other examples used in this book.

"""Convert to and from Roman numerals"""

#Define exceptions
class RomanError(Exception): pass                
class OutOfRangeError(RomanError): pass          
class NotIntegerError(RomanError): pass
class InvalidRomanNumeralError(RomanError): pass 

def toRoman(n):
    """convert integer to Roman numeral"""
    pass     

def fromRoman(s):
    """convert Roman numeral to integer"""
    pass

	This is how you define your own custom exceptions in Python. Exceptions are classes, and you create your own by subclassing existing exceptions. It is strongly recommended (but not required) that you subclass Exception, which is the base class that all built-in exceptions inherit from. Here I am defining RomanError (inherited from Exception) to act as the base class for all my other custom exceptions to follow. This is a matter of style; I could just as easily have inherited each individual exception from the Exception class directly.
	The OutOfRangeError and NotIntegerError exceptions will eventually be used by toRoman to flag various forms of invalid input, as specified in ToRomanBadInput.
	The InvalidRomanNumeralError exception will eventually be used by fromRoman to flag invalid input, as specified in FromRomanBadInput.
	At this stage, you want to define the API of each of your functions, but you don't want to code them yet, so you stub them out using the Python reserved word pass.

Now for the big moment (drum roll please): you're finally going to run the unit test against this stubby little module. At this point, every test case should fail. In fact, if any test case passes in stage 1, you should go back to romantest.py and re-evaluate why you coded a test so useless that it passes with do-nothing functions.

Run romantest1.py with the -v command-line option, which will give more verbose output so you can see exactly what's going on as each test case runs. With any luck, your output should look like this:

Example 14.2. Output of romantest1.py against roman1.py

fromRoman should only accept uppercase input ... ERROR
toRoman should always return uppercase ... ERROR
fromRoman should fail with malformed antecedents ... FAIL
fromRoman should fail with repeated pairs of numerals ... FAIL
fromRoman should fail with too many repeated numerals ... FAIL
fromRoman should give known result with known input ... FAIL
toRoman should give known result with known input ... FAIL
fromRoman(toRoman(n))==n for all n ... FAIL
toRoman should fail with non-integer input ... FAIL
toRoman should fail with negative input ... FAIL
toRoman should fail with large input ... FAIL
toRoman should fail with 0 input ... FAIL

======================================================================
ERROR: fromRoman should only accept uppercase input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 154, in testFromRomanCase
    roman1.fromRoman(numeral.upper())
AttributeError: 'None' object has no attribute 'upper'
======================================================================
ERROR: toRoman should always return uppercase
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 148, in testToRomanCase
    self.assertEqual(numeral, numeral.upper())
AttributeError: 'None' object has no attribute 'upper'
======================================================================
FAIL: fromRoman should fail with malformed antecedents
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 133, in testMalformedAntecedent
    self.assertRaises(roman1.InvalidRomanNumeralError, roman1.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should fail with repeated pairs of numerals
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 127, in testRepeatedPairs
    self.assertRaises(roman1.InvalidRomanNumeralError, roman1.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should fail with too many repeated numerals
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 122, in testTooManyRepeatedNumerals
    self.assertRaises(roman1.InvalidRomanNumeralError, roman1.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should give known result with known input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 99, in testFromRomanKnownValues
    self.assertEqual(integer, result)
  File "c:\python21\lib\unittest.py", line 273, in failUnlessEqual
    raise self.failureException, (msg or '%s != %s' % (first, second))
AssertionError: 1 != None
======================================================================
FAIL: toRoman should give known result with known input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 93, in testToRomanKnownValues
    self.assertEqual(numeral, result)
  File "c:\python21\lib\unittest.py", line 273, in failUnlessEqual
    raise self.failureException, (msg or '%s != %s' % (first, second))
AssertionError: I != None
======================================================================
FAIL: fromRoman(toRoman(n))==n for all n
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 141, in testSanity
    self.assertEqual(integer, result)
  File "c:\python21\lib\unittest.py", line 273, in failUnlessEqual
    raise self.failureException, (msg or '%s != %s' % (first, second))
AssertionError: 1 != None
======================================================================
FAIL: toRoman should fail with non-integer input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 116, in testNonInteger
    self.assertRaises(roman1.NotIntegerError, roman1.toRoman, 0.5)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: NotIntegerError
======================================================================
FAIL: toRoman should fail with negative input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 112, in testNegative
    self.assertRaises(roman1.OutOfRangeError, roman1.toRoman, -1)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: OutOfRangeError
======================================================================
FAIL: toRoman should fail with large input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 104, in testTooLarge
    self.assertRaises(roman1.OutOfRangeError, roman1.toRoman, 4000)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: OutOfRangeError
======================================================================
FAIL: toRoman should fail with 0 input               
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage1\romantest1.py", line 108, in testZero
    self.assertRaises(roman1.OutOfRangeError, roman1.toRoman, 0)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: OutOfRangeError    
----------------------------------------------------------------------
Ran 12 tests in 0.040s             

FAILED (failures=10, errors=2)     

	Running the script runs unittest.main(), which runs each test case, which is to say each method defined in each class within romantest.py. For each test case, it prints out the docstring of the method and whether that test passed or failed. As expected, none of the test cases passed.
	For each failed test case, unittest displays the trace information showing exactly what happened. In this case, the call to assertRaises (also called failUnlessRaises) raised an AssertionError because it was expecting toRoman to raise an OutOfRangeError and it didn't.
	After the detail, unittest displays a summary of how many tests were performed and how long it took.
	Overall, the unit test failed because at least one test case did not pass. When a test case doesn't pass, unittest distinguishes between failures and errors. A failure is a call to an assertXYZ method, like assertEqual or assertRaises, that fails because the asserted condition is not true or the expected exception was not raised. An error is any other sort of exception raised in the code you're testing or the unit test case itself. For instance, the testFromRomanCase method (“fromRoman should only accept uppercase input”) was an error, because the call to numeral.upper() raised an AttributeError exception, because toRoman was supposed to return a string but didn't. But testZero (“toRoman should fail with 0 input”) was a failure, because the call to fromRoman did not raise the InvalidRomanNumeral exception that assertRaises was looking for.

14.2. roman.py, stage 2

Now that you have the framework of the roman module laid out, it's time to start writing code and passing test cases.

Example 14.3. roman2.py

This file is available in py/roman/stage2/ in the examples directory.

If you have not already done so, you can download this and other examples used in this book.

"""Convert to and from Roman numerals"""

#Define exceptions
class RomanError(Exception): pass
class OutOfRangeError(RomanError): pass
class NotIntegerError(RomanError): pass
class InvalidRomanNumeralError(RomanError): pass

#Define digit mapping
romanNumeralMap = (('M',  1000), 
 ('CM', 900),
 ('D',  500),
 ('CD', 400),
 ('C',  100),
 ('XC', 90),
 ('L',  50),
 ('XL', 40),
 ('X',  10),
 ('IX', 9),
 ('V',  5),
 ('IV', 4),
 ('I',  1))

def toRoman(n):
    """convert integer to Roman numeral"""
    result = ""
    for numeral, integer in romanNumeralMap:
        while n >= integer:      
            result += numeral
            n -= integer
    return result

def fromRoman(s):
    """convert Roman numeral to integer"""
    pass

romanNumeralMap is a tuple of tuples which defines three things: The character representations of the most basic Roman numerals. Note that this is not just the single-character Roman numerals; you're also defining two-character pairs like CM (“one hundred less than one thousand”); this will make the toRoman code simpler later. The order of the Roman numerals. They are listed in descending value order, from M all the way down to I. The value of each Roman numeral. Each inner tuple is a pair of (numeral, value).

Here's where your rich data structure pays off, because you don't need any special logic to handle the subtraction rule. To convert to Roman numerals, you simply iterate through romanNumeralMap looking for the largest integer value less than or equal to the input. Once found, you add the Roman numeral representation to the end of the output, subtract the corresponding integer value from the input, lather, rinse, repeat.

Example 14.4. How toRoman works

If you're not clear how toRoman works, add a print statement to the end of the while loop:

        while n >= integer:
            result += numeral
            n -= integer
            print 'subtracting', integer, 'from input, adding', numeral, 'to output'

>>> import roman2
>>> roman2.toRoman(1424)
subtracting 1000 from input, adding M to output
subtracting 400 from input, adding CD to output
subtracting 10 from input, adding X to output
subtracting 10 from input, adding X to output
subtracting 4 from input, adding IV to output
'MCDXXIV'

So toRoman appears to work, at least in this manual spot check. But will it pass the unit testing? Well no, not entirely.

Example 14.5. Output of romantest2.py against roman2.py

Remember to run romantest2.py with the -v command-line flag to enable verbose mode.

fromRoman should only accept uppercase input ... FAIL
toRoman should always return uppercase ... ok
fromRoman should fail with malformed antecedents ... FAIL
fromRoman should fail with repeated pairs of numerals ... FAIL
fromRoman should fail with too many repeated numerals ... FAIL
fromRoman should give known result with known input ... FAIL
toRoman should give known result with known input ... ok       
fromRoman(toRoman(n))==n for all n ... FAIL
toRoman should fail with non-integer input ... FAIL            
toRoman should fail with negative input ... FAIL
toRoman should fail with large input ... FAIL
toRoman should fail with 0 input ... FAIL

	toRoman does, in fact, always return uppercase, because romanNumeralMap defines the Roman numeral representations as uppercase. So this test passes already.
	Here's the big news: this version of the toRoman function passes the known values test. Remember, it's not comprehensive, but it does put the function through its paces with a variety of good inputs, including inputs that produce every single-character Roman numeral, the largest possible input (3999), and the input that produces the longest possible Roman numeral (3888). At this point, you can be reasonably confident that the function works for any good input value you could throw at it.
	However, the function does not “work” for bad values; it fails every single bad input test. That makes sense, because you didn't include any checks for bad input. Those test cases look for specific exceptions to be raised (via assertRaises), and you're never raising them. You'll do that in the next stage.

Here's the rest of the output of the unit test, listing the details of all the failures. You're down to 10.

======================================================================
FAIL: fromRoman should only accept uppercase input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage2\romantest2.py", line 156, in testFromRomanCase
    roman2.fromRoman, numeral.lower())
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should fail with malformed antecedents
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage2\romantest2.py", line 133, in testMalformedAntecedent
    self.assertRaises(roman2.InvalidRomanNumeralError, roman2.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should fail with repeated pairs of numerals
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage2\romantest2.py", line 127, in testRepeatedPairs
    self.assertRaises(roman2.InvalidRomanNumeralError, roman2.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should fail with too many repeated numerals
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage2\romantest2.py", line 122, in testTooManyRepeatedNumerals
    self.assertRaises(roman2.InvalidRomanNumeralError, roman2.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should give known result with known input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage2\romantest2.py", line 99, in testFromRomanKnownValues
    self.assertEqual(integer, result)
  File "c:\python21\lib\unittest.py", line 273, in failUnlessEqual
    raise self.failureException, (msg or '%s != %s' % (first, second))
AssertionError: 1 != None
======================================================================
FAIL: fromRoman(toRoman(n))==n for all n
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage2\romantest2.py", line 141, in testSanity
    self.assertEqual(integer, result)
  File "c:\python21\lib\unittest.py", line 273, in failUnlessEqual
    raise self.failureException, (msg or '%s != %s' % (first, second))
AssertionError: 1 != None
======================================================================
FAIL: toRoman should fail with non-integer input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage2\romantest2.py", line 116, in testNonInteger
    self.assertRaises(roman2.NotIntegerError, roman2.toRoman, 0.5)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: NotIntegerError
======================================================================
FAIL: toRoman should fail with negative input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage2\romantest2.py", line 112, in testNegative
    self.assertRaises(roman2.OutOfRangeError, roman2.toRoman, -1)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: OutOfRangeError
======================================================================
FAIL: toRoman should fail with large input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage2\romantest2.py", line 104, in testTooLarge
    self.assertRaises(roman2.OutOfRangeError, roman2.toRoman, 4000)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: OutOfRangeError
======================================================================
FAIL: toRoman should fail with 0 input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage2\romantest2.py", line 108, in testZero
    self.assertRaises(roman2.OutOfRangeError, roman2.toRoman, 0)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: OutOfRangeError
----------------------------------------------------------------------
Ran 12 tests in 0.320s

FAILED (failures=10)

14.3. roman.py, stage 3

Now that toRoman behaves correctly with good input (integers from 1 to 3999), it's time to make it behave correctly with bad input (everything else).

Example 14.6. roman3.py

This file is available in py/roman/stage3/ in the examples directory.

If you have not already done so, you can download this and other examples used in this book.

"""Convert to and from Roman numerals"""

#Define exceptions
class RomanError(Exception): pass
class OutOfRangeError(RomanError): pass
class NotIntegerError(RomanError): pass
class InvalidRomanNumeralError(RomanError): pass

#Define digit mapping
romanNumeralMap = (('M',  1000),
 ('CM', 900),
 ('D',  500),
 ('CD', 400),
 ('C',  100),
 ('XC', 90),
 ('L',  50),
 ('XL', 40),
 ('X',  10),
 ('IX', 9),
 ('V',  5),
 ('IV', 4),
 ('I',  1))

def toRoman(n):
    """convert integer to Roman numeral"""
    if not (0 < n < 4000):         
        raise OutOfRangeError, "number out of range (must be 1..3999)" 
    if int(n) <> n:                
        raise NotIntegerError, "non-integers can not be converted"

    result = ""  
    for numeral, integer in romanNumeralMap:
        while n >= integer:
            result += numeral
            n -= integer
    return result

def fromRoman(s):
    """convert Roman numeral to integer"""
    pass

	This is a nice Pythonic shortcut: multiple comparisons at once. This is equivalent to if not ((0 < n) and (n < 4000)), but it's much easier to read. This is the range check, and it should catch inputs that are too large, negative, or zero.
	You raise exceptions yourself with the raise statement. You can raise any of the built-in exceptions, or you can raise any of your custom exceptions that you've defined. The second parameter, the error message, is optional; if given, it is displayed in the traceback that is printed if the exception is never handled.
	This is the non-integer check. Non-integers can not be converted to Roman numerals.
	The rest of the function is unchanged.

Example 14.7. Watching toRoman handle bad input

>>> import roman3
>>> roman3.toRoman(4000)
Traceback (most recent call last):
  File "<interactive input>", line 1, in ?
  File "roman3.py", line 27, in toRoman
    raise OutOfRangeError, "number out of range (must be 1..3999)"
OutOfRangeError: number out of range (must be 1..3999)
>>> roman3.toRoman(1.5)
Traceback (most recent call last):
  File "<interactive input>", line 1, in ?
  File "roman3.py", line 29, in toRoman
    raise NotIntegerError, "non-integers can not be converted"
NotIntegerError: non-integers can not be converted

Example 14.8. Output of romantest3.py against roman3.py

fromRoman should only accept uppercase input ... FAIL
toRoman should always return uppercase ... ok
fromRoman should fail with malformed antecedents ... FAIL
fromRoman should fail with repeated pairs of numerals ... FAIL
fromRoman should fail with too many repeated numerals ... FAIL
fromRoman should give known result with known input ... FAIL
toRoman should give known result with known input ... ok 
fromRoman(toRoman(n))==n for all n ... FAIL
toRoman should fail with non-integer input ... ok        
toRoman should fail with negative input ... ok           
toRoman should fail with large input ... ok
toRoman should fail with 0 input ... ok

	toRoman still passes the known values test, which is comforting. All the tests that passed in stage 2 still pass, so the latest code hasn't broken anything.
	More exciting is the fact that all of the bad input tests now pass. This test, testNonInteger, passes because of the int(n) <> n check. When a non-integer is passed to toRoman, the int(n) <> n check notices it and raises the NotIntegerError exception, which is what testNonInteger is looking for.
	This test, testNegative, passes because of the not (0 < n < 4000) check, which raises an OutOfRangeError exception, which is what testNegative is looking for.

======================================================================
FAIL: fromRoman should only accept uppercase input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage3\romantest3.py", line 156, in testFromRomanCase
    roman3.fromRoman, numeral.lower())
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should fail with malformed antecedents
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage3\romantest3.py", line 133, in testMalformedAntecedent
    self.assertRaises(roman3.InvalidRomanNumeralError, roman3.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should fail with repeated pairs of numerals
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage3\romantest3.py", line 127, in testRepeatedPairs
    self.assertRaises(roman3.InvalidRomanNumeralError, roman3.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should fail with too many repeated numerals
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage3\romantest3.py", line 122, in testTooManyRepeatedNumerals
    self.assertRaises(roman3.InvalidRomanNumeralError, roman3.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should give known result with known input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage3\romantest3.py", line 99, in testFromRomanKnownValues
    self.assertEqual(integer, result)
  File "c:\python21\lib\unittest.py", line 273, in failUnlessEqual
    raise self.failureException, (msg or '%s != %s' % (first, second))
AssertionError: 1 != None
======================================================================
FAIL: fromRoman(toRoman(n))==n for all n
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage3\romantest3.py", line 141, in testSanity
    self.assertEqual(integer, result)
  File "c:\python21\lib\unittest.py", line 273, in failUnlessEqual
    raise self.failureException, (msg or '%s != %s' % (first, second))
AssertionError: 1 != None
----------------------------------------------------------------------
Ran 12 tests in 0.401s

FAILED (failures=6) 

You're down to 6 failures, and all of them involve fromRoman: the known values test, the three separate bad input tests, the case check, and the sanity check. That means that toRoman has passed all the tests it can pass by itself. (It's involved in the sanity check, but that also requires that fromRoman be written, which it isn't yet.) Which means that you must stop coding toRoman now. No tweaking, no twiddling, no extra checks “just in case”. Stop. Now. Back away from the keyboard.

