coding-standard.tex [plain text]
\section{Twisted Coding Standard\label{doc/howto/policy/coding-standard.xhtml}}
\subsection{Naming}
Try to choose names which are both easy to remember and meaningful. Some silliness is OK at the module naming level (see \texttt{twisted.\linebreak[1]spread}...) but when choosing class names, be as precise as possible. Write code with a dictionary and thesaurus open on the table next to you.
Try to avoid overloaded terms. This rule is often broken, since it is incredibly difficult, as most normal words have already been taken by some other software. More importantly, try to avoid meaningless words. In particular, words like ``handler'', ``processor'', ``engine'', ``manager'' and ``component'' don't really indicate what something does, only that it does \begin{em}something\end{em}.
Use American spelling in both names and docstrings. For compound technical terms such as 'filesystem', use a non-hyphenated spelling in both docstrings and code in order to avoid unnecessary capitalization.
\subsection{Testing}
Unit tests are written using the \texttt{twisted.\linebreak[1]trial} framework. Many examples are in the \texttt{twisted.\linebreak[1]test} package. Test modules should start with 'test\_' in their name. Source files should have \textit{\texttt{test-case-name}}\loreref{doc/howto/policy/test-standard.xhtml} tags that point to their related tests.
Acceptance tests are all automated by the admin/accepttests script currently. (TODO: real acceptance tests strategy!)
Run the unit tests tests before you check anything in.
Let me repeat that, for emphasis: \begin{em}run the unit tests before you check \begin{em}anything\end{em} in\end{em}. Code which breaks functionality is unfortunate and unavoidable. The acceptance tests are highly nonportable and sometimes a pain to run, so this is pardonable. Code which breaks the unit tests in a way that you could have prevented by running them yourself, however, may be grounds for anything from merciless taunting through revertion of the breakage to revocation of cvs commit privileges.
It is strongly suggested that developers learn to use Emacs, and use the \texttt{twisted-dev.\linebreak[1]el} file included in the TwistedEmacs package to bind the F9 key to ``run unit tests'' and bang on it frequently. Support for other editors is unavailable at this time but we would love to provide it.
If you modify, or write a new, HOWTO, please read the \textit{Lore documentation}\loreref{doc/howto/lore.xhtml} to learn the format the docs.
\subsection{Whitespace}
Indentation is 4 spaces per indent. Tabs are not allowed. It is preferred that every block appear on a new line, so that control structure indentation is always visible.
\subsection{Modules}
Modules must be named in all lower-case, preferably short, single words. If a module name contains multiple words, they may be separated by underscores or not separated at all.
In most cases, modules should contain more than one class, function, or method; if a module contains only one object, consider refactoring to include more related functionality in that module.
Depending on the situation, it is acceptable to have imports that look like this: \begin{verbatim}
from twisted.internet.defer import Deferred
\end{verbatim}
or like this: \begin{verbatim}
from twisted.internet import defer
\end{verbatim}
That is, modules should import \begin{em}modules\end{em} or \begin{em}classes and functions\end{em}, but not \begin{em}packages\end{em}.
\subsection{Packages}
Package names should follow the same conventions as module names. All modules must be encapsulated in some package. Nested packages may be used to further organize related modules.
\texttt{\_\_init\_\_.\linebreak[1]py} must never contain anything other than a docstring and (optionally) an \texttt{\_\_all\_\_} attribute. Packages are not modules and should be treated differently. This rule may be broken to preserve backwards compatibility if a module is made into a nested package as part of a refactoring.
If you wish to promote code from a module to a package, for example, to break a large module out into several smaller files, the accepted way to do this is to promote from within the module. For example,\begin{verbatim}
# parent/
# --- __init__.py ---
import child
# --- child.py ---
import parent
class Foo:
pass
parent.Foo = Foo
\end{verbatim}
Every package should be added to the list in \texttt{setup.\linebreak[1]py}.
Packages must not depend circularly upon each other. To simplify maintaining this state, packages must also not import each other circularly. While this applies to all packages within Twisted, one \texttt{twisted.\linebreak[1]python} deserves particular attention, as it may not depend on any other Twisted package.
\subsection{Docstrings}
Wherever possible, docstrings should be used to describe the purpose of methods, functions, classes, and modules. In cases where it's desirable to avoid documenting thoroughly -- for example, and evolving interface -- insert a placeholder docstring (\texttt{"UNDOCUMENTED"} is preferred), so that the auto-generated API documentation will not pick up an extraneous comment as the documentation for that module/class/function.
Docstrings are \begin{em}never\end{em} to be used to provide semantic information about an object; this rule may be violated if the code in question is to be used in a system where this is a requirement (such as Zope).
Docstrings should be indented to the level of the code they are documenting.
Docstrings should be triple-quoted.
Docstrings should be written in epytext format; more documentation is available in the Epytext Markup Language documentation\footnote{http://epydoc.sourceforge.net/epytext.html}.
Additionally, to accommodate emacs users:\begin{itemize}
\item Single quotes of the type of the docstring's triple-quote should be escaped. This will prevent font-lock from accidentally fontifying large portions of the file as a string.
