Development/Tutorials/API Documentation

    From KDE TechBase
    Revision as of 17:19, 27 August 2012 by Claus chr (talk | contribs) (Box template should not be used anymore)
    Definition
    API (Application Programming Interface) documentation is documentation that applies to a program and its interfaces. It is documentation that explains to a developer who is going to work with the program, or a developer who is going to work on a program, how things work and what methods are there to call.

    API documentation is sometimes called an API reference manual. It needn't just be a reference manual, though, there can be extensive additional material, like tutorials, histories, etc. In this page, we will refer to all the material that documents and explains the API of a piece of code as "apidox."

    Good apidox take some effort to write -- but I would say less than good user documentation, since you as a developer are writing for another developer, and explaining how code works and what it is supposed to do. So your audience is clear to start with. The benefits of apidox come when someone else (or maybe even you several months later) needs to (re)use the code, or extend it. Good apidox means that someone new can show up, understand your code, and produce meaningful patches (that is, help) all the more quickly. Especially for libraries, good apidox make the library usable as a black box (and that's how they should be) because all the needed documentation is available in the apidox and not hidden in the (perhaps inaccessible) C code.


    Preamble

    APIDOX are what make a program accessible to other contributors. They are not essential, but they certainly help a great deal for new people who want to work on your code, or better yet, re-use it elsewhere with a minimum of modification.

    Look at the Qt documentation to get a feeling of what good apidox look like. There is a consistency of style, a thoroughness which permeates all the documentation. It is possible to learn a lot about Qt just from reading the documentation. You do not necessarily need to run the tutorial programs or read the source code of the library to find out what a parameter flags does in some method of the library. It is all spelled out for you.

    Writing apidox is a two-part process. One part is technical: you need to understand the code you are documenting -- or at least know what it is supposed to do and how it is meant to be used. The other part is plain discipline: apidox are most useful when they are exhaustive.

    This document is going to try to prevent the apidox process from becoming exhausting, by giving straightforward tips about how to write apidox. There are also the library apidox policies, which are much stricter, which provide measurable properties that apidox must adhere to. It does not hurt to glance at them now.

    APIDOX Basics

    APIDOX are processed by the Doxygen documentation tool. This tool reads source code for your application (or library) and produces nicely formatted documentation from it. There is a good reference manual available -- but let's hope to make it unnecessary to read, for basic use anyway.

    Tip
    You don't even need to have Doxygen installed to work on apidox for your application. Every few hours, the EnglishBreakfastNetwork compiles apidox for all of the KDE modules we know about. Logfiles are kept and you can see your apidox on the site, or read the error messages and fix those. ...it might not be the fastest way to write and fix apidox, but it gets the job done, and if you spend just an hour at the end of the day writing some apidox, it's sufficiently useful.


    Tip
    If you do have Doxygen installed and want to build the apidox, use the method described in apidox.


    Basic apidox are fun and simple: you write comments in your code explaining what things are supposed to be for. These comments are nearly indistinguishable from stuff you would be writing in the headers of your code anyway, so that's not hard to do.

    APIDOX are written in special comments. These comments start with /** (slash, star, star) -- that's what makes them special. The rest of the content of the comment is just plain text describing a part of your program. The plain text is interpreted by the Doxygen processor, so that you can write accurate descriptions of parameters, return types, and do some basic text markup. But documentation can be very straighforward: just write down what a method does, surrounded by /** and */, like this:

    /** This method increases the sexyness of Kontact and should as
     * such be called whenever possible (i.e. instead of having idle
     * time, you might think of calling this method and helping
     * Kontact gain even more popularity). You might even insert it
     * into your own event loop to ensure it is called as often as
     * possible. If these calls decrease the number of new features,
     * it's still no problem to call it. 
     */
    void incSexyness(KInstance *instance);
    

    For proper apidox, you need to document every "thing" in your program. "Things" here are directories, files, namespaces, classes, methods, enums, and member variables. You can actually leave out files and directories, and concentrate on the latter. Complete apidox looks something like this:

    /** Namespace for KDE network-related classes */
    namespace KDENetwork {
      /** Wrapper for a TCP/IP socket */
      class Socket {
      public:
        /** Constructor. Calls listen() on some random high port. */
        Socket();
      private:
        int fd;
      };
    }
    

    You can see here that each nesting level of "things" is documented with a comment -- the namespace, the class and the method. Things that are private do not need apidox, but things that are protected do.

    It is important to document each level of nesting, because if you leave one level out, Doxygen will ignore the documentation in inner nesting levels, and you will be left wondering where it has gone. For instance, if we leave out the class documentation, then the method documentation for Socket() will vanish as well. This is one of the common pitfalls to writing apidox.