	The most important thing that comprehensive unit testing can tell you is when to stop coding. When all the unit tests for a function pass, stop coding the function. When all the unit tests for an entire module pass, stop coding the module.

14.4. roman.py, stage 4

Now that toRoman is done, it's time to start coding fromRoman. Thanks to the rich data structure that maps individual Roman numerals to integer values, this is no more difficult than the toRoman function.

Example 14.9. roman4.py

This file is available in py/roman/stage4/ in the examples directory.

If you have not already done so, you can download this and other examples used in this book.

"""Convert to and from Roman numerals"""

#Define exceptions
class RomanError(Exception): pass
class OutOfRangeError(RomanError): pass
class NotIntegerError(RomanError): pass
class InvalidRomanNumeralError(RomanError): pass

#Define digit mapping
romanNumeralMap = (('M',  1000),
 ('CM', 900),
 ('D',  500),
 ('CD', 400),
 ('C',  100),
 ('XC', 90),
 ('L',  50),
 ('XL', 40),
 ('X',  10),
 ('IX', 9),
 ('V',  5),
 ('IV', 4),
 ('I',  1))

# toRoman function omitted for clarity (it hasn't changed)

def fromRoman(s):
    """convert Roman numeral to integer"""
    result = 0
    index = 0
    for numeral, integer in romanNumeralMap:
        while s[index:index+len(numeral)] == numeral: 
            result += integer
            index += len(numeral)
    return result

The pattern here is the same as toRoman. You iterate through your Roman numeral data structure (a tuple of tuples), and instead of matching the highest integer values as often as possible, you match the “highest” Roman numeral character strings as often as possible.

Example 14.10. How fromRoman works

If you're not clear how fromRoman works, add a print statement to the end of the while loop:

        while s[index:index+len(numeral)] == numeral:
            result += integer
            index += len(numeral)
            print 'found', numeral, 'of length', len(numeral), ', adding', integer

>>> import roman4
>>> roman4.fromRoman('MCMLXXII')
found M , of length 1, adding 1000
found CM , of length 2, adding 900
found L , of length 1, adding 50
found X , of length 1, adding 10
found X , of length 1, adding 10
found I , of length 1, adding 1
found I , of length 1, adding 1
1972

Example 14.11. Output of romantest4.py against roman4.py

fromRoman should only accept uppercase input ... FAIL
toRoman should always return uppercase ... ok
fromRoman should fail with malformed antecedents ... FAIL
fromRoman should fail with repeated pairs of numerals ... FAIL
fromRoman should fail with too many repeated numerals ... FAIL
fromRoman should give known result with known input ... ok 
toRoman should give known result with known input ... ok
fromRoman(toRoman(n))==n for all n ... ok
toRoman should fail with non-integer input ... ok
toRoman should fail with negative input ... ok
toRoman should fail with large input ... ok
toRoman should fail with 0 input ... ok

Two pieces of exciting news here. The first is that fromRoman works for good input, at least for all the known values you test.

The second is that the sanity check also passed. Combined with the known values tests, you can be reasonably sure that both toRoman and fromRoman work properly for all possible good values. (This is not guaranteed; it is theoretically possible that toRoman has a bug that produces the wrong Roman numeral for some particular set of inputs, and that fromRoman has a reciprocal bug that produces the same wrong integer values for exactly that set of Roman numerals that toRoman generated incorrectly. Depending on your application and your requirements, this possibility may bother you; if so, write more comprehensive test cases until it doesn't bother you.)

======================================================================
FAIL: fromRoman should only accept uppercase input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage4\romantest4.py", line 156, in testFromRomanCase
    roman4.fromRoman, numeral.lower())
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should fail with malformed antecedents
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage4\romantest4.py", line 133, in testMalformedAntecedent
    self.assertRaises(roman4.InvalidRomanNumeralError, roman4.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should fail with repeated pairs of numerals
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage4\romantest4.py", line 127, in testRepeatedPairs
    self.assertRaises(roman4.InvalidRomanNumeralError, roman4.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
======================================================================
FAIL: fromRoman should fail with too many repeated numerals
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage4\romantest4.py", line 122, in testTooManyRepeatedNumerals
    self.assertRaises(roman4.InvalidRomanNumeralError, roman4.fromRoman, s)
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
----------------------------------------------------------------------
Ran 12 tests in 1.222s

FAILED (failures=4)

14.5. roman.py, stage 5

Now that fromRoman works properly with good input, it's time to fit in the last piece of the puzzle: making it work properly with bad input. That means finding a way to look at a string and determine if it's a valid Roman numeral. This is inherently more difficult than validating numeric input in toRoman, but you have a powerful tool at your disposal: regular expressions.

If you're not familiar with regular expressions and didn't read Chapter 7, Regular Expressions, now would be a good time.

As you saw in Section 7.3, “Case Study: Roman Numerals”, there are several simple rules for constructing a Roman numeral, using the letters M, D, C, L, X, V, and I. Let's review the rules:

1. Characters are additive. I is 1, II is 2, and III is 3. VI is 6 (literally, “5 and 1”), VII is 7, and VIII is 8.
2. The tens characters (I, X, C, and M) can be repeated up to three times. At 4, you need to subtract from the next highest fives character. You can't represent 4 as IIII; instead, it is represented as IV (“1 less than 5”). 40 is written as XL (“10 less than 50”), 41 as XLI, 42 as XLII, 43 as XLIII, and then 44 as XLIV (“10 less than 50, then 1 less than 5”).
3. Similarly, at 9, you need to subtract from the next highest tens character: 8 is VIII, but 9 is IX (“1 less than 10”), not VIIII (since the I character can not be repeated four times). 90 is XC, 900 is CM.
4. The fives characters can not be repeated. 10 is always represented as X, never as VV. 100 is always C, never LL.
5. Roman numerals are always written highest to lowest, and read left to right, so order of characters matters very much. DC is 600; CD is a completely different number (400, “100 less than 500”). CI is 101; IC is not even a valid Roman numeral (because you can't subtract 1 directly from 100; you would need to write it as XCIX, “10 less than 100, then 1 less than 10”).

Example 14.12. roman5.py

This file is available in py/roman/stage5/ in the examples directory.

If you have not already done so, you can download this and other examples used in this book.

"""Convert to and from Roman numerals"""
import re

#Define exceptions
class RomanError(Exception): pass
class OutOfRangeError(RomanError): pass
class NotIntegerError(RomanError): pass
class InvalidRomanNumeralError(RomanError): pass

#Define digit mapping
romanNumeralMap = (('M',  1000),
 ('CM', 900),
 ('D',  500),
 ('CD', 400),
 ('C',  100),
 ('XC', 90),
 ('L',  50),
 ('XL', 40),
 ('X',  10),
 ('IX', 9),
 ('V',  5),
 ('IV', 4),
 ('I',  1))

def toRoman(n):
    """convert integer to Roman numeral"""
    if not (0 < n < 4000):
        raise OutOfRangeError, "number out of range (must be 1..3999)"
    if int(n) <> n:
        raise NotIntegerError, "non-integers can not be converted"

    result = ""
    for numeral, integer in romanNumeralMap:
        while n >= integer:
            result += numeral
            n -= integer
    return result

#Define pattern to detect valid Roman numerals
romanNumeralPattern = '^M?M?M?(CM|CD|D?C?C?C?)(XC|XL|L?X?X?X?)(IX|IV|V?I?I?I?)$' 

def fromRoman(s):
    """convert Roman numeral to integer"""
    if not re.search(romanNumeralPattern, s):
        raise InvalidRomanNumeralError, 'Invalid Roman numeral: %s' % s

    result = 0
    index = 0
    for numeral, integer in romanNumeralMap:
        while s[index:index+len(numeral)] == numeral:
            result += integer
            index += len(numeral)
    return result

	This is just a continuation of the pattern you discussed in Section 7.3, “Case Study: Roman Numerals”. The tens places is either XC (90), XL (40), or an optional L followed by 0 to 3 optional X characters. The ones place is either IX (9), IV (4), or an optional V followed by 0 to 3 optional I characters.
	Having encoded all that logic into a regular expression, the code to check for invalid Roman numerals becomes trivial. If re.search returns an object, then the regular expression matched and the input is valid; otherwise, the input is invalid.

At this point, you are allowed to be skeptical that that big ugly regular expression could possibly catch all the types of invalid Roman numerals. But don't take my word for it, look at the results:

Example 14.13. Output of romantest5.py against roman5.py

fromRoman should only accept uppercase input ... ok          
toRoman should always return uppercase ... ok
fromRoman should fail with malformed antecedents ... ok      
fromRoman should fail with repeated pairs of numerals ... ok 
fromRoman should fail with too many repeated numerals ... ok
fromRoman should give known result with known input ... ok
toRoman should give known result with known input ... ok
fromRoman(toRoman(n))==n for all n ... ok
toRoman should fail with non-integer input ... ok
toRoman should fail with negative input ... ok
toRoman should fail with large input ... ok
toRoman should fail with 0 input ... ok

----------------------------------------------------------------------
Ran 12 tests in 2.864s

OK     

	One thing I didn't mention about regular expressions is that, by default, they are case-sensitive. Since the regular expression romanNumeralPattern was expressed in uppercase characters, the re.search check will reject any input that isn't completely uppercase. So the uppercase input test passes.
	More importantly, the bad input tests pass. For instance, the malformed antecedents test checks cases like MCMC. As you've seen, this does not match the regular expression, so fromRoman raises an InvalidRomanNumeralError exception, which is what the malformed antecedents test case is looking for, so the test passes.
	In fact, all the bad input tests pass. This regular expression catches everything you could think of when you made your test cases.
	And the anticlimax award of the year goes to the word “OK”, which is printed by the unittest module when all the tests pass.

	When all of your tests pass, stop coding.

Chapter 15. Refactoring

15.1. Handling bugs

Despite your best efforts to write comprehensive unit tests, bugs happen. What do I mean by “bug”? A bug is a test case you haven't written yet.

Example 15.1. The bug

>>> import roman5
>>> roman5.fromRoman("") 
0

Remember in the previous section when you kept seeing that an empty string would match the regular expression you were using to check for valid Roman numerals? Well, it turns out that this is still true for the final version of the regular expression. And that's a bug; you want an empty string to raise an InvalidRomanNumeralError exception just like any other sequence of characters that don't represent a valid Roman numeral.

After reproducing the bug, and before fixing it, you should write a test case that fails, thus illustrating the bug.

Example 15.2. Testing for the bug (romantest61.py)

class FromRomanBadInput(unittest.TestCase):  

    # previous test cases omitted for clarity (they haven't changed)

    def testBlank(self):
        """fromRoman should fail with blank string"""
        self.assertRaises(roman.InvalidRomanNumeralError, roman.fromRoman, "") 

Pretty simple stuff here. Call fromRoman with an empty string and make sure it raises an InvalidRomanNumeralError exception. The hard part was finding the bug; now that you know about it, testing for it is the easy part.

Since your code has a bug, and you now have a test case that tests this bug, the test case will fail:

Example 15.3. Output of romantest61.py against roman61.py

fromRoman should only accept uppercase input ... ok
toRoman should always return uppercase ... ok
fromRoman should fail with blank string ... FAIL
fromRoman should fail with malformed antecedents ... ok
fromRoman should fail with repeated pairs of numerals ... ok
fromRoman should fail with too many repeated numerals ... ok
fromRoman should give known result with known input ... ok
toRoman should give known result with known input ... ok
fromRoman(toRoman(n))==n for all n ... ok
toRoman should fail with non-integer input ... ok
toRoman should fail with negative input ... ok
toRoman should fail with large input ... ok
toRoman should fail with 0 input ... ok

======================================================================
FAIL: fromRoman should fail with blank string
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage6\romantest61.py", line 137, in testBlank
    self.assertRaises(roman61.InvalidRomanNumeralError, roman61.fromRoman, "")
  File "c:\python21\lib\unittest.py", line 266, in failUnlessRaises
    raise self.failureException, excName
AssertionError: InvalidRomanNumeralError
----------------------------------------------------------------------
Ran 13 tests in 2.864s

FAILED (failures=1)

Now you can fix the bug.

Example 15.4. Fixing the bug (roman62.py)

This file is available in py/roman/stage6/ in the examples directory.

def fromRoman(s):
    """convert Roman numeral to integer"""
    if not s: 
        raise InvalidRomanNumeralError, 'Input can not be blank'
    if not re.search(romanNumeralPattern, s):
        raise InvalidRomanNumeralError, 'Invalid Roman numeral: %s' % s

    result = 0
    index = 0
    for numeral, integer in romanNumeralMap:
        while s[index:index+len(numeral)] == numeral:
            result += integer
            index += len(numeral)
    return result

Only two lines of code are required: an explicit check for an empty string, and a raise statement.

Example 15.5. Output of romantest62.py against roman62.py

fromRoman should only accept uppercase input ... ok
toRoman should always return uppercase ... ok
fromRoman should fail with blank string ... ok 
fromRoman should fail with malformed antecedents ... ok
fromRoman should fail with repeated pairs of numerals ... ok
fromRoman should fail with too many repeated numerals ... ok
fromRoman should give known result with known input ... ok
toRoman should give known result with known input ... ok
fromRoman(toRoman(n))==n for all n ... ok
toRoman should fail with non-integer input ... ok
toRoman should fail with negative input ... ok
toRoman should fail with large input ... ok
toRoman should fail with 0 input ... ok

----------------------------------------------------------------------
Ran 13 tests in 2.834s

OK 

	The blank string test case now passes, so the bug is fixed.
	All the other test cases still pass, which means that this bug fix didn't break anything else. Stop coding.

Coding this way does not make fixing bugs any easier. Simple bugs (like this one) require simple test cases; complex bugs will require complex test cases. In a testing-centric environment, it may seem like it takes longer to fix a bug, since you need to articulate in code exactly what the bug is (to write the test case), then fix the bug itself. Then if the test case doesn't pass right away, you need to figure out whether the fix was wrong, or whether the test case itself has a bug in it. However, in the long run, this back-and-forth between test code and code tested pays for itself, because it makes it more likely that bugs are fixed correctly the first time. Also, since you can easily re-run all the test cases along with your new one, you are much less likely to break old code when fixing new code. Today's unit test is tomorrow's regression test.

15.2. Handling changing requirements

Despite your best efforts to pin your customers to the ground and extract exact requirements from them on pain of horrible nasty things involving scissors and hot wax, requirements will change. Most customers don't know what they want until they see it, and even if they do, they aren't that good at articulating what they want precisely enough to be useful. And even if they do, they'll want more in the next release anyway. So be prepared to update your test cases as requirements change.

Suppose, for instance, that you wanted to expand the range of the Roman numeral conversion functions. Remember the rule that said that no character could be repeated more than three times? Well, the Romans were willing to make an exception to that rule by having 4 M characters in a row to represent 4000. If you make this change, you'll be able to expand the range of convertible numbers from 1..3999 to 1..4999. But first, you need to make some changes to the test cases.

Example 15.6. Modifying test cases for new requirements (romantest71.py)

This file is available in py/roman/stage7/ in the examples directory.

