source: trunk/essentials/dev-lang/python/Doc/howto/advocacy.tex@ 3315

\documentclass{howto}

\title{Python Advocacy HOWTO}

\release{0.03}

\author{A.M. Kuchling}
\authoraddress{\email{[email protected]}}

\begin{document}
\maketitle

\begin{abstract}
\noindent
It's usually difficult to get your management to accept open source
software, and Python is no exception to this rule.  This document
discusses reasons to use Python, strategies for winning acceptance,
facts and arguments you can use, and cases where you \emph{shouldn't}
try to use Python.

This document is available from the Python HOWTO page at
\url{http://www.python.org/doc/howto}.

\end{abstract}

\tableofcontents

\section{Reasons to Use Python}

There are several reasons to incorporate a scripting language into
your development process, and this section will discuss them, and why
Python has some properties that make it a particularly good choice.

 \subsection{Programmability}

Programs are often organized in a modular fashion.  Lower-level
operations are grouped together, and called by higher-level functions,
which may in turn be used as basic operations by still further upper
levels.  

For example, the lowest level might define a very low-level
set of functions for accessing a hash table.  The next level might use
hash tables to store the headers of a mail message, mapping a header
name like \samp{Date} to a value such as \samp{Tue, 13 May 1997
20:00:54 -0400}.  A yet higher level may operate on message objects,
without knowing or caring that message headers are stored in a hash
table, and so forth.  

Often, the lowest levels do very simple things; they implement a data
structure such as a binary tree or hash table, or they perform some
simple computation, such as converting a date string to a number.  The
higher levels then contain logic connecting these primitive
operations.  Using the approach, the primitives can be seen as basic
building blocks which are then glued together to produce the complete
product.  

Why is this design approach relevant to Python?  Because Python is
well suited to functioning as such a glue language.  A common approach
is to write a Python module that implements the lower level
operations; for the sake of speed, the implementation might be in C,
Java, or even Fortran.  Once the primitives are available to Python
programs, the logic underlying higher level operations is written in
the form of Python code.  The high-level logic is then more
understandable, and easier to modify.

John Ousterhout wrote a paper that explains this idea at greater
length, entitled ``Scripting: Higher Level Programming for the 21st
Century''.  I recommend that you read this paper; see the references
for the URL.  Ousterhout is the inventor of the Tcl language, and
therefore argues that Tcl should be used for this purpose; he only
briefly refers to other languages such as Python, Perl, and
Lisp/Scheme, but in reality, Ousterhout's argument applies to
scripting languages in general, since you could equally write
extensions for any of the languages mentioned above.

 \subsection{Prototyping}

In \emph{The Mythical Man-Month}, Fredrick Brooks suggests the
following rule when planning software projects: ``Plan to throw one
away; you will anyway.''  Brooks is saying that the first attempt at a
software design often turns out to be wrong; unless the problem is
very simple or you're an extremely good designer, you'll find that new
requirements and features become apparent once development has
actually started.  If these new requirements can't be cleanly
incorporated into the program's structure, you're presented with two
unpleasant choices: hammer the new features into the program somehow,
or scrap everything and write a new version of the program, taking the
new features into account from the beginning.

Python provides you with a good environment for quickly developing an
initial prototype.  That lets you get the overall program structure
and logic right, and you can fine-tune small details in the fast
development cycle that Python provides.  Once you're satisfied with
the GUI interface or program output, you can translate the Python code
into C++, Fortran, Java, or some other compiled language.

Prototyping means you have to be careful not to use too many Python
features that are hard to implement in your other language.  Using
\code{eval()}, or regular expressions, or the \module{pickle} module,
means that you're going to need C or Java libraries for formula
evaluation, regular expressions, and serialization, for example.  But
it's not hard to avoid such tricky code, and in the end the
translation usually isn't very difficult.  The resulting code can be
rapidly debugged, because any serious logical errors will have been
removed from the prototype, leaving only more minor slip-ups in the
translation to track down.  

This strategy builds on the earlier discussion of programmability.
Using Python as glue to connect lower-level components has obvious
relevance for constructing prototype systems.  In this way Python can
help you with development, even if end users never come in contact
with Python code at all.  If the performance of the Python version is
adequate and corporate politics allow it, you may not need to do a
translation into C or Java, but it can still be faster to develop a
prototype and then translate it, instead of attempting to produce the
final version immediately.