    If you just do this -- write an explanation for every part of your program -- then you're already a long way on the road to having complete apidox. Writing all those explanations makes your program more accessible to other developers -- and often shows you where the design or choice of names is sub-optimal. So it's a win for you both ways.

    You can consult the list of supported tags for examples of more fancy apidox -- explaining parameters, for instance, and annotating the apidox with credits and examples. It's also worthwhile to take a look at the section on enabling apidox, but it's also fine to divide the work: you write the apidox themselves, and ask me (mailto:[email protected]) to enable apidox generation for your module. And I will.

    Writing APIDOX in New Code

    If you are writing new code and want to write apidox, use KDevelop. Really. Properly configured, it can automatically set up all the skeletons for apidox that you need, and what you do is write the descriptions. No messing with apidox commands.

    If you're not in that fortunate position, the following little checklist should get you through the worst of writing apidox.

    1. Write apidox as you code

    The discipline it takes to write down the apidox for function foo() now as you are thinking of foo() and what it needs to do more than compensates the effort later where you have to remember what foo() was supposed to do, anyway.

    This isn't to say you have to do it this way, but it is convenient. The apidox also document design decisions. They also document what you want a particular piece of code to do, regardless of what it actually does. That makes it possible to independently check that the code does what it's supposed to: because it's written down.

    2. Document your headers completely

    The headers are what's most visible to users (in this context, users are developers who are re-using) of your code, and they should be complete. Document each structural bit of the headers as you go along. This means:

    • Every file should have a file-level comment. This is suggested in the KDE guidelines anyway -- near the top of your file, write down what the file is for, vaguely what it defines. Wrap this up in a /** @file */ comment and you are set.
    • Every namespace should have a comment. A given namespace only needs a comment once in your source tree (or within one bunch of files that generate apidox together), but it can't hurt to repeat the description on each occasion -- just make it brief. These comments are just plain apidox comments wrapped up in /** */ ; there are no special markups needed like the @file for file comments.
      Do make sure the comment is just before the namespace and doesn't get distanced from the namespace it documents. The following is fine:
    /** The Ungulate namespace collects classes and methods
    pertaining to hooved animals. */
    
    namespace Ungulate {
    
    In the next example, someone has snuck in some extra stuff between the apidox comment and the namespace it is documenting. This may cause Doxygen errors (so then it is easy to spot) or it may cause the namespace documentation to attach to something wildly different (and then it's hard to spot).
    /** The Ungulate namespace collects classes and methods
    pertaining to hooved animals. */
    class Mammal;
    namespace Ungulate {
    
    There is not much you can do about this except to be watching when you insert code -- don't separate apidox from the thing they are documenting.
    • Every class should have a comment. Classes are the important building blocks of your application or library, so this is one place where writing lots helps. Write down why the class exists. Write down what it is supposed to do. Give an example of how to use it. Explain how not to use it, or what prerequisites it has (for instance, lots of classes need a KInstance object, but don't document that explicitly).
      The same caveats apply as with namespace apidox: make sure the class follows its apidox immediately.
    • Every method should have a comment explaining what it does and what the parameters are for. Method parameters should be documented using @param. Don't rely on the name of the method or the parameters to be fully self-documenting. Besides, writing these things down properly will trigger Doxygen errors if you change them in an incompatible way later -- and that is going to save you lots of time in finding source and binary incompatibilities and explaining to users why their code suddenly doesn't do what they expect (assuming it compiles at all). So good method apidox is an additional insurance against making bad changes. Same caveats apply.
    • Every enumeration type should have a comment explaining what the enumeration is for, even if it's just /** Various constants */.
    • Every enumeration value should have a comment too, to explain what it represents. Don't rely on the name of the enumeration value being sufficiently expressive.
      For the purposes of readability, I suggest that you document enumeration values after the value, as opposed to the documentation format for namespaces, classes and methods where you write the documentation in front of the thing you are documenting. The format of the documentation is slightly different. Instead of writing /** Documentation */ in front, you write /**< Documentation afterards */, where the < denotes that the documentation applies to the thing just past.
      It looks like this:
    enum State {
      none,    /**< No bracket seen */
      bracket, /**< Found a ( for grouping */
      squote,  /**< Found a single quote */
      dquote   /**< Found double quotes */
    };
    

    3. Watch this space!

    Watch the English Breakfast Network for the results of your apidox work. Check the log files for errors -- Doxygen can complain quite loudly.

    4. Write a main page for your application.

    This is usually done in a separate file Mainpage.dox in the top-level of a library or application. The file's content is just a single apidox comment that starts with /** @mainpage title ; the rest of the file is just a long comment about what the library or application is for.