If you have not already done so, you can download this and other examples used in this book.

import roman71
import unittest

class KnownValues(unittest.TestCase):
    knownValues = ( (1, 'I'),
  (2, 'II'),
  (3, 'III'),
  (4, 'IV'),
  (5, 'V'),
  (6, 'VI'),
  (7, 'VII'),
  (8, 'VIII'),
  (9, 'IX'),
  (10, 'X'),
  (50, 'L'),
  (100, 'C'),
  (500, 'D'),
  (1000, 'M'),
  (31, 'XXXI'),
  (148, 'CXLVIII'),
  (294, 'CCXCIV'),
  (312, 'CCCXII'),
  (421, 'CDXXI'),
  (528, 'DXXVIII'),
  (621, 'DCXXI'),
  (782, 'DCCLXXXII'),
  (870, 'DCCCLXX'),
  (941, 'CMXLI'),
  (1043, 'MXLIII'),
  (1110, 'MCX'),
  (1226, 'MCCXXVI'),
  (1301, 'MCCCI'),
  (1485, 'MCDLXXXV'),
  (1509, 'MDIX'),
  (1607, 'MDCVII'),
  (1754, 'MDCCLIV'),
  (1832, 'MDCCCXXXII'),
  (1993, 'MCMXCIII'),
  (2074, 'MMLXXIV'),
  (2152, 'MMCLII'),
  (2212, 'MMCCXII'),
  (2343, 'MMCCCXLIII'),
  (2499, 'MMCDXCIX'),
  (2574, 'MMDLXXIV'),
  (2646, 'MMDCXLVI'),
  (2723, 'MMDCCXXIII'),
  (2892, 'MMDCCCXCII'),
  (2975, 'MMCMLXXV'),
  (3051, 'MMMLI'),
  (3185, 'MMMCLXXXV'),
  (3250, 'MMMCCL'),
  (3313, 'MMMCCCXIII'),
  (3408, 'MMMCDVIII'),
  (3501, 'MMMDI'),
  (3610, 'MMMDCX'),
  (3743, 'MMMDCCXLIII'),
  (3844, 'MMMDCCCXLIV'),
  (3888, 'MMMDCCCLXXXVIII'),
  (3940, 'MMMCMXL'),
  (3999, 'MMMCMXCIX'),
  (4000, 'MMMM'),   
  (4500, 'MMMMD'),
  (4888, 'MMMMDCCCLXXXVIII'),
  (4999, 'MMMMCMXCIX'))

    def testToRomanKnownValues(self):
        """toRoman should give known result with known input"""
        for integer, numeral in self.knownValues:
            result = roman71.toRoman(integer)
            self.assertEqual(numeral, result)

    def testFromRomanKnownValues(self):
        """fromRoman should give known result with known input"""
        for integer, numeral in self.knownValues:
            result = roman71.fromRoman(numeral)
            self.assertEqual(integer, result)

class ToRomanBadInput(unittest.TestCase):
    def testTooLarge(self):
        """toRoman should fail with large input"""
        self.assertRaises(roman71.OutOfRangeError, roman71.toRoman, 5000) 

    def testZero(self):
        """toRoman should fail with 0 input"""
        self.assertRaises(roman71.OutOfRangeError, roman71.toRoman, 0)

    def testNegative(self):
        """toRoman should fail with negative input"""
        self.assertRaises(roman71.OutOfRangeError, roman71.toRoman, -1)

    def testNonInteger(self):
        """toRoman should fail with non-integer input"""
        self.assertRaises(roman71.NotIntegerError, roman71.toRoman, 0.5)

class FromRomanBadInput(unittest.TestCase):
    def testTooManyRepeatedNumerals(self):
        """fromRoman should fail with too many repeated numerals"""
        for s in ('MMMMM', 'DD', 'CCCC', 'LL', 'XXXX', 'VV', 'IIII'):     
            self.assertRaises(roman71.InvalidRomanNumeralError, roman71.fromRoman, s)

    def testRepeatedPairs(self):
        """fromRoman should fail with repeated pairs of numerals"""
        for s in ('CMCM', 'CDCD', 'XCXC', 'XLXL', 'IXIX', 'IVIV'):
            self.assertRaises(roman71.InvalidRomanNumeralError, roman71.fromRoman, s)

    def testMalformedAntecedent(self):
        """fromRoman should fail with malformed antecedents"""
        for s in ('IIMXCC', 'VX', 'DCM', 'CMM', 'IXIV',
'MCMC', 'XCX', 'IVI', 'LM', 'LD', 'LC'):
            self.assertRaises(roman71.InvalidRomanNumeralError, roman71.fromRoman, s)

    def testBlank(self):
        """fromRoman should fail with blank string"""
        self.assertRaises(roman71.InvalidRomanNumeralError, roman71.fromRoman, "")

class SanityCheck(unittest.TestCase):
    def testSanity(self):
        """fromRoman(toRoman(n))==n for all n"""
        for integer in range(1, 5000):
            numeral = roman71.toRoman(integer)
            result = roman71.fromRoman(numeral)
            self.assertEqual(integer, result)

class CaseCheck(unittest.TestCase):
    def testToRomanCase(self):
        """toRoman should always return uppercase"""
        for integer in range(1, 5000):
            numeral = roman71.toRoman(integer)
            self.assertEqual(numeral, numeral.upper())

    def testFromRomanCase(self):
        """fromRoman should only accept uppercase input"""
        for integer in range(1, 5000):
            numeral = roman71.toRoman(integer)
            roman71.fromRoman(numeral.upper())
            self.assertRaises(roman71.InvalidRomanNumeralError,
            roman71.fromRoman, numeral.lower())

if __name__ == "__main__":
    unittest.main()

	The existing known values don't change (they're all still reasonable values to test), but you need to add a few more in the 4000 range. Here I've included 4000 (the shortest), 4500 (the second shortest), 4888 (the longest), and 4999 (the largest).
	The definition of “large input” has changed. This test used to call toRoman with 4000 and expect an error; now that 4000-4999 are good values, you need to bump this up to 5000.
	The definition of “too many repeated numerals” has also changed. This test used to call fromRoman with 'MMMM' and expect an error; now that MMMM is considered a valid Roman numeral, you need to bump this up to 'MMMMM'.
	The sanity check and case checks loop through every number in the range, from 1 to 3999. Since the range has now expanded, these for loops need to be updated as well to go up to 4999.

Now your test cases are up to date with the new requirements, but your code is not, so you expect several of the test cases to fail.

Example 15.7. Output of romantest71.py against roman71.py

fromRoman should only accept uppercase input ... ERROR        
toRoman should always return uppercase ... ERROR
fromRoman should fail with blank string ... ok
fromRoman should fail with malformed antecedents ... ok
fromRoman should fail with repeated pairs of numerals ... ok
fromRoman should fail with too many repeated numerals ... ok
fromRoman should give known result with known input ... ERROR 
toRoman should give known result with known input ... ERROR   
fromRoman(toRoman(n))==n for all n ... ERROR
toRoman should fail with non-integer input ... ok
toRoman should fail with negative input ... ok
toRoman should fail with large input ... ok
toRoman should fail with 0 input ... ok

	Our case checks now fail because they loop from 1 to 4999, but toRoman only accepts numbers from 1 to 3999, so it will fail as soon the test case hits 4000.
	The fromRoman known values test will fail as soon as it hits 'MMMM', because fromRoman still thinks this is an invalid Roman numeral.
	The toRoman known values test will fail as soon as it hits 4000, because toRoman still thinks this is out of range.
	The sanity check will also fail as soon as it hits 4000, because toRoman still thinks this is out of range.

======================================================================
ERROR: fromRoman should only accept uppercase input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage7\romantest71.py", line 161, in testFromRomanCase
    numeral = roman71.toRoman(integer)
  File "roman71.py", line 28, in toRoman
    raise OutOfRangeError, "number out of range (must be 1..3999)"
OutOfRangeError: number out of range (must be 1..3999)
======================================================================
ERROR: toRoman should always return uppercase
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage7\romantest71.py", line 155, in testToRomanCase
    numeral = roman71.toRoman(integer)
  File "roman71.py", line 28, in toRoman
    raise OutOfRangeError, "number out of range (must be 1..3999)"
OutOfRangeError: number out of range (must be 1..3999)
======================================================================
ERROR: fromRoman should give known result with known input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage7\romantest71.py", line 102, in testFromRomanKnownValues
    result = roman71.fromRoman(numeral)
  File "roman71.py", line 47, in fromRoman
    raise InvalidRomanNumeralError, 'Invalid Roman numeral: %s' % s
InvalidRomanNumeralError: Invalid Roman numeral: MMMM
======================================================================
ERROR: toRoman should give known result with known input
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage7\romantest71.py", line 96, in testToRomanKnownValues
    result = roman71.toRoman(integer)
  File "roman71.py", line 28, in toRoman
    raise OutOfRangeError, "number out of range (must be 1..3999)"
OutOfRangeError: number out of range (must be 1..3999)
======================================================================
ERROR: fromRoman(toRoman(n))==n for all n
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\docbook\dip\py\roman\stage7\romantest71.py", line 147, in testSanity
    numeral = roman71.toRoman(integer)
  File "roman71.py", line 28, in toRoman
    raise OutOfRangeError, "number out of range (must be 1..3999)"
OutOfRangeError: number out of range (must be 1..3999)
----------------------------------------------------------------------
Ran 13 tests in 2.213s

FAILED (errors=5)

Now that you have test cases that fail due to the new requirements, you can think about fixing the code to bring it in line with the test cases. (One thing that takes some getting used to when you first start coding unit tests is that the code being tested is never “ahead” of the test cases. While it's behind, you still have some work to do, and as soon as it catches up to the test cases, you stop coding.)

Example 15.8. Coding the new requirements (roman72.py)

This file is available in py/roman/stage7/ in the examples directory.

"""Convert to and from Roman numerals"""
import re

#Define exceptions
class RomanError(Exception): pass
class OutOfRangeError(RomanError): pass
class NotIntegerError(RomanError): pass
class InvalidRomanNumeralError(RomanError): pass

#Define digit mapping
romanNumeralMap = (('M',  1000),
 ('CM', 900),
 ('D',  500),
 ('CD', 400),
 ('C',  100),
 ('XC', 90),
 ('L',  50),
 ('XL', 40),
 ('X',  10),
 ('IX', 9),
 ('V',  5),
 ('IV', 4),
 ('I',  1))

def toRoman(n):
    """convert integer to Roman numeral"""
    if not (0 < n < 5000):   
        raise OutOfRangeError, "number out of range (must be 1..4999)"
    if int(n) <> n:
        raise NotIntegerError, "non-integers can not be converted"

    result = ""
    for numeral, integer in romanNumeralMap:
        while n >= integer:
            result += numeral
            n -= integer
    return result

#Define pattern to detect valid Roman numerals
romanNumeralPattern = '^M?M?M?M?(CM|CD|D?C?C?C?)(XC|XL|L?X?X?X?)(IX|IV|V?I?I?I?)$' 

def fromRoman(s):
    """convert Roman numeral to integer"""
    if not s:
        raise InvalidRomanNumeralError, 'Input can not be blank'
    if not re.search(romanNumeralPattern, s):
        raise InvalidRomanNumeralError, 'Invalid Roman numeral: %s' % s

    result = 0
    index = 0
    for numeral, integer in romanNumeralMap:
        while s[index:index+len(numeral)] == numeral:
            result += integer
            index += len(numeral)
    return result

toRoman only needs one small change, in the range check. Where you used to check 0 < n < 4000, you now check 0 < n < 5000. And you change the error message that you raise to reflect the new acceptable range (1..4999 instead of 1..3999). You don't need to make any changes to the rest of the function; it handles the new cases already. (It merrily adds 'M' for each thousand that it finds; given 4000, it will spit out 'MMMM'. The only reason it didn't do this before is that you explicitly stopped it with the range check.)

You don't need to make any changes to fromRoman at all. The only change is to romanNumeralPattern; if you look closely, you'll notice that you added another optional M in the first section of the regular expression. This will allow up to 4 M characters instead of 3, meaning you will allow the Roman numeral equivalents of 4999 instead of 3999. The actual fromRoman function is completely general; it just looks for repeated Roman numeral characters and adds them up, without caring how many times they repeat. The only reason it didn't handle 'MMMM' before is that you explicitly stopped it with the regular expression pattern matching.

You may be skeptical that these two small changes are all that you need. Hey, don't take my word for it; see for yourself:

Example 15.9. Output of romantest72.py against roman72.py

fromRoman should only accept uppercase input ... ok
toRoman should always return uppercase ... ok
fromRoman should fail with blank string ... ok
fromRoman should fail with malformed antecedents ... ok
fromRoman should fail with repeated pairs of numerals ... ok
fromRoman should fail with too many repeated numerals ... ok
fromRoman should give known result with known input ... ok
toRoman should give known result with known input ... ok
fromRoman(toRoman(n))==n for all n ... ok
toRoman should fail with non-integer input ... ok
toRoman should fail with negative input ... ok
toRoman should fail with large input ... ok
toRoman should fail with 0 input ... ok

----------------------------------------------------------------------
Ran 13 tests in 3.685s

OK 

All the test cases pass. Stop coding.

Comprehensive unit testing means never having to rely on a programmer who says “Trust me.”

15.3. Refactoring

The best thing about comprehensive unit testing is not the feeling you get when all your test cases finally pass, or even the feeling you get when someone else blames you for breaking their code and you can actually prove that you didn't. The best thing about unit testing is that it gives you the freedom to refactor mercilessly.

Refactoring is the process of taking working code and making it work better. Usually, “better” means “faster”, although it can also mean “using less memory”, or “using less disk space”, or simply “more elegantly”. Whatever it means to you, to your project, in your environment, refactoring is important to the long-term health of any program.

Here, “better” means “faster”. Specifically, the fromRoman function is slower than it needs to be, because of that big nasty regular expression that you use to validate Roman numerals. It's probably not worth trying to do away with the regular expression altogether (it would be difficult, and it might not end up any faster), but you can speed up the function by precompiling the regular expression.

Example 15.10. Compiling regular expressions

>>> import re
>>> pattern = '^M?M?M?$'
>>> re.search(pattern, 'M')               
<SRE_Match object at 01090490>
>>> compiledPattern = re.compile(pattern) 
>>> compiledPattern
<SRE_Pattern object at 00F06E28>
>>> dir(compiledPattern)
['findall', 'match', 'scanner', 'search', 'split', 'sub', 'subn']
>>> compiledPattern.search('M')           
<SRE_Match object at 01104928>

	This is the syntax you've seen before: re.search takes a regular expression as a string (pattern) and a string to match against it ('M'). If the pattern matches, the function returns a match object which can be queried to find out exactly what matched and how.
	This is the new syntax: re.compile takes a regular expression as a string and returns a pattern object. Note there is no string to match here. Compiling a regular expression has nothing to do with matching it against any specific strings (like 'M'); it only involves the regular expression itself.
	The compiled pattern object returned from re.compile has several useful-looking functions, including several (like search and sub) that are available directly in the re module.
	Calling the compiled pattern object's search function with the string 'M' accomplishes the same thing as calling re.search with both the regular expression and the string 'M'. Only much, much faster. (In fact, the re.search function simply compiles the regular expression and calls the resulting pattern object's search method for you.)

	Whenever you are going to use a regular expression more than once, you should compile it to get a pattern object, then call the methods on the pattern object directly.

Example 15.11. Compiled regular expressions in roman81.py

This file is available in py/roman/stage8/ in the examples directory.

If you have not already done so, you can download this and other examples used in this book.

# toRoman and rest of module omitted for clarity

romanNumeralPattern = \
    re.compile('^M?M?M?M?(CM|CD|D?C?C?C?)(XC|XL|L?X?X?X?)(IX|IV|V?I?I?I?)$') 

def fromRoman(s):
    """convert Roman numeral to integer"""
    if not s:
        raise InvalidRomanNumeralError, 'Input can not be blank'
    if not romanNumeralPattern.search(s):
        raise InvalidRomanNumeralError, 'Invalid Roman numeral: %s' % s

    result = 0
    index = 0
    for numeral, integer in romanNumeralMap:
        while s[index:index+len(numeral)] == numeral:
            result += integer
            index += len(numeral)
    return result

This looks very similar, but in fact a lot has changed. romanNumeralPattern is no longer a string; it is a pattern object which was returned from re.compile.

That means that you can call methods on romanNumeralPattern directly. This will be much, much faster than calling re.search every time. The regular expression is compiled once and stored in romanNumeralPattern when the module is first imported; then, every time you call fromRoman, you can immediately match the input string against the regular expression, without any intermediate steps occurring under the covers.

So how much faster is it to compile regular expressions? See for yourself:

Example 15.12. Output of romantest81.py against roman81.py

.............          
----------------------------------------------------------------------
Ran 13 tests in 3.385s 

OK   

	Just a note in passing here: this time, I ran the unit test without the -v option, so instead of the full docstring for each test, you only get a dot for each test that passes. (If a test failed, you'd get an F, and if it had an error, you'd get an E. You'd still get complete tracebacks for each failure and error, so you could track down any problems.)
	You ran 13 tests in 3.385 seconds, compared to 3.685 seconds without precompiling the regular expressions. That's an 8% improvement overall, and remember that most of the time spent during the unit test is spent doing other things. (Separately, I time-tested the regular expressions by themselves, apart from the rest of the unit tests, and found that compiling this regular expression speeds up the search by an average of 54%.) Not bad for such a simple fix.
	Oh, and in case you were wondering, precompiling the regular expression didn't break anything, and you just proved it.

There is one other performance optimization that I want to try. Given the complexity of regular expression syntax, it should come as no surprise that there is frequently more than one way to write the same expression. After some discussion about this module on comp.lang.python, someone suggested that I try using the {m,n} syntax for the optional repeated characters.

Example 15.13. roman82.py

This file is available in py/roman/stage8/ in the examples directory.

If you have not already done so, you can download this and other examples used in this book.

# rest of program omitted for clarity

#old version
#romanNumeralPattern = \
#   re.compile('^M?M?M?M?(CM|CD|D?C?C?C?)(XC|XL|L?X?X?X?)(IX|IV|V?I?I?I?)$')

#new version
romanNumeralPattern = \
    re.compile('^M{0,4}(CM|CD|D?C{0,3})(XC|XL|L?X{0,3})(IX|IV|V?I{0,3})$') 

You have replaced M?M?M?M? with M{0,4}. Both mean the same thing: “match 0 to 4 M characters”. Similarly, C?C?C? became C{0,3} (“match 0 to 3 C characters”) and so forth for X and I.

This form of the regular expression is a little shorter (though not any more readable). The big question is, is it any faster?

Example 15.14. Output of romantest82.py against roman82.py

.............
----------------------------------------------------------------------
Ran 13 tests in 3.315s 

OK   

Overall, the unit tests run 2% faster with this form of regular expression. That doesn't sound exciting, but remember that the search function is a small part of the overall unit test; most of the time is spent doing other things. (Separately, I time-tested just the regular expressions, and found that the search function is 11% faster with this syntax.) By precompiling the regular expression and rewriting part of it to use this new syntax, you've improved the regular expression performance by over 60%, and improved the overall performance of the entire unit test by over 10%.

More important than any performance boost is the fact that the module still works perfectly. This is the freedom I was talking about earlier: the freedom to tweak, change, or rewrite any piece of it and verify that you haven't messed anything up in the process. This is not a license to endlessly tweak your code just for the sake of tweaking it; you had a very specific objective (“make fromRoman faster”), and you were able to accomplish that objective without any lingering doubts about whether you introduced new bugs in the process.

