The Artima Developer Community
Sponsored Link

The Explorer
Generic functions vs mixins: a case study
by Michele Simionato
September 3, 2008
Just yesterday at work I had a good real-life use case for generic functions which deserved a blog post.


In the last few weeks my collegues and me have been involved in a project which required a command line interface. We did so by leveraging on the cmd module in the standard Python library, to which we added a network layer using Twisted. In the end, we had classes interacting with the standard streams stdin, stdout, stderr and classes interacting with nonstandard streams such as Twisted transports. All the I/O was line oriented and we basically needed three methods:

Depending on the type of self, self.stdout was sys.stdout, a Twisted transport, a log file or a file-like wrapper to a database. Likewise for self.stderr and self.stdin.

This is a problem that begs for generic functions. Unfortunately, nobody in the Python world uses them (with the exception of P. J. Eby) so for the moment we are using a suboptimal design involving mixins instead. I am not really happy with that. The aim of this blog post is to explain why a mixin solution is inferior to a generic functions solution.

A mixin-oriented solution

In the mixin solution, instead of generic functions one uses plain old methods, stored into a mixin class. In this specific case let me call the class StdIOMixin:

class StdIOMixin(object):
    "A mixin implementing line-oriented I/O"
    stdin = sys.stdin
    stdout = sys.stdout
    stderr = sys.stderr
    linesep = os.linesep

    def print_out(self, text, *args):
        "Write on self.stdout by flushing"
        write(self.stdout, str(text) + self.linesep, *args)

    def print_err(self, text, *args):
        "Write on self.stderr by flushing"
        write(self.stderr, str(text) + self.linesep, *args)

    def readln_in(self):
        "Read a line from self.stdin (without trailing newline) or None"
        line = self.stdin.readline()
        if line:
            return line[:-1] # strip trailing newline

where write is the following helper function:

def write(stream, text, *args):
    'Write on a stream by flushing if possible'
    if args: # when no args, do not consider '%' a special char
        text = text % args
    flush = getattr(stream, 'flush', False)
    if flush:

StdIOMixin is there to be mixed with other classes, providing them with the ability to perform line-oriented I/O. By default, it works on the standard streams, but if the client class overrides the attributes stdout, stderr, stdin with suitable file-like objects, it can be made to work with Twisted transports, files and databases. For instance, here is an example where stdout and stderr are overridden as files:

class FileIO(StdIOMixin):
    def __init__(self):
        self.stdout = file('out.txt', 'w')
        self.stderr = file('err.txt', 'w')
>>> FileIO().print_out('hello!') # prints a line on out.txt

The design works and it looks elegant, but still I say that it is sub-optimal compared to generic functions.

The basic problem of this design is that it adds methods to the client classes and therefore it adds to the learning curve. Suppose you have four client classes - one managing standard stream, one managing files, one managing Twisted transports and one managing database connections - then you have to add the mixin four times. If you generate the documentation for your classes, the methods print_out, print_err and readln_in will be documented four times. And this is not a shortcoming of pydoc: the three methods are effectively cluttering your application in a linear way, proportionally to the number of classes you have.

Moreover, those methods will add to the pollution of your class namespace, with the potential risk on name collisions, especially in large frameworks. In large frameworks (i.e. Plone, where a class my have 700+ attributes) this is a serious problem: for instance, you cannot even use auto-completion, since there are just too many completions. You must know that I am very sensitive to namespace pollution so I always favor approaches that can avoid it.

Also, suppose you only need the print_out functionality; the mixin approach naturally would invite you to include the entire StdIOMixin, importing in your class methods you don't need. The alternative would be to create three mixin classes StdinMixin, StdoutMixin, StderrMixin, but most of the time you would need all of them; it seems overkill to complicate so much your inheritance hierarchy for a very simple functionality.

As you may know, I am always looking for solutions avoiding (multiple) inheritance and generic functions fit the bill perfectly.