\item Code examples in docstrings should be prefixed by the | character. This will prevent IM-Python from regarding sample code as real functions, methods, and classes.
\end{itemize}
For example,\begin{verbatim}
def foo2bar(f):
"""I am a function to convert foos to bars.
I should be used when you have a foo but you want a bar; note that this is
a non-destructive operation. If I can\'t convert the foo to a bar I will
raise a FooException().
For example::
| import wombat
| def sample(something):
| f = something.getFoo()
| f.doFooThing()
| b = wombat.foo2bar(f)
| b.doBarThing()
| return b
"""
# Optionally, actual code can go here.
\end{verbatim}
\subsection{Scripts}
For each ``script'', that is, a program you expect a Twisted user to run from the command-line, the following things must be done:\begin{enumerate}
\item Write a module in \texttt{twisted.\linebreak[1]scripts} which contains a callable global named \texttt{run}. This will be called by the command line part with no arguments (it will usually read \texttt{sys.\linebreak[1]argv}). Feel free to write more functions or classes in this module, if you feel they are useful to others.
\item Write a file in \texttt{bin/} which contains the Twisted running-from-CVS header, and ending with \begin{verbatim}
from twisted.scripts.yourmodule import run
run()
\end{verbatim}
\item Write a manpage in \texttt{doc/man}. On debian systems you can find a skeleton example of a manpage in \texttt{/usr/share/doc/man-db/examples/manpage.\linebreak[1]example}.
\item Add your script to the script list in \texttt{setup.\linebreak[1]py}.
\end{enumerate}
This will insure your program will work correctly for users of CVS, Windows releases and Debian packages.
\subsection{Standard Library Extension Modules}
When using the extension version of a module for which there is also a Python version, place the import statement inside a try/except block, and import the Python version if the import fails. This allows code to work on platforms where the extension version is not available. For example: \begin{verbatim}
try:
import cPickle as pickle
except ImportError:
import pickle
\end{verbatim}
Use the "as" syntax of the import statement as well, to set the name of the extension module to the name of the Python module.
\subsection{ChangeLog}
All changes that will affect the way end-users see Twisted should come with an appropriate entry in the ChangeLog that summarizes that impact.
The correct format for the ChangeLog is GNU changelog format. There is an emacs mode for editing this, use \texttt{M-x add-change-log-entry}. If you are, for whatever absurd reason, using an editor other than emacs to edit Twisted, you can use Moshe Zadka's helpfully provided \texttt{admin/change} script to add a properly-formatted entry.
\subsection{Classes}
Classes are to be named in mixed case, with the first letter capitalized; each word separated by having its first letter capitalized. Acronyms should be capitalized in their entirety. Class names should not be prefixed with the name of the module they are in. Examples of classes meeting this criteria:\begin{itemize}
\item twisted.spread.pb.ViewPoint
\item twisted.parser.patterns.Pattern
\end{itemize}
Examples of classes \begin{em}not\end{em} meeting this criteria:\begin{itemize}
\item event.EventHandler
\item main.MainGadget
\end{itemize}
An effort should be made to prevent class names from clashing with each other between modules, to reduce the need for qualification when importing. For example, a Service subclass for Forums might be named twisted.forum.service.ForumService, and a Service subclass for Words might be twisted.words.service.WordsService. Since neither of these modules are volatile \begin{em}(see above)\end{em} the classes may be imported directly into the user's namespace and not cause confusion.
\subsection{Methods}
Methods should be in mixed case, with the first letter lower case, each word separated by having its first letter capitalized. For example, \texttt{some\linebreak[1]Method\linebreak[1]Name}, \texttt{method}.
Sometimes, a class will dispatch to a specialized sort of method using its name; for example, twisted.reflect.Accessor. In those cases, the type of method should be a prefix in all lower-case with a trailing underscore, so method names will have an underscore in them. For example, \texttt{get\_some\linebreak[1]Attribute}. Underscores in method names in twisted code are therefore expected to have some semantic associated with them.
Some methods, in particular \texttt{add\linebreak[1]Callback} and its cousins return self to allow for chaining calls. In this case, wrap the chain in parenthesis, and start each chained call on a separate line, for example:\begin{verbatim}
return (foo()
.addCallback(bar)
.addCallback(thud)
.addCallback(wozers))
\end{verbatim}
\subsection{Callback Arguments}
There are several methods whose purpose is to help the user set up callback functions, for example \texttt{Deferred.\linebreak[1]add\linebreak[1]Callback} or the reactor's \texttt{call\linebreak[1]Later} method. To make access to the callback as transparent as possible, most of these methods use \texttt{**kwargs} to capture arbitrary arguments that are destined for the user's callback. This allows the call to the setup function to look very much like the eventual call to the target callback function.
In these methods, take care to not have other argument names that will ``steal'' the user's callback's arguments. When sensible, prefix these ``internal'' argument names with an underscore. For example, \texttt{Remote\linebreak[1]Reference.\linebreak[1]call\linebreak[1]Remote} is meant to be called like this:\begin{verbatim}
myref.callRemote("addUser", "bob", "555-1212")
# on the remote end, the following method is invoked:
def addUser(name, phone):
...