One other tweak I would like to make, and then I promise I'll stop refactoring and put this module to bed. As you've seen repeatedly, regular expressions can get pretty hairy and unreadable pretty quickly. I wouldn't like to come back to this module in six months and try to maintain it. Sure, the test cases pass, so I know that it works, but if I can't figure out how it works, it's still going to be difficult to add new features, fix new bugs, or otherwise maintain it. As you saw in Section 7.5, “Verbose Regular Expressions”, Python provides a way to document your logic line-by-line.

Example 15.15. roman83.py

This file is available in py/roman/stage8/ in the examples directory.

If you have not already done so, you can download this and other examples used in this book.

# rest of program omitted for clarity

#old version
#romanNumeralPattern = \
#   re.compile('^M{0,4}(CM|CD|D?C{0,3})(XC|XL|L?X{0,3})(IX|IV|V?I{0,3})$')

#new version
romanNumeralPattern = re.compile('''
    ^ # beginning of string
    M{0,4}              # thousands - 0 to 4 M's
    (CM|CD|D?C{0,3})    # hundreds - 900 (CM), 400 (CD), 0-300 (0 to 3 C's),
      #            or 500-800 (D, followed by 0 to 3 C's)
    (XC|XL|L?X{0,3})    # tens - 90 (XC), 40 (XL), 0-30 (0 to 3 X's),
      #        or 50-80 (L, followed by 0 to 3 X's)
    (IX|IV|V?I{0,3})    # ones - 9 (IX), 4 (IV), 0-3 (0 to 3 I's),
      #        or 5-8 (V, followed by 0 to 3 I's)
    $ # end of string
    ''', re.VERBOSE) 

The re.compile function can take an optional second argument, which is a set of one or more flags that control various options about the compiled regular expression. Here you're specifying the re.VERBOSE flag, which tells Python that there are in-line comments within the regular expression itself. The comments and all the whitespace around them are not considered part of the regular expression; the re.compile function simply strips them all out when it compiles the expression. This new, “verbose” version is identical to the old version, but it is infinitely more readable.

Example 15.16. Output of romantest83.py against roman83.py

.............
----------------------------------------------------------------------
Ran 13 tests in 3.315s 

OK   

	This new, “verbose” version runs at exactly the same speed as the old version. In fact, the compiled pattern objects are the same, since the re.compile function strips out all the stuff you added.
	This new, “verbose” version passes all the same tests as the old version. Nothing has changed, except that the programmer who comes back to this module in six months stands a fighting chance of understanding how the function works.

15.4. Postscript

A clever reader read the previous section and took it to the next level. The biggest headache (and performance drain) in the program as it is currently written is the regular expression, which is required because you have no other way of breaking down a Roman numeral. But there's only 5000 of them; why don't you just build a lookup table once, then simply read that? This idea gets even better when you realize that you don't need to use regular expressions at all. As you build the lookup table for converting integers to Roman numerals, you can build the reverse lookup table to convert Roman numerals to integers.

And best of all, he already had a complete set of unit tests. He changed over half the code in the module, but the unit tests stayed the same, so he could prove that his code worked just as well as the original.

Example 15.17. roman9.py

This file is available in py/roman/stage9/ in the examples directory.

If you have not already done so, you can download this and other examples used in this book.

#Define exceptions
class RomanError(Exception): pass
class OutOfRangeError(RomanError): pass
class NotIntegerError(RomanError): pass
class InvalidRomanNumeralError(RomanError): pass

#Roman numerals must be less than 5000
MAX_ROMAN_NUMERAL = 4999

#Define digit mapping
romanNumeralMap = (('M',  1000),
 ('CM', 900),
 ('D',  500),
 ('CD', 400),
 ('C',  100),
 ('XC', 90),
 ('L',  50),
 ('XL', 40),
 ('X',  10),
 ('IX', 9),
 ('V',  5),
 ('IV', 4),
 ('I',  1))

#Create tables for fast conversion of roman numerals.
#See fillLookupTables() below.
toRomanTable = [ None ]  # Skip an index since Roman numerals have no zero
fromRomanTable = {}

def toRoman(n):
    """convert integer to Roman numeral"""
    if not (0 < n <= MAX_ROMAN_NUMERAL):
        raise OutOfRangeError, "number out of range (must be 1..%s)" % MAX_ROMAN_NUMERAL
    if int(n) <> n:
        raise NotIntegerError, "non-integers can not be converted"
    return toRomanTable[n]

def fromRoman(s):
    """convert Roman numeral to integer"""
    if not s:
        raise InvalidRomanNumeralError, "Input can not be blank"
    if not fromRomanTable.has_key(s):
        raise InvalidRomanNumeralError, "Invalid Roman numeral: %s" % s
    return fromRomanTable[s]

def toRomanDynamic(n):
    """convert integer to Roman numeral using dynamic programming"""
    result = ""
    for numeral, integer in romanNumeralMap:
        if n >= integer:
            result = numeral
            n -= integer
            break
    if n > 0:
        result += toRomanTable[n]
    return result

def fillLookupTables():
    """compute all the possible roman numerals"""
    #Save the values in two global tables to convert to and from integers.
    for integer in range(1, MAX_ROMAN_NUMERAL + 1):
        romanNumber = toRomanDynamic(integer)
        toRomanTable.append(romanNumber)
        fromRomanTable[romanNumber] = integer

fillLookupTables()

So how fast is it?

Example 15.18. Output of romantest9.py against roman9.py

.............
----------------------------------------------------------------------
Ran 13 tests in 0.791s

OK

Remember, the best performance you ever got in the original version was 13 tests in 3.315 seconds. Of course, it's not entirely a fair comparison, because this version will take longer to import (when it fills the lookup tables). But since import is only done once, this is negligible in the long run.

The moral of the story?

• Simplicity is a virtue.
• Especially when regular expressions are involved.
• And unit tests can give you the confidence to do large-scale refactoring... even if you didn't write the original code.

15.5. Summary

Unit testing is a powerful concept which, if properly implemented, can both reduce maintenance costs and increase flexibility in any long-term project. It is also important to understand that unit testing is not a panacea, a Magic Problem Solver, or a silver bullet. Writing good test cases is hard, and keeping them up to date takes discipline (especially when customers are screaming for critical bug fixes). Unit testing is not a replacement for other forms of testing, including functional testing, integration testing, and user acceptance testing. But it is feasible, and it does work, and once you've seen it work, you'll wonder how you ever got along without it.

This chapter covered a lot of ground, and much of it wasn't even Python-specific. There are unit testing frameworks for many languages, all of which require you to understand the same basic concepts:

• Designing test cases that are specific, automated, and independent
• Writing test cases before the code they are testing
• Writing tests that test good input and check for proper results
• Writing tests that test bad input and check for proper failures
• Writing and updating test cases to illustrate bugs or reflect new requirements
• Refactoring mercilessly to improve performance, scalability, readability, maintainability, or whatever other -ility you're lacking

Additionally, you should be comfortable doing all of the following Python-specific things:

• Subclassing unittest.TestCase and writing methods for individual test cases
• Using assertEqual to check that a function returns a known value
• Using assertRaises to check that a function raises a known exception
• Calling unittest.main() in your if __name__ clause to run all your test cases at once
• Running unit tests in verbose or regular mode

Further reading

• XProgramming.com has links to download unit testing frameworks for many different languages.

Chapter 16. Functional Programming

16.1. Diving in

In Chapter 13, Unit Testing, you learned about the philosophy of unit testing. In Chapter 14, Test-First Programming, you stepped through the implementation of basic unit tests in Python. In Chapter 15, Refactoring, you saw how unit testing makes large-scale refactoring easier. This chapter will build on those sample programs, but here we will focus more on advanced Python-specific techniques, rather than on unit testing itself.

The following is a complete Python program that acts as a cheap and simple regression testing framework. It takes unit tests that you've written for individual modules, collects them all into one big test suite, and runs them all at once. I actually use this script as part of the build process for this book; I have unit tests for several of the example programs (not just the roman.py module featured in Chapter 13, Unit Testing), and the first thing my automated build script does is run this program to make sure all my examples still work. If this regression test fails, the build immediately stops. I don't want to release non-working examples any more than you want to download them and sit around scratching your head and yelling at your monitor and wondering why they don't work.

Example 16.1. regression.py

If you have not already done so, you can download this and other examples used in this book.

"""Regression testing framework

This module will search for scripts in the same directory named
XYZtest.py.  Each such script should be a test suite that tests a
module through PyUnit.  (As of Python 2.1, PyUnit is included in
the standard library as "unittest".)  This script will aggregate all
found test suites into one big test suite and run them all at once.
"""

import sys, os, re, unittest

def regressionTest():
    path = os.path.abspath(os.path.dirname(sys.argv[0]))   
    files = os.listdir(path)             
    test = re.compile("test\.py$", re.IGNORECASE)          
    files = filter(test.search, files)   
    filenameToModuleName = lambda f: os.path.splitext(f)[0]
    moduleNames = map(filenameToModuleName, files)         
    modules = map(__import__, moduleNames)                 
    load = unittest.defaultTestLoader.loadTestsFromModule  
    return unittest.TestSuite(map(load, modules))          

if __name__ == "__main__": 
    unittest.main(defaultTest="regressionTest")

Running this script in the same directory as the rest of the example scripts that come with this book will find all the unit tests, named moduletest.py, run them as a single test, and pass or fail them all at once.

Example 16.2. Sample output of regression.py

[you@localhost py]$ python regression.py -v
help should fail with no object ... ok           
help should return known result for apihelper ... ok
help should honor collapse argument ... ok
help should honor spacing argument ... ok
buildConnectionString should fail with list input ... ok           
buildConnectionString should fail with string input ... ok
buildConnectionString should fail with tuple input ... ok
buildConnectionString handles empty dictionary ... ok
buildConnectionString returns known result with known input ... ok
fromRoman should only accept uppercase input ... ok                
toRoman should always return uppercase ... ok
fromRoman should fail with blank string ... ok
fromRoman should fail with malformed antecedents ... ok
fromRoman should fail with repeated pairs of numerals ... ok
fromRoman should fail with too many repeated numerals ... ok
fromRoman should give known result with known input ... ok
toRoman should give known result with known input ... ok
fromRoman(toRoman(n))==n for all n ... ok
toRoman should fail with non-integer input ... ok
toRoman should fail with negative input ... ok
toRoman should fail with large input ... ok
toRoman should fail with 0 input ... ok
kgp a ref test ... ok
kgp b ref test ... ok
kgp c ref test ... ok
kgp d ref test ... ok
kgp e ref test ... ok
kgp f ref test ... ok
kgp g ref test ... ok

----------------------------------------------------------------------
Ran 29 tests in 2.799s

OK

	The first 5 tests are from apihelpertest.py, which tests the example script from Chapter 4, The Power Of Introspection.
	The next 5 tests are from odbchelpertest.py, which tests the example script from Chapter 2, Your First Python Program.
	The rest are from romantest.py, which you studied in depth in Chapter 13, Unit Testing.

16.2. Finding the path

When running Python scripts from the command line, it is sometimes useful to know where the currently running script is located on disk.

This is one of those obscure little tricks that is virtually impossible to figure out on your own, but simple to remember once you see it. The key to it is sys.argv. As you saw in Chapter 9, XML Processing, this is a list that holds the list of command-line arguments. However, it also holds the name of the running script, exactly as it was called from the command line, and this is enough information to determine its location.

Example 16.3. fullpath.py

If you have not already done so, you can download this and other examples used in this book.

import sys, os

print 'sys.argv[0] =', sys.argv[0]             
pathname = os.path.dirname(sys.argv[0])        
print 'path =', pathname
print 'full path =', os.path.abspath(pathname) 

	Regardless of how you run a script, sys.argv[0] will always contain the name of the script, exactly as it appears on the command line. This may or may not include any path information, as you'll see shortly.
	os.path.dirname takes a filename as a string and returns the directory path portion. If the given filename does not include any path information, os.path.dirname returns an empty string.
	os.path.abspath is the key here. It takes a pathname, which can be partial or even blank, and returns a fully qualified pathname.

os.path.abspath deserves further explanation. It is very flexible; it can take any kind of pathname.

Example 16.4. Further explanation of os.path.abspath

>>> import os
>>> os.getcwd()      
/home/you
>>> os.path.abspath('')                
/home/you
>>> os.path.abspath('.ssh')            
/home/you/.ssh
>>> os.path.abspath('/home/you/.ssh') 
/home/you/.ssh
>>> os.path.abspath('.ssh/../foo/')    
/home/you/foo

	os.getcwd() returns the current working directory.
	Calling os.path.abspath with an empty string returns the current working directory, same as os.getcwd().
	Calling os.path.abspath with a partial pathname constructs a fully qualified pathname out of it, based on the current working directory.
	Calling os.path.abspath with a full pathname simply returns it.
	os.path.abspath also normalizes the pathname it returns. Note that this example worked even though I don't actually have a 'foo' directory. os.path.abspath never checks your actual disk; this is all just string manipulation.

	The pathnames and filenames you pass to os.path.abspath do not need to exist.

	os.path.abspath not only constructs full path names, it also normalizes them. That means that if you are in the /usr/ directory, os.path.abspath('bin/../local/bin') will return /usr/local/bin. It normalizes the path by making it as simple as possible. If you just want to normalize a pathname like this without turning it into a full pathname, use os.path.normpath instead.

Example 16.5. Sample output from fullpath.py

[you@localhost py]$ python /home/you/diveintopython3/common/py/fullpath.py 
sys.argv[0] = /home/you/diveintopython3/common/py/fullpath.py
path = /home/you/diveintopython3/common/py
full path = /home/you/diveintopython3/common/py
[you@localhost diveintopython3]$ python common/py/fullpath.py               
sys.argv[0] = common/py/fullpath.py
path = common/py
full path = /home/you/diveintopython3/common/py
[you@localhost diveintopython3]$ cd common/py
[you@localhost py]$ python fullpath.py 
sys.argv[0] = fullpath.py
path = 
full path = /home/you/diveintopython3/common/py

	In the first case, sys.argv[0] includes the full path of the script. You can then use the os.path.dirname function to strip off the script name and return the full directory name, and os.path.abspath simply returns what you give it.
	If the script is run by using a partial pathname, sys.argv[0] will still contain exactly what appears on the command line. os.path.dirname will then give you a partial pathname (relative to the current directory), and os.path.abspath will construct a full pathname from the partial pathname.
	If the script is run from the current directory without giving any path, os.path.dirname will simply return an empty string. Given an empty string, os.path.abspath returns the current directory, which is what you want, since the script was run from the current directory.

	Like the other functions in the os and os.path modules, os.path.abspath is cross-platform. Your results will look slightly different than my examples if you're running on Windows (which uses backslash as a path separator) or Mac OS (which uses colons), but they'll still work. That's the whole point of the os module.

Addendum. One reader was dissatisfied with this solution, and wanted to be able to run all the unit tests in the current directory, not the directory where regression.py is located. He suggests this approach instead:

Example 16.6. Running scripts in the current directory

import sys, os, re, unittest

def regressionTest():
    path = os.getcwd()       
    sys.path.append(path)    
    files = os.listdir(path) 

	Instead of setting path to the directory where the currently running script is located, you set it to the current working directory instead. This will be whatever directory you were in before you ran the script, which is not necessarily the same as the directory the script is in. (Read that sentence a few times until you get it.)
	Append this directory to the Python library search path, so that when you dynamically import the unit test modules later, Python can find them. You didn't need to do this when path was the directory of the currently running script, because Python always looks in that directory.
	The rest of the function is the same.

This technique will allow you to re-use this regression.py script on multiple projects. Just put the script in a common directory, then change to the project's directory before running it. All of that project's unit tests will be found and tested, instead of the unit tests in the common directory where regression.py is located.

16.3. Filtering lists revisited

You're already familiar with using list comprehensions to filter lists. There is another way to accomplish this same thing, which some people feel is more expressive.

Python has a built-in filter function which takes two arguments, a function and a list, and returns a list.^[7] The function passed as the first argument to filter must itself take one argument, and the list that filter returns will contain all the elements from the list passed to filter for which the function passed to filter returns true.

Got all that? It's not as difficult as it sounds.

Example 16.7. Introducing filter

>>> def odd(n):                 
...     return n % 2
...     
>>> li = [1, 2, 3, 5, 9, 10, 256, -3]
>>> filter(odd, li)             
[1, 3, 5, 9, -3]
>>> [e for e in li if odd(e)]   
>>> filteredList = []
>>> for n in li:                
...     if odd(n):
...         filteredList.append(n)
...     
>>> filteredList
[1, 3, 5, 9, -3]

	odd uses the built-in mod function “%” to return True if n is odd and False if n is even.
	filter takes two arguments, a function (odd) and a list (li). It loops through the list and calls odd with each element. If odd returns a true value (remember, any non-zero value is true in Python), then the element is included in the returned list, otherwise it is filtered out. The result is a list of only the odd numbers from the original list, in the same order as they appeared in the original.
	You could accomplish the same thing using list comprehensions, as you saw in Section 4.5, “Filtering Lists”.
	You could also accomplish the same thing with a for loop. Depending on your programming background, this may seem more “straightforward”, but functions like filter are much more expressive. Not only is it easier to write, it's easier to read, too. Reading the for loop is like standing too close to a painting; you see all the details, but it may take a few seconds to be able to step back and see the bigger picture: “Oh, you're just filtering the list!”