One example of this development strategy is Microsoft Merchant Server.
Version 1.0 was written in pure Python, by a company that subsequently
was purchased by Microsoft.  Version 2.0 began to translate the code
into \Cpp, shipping with some \Cpp code and some Python code.  Version
3.0 didn't contain any Python at all; all the code had been translated
into \Cpp.  Even though the product doesn't contain a Python
interpreter, the Python language has still served a useful purpose by
speeding up development.  

This is a very common use for Python.  Past conference papers have
also described this approach for developing high-level numerical
algorithms; see David M. Beazley and Peter S. Lomdahl's paper
``Feeding a Large-scale Physics Application to Python'' in the
references for a good example.  If an algorithm's basic operations are
things like "Take the inverse of this 4000x4000 matrix", and are
implemented in some lower-level language, then Python has almost no
additional performance cost; the extra time required for Python to
evaluate an expression like \code{m.invert()} is dwarfed by the cost
of the actual computation.  It's particularly good for applications
where seemingly endless tweaking is required to get things right. GUI
interfaces and Web sites are prime examples.

The Python code is also shorter and faster to write (once you're
familiar with Python), so it's easier to throw it away if you decide
your approach was wrong; if you'd spent two weeks working on it
instead of just two hours, you might waste time trying to patch up
what you've got out of a natural reluctance to admit that those two
weeks were wasted.  Truthfully, those two weeks haven't been wasted,
since you've learnt something about the problem and the technology
you're using to solve it, but it's human nature to view this as a
failure of some sort.

 \subsection{Simplicity and Ease of Understanding}

Python is definitely \emph{not} a toy language that's only usable for
small tasks.  The language features are general and powerful enough to
enable it to be used for many different purposes.  It's useful at the
small end, for 10- or 20-line scripts, but it also scales up to larger
systems that contain thousands of lines of code.

However, this expressiveness doesn't come at the cost of an obscure or
tricky syntax.  While Python has some dark corners that can lead to
obscure code, there are relatively few such corners, and proper design
can isolate their use to only a few classes or modules.  It's
certainly possible to write confusing code by using too many features
with too little concern for clarity, but most Python code can look a
lot like a slightly-formalized version of human-understandable
pseudocode.

In \emph{The New Hacker's Dictionary}, Eric S. Raymond gives the following
definition for "compact":

\begin{quotation}
        Compact \emph{adj.}  Of a design, describes the valuable property
        that it can all be apprehended at once in one's head. This
        generally means the thing created from the design can be used
        with greater facility and fewer errors than an equivalent tool
        that is not compact. Compactness does not imply triviality or
        lack of power; for example, C is compact and FORTRAN is not,
        but C is more powerful than FORTRAN. Designs become
        non-compact through accreting features and cruft that don't
        merge cleanly into the overall design scheme (thus, some fans
        of Classic C maintain that ANSI C is no longer compact).
\end{quotation}

(From \url{http://www.catb.org/~esr/jargon/html/C/compact.html})

In this sense of the word, Python is quite compact, because the
language has just a few ideas, which are used in lots of places.  Take
namespaces, for example.  Import a module with \code{import math}, and
you create a new namespace called \samp{math}.  Classes are also
namespaces that share many of the properties of modules, and have a
few of their own; for example, you can create instances of a class.
Instances?  They're yet another namespace.  Namespaces are currently
implemented as Python dictionaries, so they have the same methods as
the standard dictionary data type: .keys() returns all the keys, and
so forth.

This simplicity arises from Python's development history.  The
language syntax derives from different sources; ABC, a relatively
obscure teaching language, is one primary influence, and Modula-3 is
another.  (For more information about ABC and Modula-3, consult their
respective Web sites at \url{http://www.cwi.nl/~steven/abc/} and
\url{http://www.m3.org}.)  Other features have come from C, Icon,
Algol-68, and even Perl.  Python hasn't really innovated very much,
but instead has tried to keep the language small and easy to learn,
building on ideas that have been tried in other languages and found
useful.

Simplicity is a virtue that should not be underestimated.  It lets you
learn the language more quickly, and then rapidly write code, code
that often works the first time you run it.

 \subsection{Java Integration}

If you're working with Java, Jython