\end{verbatim}
where ``addUser'' is the remote method name. The user might also choose to call it with named parameters like this:\begin{verbatim}
myref.callRemote("addUser", name="bob", phone="555-1212")
\end{verbatim}
In this case, \texttt{call\linebreak[1]Remote} (and any code that uses the **kwargs syntax) must be careful to not use ``name'', ``phone'', or any other name that might overlap with a user-provided named parameter. Therefore, \texttt{call\linebreak[1]Remote} is implemented with the following signature:\begin{verbatim}
def callRemote(self, _name, *args, **kw):
...
\end{verbatim}
Do whatever you can to reduce user confusion. It may also be appropriate to \texttt{assert} that the kwargs dictionary does not contain parameters with names that will eventually cause problems.
\subsection{Special Methods}
The augmented assignment protocol, defined by \_\_iadd\_\_ and other similarly named methods, can be used to allow objects to be modified in place or to rebind names if an object is immutable -- both through use of the same operator. This can lead to confusing code, which in turn leads to buggy code. For this reason, methods of the augmented assignment protocol should not be used in Twisted.
\subsection{Functions}
Functions should be named similiarly to methods.
Functions or methods which are responding to events to complete a callback or errback should be named \texttt{\_cb\linebreak[1]Method\linebreak[1]Name} or \texttt{\_eb\linebreak[1]Method\linebreak[1]Name}, in order to distinguish them from normal methods.
\subsection{Attributes}
Attributes should be named similarly to functions and methods. Attributes should be named descriptively; attribute names like \texttt{mode}, \texttt{type}, and \texttt{buf} are generally discouraged. Instead, use \texttt{display\linebreak[1]Mode}, \texttt{player\linebreak[1]Type}, or \texttt{input\linebreak[1]Buffer}.
Do not use Python's ``private'' attribute syntax; prefix non-public attributes with a single leading underscore. Since several classes have the same name in Twisted, and they are distinguished by which package they come from, Python's double-underscore name mangling will not work reliably in some cases. Also, name-mangled private variables are more difficult to address when unit testing or persisting a class.
An attribute (or function, method or class) should be considered private when one or more of the following conditions are true:\begin{itemize}
\item The attribute represents intermediate state which is not always kept up-to-date.
\item Referring to the contents of the attribute or otherwise maintaining a reference to it may cause resources to leak.
\item Assigning to the attribute will break internal assumptions.
\item The attribute is part of a known-to-be-sub-optimal interface and will certainly be removed in a future release.
\end{itemize}
\subsection{Database}
Database tables will be named with plural nouns.
Database columns will be named with underscores between words, all lower case, since most databases do not distinguish between case.
Any attribute, method argument, or method name that corresponds \begin{em}directly\end{em} to a column in the database will be named exactly the same as that column, regardless of other coding conventions surrounding that circumstance.
All SQL keywords should be in upper case.
\subsection{C Code}
Wherever possible, C code should be optional, and the default python implementation should be maintained in tandem with it. C code should be strict ANSI C, and \begin{em}must\end{em} build using GCC as well as Visual Studio for Windows, and really shouldn't have any problems with other compilers either. Don't do anything tricky.
C code should only be used for efficiency, not for binding to external libraries. If your particular code is not frequently run, write it in Python. If you require the use of an external library, develop a separate, external bindings package and make your twisted code depend on it.
\subsection{Checkin Messages}
Thanks to CVSToys, the checkin messages are being used in a myriad of ways. Because of that, you need to observe a few simple rules when writing a checkin message.
The first line of the message is being used as both the subject of the commit e-mail and the announcement on \#twisted. Therefore, it should be short (aim for $<$ 80 characters) and descriptive -- and must be able to stand alone (it is best if it is a complete sentence). The rest of the e-mail should be separated with \begin{em}hard line breaks\end{em} into short lines ($<$ 70 characters). This is free-format, so you can do whatever you like here.
Checkin messages should be about \begin{em}what\end{em}, not \begin{em}how\end{em}: we can get how from CVS diff. Explain reasons for checkins, and what they affect.
Each commit should be a single logical change, which is internally consistent. If you can't summarize your changes in one short line, this is probably a sign that they should be broken into multiple checkins.
\subsection{Recommendations}
These things aren't necessarily standardizeable (in that code can't be easily checked for compliance) but are a good idea to keep in mind while working on Twisted.
If you're going to work on a fragment of the Twisted codebase, please consider finding a way that you would \begin{em}use\end{em} such a fragment in daily life. I use the Twisted Web server on the main TML website, and aside from being good PR, this encourages you to actively maintain and improve your code, as the little everyday issues with using it become apparent.
Twisted is a \begin{em}big\end{em} codebase! If you're refactoring something, please make sure to recursively grep for the names of functions you're changing. You may be surprised to learn where something is called. Especially if you are moving or renaming a function, class, method, or module, make sure that it won't instantly break other code.