Example 16.8. filter in regression.py

    files = os.listdir(path)              
    test = re.compile("test\.py$", re.IGNORECASE)           
    files = filter(test.search, files)    

	As you saw in Section 16.2, “Finding the path”, path may contain the full or partial pathname of the directory of the currently running script, or it may contain an empty string if the script is being run from the current directory. Either way, files will end up with the names of the files in the same directory as this script you're running.
	This is a compiled regular expression. As you saw in Section 15.3, “Refactoring”, if you're going to use the same regular expression over and over, you should compile it for faster performance. The compiled object has a search method which takes a single argument, the string to search. If the regular expression matches the string, the search method returns a Match object containing information about the regular expression match; otherwise it returns None, the Python null value.
	For each element in the files list, you're going to call the search method of the compiled regular expression object, test. If the regular expression matches, the method will return a Match object, which Python considers to be true, so the element will be included in the list returned by filter. If the regular expression does not match, the search method will return None, which Python considers to be false, so the element will not be included.

Historical note. Versions of Python prior to 2.0 did not have list comprehensions, so you couldn't filter using list comprehensions; the filter function was the only game in town. Even with the introduction of list comprehensions in 2.0, some people still prefer the old-style filter (and its companion function, map, which you'll see later in this chapter). Both techniques work at the moment, so which one you use is a matter of style. There is discussion that map and filter might be deprecated in a future version of Python, but no decision has been made.

Example 16.9. Filtering using list comprehensions instead

    files = os.listdir(path)             
    test = re.compile("test\.py$", re.IGNORECASE)          
    files = [f for f in files if test.search(f)] 

This will accomplish exactly the same result as using the filter function. Which way is more expressive? That's up to you.

16.4. Mapping lists revisited

You're already familiar with using list comprehensions to map one list into another. There is another way to accomplish the same thing, using the built-in map function. It works much the same way as the filter function.

Example 16.10. Introducing map

>>> def double(n):
...     return n*2
...     
>>> li = [1, 2, 3, 5, 9, 10, 256, -3]
>>> map(double, li)     
[2, 4, 6, 10, 18, 20, 512, -6]
>>> [double(n) for n in li]               
[2, 4, 6, 10, 18, 20, 512, -6]
>>> newlist = []
>>> for n in li:        
...     newlist.append(double(n))
...     
>>> newlist
[2, 4, 6, 10, 18, 20, 512, -6]

	map takes a function and a list[8] and returns a new list by calling the function with each element of the list in order. In this case, the function simply multiplies each element by 2.
	You could accomplish the same thing with a list comprehension. List comprehensions were first introduced in Python 2.0; map has been around forever.
	You could, if you insist on thinking like a Visual Basic programmer, use a for loop to accomplish the same thing.

Example 16.11. map with lists of mixed datatypes

>>> li = [5, 'a', (2, 'b')]
>>> map(double, li)     
[10, 'aa', (2, 'b', 2, 'b')]

As a side note, I'd like to point out that map works just as well with lists of mixed datatypes, as long as the function you're using correctly handles each type. In this case, the double function simply multiplies the given argument by 2, and Python Does The Right Thing depending on the datatype of the argument. For integers, this means actually multiplying it by 2; for strings, it means concatenating the string with itself; for tuples, it means making a new tuple that has all of the elements of the original, then all of the elements of the original again.

All right, enough play time. Let's look at some real code.

Example 16.12. map in regression.py

    filenameToModuleName = lambda f: os.path.splitext(f)[0] 
    moduleNames = map(filenameToModuleName, files)          

	As you saw in Section 4.7, “Using lambda Functions”, lambda defines an inline function. And as you saw in Example 6.17, “Splitting Pathnames”, os.path.splitext takes a filename and returns a tuple (name, extension). So filenameToModuleName is a function which will take a filename and strip off the file extension, and return just the name.
	Calling map takes each filename listed in files, passes it to the function filenameToModuleName, and returns a list of the return values of each of those function calls. In other words, you strip the file extension off of each filename, and store the list of all those stripped filenames in moduleNames.

As you'll see in the rest of the chapter, you can extend this type of data-centric thinking all the way to the final goal, which is to define and execute a single test suite that contains the tests from all of those individual test suites.

16.5. Data-centric programming

By now you're probably scratching your head wondering why this is better than using for loops and straight function calls. And that's a perfectly valid question. Mostly, it's a matter of perspective. Using map and filter forces you to center your thinking around your data.

In this case, you started with no data at all; the first thing you did was get the directory path of the current script, and got a list of files in that directory. That was the bootstrap, and it gave you real data to work with: a list of filenames.

However, you knew you didn't care about all of those files, only the ones that were actually test suites. You had too much data, so you needed to filter it. How did you know which data to keep? You needed a test to decide, so you defined one and passed it to the filter function. In this case you used a regular expression to decide, but the concept would be the same regardless of how you constructed the test.

Now you had the filenames of each of the test suites (and only the test suites, since everything else had been filtered out), but you really wanted module names instead. You had the right amount of data, but it was in the wrong format. So you defined a function that would transform a single filename into a module name, and you mapped that function onto the entire list. From one filename, you can get a module name; from a list of filenames, you can get a list of module names.

Instead of filter, you could have used a for loop with an if statement. Instead of map, you could have used a for loop with a function call. But using for loops like that is busywork. At best, it simply wastes time; at worst, it introduces obscure bugs. For instance, you need to figure out how to test for the condition “is this file a test suite?” anyway; that's the application-specific logic, and no language can write that for us. But once you've figured that out, do you really want go to all the trouble of defining a new empty list and writing a for loop and an if statement and manually calling append to add each element to the new list if it passes the condition and then keeping track of which variable holds the new filtered data and which one holds the old unfiltered data? Why not just define the test condition, then let Python do the rest of that work for us?

Oh sure, you could try to be fancy and delete elements in place without creating a new list. But you've been burned by that before. Trying to modify a data structure that you're looping through can be tricky. You delete an element, then loop to the next element, and suddenly you've skipped one. Is Python one of the languages that works that way? How long would it take you to figure it out? Would you remember for certain whether it was safe the next time you tried? Programmers spend so much time and make so many mistakes dealing with purely technical issues like this, and it's all pointless. It doesn't advance your program at all; it's just busywork.

I resisted list comprehensions when I first learned Python, and I resisted filter and map even longer. I insisted on making my life more difficult, sticking to the familiar way of for loops and if statements and step-by-step code-centric programming. And my Python programs looked a lot like Visual Basic programs, detailing every step of every operation in every function. And they had all the same types of little problems and obscure bugs. And it was all pointless.

Let it all go. Busywork code is not important. Data is important. And data is not difficult. It's only data. If you have too much, filter it. If it's not what you want, map it. Focus on the data; leave the busywork behind.

16.6. Dynamically importing modules

OK, enough philosophizing. Let's talk about dynamically importing modules.

First, let's look at how you normally import modules. The import module syntax looks in the search path for the named module and imports it by name. You can even import multiple modules at once this way, with a comma-separated list. You did this on the very first line of this chapter's script.

Example 16.13. Importing multiple modules at once

import sys, os, re, unittest 

This imports four modules at once: sys (for system functions and access to the command line parameters), os (for operating system functions like directory listings), re (for regular expressions), and unittest (for unit testing).

Now let's do the same thing, but with dynamic imports.

Example 16.14. Importing modules dynamically

>>> sys = __import__('sys')           
>>> os = __import__('os')
>>> re = __import__('re')
>>> unittest = __import__('unittest')
>>> sys             
>>> <module 'sys' (built-in)>
>>> os
>>> <module 'os' from '/usr/local/lib/python2.2/os.pyc'>

	The built-in __import__ function accomplishes the same goal as using the import statement, but it's an actual function, and it takes a string as an argument.
	The variable sys is now the sys module, just as if you had said import sys. The variable os is now the os module, and so forth.

So __import__ imports a module, but takes a string argument to do it. In this case the module you imported was just a hard-coded string, but it could just as easily be a variable, or the result of a function call. And the variable that you assign the module to doesn't need to match the module name, either. You could import a series of modules and assign them to a list.

Example 16.15. Importing a list of modules dynamically

>>> moduleNames = ['sys', 'os', 're', 'unittest'] 
>>> moduleNames
['sys', 'os', 're', 'unittest']
>>> modules = map(__import__, moduleNames)        
>>> modules   
[<module 'sys' (built-in)>,
<module 'os' from 'c:\Python22\lib\os.pyc'>,
<module 're' from 'c:\Python22\lib\re.pyc'>,
<module 'unittest' from 'c:\Python22\lib\unittest.pyc'>]
>>> modules[0].version          
'2.2.2 (#37, Nov 26 2002, 10:24:37) [MSC 32 bit (Intel)]'
>>> import sys
>>> sys.version
'2.2.2 (#37, Nov 26 2002, 10:24:37) [MSC 32 bit (Intel)]'

	moduleNames is just a list of strings. Nothing fancy, except that the strings happen to be names of modules that you could import, if you wanted to.
	Surprise, you wanted to import them, and you did, by mapping the __import__ function onto the list. Remember, this takes each element of the list (moduleNames) and calls the function (__import__) over and over, once with each element of the list, builds a list of the return values, and returns the result.
	So now from a list of strings, you've created a list of actual modules. (Your paths may be different, depending on your operating system, where you installed Python, the phase of the moon, etc.)
	To drive home the point that these are real modules, let's look at some module attributes. Remember, modules[0] is the sys module, so modules[0].version is sys.version. All the other attributes and methods of these modules are also available. There's nothing magic about the import statement, and there's nothing magic about modules. Modules are objects. Everything is an object.

Now you should be able to put this all together and figure out what most of this chapter's code sample is doing.

16.7. Putting it all together

You've learned enough now to deconstruct the first seven lines of this chapter's code sample: reading a directory and importing selected modules within it.

Example 16.16. The regressionTest function

def regressionTest():
    path = os.path.abspath(os.path.dirname(sys.argv[0]))   
    files = os.listdir(path)             
    test = re.compile("test\.py$", re.IGNORECASE)          
    files = filter(test.search, files)   
    filenameToModuleName = lambda f: os.path.splitext(f)[0]
    moduleNames = map(filenameToModuleName, files)         
    modules = map(__import__, moduleNames)                 
load = unittest.defaultTestLoader.loadTestsFromModule  
return unittest.TestSuite(map(load, modules))          

Let's look at it line by line, interactively. Assume that the current directory is c:\diveintopython3\py, which contains the examples that come with this book, including this chapter's script. As you saw in Section 16.2, “Finding the path”, the script directory will end up in the path variable, so let's start hard-code that and go from there.

Example 16.17. Step 1: Get all the files

>>> import sys, os, re, unittest
>>> path = r'c:\diveintopython3\py'
>>> files = os.listdir(path)             
>>> files 
['BaseHTMLProcessor.py', 'LICENSE.txt', 'apihelper.py', 'apihelpertest.py',
'argecho.py', 'autosize.py', 'builddialectexamples.py', 'dialect.py',
'fileinfo.py', 'fullpath.py', 'kgptest.py', 'makerealworddoc.py',
'odbchelper.py', 'odbchelpertest.py', 'parsephone.py', 'piglatin.py',
'plural.py', 'pluraltest.py', 'pyfontify.py', 'regression.py', 'roman.py', 'romantest.py',
'uncurly.py', 'unicode2koi8r.py', 'urllister.py', 'kgp', 'plural', 'roman',
'colorize.py']

files is a list of all the files and directories in the script's directory. (If you've been running some of the examples already, you may also see some .pyc files in there as well.)

Example 16.18. Step 2: Filter to find the files you care about

>>> test = re.compile("test\.py$", re.IGNORECASE)           
>>> files = filter(test.search, files)    
>>> files               
['apihelpertest.py', 'kgptest.py', 'odbchelpertest.py', 'pluraltest.py', 'romantest.py']

	This regular expression will match any string that ends with test.py. Note that you need to escape the period, since a period in a regular expression usually means “match any single character”, but you actually want to match a literal period instead.
	The compiled regular expression acts like a function, so you can use it to filter the large list of files and directories, to find the ones that match the regular expression.
	And you're left with the list of unit testing scripts, because they were the only ones named SOMETHINGtest.py.

Example 16.19. Step 3: Map filenames to module names

>>> filenameToModuleName = lambda f: os.path.splitext(f)[0] 
>>> filenameToModuleName('romantest.py')  
'romantest'
>>> filenameToModuleName('odchelpertest.py')
'odbchelpertest'
>>> moduleNames = map(filenameToModuleName, files)          
>>> moduleNames         
['apihelpertest', 'kgptest', 'odbchelpertest', 'pluraltest', 'romantest']

	As you saw in Section 4.7, “Using lambda Functions”, lambda is a quick-and-dirty way of creating an inline, one-line function. This one takes a filename with an extension and returns just the filename part, using the standard library function os.path.splitext that you saw in Example 6.17, “Splitting Pathnames”.
	filenameToModuleName is a function. There's nothing magic about lambda functions as opposed to regular functions that you define with a def statement. You can call the filenameToModuleName function like any other, and it does just what you wanted it to do: strips the file extension off of its argument.
	Now you can apply this function to each file in the list of unit test files, using map.
	And the result is just what you wanted: a list of modules, as strings.

Example 16.20. Step 4: Mapping module names to modules

>>> modules = map(__import__, moduleNames)
>>> modules             
[<module 'apihelpertest' from 'apihelpertest.py'>,
<module 'kgptest' from 'kgptest.py'>,
<module 'odbchelpertest' from 'odbchelpertest.py'>,
<module 'pluraltest' from 'pluraltest.py'>,
<module 'romantest' from 'romantest.py'>]
>>> modules[-1]         
<module 'romantest' from 'romantest.py'>

	As you saw in Section 16.6, “Dynamically importing modules”, you can use a combination of map and __import__ to map a list of module names (as strings) into actual modules (which you can call or access like any other module).
	modules is now a list of modules, fully accessible like any other module.
	The last module in the list is the romantest module, just as if you had said import romantest.

Example 16.21. Step 5: Loading the modules into a test suite

>>> load = unittest.defaultTestLoader.loadTestsFromModule  
>>> map(load, modules)   
[<unittest.TestSuite tests=[
  <unittest.TestSuite tests=[<apihelpertest.BadInput testMethod=testNoObject>]>,
  <unittest.TestSuite tests=[<apihelpertest.KnownValues testMethod=testApiHelper>]>,
  <unittest.TestSuite tests=[
    <apihelpertest.ParamChecks testMethod=testCollapse>, 
    <apihelpertest.ParamChecks testMethod=testSpacing>]>, 
    ...
  ]
]
>>> unittest.TestSuite(map(load, modules)) 

These are real module objects. Not only can you access them like any other module, instantiate classes and call functions, you can also introspect into the module to figure out which classes and functions it has in the first place. That's what the loadTestsFromModule method does: it introspects into each module and returns a unittest.TestSuite object for each module. Each TestSuite object actually contains a list of TestSuite objects, one for each TestCase class in your module, and each of those TestSuite objects contains a list of tests, one for each test method in your module.

Finally, you wrap the list of TestSuite objects into one big test suite. The unittest module has no problem traversing this tree of nested test suites within test suites; eventually it gets down to an individual test method and executes it, verifies that it passes or fails, and moves on to the next one.

This introspection process is what the unittest module usually does for us. Remember that magic-looking unittest.main() function that our individual test modules called to kick the whole thing off? unittest.main() actually creates an instance of unittest.TestProgram, which in turn creates an instance of a unittest.defaultTestLoader and loads it up with the module that called it. (How does it get a reference to the module that called it if you don't give it one? By using the equally-magic __import__('__main__') command, which dynamically imports the currently-running module. I could write a book on all the tricks and techniques used in the unittest module, but then I'd never finish this one.)

Example 16.22. Step 6: Telling unittest to use your test suite

if __name__ == "__main__": 
    unittest.main(defaultTest="regressionTest") 

Instead of letting the unittest module do all its magic for us, you've done most of it yourself. You've created a function (regressionTest) that imports the modules yourself, calls unittest.defaultTestLoader yourself, and wraps it all up in a test suite. Now all you need to do is tell unittest that, instead of looking for tests and building a test suite in the usual way, it should just call the regressionTest function, which returns a ready-to-use TestSuite.

16.8. Summary

The regression.py program and its output should now make perfect sense.

You should now feel comfortable doing all of these things:

• Manipulating path information from the command line.
• Filtering lists using filter instead of list comprehensions.
• Mapping lists using map instead of list comprehensions.
• Dynamically importing modules.

---

^[7] Technically, the second argument to filter can be any sequence, including lists, tuples, and custom classes that act like lists by defining the __getitem__ special method. If possible, filter will return the same datatype as you give it, so filtering a list returns a list, but filtering a tuple returns a tuple.

^[8] Again, I should point out that map can take a list, a tuple, or any object that acts like a sequence. See previous footnote about filter.

Chapter 17. Dynamic functions

17.1. Diving in

I want to talk about plural nouns. Also, functions that return other functions, advanced regular expressions, and generators. Generators are new in Python 2.3. But first, let's talk about how to make plural nouns.

If you haven't read Chapter 7, Regular Expressions, now would be a good time. This chapter assumes you understand the basics of regular expressions, and quickly descends into more advanced uses.

English is a schizophrenic language that borrows from a lot of other languages, and the rules for making singular nouns into plural nouns are varied and complex. There are rules, and then there are exceptions to those rules, and then there are exceptions to the exceptions.