A generic functions solution

I am sure most people do not know about it, but Python 2.5 ships with an implementation of generic functions in the standard library, in the pkgutil module (by P.J. Eby). Currently, the implementation is only used internally in pkgutil and it is completely undocumented; therefore I never had the courage to use it in production, but it works well. Even if it is simple, it is able to cover most practical uses of generic functions. For instance, in our case we need three generic functions:

from pkgutil import simplegeneric

def print_out(self, text, *args):
    if args:
        text = text % args
    print >> self.stdout, text

def print_err(self, text, *args):
    if args:
        text = text % args
    print >> self.stderr, text

def readln_in(self):
    "Read a line from self.stdin (without trailing newline)"
    line = self.stdin.readline()
    if line:
        return line[:-1] # strip trailing newline

The power of generic functions is that you don't need to use inheritance: print_out will work on any object with a .stdout attribute even if it does not derive from StdIOMixin. For instance, if you define the class

class FileOut(object):
    def __init__(self):
        self.stdout = file('out.txt', 'w')

the following will print a message on the file out.txt:

>>> print_out(FileOut(), 'writing on file') # prints a line on out.txt

Simple, isn't it?

Extending generic functions

One advantage of methods with respect to ordinary functions is that they can be overridden in subclasses; however, generic functions can be overridden too - this is why they are also called multimethods. For instance, you could define a class AddTimeStamp and override print_out to add a time stamp when applied to instances of AddTimeStamp. Here is how you would do it:

class AddTimeStamp(object):
    stdout = sys.stdout
@print_out.register(AddTimeStamp) # add an implementation to print_out
def impl(self, text, *args):
    "Implementation of print_out for AddTimeStamp instances"
    if args:
        text = text % args
    print >> self.stdout,, text

and here in an example of use:

>>> print_out(AddTimeStamp(), 'writing on stdout')
2008-09-02T07:28:46.863932 writing on stdout

The syntax @print_out.register(AddTimeStamp) is not the most beatiful in the world, but its purposes should be clear: we are registering the implementation of print_out to be used for instances of AddTimeStamp. When print_out is invoked on an instance of AddTimeStamp a time stamp is printed; otherwise, the default implementation is used.

Notice that since the implementation of simplegeneric is simple, the internal registry of implementations is not exposed and there is no introspection API; moreover, simplegeneric works for single dispatch only and there is no explicit support for multimethod cooperation (i.e. call-next-method, for the ones familiar with Common Lisp). Yet, you cannot pretend too much from thirty lines of code ;)

In this example I have named the AddTimeStamp implementation of print_out impl, but you could have used any valid Python identifier, including print_out_AddTimeStamp or _, if you felt so. Since the name print_out is explicit in the decorator and since in practice you do not need to access the explicit implementation directly, I have settled for a generic name like impl. There is no standard convention since nobody uses generic functions in Python (yet).

There were plan to add generic functions to Python 3.0, but the proposal have been shifted to Python 3.1, with a syntax yet to define. Nevertheless, for people who don't want to wait, pkgutil.simplegeneric is already there and you can start experimenting with generic functions right now. Have fun!

Talk Back!

Have an opinion? Readers have already posted 11 comments about this weblog entry. Why not add yours?

RSS Feed

If you'd like to be notified whenever Michele Simionato adds a new entry to his weblog, subscribe to his RSS feed.

About the Blogger

Michele Simionato started his career as a Theoretical Physicist, working in Italy, France and the U.S. He turned to programming in 2003; since then he has been working professionally as a Python developer and now he lives in Milan, Italy. Michele is well known in the Python community for his posts in the newsgroup(s), his articles and his Open Source libraries and recipes. His interests include object oriented programming, functional programming, and in general programming metodologies that enable us to manage the complexity of modern software developement.

This weblog entry is Copyright © 2008 Michele Simionato. All rights reserved.

Sponsored Links


Copyright © 1996-2014 Artima, Inc. All Rights Reserved. - Privacy Policy - Terms of Use - Advertise with Us