If you grew up in an English-speaking country or learned English in a formal school setting, you're probably familiar with the basic rules:

1. If a word ends in S, X, or Z, add ES. “Bass” becomes “basses”, “fax” becomes “faxes”, and “waltz” becomes “waltzes”.
2. If a word ends in a noisy H, add ES; if it ends in a silent H, just add S. What's a noisy H? One that gets combined with other letters to make a sound that you can hear. So “coach” becomes “coaches” and “rash” becomes “rashes”, because you can hear the CH and SH sounds when you say them. But “cheetah” becomes “cheetahs”, because the H is silent.
3. If a word ends in Y that sounds like I, change the Y to IES; if the Y is combined with a vowel to sound like something else, just add S. So “vacancy” becomes “vacancies”, but “day” becomes “days”.
4. If all else fails, just add S and hope for the best.

(I know, there are a lot of exceptions. “Man” becomes “men” and “woman” becomes “women”, but “human” becomes “humans”. “Mouse” becomes “mice” and “louse” becomes “lice”, but “house” becomes “houses”. “Knife” becomes “knives” and “wife” becomes “wives”, but “lowlife” becomes “lowlifes”. And don't even get me started on words that are their own plural, like “sheep”, “deer”, and “haiku”.)

Other languages are, of course, completely different.

Let's design a module that pluralizes nouns. Start with just English nouns, and just these four rules, but keep in mind that you'll inevitably need to add more rules, and you may eventually need to add more languages.

17.2. plural.py, stage 1

So you're looking at words, which at least in English are strings of characters. And you have rules that say you need to find different combinations of characters, and then do different things to them. This sounds like a job for regular expressions.

Example 17.1. plural1.py

import re

def plural(noun):          
    if re.search('[sxz]$', noun):             
        return re.sub('$', 'es', noun)        
    elif re.search('[^aeioudgkprt]h$', noun):
        return re.sub('$', 'es', noun)       
    elif re.search('[^aeiou]y$', noun):      
        return re.sub('y$', 'ies', noun)     
    else:
        return noun + 's'  

	OK, this is a regular expression, but it uses a syntax you didn't see in Chapter 7, Regular Expressions. The square brackets mean “match exactly one of these characters”. So [sxz] means “s, or x, or z”, but only one of them. The $ should be familiar; it matches the end of string. So you're checking to see if noun ends with s, x, or z.
	This re.sub function performs regular expression-based string substitutions. Let's look at it in more detail.

Example 17.2. Introducing re.sub

>>> import re
>>> re.search('[abc]', 'Mark')   
<_sre.SRE_Match object at 0x001C1FA8>
>>> re.sub('[abc]', 'o', 'Mark') 
'Mork'
>>> re.sub('[abc]', 'o', 'rock') 
'rook'
>>> re.sub('[abc]', 'o', 'caps') 
'oops'

	Does the string Mark contain a, b, or c? Yes, it contains a.
	OK, now find a, b, or c, and replace it with o. Mark becomes Mork.
	The same function turns rock into rook.
	You might think this would turn caps into oaps, but it doesn't. re.sub replaces all of the matches, not just the first one. So this regular expression turns caps into oops, because both the c and the a get turned into o.

Example 17.3. Back to plural1.py

import re

def plural(noun):          
    if re.search('[sxz]$', noun):            
        return re.sub('$', 'es', noun)        
    elif re.search('[^aeioudgkprt]h$', noun): 
        return re.sub('$', 'es', noun)        
    elif re.search('[^aeiou]y$', noun):      
        return re.sub('y$', 'ies', noun)     
    else:
        return noun + 's'  

	Back to the plural function. What are you doing? You're replacing the end of string with es. In other words, adding es to the string. You could accomplish the same thing with string concatenation, for example noun + 'es', but I'm using regular expressions for everything, for consistency, for reasons that will become clear later in the chapter.
	Look closely, this is another new variation. The ^ as the first character inside the square brackets means something special: negation. [^abc] means “any single character except a, b, or c”. So [^aeioudgkprt] means any character except a, e, i, o, u, d, g, k, p, r, or t. Then that character needs to be followed by h, followed by end of string. You're looking for words that end in H where the H can be heard.
	Same pattern here: match words that end in Y, where the character before the Y is not a, e, i, o, or u. You're looking for words that end in Y that sounds like I.

Example 17.4. More on negation regular expressions

>>> import re
>>> re.search('[^aeiou]y$', 'vacancy') 
<_sre.SRE_Match object at 0x001C1FA8>
>>> re.search('[^aeiou]y$', 'boy')     
>>> 
>>> re.search('[^aeiou]y$', 'day')
>>> 
>>> re.search('[^aeiou]y$', 'pita')    
>>> 

	vacancy matches this regular expression, because it ends in cy, and c is not a, e, i, o, or u.
	boy does not match, because it ends in oy, and you specifically said that the character before the y could not be o. day does not match, because it ends in ay.
	pita does not match, because it does not end in y.

Example 17.5. More on re.sub

>>> re.sub('y$', 'ies', 'vacancy')              
'vacancies'
>>> re.sub('y$', 'ies', 'agency')
'agencies'
>>> re.sub('([^aeiou])y$', r'\1ies', 'vacancy') 
'vacancies'

This regular expression turns vacancy into vacancies and agency into agencies, which is what you wanted. Note that it would also turn boy into boies, but that will never happen in the function because you did that re.search first to find out whether you should do this re.sub.

Just in passing, I want to point out that it is possible to combine these two regular expressions (one to find out if the rule applies, and another to actually apply it) into a single regular expression. Here's what that would look like. Most of it should look familiar: you're using a remembered group, which you learned in Section 7.6, “Case study: Parsing Phone Numbers”, to remember the character before the y. Then in the substitution string, you use a new syntax, \1, which means “hey, that first group you remembered? put it here”. In this case, you remember the c before the y, and then when you do the substitution, you substitute c in place of c, and ies in place of y. (If you have more than one remembered group, you can use \2 and \3 and so on.)

Regular expression substitutions are extremely powerful, and the \1 syntax makes them even more powerful. But combining the entire operation into one regular expression is also much harder to read, and it doesn't directly map to the way you first described the pluralizing rules. You originally laid out rules like “if the word ends in S, X, or Z, then add ES”. And if you look at this function, you have two lines of code that say “if the word ends in S, X, or Z, then add ES”. It doesn't get much more direct than that.

17.3. plural.py, stage 2

Now you're going to add a level of abstraction. You started by defining a list of rules: if this, then do that, otherwise go to the next rule. Let's temporarily complicate part of the program so you can simplify another part.

Example 17.6. plural2.py

import re

def match_sxz(noun):        
    return re.search('[sxz]$', noun)          

def apply_sxz(noun):        
    return re.sub('$', 'es', noun)            

def match_h(noun):          
    return re.search('[^aeioudgkprt]h$', noun)

def apply_h(noun):          
    return re.sub('$', 'es', noun)            

def match_y(noun):          
    return re.search('[^aeiou]y$', noun)      
        
def apply_y(noun):          
    return re.sub('y$', 'ies', noun)          

def match_default(noun):    
    return 1                
        
def apply_default(noun):    
    return noun + 's'       

rules = ((match_sxz, apply_sxz),
         (match_h, apply_h),
         (match_y, apply_y),
         (match_default, apply_default)
         ) 

def plural(noun):           
    for matchesRule, applyRule in rules:       
        if matchesRule(noun):
            return applyRule(noun)             

	This version looks more complicated (it's certainly longer), but it does exactly the same thing: try to match four different rules, in order, and apply the appropriate regular expression when a match is found. The difference is that each individual match and apply rule is defined in its own function, and the functions are then listed in this rules variable, which is a tuple of tuples.
	Using a for loop, you can pull out the match and apply rules two at a time (one match, one apply) from the rules tuple. On the first iteration of the for loop, matchesRule will get match_sxz, and applyRule will get apply_sxz. On the second iteration (assuming you get that far), matchesRule will be assigned match_h, and applyRule will be assigned apply_h.
	Remember that everything in Python is an object, including functions. rules contains actual functions; not names of functions, but actual functions. When they get assigned in the for loop, then matchesRule and applyRule are actual functions that you can call. So on the first iteration of the for loop, this is equivalent to calling matches_sxz(noun).
	On the first iteration of the for loop, this is equivalent to calling apply_sxz(noun), and so forth.

If this additional level of abstraction is confusing, try unrolling the function to see the equivalence. This for loop is equivalent to the following:

Example 17.7. Unrolling the plural function

def plural(noun):
    if match_sxz(noun):
        return apply_sxz(noun)
    if match_h(noun):
        return apply_h(noun)
    if match_y(noun):
        return apply_y(noun)
    if match_default(noun):
        return apply_default(noun)

The benefit here is that that plural function is now simplified. It takes a list of rules, defined elsewhere, and iterates through them in a generic fashion. Get a match rule; does it match? Then call the apply rule. The rules could be defined anywhere, in any way. The plural function doesn't care.

Now, was adding this level of abstraction worth it? Well, not yet. Let's consider what it would take to add a new rule to the function. Well, in the previous example, it would require adding an if statement to the plural function. In this example, it would require adding two functions, match_foo and apply_foo, and then updating the rules list to specify where in the order the new match and apply functions should be called relative to the other rules.

This is really just a stepping stone to the next section. Let's move on.

17.4. plural.py, stage 3

Defining separate named functions for each match and apply rule isn't really necessary. You never call them directly; you define them in the rules list and call them through there. Let's streamline the rules definition by anonymizing those functions.

Example 17.8. plural3.py

import re

rules = \
  (
    (
     lambda word: re.search('[sxz]$', word),
     lambda word: re.sub('$', 'es', word)
    ),
    (
     lambda word: re.search('[^aeioudgkprt]h$', word),
     lambda word: re.sub('$', 'es', word)
    ),
    (
     lambda word: re.search('[^aeiou]y$', word),
     lambda word: re.sub('y$', 'ies', word)
    ),
    (
     lambda word: re.search('$', word),
     lambda word: re.sub('$', 's', word)
    )
   )       

def plural(noun):           
    for matchesRule, applyRule in rules:       
        if matchesRule(noun):                 
            return applyRule(noun)            

This is the same set of rules as you defined in stage 2. The only difference is that instead of defining named functions like match_sxz and apply_sxz, you have “inlined” those function definitions directly into the rules list itself, using lambda functions.

Note that the plural function hasn't changed at all. It iterates through a set of rule functions, checks the first rule, and if it returns a true value, calls the second rule and returns the value. Same as above, word for word. The only difference is that the rule functions were defined inline, anonymously, using lambda functions. But the plural function doesn't care how they were defined; it just gets a list of rules and blindly works through them.

Now to add a new rule, all you need to do is define the functions directly in the rules list itself: one match rule, and one apply rule. But defining the rule functions inline like this makes it very clear that you have some unnecessary duplication here. You have four pairs of functions, and they all follow the same pattern. The match function is a single call to re.search, and the apply function is a single call to re.sub. Let's factor out these similarities.

17.5. plural.py, stage 4

Let's factor out the duplication in the code so that defining new rules can be easier.

Example 17.9. plural4.py

import re

def buildMatchAndApplyFunctions((pattern, search, replace)):  
    matchFunction = lambda word: re.search(pattern, word)      
    applyFunction = lambda word: re.sub(search, replace, word) 
    return (matchFunction, applyFunction)    

	buildMatchAndApplyFunctions is a function that builds other functions dynamically. It takes pattern, search and replace (actually it takes a tuple, but more on that in a minute), and you can build the match function using the lambda syntax to be a function that takes one parameter (word) and calls re.search with the pattern that was passed to the buildMatchAndApplyFunctions function, and the word that was passed to the match function you're building. Whoa.
	Building the apply function works the same way. The apply function is a function that takes one parameter, and calls re.sub with the search and replace parameters that were passed to the buildMatchAndApplyFunctions function, and the word that was passed to the apply function you're building. This technique of using the values of outside parameters within a dynamic function is called closures. You're essentially defining constants within the apply function you're building: it takes one parameter (word), but it then acts on that plus two other values (search and replace) which were set when you defined the apply function.
	Finally, the buildMatchAndApplyFunctions function returns a tuple of two values: the two functions you just created. The constants you defined within those functions (pattern within matchFunction, and search and replace within applyFunction) stay with those functions, even after you return from buildMatchAndApplyFunctions. That's insanely cool.

If this is incredibly confusing (and it should be, this is weird stuff), it may become clearer when you see how to use it.

Example 17.10. plural4.py continued

patterns = \
  (
    ('[sxz]$', '$', 'es'),
    ('[^aeioudgkprt]h$', '$', 'es'),
    ('(qu|[^aeiou])y$', 'y$', 'ies'),
    ('$', '$', 's')
  )             
rules = map(buildMatchAndApplyFunctions, patterns)  

Our pluralization rules are now defined as a series of strings (not functions). The first string is the regular expression that you would use in re.search to see if this rule matches; the second and third are the search and replace expressions you would use in re.sub to actually apply the rule to turn a noun into its plural.

This line is magic. It takes the list of strings in patterns and turns them into a list of functions. How? By mapping the strings to the buildMatchAndApplyFunctions function, which just happens to take three strings as parameters and return a tuple of two functions. This means that rules ends up being exactly the same as the previous example: a list of tuples, where each tuple is a pair of functions, where the first function is the match function that calls re.search, and the second function is the apply function that calls re.sub.

I swear I am not making this up: rules ends up with exactly the same list of functions as the previous example. Unroll the rules definition, and you'll get this:

Example 17.11. Unrolling the rules definition

rules = \
  (
    (
     lambda word: re.search('[sxz]$', word),
     lambda word: re.sub('$', 'es', word)
    ),
    (
     lambda word: re.search('[^aeioudgkprt]h$', word),
     lambda word: re.sub('$', 'es', word)
    ),
    (
     lambda word: re.search('[^aeiou]y$', word),
     lambda word: re.sub('y$', 'ies', word)
    ),
    (
     lambda word: re.search('$', word),
     lambda word: re.sub('$', 's', word)
    )
   )      

Example 17.12. plural4.py, finishing up

def plural(noun):                
    for matchesRule, applyRule in rules:            
        if matchesRule(noun):    
            return applyRule(noun)                 

Since the rules list is the same as the previous example, it should come as no surprise that the plural function hasn't changed. Remember, it's completely generic; it takes a list of rule functions and calls them in order. It doesn't care how the rules are defined. In stage 2, they were defined as seperate named functions. In stage 3, they were defined as anonymous lambda functions. Now in stage 4, they are built dynamically by mapping the buildMatchAndApplyFunctions function onto a list of raw strings. Doesn't matter; the plural function still works the same way.

Just in case that wasn't mind-blowing enough, I must confess that there was a subtlety in the definition of buildMatchAndApplyFunctions that I skipped over. Let's go back and take another look.

Example 17.13. Another look at buildMatchAndApplyFunctions

def buildMatchAndApplyFunctions((pattern, search, replace)):   

Notice the double parentheses? This function doesn't actually take three parameters; it actually takes one parameter, a tuple of three elements. But the tuple is expanded when the function is called, and the three elements of the tuple are each assigned to different variables: pattern, search, and replace. Confused yet? Let's see it in action.

Example 17.14. Expanding tuples when calling functions

>>> def foo((a, b, c)):
...     print c
...     print b
...     print a
>>> parameters = ('apple', 'bear', 'catnap')
>>> foo(parameters) 
catnap
bear
apple

The proper way to call the function foo is with a tuple of three elements. When the function is called, the elements are assigned to different local variables within foo.

Now let's go back and see why this auto-tuple-expansion trick was necessary. patterns was a list of tuples, and each tuple had three elements. When you called map(buildMatchAndApplyFunctions, patterns), that means that buildMatchAndApplyFunctions is not getting called with three parameters. Using map to map a single list onto a function always calls the function with a single parameter: each element of the list. In the case of patterns, each element of the list is a tuple, so buildMatchAndApplyFunctions always gets called with the tuple, and you use the auto-tuple-expansion trick in the definition of buildMatchAndApplyFunctions to assign the elements of that tuple to named variables that you can work with.

17.6. plural.py, stage 5

You've factored out all the duplicate code and added enough abstractions so that the pluralization rules are defined in a list of strings. The next logical step is to take these strings and put them in a separate file, where they can be maintained separately from the code that uses them.

First, let's create a text file that contains the rules you want. No fancy data structures, just space- (or tab-)delimited strings in three columns. You'll call it rules.en; “en” stands for English. These are the rules for pluralizing English nouns. You could add other rule files for other languages later.

Example 17.15. rules.en

[sxz]$$               es
[^aeioudgkprt]h$        $               es
[^aeiou]y$              y$              ies
$     $               s

Now let's see how you can use this rules file.

Example 17.16. plural5.py

import re
import string               

def buildRule((pattern, search, replace)):    
    return lambda word: re.search(pattern, word) and re.sub(search, replace, word) 

def plural(noun, language='en'):           
    lines = file('rules.%s' % language).readlines()          
    patterns = map(string.split, lines)    
    rules = map(buildRule, patterns)       
    for rule in rules:  
        result = rule(noun)                
        if result: return result          

	You're still using the closures technique here (building a function dynamically that uses variables defined outside the function), but now you've combined the separate match and apply functions into one. (The reason for this change will become clear in the next section.) This will let you accomplish the same thing as having two functions, but you'll need to call it differently, as you'll see in a minute.
	Our plural function now takes an optional second parameter, language, which defaults to en.
	You use the language parameter to construct a filename, then open the file and read the contents into a list. If language is en, then you'll open the rules.en file, read the entire thing, break it up by carriage returns, and return a list. Each line of the file will be one element in the list.
	As you saw, each line in the file really has three values, but they're separated by whitespace (tabs or spaces, it makes no difference). Mapping the string.split function onto this list will create a new list where each element is a tuple of three strings. So a line like [sxz]$ $ es will be broken up into the tuple ('[sxz]$', '$', 'es'). This means that patterns will end up as a list of tuples, just like you hard-coded it in stage 4.
	If patterns is a list of tuples, then rules will be a list of the functions created dynamically by each call to buildRule. Calling buildRule(('[sxz]$', '$', 'es')) returns a function that takes a single parameter, word. When this returned function is called, it will execute re.search('[sxz]$', word) and re.sub('$', 'es', word).
	Because you're now building a combined match-and-apply function, you need to call it differently. Just call the function, and if it returns something, then that's the plural; if it returns nothing (None), then the rule didn't match and you need to try another rule.

So the improvement here is that you've completely separated the pluralization rules into an external file. Not only can the file be maintained separately from the code, but you've set up a naming scheme where the same plural function can use different rule files, based on the language parameter.

The downside here is that you're reading that file every time you call the plural function. I thought I could get through this entire book without using the phrase “left as an exercise for the reader”, but here you go: building a caching mechanism for the language-specific rule files that auto-refreshes itself if the rule files change between calls is left as an exercise for the reader. Have fun.

17.7. plural.py, stage 6

Now you're ready to talk about generators.

Example 17.17. plural6.py

import re

def rules(language):           
    for line in file('rules.%s' % language):     
        pattern, search, replace = line.split()  
        yield lambda word: re.search(pattern, word) and re.sub(search, replace, word)

def plural(noun, language='en'):      
    for applyRule in rules(language): 
        result = applyRule(noun)      
        if result: return result      

This uses a technique called generators, which I'm not even going to try to explain until you look at a simpler example first.

Example 17.18. Introducing generators

>>> def make_counter(x):
...     print 'entering make_counter'
...     while 1:
...         yield x               
...         print 'incrementing x'
...         x = x + 1
...     
>>> counter = make_counter(2) 
>>> counter 
<generator object at 0x001C9C10>
>>> counter.next()            
entering make_counter
2
>>> counter.next()            
incrementing x
3
>>> counter.next()            
incrementing x
4

	The presence of the yield keyword in make_counter means that this is not a normal function. It is a special kind of function which generates values one at a time. You can think of it as a resumable function. Calling it will return a generator that can be used to generate successive values of x.
	To create an instance of the make_counter generator, just call it like any other function. Note that this does not actually execute the function code. You can tell this because the first line of make_counter is a print statement, but nothing has been printed yet.
	The make_counter function returns a generator object.
	The first time you call the next() method on the generator object, it executes the code in make_counter up to the first yield statement, and then returns the value that was yielded. In this case, that will be 2, because you originally created the generator by calling make_counter(2).
	Repeatedly calling next() on the generator object resumes where you left off and continues until you hit the next yield statement. The next line of code waiting to be executed is the print statement that prints incrementing x, and then after that the x = x + 1 statement that actually increments it. Then you loop through the while loop again, and the first thing you do is yield x, which returns the current value of x (now 3).
	The second time you call counter.next(), you do all the same things again, but this time x is now 4. And so forth. Since make_counter sets up an infinite loop, you could theoretically do this forever, and it would just keep incrementing x and spitting out values. But let's look at more productive uses of generators instead.

Example 17.19. Using generators instead of recursion

def fibonacci(max):
    a, b = 0, 1       
    while a < max:
        yield a       
        a, b = b, a+b 

	The Fibonacci sequence is a sequence of numbers where each number is the sum of the two numbers before it. It starts with 0 and 1, goes up slowly at first, then more and more rapidly. To start the sequence, you need two variables: a starts at 0, and b starts at 1.
	a is the current number in the sequence, so yield it.
	b is the next number in the sequence, so assign that to a, but also calculate the next value (a+b) and assign that to b for later use. Note that this happens in parallel; if a is 3 and b is 5, then a, b = b, a+b will set a to 5 (the previous value of b) and b to 8 (the sum of the previous values of a and b).

So you have a function that spits out successive Fibonacci numbers. Sure, you could do that with recursion, but this way is easier to read. Also, it works well with for loops.

Example 17.20. Generators in for loops

>>> for n in fibonacci(1000): 
...     print n,              
0 1 1 2 3 5 8 13 21 34 55 89 144 233 377 610 987

	You can use a generator like fibonacci in a for loop directly. The for loop will create the generator object and successively call the next() method to get values to assign to the for loop index variable (n).
	Each time through the for loop, n gets a new value from the yield statement in fibonacci, and all you do is print it out. Once fibonacci runs out of numbers (a gets bigger than max, which in this case is 1000), then the for loop exits gracefully.

OK, let's go back to the plural function and see how you're using this.

Example 17.21. Generators that generate dynamic functions

def rules(language):           
    for line in file('rules.%s' % language):      
        pattern, search, replace = line.split()   
        yield lambda word: re.search(pattern, word) and re.sub(search, replace, word) 

def plural(noun, language='en'):      
    for applyRule in rules(language):  
        result = applyRule(noun)      
        if result: return result      

	for line in file(...) is a common idiom for reading lines from a file, one line at a time. It works because file actually returns a generator whose next() method returns the next line of the file. That is so insanely cool, I wet myself just thinking about it.
	No magic here. Remember that the lines of the rules file have three values separated by whitespace, so line.split() returns a tuple of 3 values, and you assign those values to 3 local variables.
	And then you yield. What do you yield? A function, built dynamically with lambda, that is actually a closure (it uses the local variables pattern, search, and replace as constants). In other words, rules is a generator that spits out rule functions.
	Since rules is a generator, you can use it directly in a for loop. The first time through the for loop, you will call the rules function, which will open the rules file, read the first line out of it, dynamically build a function that matches and applies the first rule defined in the rules file, and yields the dynamically built function. The second time through the for loop, you will pick up where you left off in rules (which was in the middle of the for line in file(...) loop), read the second line of the rules file, dynamically build another function that matches and applies the second rule defined in the rules file, and yields it. And so forth.

What have you gained over stage 5? In stage 5, you read the entire rules file and built a list of all the possible rules before you even tried the first one. Now with generators, you can do everything lazily: you open the first and read the first rule and create a function to try it, but if that works you don't ever read the rest of the file or create any other functions.

Further reading

• PEP 255 defines generators.
• Python Cookbook has many more examples of generators.

17.8. Summary

You talked about several different advanced techniques in this chapter. Not all of them are appropriate for every situation.

You should now be comfortable with all of these techniques:

• Performing string substitution with regular expressions.
• Treating functions as objects, storing them in lists, assigning them to variables, and calling them through those variables.
• Building dynamic functions with lambda.
• Building closures, dynamic functions that contain surrounding variables as constants.
• Building generators, resumable functions that perform incremental logic and return different values each time you call them.

Adding abstractions, building functions dynamically, building closures, and using generators can all make your code simpler, more readable, and more flexible. But they can also end up making it more difficult to debug later. It's up to you to find the right balance between simplicity and power.

Chapter 18. Performance Tuning

Performance tuning is a many-splendored thing. Just because Python is an interpreted language doesn't mean you shouldn't worry about code optimization. But don't worry about it too much.

18.1. Diving in

There are so many pitfalls involved in optimizing your code, it's hard to know where to start.

Let's start here: are you sure you need to do it at all? Is your code really so bad? Is it worth the time to tune it? Over the lifetime of your application, how much time is going to be spent running that code, compared to the time spent waiting for a remote database server, or waiting for user input?

Second, are you sure you're done coding? Premature optimization is like spreading frosting on a half-baked cake. You spend hours or days (or more) optimizing your code for performance, only to discover it doesn't do what you need it to do. That's time down the drain.

This is not to say that code optimization is worthless, but you need to look at the whole system and decide whether it's the best use of your time. Every minute you spend optimizing code is a minute you're not spending adding new features, or writing documentation, or playing with your kids, or writing unit tests.

Oh yes, unit tests. It should go without saying that you need a complete set of unit tests before you begin performance tuning. The last thing you need is to introduce new bugs while fiddling with your algorithms.

With these caveats in place, let's look at some techniques for optimizing Python code. The code in question is an implementation of the Soundex algorithm. Soundex was a method used in the early 20th century for categorizing surnames in the United States census. It grouped similar-sounding names together, so even if a name was misspelled, researchers had a chance of finding it. Soundex is still used today for much the same reason, although of course we use computerized database servers now. Most database servers include a Soundex function.

There are several subtle variations of the Soundex algorithm. This is the one used in this chapter:

1. Keep the first letter of the name as-is.
2. Convert the remaining letters to digits, according to a specific table:
   • B, F, P, and V become 1.
   • C, G, J, K, Q, S, X, and Z become 2.
   • D and T become 3.
   • L becomes 4.
   • M and N become 5.
   • R becomes 6.
   • All other letters become 9.
3. Remove consecutive duplicates.
4. Remove all 9s altogether.
5. If the result is shorter than four characters (the first letter plus three digits), pad the result with trailing zeros.
6. if the result is longer than four characters, discard everything after the fourth character.

For example, my name, Pilgrim, becomes P942695. That has no consecutive duplicates, so nothing to do there. Then you remove the 9s, leaving P4265. That's too long, so you discard the excess character, leaving P426.

Another example: Woo becomes W99, which becomes W9, which becomes W, which gets padded with zeros to become W000.

Here's a first attempt at a Soundex function:

Example 18.1. soundex/stage1/soundex1a.py

If you have not already done so, you can download this and other examples used in this book.

import string, re

charToSoundex = {"A": "9",
                 "B": "1",
                 "C": "2",
                 "D": "3",
                 "E": "9",
                 "F": "1",
                 "G": "2",
                 "H": "9",
                 "I": "9",
                 "J": "2",
                 "K": "2",
                 "L": "4",
                 "M": "5",
                 "N": "5",
                 "O": "9",
                 "P": "1",
                 "Q": "2",
                 "R": "6",
                 "S": "2",
                 "T": "3",
                 "U": "9",
                 "V": "1",
                 "W": "9",
                 "X": "2",
                 "Y": "9",
                 "Z": "2"}

def soundex(source):
    "convert string to Soundex equivalent"

    # Soundex requirements:
    # source string must be at least 1 character
    # and must consist entirely of letters
    allChars = string.uppercase + string.lowercase
    if not re.search('^[%s]+$' % allChars, source):
        return "0000"

    # Soundex algorithm:
    # 1. make first character uppercase
    source = source[0].upper() + source[1:]
    
    # 2. translate all other characters to Soundex digits
    digits = source[0]
    for s in source[1:]:
        s = s.upper()
        digits += charToSoundex[s]

    # 3. remove consecutive duplicates
    digits2 = digits[0]
    for d in digits[1:]:
        if digits2[-1] != d:
            digits2 += d
        
    # 4. remove all "9"s
    digits3 = re.sub('9', '', digits2)
    
    # 5. pad end with "0"s to 4 characters
    while len(digits3) < 4:
        digits3 += "0"
        
    # 6. return first 4 characters
    return digits3[:4]

if __name__ == '__main__':
    from timeit import Timer
    names = ('Woo', 'Pilgrim', 'Flingjingwaller')
    for name in names:
        statement = "soundex('%s')" % name
        t = Timer(statement, "from __main__ import soundex")
        print name.ljust(15), soundex(name), min(t.repeat())

Further Reading on Soundex

• Soundexing and Genealogy gives a chronology of the evolution of the Soundex and its regional variations.

18.2. Using the timeit Module

The most important thing you need to know about optimizing Python code is that you shouldn't write your own timing function.

Timing short pieces of code is incredibly complex. How much processor time is your computer devoting to running this code? Are there things running in the background? Are you sure? Every modern computer has background processes running, some all the time, some intermittently. Cron jobs fire off at consistent intervals; background services occasionally “wake up” to do useful things like check for new mail, connect to instant messaging servers, check for application updates, scan for viruses, check whether a disk has been inserted into your CD drive in the last 100 nanoseconds, and so on. Before you start your timing tests, turn everything off and disconnect from the network. Then turn off all the things you forgot to turn off the first time, then turn off the service that's incessantly checking whether the network has come back yet, then ...

And then there's the matter of the variations introduced by the timing framework itself. Does the Python interpreter cache method name lookups? Does it cache code block compilations? Regular expressions? Will your code have side effects if run more than once? Don't forget that you're dealing with small fractions of a second, so small mistakes in your timing framework will irreparably skew your results.

The Python community has a saying: “Python comes with batteries included.” Don't write your own timing framework. Python 2.3 comes with a perfectly good one called timeit.

Example 18.2. Introducing timeit

If you have not already done so, you can download this and other examples used in this book.

>>> import timeit
>>> t = timeit.Timer("soundex.soundex('Pilgrim')",
...     "import soundex")   
>>> t.timeit()              
8.21683733547
>>> t.repeat(3, 2000000)    
[16.48319309109, 16.46128984923, 16.44203948912]

	The timeit module defines one class, Timer, which takes two arguments. Both arguments are strings. The first argument is the statement you wish to time; in this case, you are timing a call to the Soundex function within the soundex with an argument of 'Pilgrim'. The second argument to the Timer class is the import statement that sets up the environment for the statement. Internally, timeit sets up an isolated virtual environment, manually executes the setup statement (importing the soundex module), then manually compiles and executes the timed statement (calling the Soundex function).
	Once you have the Timer object, the easiest thing to do is call timeit(), which calls your function 1 million times and returns the number of seconds it took to do it.
	The other major method of the Timer object is repeat(), which takes two optional arguments. The first argument is the number of times to repeat the entire test, and the second argument is the number of times to call the timed statement within each test. Both arguments are optional, and they default to 3 and 1000000 respectively. The repeat() method returns a list of the times each test cycle took, in seconds.

	You can use the timeit module on the command line to test an existing Python program, without modifying the code. See http://docs.python.org/lib/node396.html for documentation on the command-line flags.

Note that repeat() returns a list of times. The times will almost never be identical, due to slight variations in how much processor time the Python interpreter is getting (and those pesky background processes that you can't get rid of). Your first thought might be to say “Let's take the average and call that The True Number.”

In fact, that's almost certainly wrong. The tests that took longer didn't take longer because of variations in your code or in the Python interpreter; they took longer because of those pesky background processes, or other factors outside of the Python interpreter that you can't fully eliminate. If the different timing results differ by more than a few percent, you still have too much variability to trust the results. Otherwise, take the minimum time and discard the rest.

Python has a handy min function that takes a list and returns the smallest value:

>>> min(t.repeat(3, 1000000))
8.22203948912

	The timeit module only works if you already know what piece of code you need to optimize. If you have a larger Python program and don't know where your performance problems are, check out the hotshot module.

18.3. Optimizing Regular Expressions

The first thing the Soundex function checks is whether the input is a non-empty string of letters. What's the best way to do this?

If you answered “regular expressions”, go sit in the corner and contemplate your bad instincts. Regular expressions are almost never the right answer; they should be avoided whenever possible. Not only for performance reasons, but simply because they're difficult to debug and maintain. Also for performance reasons.

This code fragment from soundex/stage1/soundex1a.py checks whether the function argument source is a word made entirely of letters, with at least one letter (not the empty string):

    allChars = string.uppercase + string.lowercase
    if not re.search('^[%s]+$' % allChars, source):
        return "0000"

How does soundex1a.py perform? For convenience, the __main__ section of the script contains this code that calls the timeit module, sets up a timing test with three different names, tests each name three times, and displays the minimum time for each:

if __name__ == '__main__':
    from timeit import Timer
    names = ('Woo', 'Pilgrim', 'Flingjingwaller')
    for name in names:
        statement = "soundex('%s')" % name
        t = Timer(statement, "from __main__ import soundex")
        print name.ljust(15), soundex(name), min(t.repeat())

So how does soundex1a.py perform with this regular expression?

C:\samples\soundex\stage1>python soundex1a.py
Woo             W000 19.3356647283
Pilgrim         P426 24.0772053431
Flingjingwaller F452 35.0463220884

As you might expect, the algorithm takes significantly longer when called with longer names. There will be a few things we can do to narrow that gap (make the function take less relative time for longer input), but the nature of the algorithm dictates that it will never run in constant time.

The other thing to keep in mind is that we are testing a representative sample of names. Woo is a kind of trivial case, in that it gets shorted down to a single letter and then padded with zeros. Pilgrim is a normal case, of average length and a mixture of significant and ignored letters. Flingjingwaller is extraordinarily long and contains consecutive duplicates. Other tests might also be helpful, but this hits a good range of different cases.

So what about that regular expression? Well, it's inefficient. Since the expression is testing for ranges of characters (A-Z in uppercase, and a-z in lowercase), we can use a shorthand regular expression syntax. Here is soundex/stage1/soundex1b.py:

    if not re.search('^[A-Za-z]+$', source):
        return "0000"

timeit says soundex1b.py is slightly faster than soundex1a.py, but nothing to get terribly excited about:

C:\samples\soundex\stage1>python soundex1b.py
Woo             W000 17.1361133887
Pilgrim         P426 21.8201693232
Flingjingwaller F452 32.7262294509

We saw in Section 15.3, “Refactoring” that regular expressions can be compiled and reused for faster results. Since this regular expression never changes across function calls, we can compile it once and use the compiled version. Here is soundex/stage1/soundex1c.py:

isOnlyChars = re.compile('^[A-Za-z]+$').search
def soundex(source):
    if not isOnlyChars(source):
        return "0000"

Using a compiled regular expression in soundex1c.py is significantly faster:

C:\samples\soundex\stage1>python soundex1c.py
Woo             W000 14.5348347346
Pilgrim         P426 19.2784703084
Flingjingwaller F452 30.0893873383

But is this the wrong path? The logic here is simple: the input source needs to be non-empty, and it needs to be composed entirely of letters. Wouldn't it be faster to write a loop checking each character, and do away with regular expressions altogether?

Here is soundex/stage1/soundex1d.py:

    if not source:
        return "0000"
    for c in source:
        if not ('A' <= c <= 'Z') and not ('a' <= c <= 'z'):
            return "0000"

It turns out that this technique in soundex1d.py is not faster than using a compiled regular expression (although it is faster than using a non-compiled regular expression):

C:\samples\soundex\stage1>python soundex1d.py
Woo             W000 15.4065058548
Pilgrim         P426 22.2753567842
Flingjingwaller F452 37.5845122774

Why isn't soundex1d.py faster? The answer lies in the interpreted nature of Python. The regular expression engine is written in C, and compiled to run natively on your computer. On the other hand, this loop is written in Python, and runs through the Python interpreter. Even though the loop is relatively simple, it's not simple enough to make up for the overhead of being interpreted. Regular expressions are never the right answer... except when they are.

It turns out that Python offers an obscure string method. You can be excused for not knowing about it, since it's never been mentioned in this book. The method is called isalpha(), and it checks whether a string contains only letters.

This is soundex/stage1/soundex1e.py:

    if (not source) and (not source.isalpha()):
        return "0000"

How much did we gain by using this specific method in soundex1e.py? Quite a bit.

C:\samples\soundex\stage1>python soundex1e.py
Woo             W000 13.5069504644
Pilgrim         P426 18.2199394057
Flingjingwaller F452 28.9975225902

Example 18.3. Best Result So Far: soundex/stage1/soundex1e.py

import string, re

charToSoundex = {"A": "9",
                 "B": "1",
                 "C": "2",
                 "D": "3",
                 "E": "9",
                 "F": "1",
                 "G": "2",
                 "H": "9",
                 "I": "9",
                 "J": "2",
                 "K": "2",
                 "L": "4",
                 "M": "5",
                 "N": "5",
                 "O": "9",
                 "P": "1",
                 "Q": "2",
                 "R": "6",
                 "S": "2",
                 "T": "3",
                 "U": "9",
                 "V": "1",
                 "W": "9",
                 "X": "2",
                 "Y": "9",
                 "Z": "2"}

def soundex(source):
    if (not source) and (not source.isalpha()):
        return "0000"
    source = source[0].upper() + source[1:]
    digits = source[0]
    for s in source[1:]:
        s = s.upper()
        digits += charToSoundex[s]
    digits2 = digits[0]
    for d in digits[1:]:
        if digits2[-1] != d:
            digits2 += d
    digits3 = re.sub('9', '', digits2)
    while len(digits3) < 4:
        digits3 += "0"
    return digits3[:4]

if __name__ == '__main__':
    from timeit import Timer
    names = ('Woo', 'Pilgrim', 'Flingjingwaller')
    for name in names:
        statement = "soundex('%s')" % name
        t = Timer(statement, "from __main__ import soundex")
        print name.ljust(15), soundex(name), min(t.repeat())

18.4. Optimizing Dictionary Lookups

The second step of the Soundex algorithm is to convert characters to digits in a specific pattern. What's the best way to do this?

The most obvious solution is to define a dictionary with individual characters as keys and their corresponding digits as values, and do dictionary lookups on each character. This is what we have in soundex/stage1/soundex1c.py (the current best result so far):

charToSoundex = {"A": "9",
                 "B": "1",
                 "C": "2",
                 "D": "3",
                 "E": "9",
                 "F": "1",
                 "G": "2",
                 "H": "9",
                 "I": "9",
                 "J": "2",
                 "K": "2",
                 "L": "4",
                 "M": "5",
                 "N": "5",
                 "O": "9",
                 "P": "1",
                 "Q": "2",
                 "R": "6",
                 "S": "2",
                 "T": "3",
                 "U": "9",
                 "V": "1",
                 "W": "9",
                 "X": "2",
                 "Y": "9",
                 "Z": "2"}

def soundex(source):
    # ... input check omitted for brevity ...
    source = source[0].upper() + source[1:]
    digits = source[0]
    for s in source[1:]:
        s = s.upper()
        digits += charToSoundex[s]

You timed soundex1c.py already; this is how it performs:

C:\samples\soundex\stage1>python soundex1c.py
Woo             W000 14.5341678901
Pilgrim         P426 19.2650071448
Flingjingwaller F452 30.1003563302

This code is straightforward, but is it the best solution? Calling upper() on each individual character seems inefficient; it would probably be better to call upper() once on the entire string.

Then there's the matter of incrementally building the digits string. Incrementally building strings like this is horribly inefficient; internally, the Python interpreter needs to create a new string each time through the loop, then discard the old one.

Python is good at lists, though. It can treat a string as a list of characters automatically. And lists are easy to combine into strings again, using the string method join().

Here is soundex/stage2/soundex2a.py, which converts letters to digits by using ↦ and lambda:

def soundex(source):
    # ...
    source = source.upper()
    digits = source[0] + "".join(map(lambda c: charToSoundex[c], source[1:]))

Surprisingly, soundex2a.py is not faster:

C:\samples\soundex\stage2>python soundex2a.py
Woo             W000 15.0097526362
Pilgrim         P426 19.254806407
Flingjingwaller F452 29.3790847719

The overhead of the anonymous lambda function kills any performance you gain by dealing with the string as a list of characters.

soundex/stage2/soundex2b.py uses a list comprehension instead of ↦ and lambda:

    source = source.upper()
    digits = source[0] + "".join([charToSoundex[c] for c in source[1:]])

Using a list comprehension in soundex2b.py is faster than using ↦ and lambda in soundex2a.py, but still not faster than the original code (incrementally building a string in soundex1c.py):

C:\samples\soundex\stage2>python soundex2b.py
Woo             W000 13.4221324219
Pilgrim         P426 16.4901234654
Flingjingwaller F452 25.8186157738

It's time for a radically different approach. Dictionary lookups are a general purpose tool. Dictionary keys can be any length string (or many other data types), but in this case we are only dealing with single-character keys and single-character values. It turns out that Python has a specialized function for handling exactly this situation: the string.maketrans function.

This is soundex/stage2/soundex2c.py:

allChar = string.uppercase + string.lowercase
charToSoundex = string.maketrans(allChar, "91239129922455912623919292" * 2)
def soundex(source):
    # ...
    digits = source[0].upper() + source[1:].translate(charToSoundex)

What the heck is going on here? string.maketrans creates a translation matrix between two strings: the first argument and the second argument. In this case, the first argument is the string ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz, and the second argument is the string 9123912992245591262391929291239129922455912623919292. See the pattern? It's the same conversion pattern we were setting up longhand with a dictionary. A maps to 9, B maps to 1, C maps to 2, and so forth. But it's not a dictionary; it's a specialized data structure that you can access using the string method translate, which translates each character into the corresponding digit, according to the matrix defined by string.maketrans.

timeit shows that soundex2c.py is significantly faster than defining a dictionary and looping through the input and building the output incrementally:

C:\samples\soundex\stage2>python soundex2c.py
Woo             W000 11.437645008
Pilgrim         P426 13.2825062962
Flingjingwaller F452 18.5570110168

You're not going to get much better than that. Python has a specialized function that does exactly what you want to do; use it and move on.

Example 18.4. Best Result So Far: soundex/stage2/soundex2c.py

import string, re

allChar = string.uppercase + string.lowercase
charToSoundex = string.maketrans(allChar, "91239129922455912623919292" * 2)
isOnlyChars = re.compile('^[A-Za-z]+$').search

def soundex(source):
    if not isOnlyChars(source):
        return "0000"
    digits = source[0].upper() + source[1:].translate(charToSoundex)
    digits2 = digits[0]
    for d in digits[1:]:
        if digits2[-1] != d:
            digits2 += d
    digits3 = re.sub('9', '', digits2)
    while len(digits3) < 4:
        digits3 += "0"
    return digits3[:4]

if __name__ == '__main__':
    from timeit import Timer
    names = ('Woo', 'Pilgrim', 'Flingjingwaller')
    for name in names:
        statement = "soundex('%s')" % name
        t = Timer(statement, "from __main__ import soundex")
        print name.ljust(15), soundex(name), min(t.repeat())

18.5. Optimizing List Operations

The third step in the Soundex algorithm is eliminating consecutive duplicate digits. What's the best way to do this?

Here's the code we have so far, in soundex/stage2/soundex2c.py:

    digits2 = digits[0]
    for d in digits[1:]:
        if digits2[-1] != d:
            digits2 += d

Here are the performance results for soundex2c.py:

C:\samples\soundex\stage2>python soundex2c.py
Woo             W000 12.6070768771
Pilgrim         P426 14.4033353401
Flingjingwaller F452 19.7774882003

The first thing to consider is whether it's efficient to check digits[-1] each time through the loop. Are list indexes expensive? Would we be better off maintaining the last digit in a separate variable, and checking that instead?

To answer this question, here is soundex/stage3/soundex3a.py:

    digits2 = ''
    last_digit = ''
    for d in digits:
        if d != last_digit:
            digits2 += d
            last_digit = d

soundex3a.py does not run any faster than soundex2c.py, and may even be slightly slower (although it's not enough of a difference to say for sure):

C:\samples\soundex\stage3>python soundex3a.py
Woo             W000 11.5346048171
Pilgrim         P426 13.3950636184
Flingjingwaller F452 18.6108927252

Why isn't soundex3a.py faster? It turns out that list indexes in Python are extremely efficient. Repeatedly accessing digits2[-1] is no problem at all. On the other hand, manually maintaining the last seen digit in a separate variable means we have two variable assignments for each digit we're storing, which wipes out any small gains we might have gotten from eliminating the list lookup.

Let's try something radically different. If it's possible to treat a string as a list of characters, it should be possible to use a list comprehension to iterate through the list. The problem is, the code needs access to the previous character in the list, and that's not easy to do with a straightforward list comprehension.

However, it is possible to create a list of index numbers using the built-in range() function, and use those index numbers to progressively search through the list and pull out each character that is different from the previous character. That will give you a list of characters, and you can use the string method join() to reconstruct a string from that.

Here is soundex/stage3/soundex3b.py:

    digits2 = "".join([digits[i] for i in range(len(digits))
     if i == 0 or digits[i-1] != digits[i]])

Is this faster? In a word, no.

C:\samples\soundex\stage3>python soundex3b.py
Woo             W000 14.2245271396
Pilgrim         P426 17.8337165757
Flingjingwaller F452 25.9954005327

It's possible that the techniques so far as have been “string-centric”. Python can convert a string into a list of characters with a single command: list('abc') returns ['a', 'b', 'c']. Furthermore, lists can be modified in place very quickly. Instead of incrementally building a new list (or string) out of the source string, why not move elements around within a single list?

Here is soundex/stage3/soundex3c.py, which modifies a list in place to remove consecutive duplicate elements:

    digits = list(source[0].upper() + source[1:].translate(charToSoundex))
    i=0
    for item in digits:
        if item==digits[i]: continue
        i+=1
        digits[i]=item
    del digits[i+1:]
    digits2 = "".join(digits)

Is this faster than soundex3a.py or soundex3b.py? No, in fact it's the slowest method yet:

C:\samples\soundex\stage3>python soundex3c.py
Woo             W000 14.1662554878
Pilgrim         P426 16.0397885765
Flingjingwaller F452 22.1789341942

We haven't made any progress here at all, except to try and rule out several “clever” techniques. The fastest code we've seen so far was the original, most straightforward method (soundex2c.py). Sometimes it doesn't pay to be clever.

Example 18.5. Best Result So Far: soundex/stage2/soundex2c.py

import string, re

allChar = string.uppercase + string.lowercase
charToSoundex = string.maketrans(allChar, "91239129922455912623919292" * 2)
isOnlyChars = re.compile('^[A-Za-z]+$').search

def soundex(source):
    if not isOnlyChars(source):
        return "0000"
    digits = source[0].upper() + source[1:].translate(charToSoundex)
    digits2 = digits[0]
    for d in digits[1:]:
        if digits2[-1] != d:
            digits2 += d
    digits3 = re.sub('9', '', digits2)
    while len(digits3) < 4:
        digits3 += "0"
    return digits3[:4]

if __name__ == '__main__':
    from timeit import Timer
    names = ('Woo', 'Pilgrim', 'Flingjingwaller')
    for name in names:
        statement = "soundex('%s')" % name
        t = Timer(statement, "from __main__ import soundex")
        print name.ljust(15), soundex(name), min(t.repeat())

18.6. Optimizing String Manipulation

The final step of the Soundex algorithm is padding short results with zeros, and truncating long results. What is the best way to do this?

This is what we have so far, taken from soundex/stage2/soundex2c.py:

    digits3 = re.sub('9', '', digits2)
    while len(digits3) < 4:
        digits3 += "0"
    return digits3[:4]

These are the results for soundex2c.py:

C:\samples\soundex\stage2>python soundex2c.py
Woo             W000 12.6070768771
Pilgrim         P426 14.4033353401
Flingjingwaller F452 19.7774882003

The first thing to consider is replacing that regular expression with a loop. This code is from soundex/stage4/soundex4a.py:

    digits3 = ''
    for d in digits2:
        if d != '9':
            digits3 += d

Is soundex4a.py faster? Yes it is:

C:\samples\soundex\stage4>python soundex4a.py
Woo             W000 6.62865531792
Pilgrim         P426 9.02247576158
Flingjingwaller F452 13.6328416042

But wait a minute. A loop to remove characters from a string? We can use a simple string method for that. Here's soundex/stage4/soundex4b.py:

    digits3 = digits2.replace('9', '')

Is soundex4b.py faster? That's an interesting question. It depends on the input:

C:\samples\soundex\stage4>python soundex4b.py
Woo             W000 6.75477414029
Pilgrim         P426 7.56652144337
Flingjingwaller F452 10.8727729362

The string method in soundex4b.py is faster than the loop for most names, but it's actually slightly slower than soundex4a.py in the trivial case (of a very short name). Performance optimizations aren't always uniform; tuning that makes one case faster can sometimes make other cases slower. In this case, the majority of cases will benefit from the change, so let's leave it at that, but the principle is an important one to remember.

Last but not least, let's examine the final two steps of the algorithm: padding short results with zeros, and truncating long results to four characters. The code you see in soundex4b.py does just that, but it's horribly inefficient. Take a look at soundex/stage4/soundex4c.py to see why:

    digits3 += '000'
    return digits3[:4]

Why do we need a while loop to pad out the result? We know in advance that we're going to truncate the result to four characters, and we know that we already have at least one character (the initial letter, which is passed unchanged from the original source variable). That means we can simply add three zeros to the output, then truncate it. Don't get stuck in a rut over the exact wording of the problem; looking at the problem slightly differently can lead to a simpler solution.

How much speed do we gain in soundex4c.py by dropping the while loop? It's significant:

C:\samples\soundex\stage4>python soundex4c.py
Woo             W000 4.89129791636
Pilgrim         P426 7.30642134685
Flingjingwaller F452 10.689832367

Finally, there is still one more thing you can do to these three lines of code to make them faster: you can combine them into one line. Take a look at soundex/stage4/soundex4d.py:

    return (digits2.replace('9', '') + '000')[:4]

Putting all this code on one line in soundex4d.py is barely faster than soundex4c.py:

C:\samples\soundex\stage4>python soundex4d.py
Woo             W000 4.93624105857
Pilgrim         P426 7.19747593619
Flingjingwaller F452 10.5490700634

It is also significantly less readable, and for not much performance gain. Is that worth it? I hope you have good comments. Performance isn't everything. Your optimization efforts must always be balanced against threats to your program's readability and maintainability.

18.7. Summary

This chapter has illustrated several important aspects of performance tuning in Python, and performance tuning in general.

• If you need to choose between regular expressions and writing a loop, choose regular expressions. The regular expression engine is compiled in C and runs natively on your computer; your loop is written in Python and runs through the Python interpreter.
• If you need to choose between regular expressions and string methods, choose string methods. Both are compiled in C, so choose the simpler one.
• General-purpose dictionary lookups are fast, but specialtiy functions such as string.maketrans and string methods such as isalpha() are faster. If Python has a custom-tailored function for you, use it.
• Don't be too clever. Sometimes the most obvious algorithm is also the fastest.
• Don't sweat it too much. Performance isn't everything.

I can't emphasize that last point strongly enough. Over the course of this chapter, you made this function three times faster and saved 20 seconds over 1 million function calls. Great. Now think: over the course of those million function calls, how many seconds will your surrounding application wait for a database connection? Or wait for disk I/O? Or wait for user input? Don't spend too much time over-optimizing one algorithm, or you'll ignore obvious improvements somewhere else. Develop an instinct for the sort of code that Python runs well, correct obvious blunders if you find them, and leave the rest alone.