    Fixing APIDOX in Old Code

    Writing apidox in old code is a lot like writing the same apidox in new code, except that there is more cruft in the way. The number 1 tip to follow is: watch the logs on English Breakfast Network. Those will show you what errors there are in the apidox. However, Doxygen can't catch everything that is wrong with the documentation on its own, so you will have to do some reading yourself. The other tips for new apidox apply equally here: you want to document everything, in a consistent style. If methods show up on the generated apidox pages with no documentation, you know that you have more apidox to write. (Doxygen may provide an error message, but doesn't do that everywhere in the current setup because there would just be too many.)

    In old apidox, you are more likely to suffer from the following symptoms:

    • Missing parameter documentation (because parameters were renamed, or added, or removed, or something).
    • Missing method documentation.
    • Missing class documentation.
    • Documentation that has wandered off on its own and is attached to the wrong thing now.

    The first of these can be fixed by correcting the parameter documentation. See the examples section. The next two -- missing documentation that you can see is there in the source files but that does not show up in the generated HTML pages, is usually a matter of missing documentation on surrounding blocks. See the common pitfalls section, and make sure that the surrounding classes, namespaces and files all have documentation.

    The last problem can best be fixed by moving the offending documentation back to where it belongs (really, it's not the documentation that is at fault, it's whatever has squeezed in -- the home-breaker -- between the documentation and the thing it was originally attached to). You could use some Doxygen special tags to avoid moving stuff around like that, but it does not help the understandability of the source much.

    Example APIDOX

    So what does documentation look like in the headers? How do you write a method documenation that describes the parameters as well? This section contains boilerplate for most common situations. Doxygen does not require a strict style -- it will ignore whitespace and asterisks at the beginning of a line, so you can make the documentation ASCII-pretty.

    Documentation for a file: The newline after @file is significant! The text after @author is listed in a special Authors section of the apidox; you can list multiple authors.

    /** @file
     * This file is part of AnApplication and defines
     * classes Ungulate and Moose.
     *
     * @author Mostly by me
     */
    

    Documentation for a namespace:

    /**
     * This namespace collects functions related to
     * counting and enumeration of mammals and their limbs.
     */
    

    Documentation for a class: Some Doxygen special commands are used here to provide additional information. @author (as with files) identifies authors of the code; these are collected in a special Authors: section of the apidox. You can list more than one author. The @since tag tells users since when the class has existed. It is usual to put a KDE release number here.

    /**
     * This class represents a Moose in the woodland
     * simulator. A single Moose object can be created,
     * but it is more useful to instantiate Moose::Factory
     * by calling Moose::factory(), and then calling spawn()
     * for each new Moose, since that maintains the ecological
     * balance far better.
     *
     * @author Adriaan de Groot <[email protected]>
     * @since  3.5
     */
    

    Method documentation: We can use @author and @since just like we do for classes. In addition, there are the parameters of the method that can be described. @p is used to refer to them in running text, and @param is used to construct a list of parameter descriptions that is specially formatted. Finally, @return entries describe the values that may be returned by the method.

    /**
     * This method names a particular Moose and as a side effect
     * sets whether or not the Moose is treated as a Reindeer.
     * When @p santa is @c true, the name of the moose is set to
     * the next of the available Reindeer (if possible).
     *
     * @param name name to assign to the Moose, which is only
     *             relevant if the moose is not a Reindeer.
     * @param santa is this Moose assigned to Santa? if so, the
     *              name is irrelevant.
     *
     * @return @c true if naming the Moose succeeded
     * @return @c false if naming the Moose failed. This only
     *            happens if the Moose is assigned to Santa duty
     *            but there are already eight named Reindeer.
     */
    bool name(const QString &name, bool santa = false);
    

    Enum documentation is described in the section on writing new apidox. The same kind of documentation as for classes applies, with the addition of the documentation for each enumerated value which belongs inline.

    Common Pitfalls

    This section lists common pitfalls in writing apidox. Typically, they are easily overlooked mistakes that produce weird error messages, but I will also include some stylistic pitfalls that should be avoided.

    Missing APIDOX: You know you wrote dox for class Moose, but after generation they are not visible. You know that the file-scope function int foo() is documented, but it's not there either! What is going on?

    The most common pitfall, the one that leads to "missing" apidox, is forgetting to document surrounding structure. The "structure" in the source comes from files, namespaces, classes and methods. In order to document a class you must document the namespace it is in (if there is one) and the file that it is in. In order to document a method, you must document the class it is in (and thus the namespace and file). So the easiest rule of thumb is to document everything.

    Note
    This hard-and-fast rule is not quite accurate:

    classes do not need the file to be documented, so you can leave out the /** @file */ comment for source files containing only classes (and namespaces). Note that classes in a namespace require the namespace to be documented, but classes in the :: namespace don't need extra documentation. I'm not sure about anonymous namespaces.

    For file-scoped functions — that is, functions that are defined in a file, not in a namespace and not in a class — you must document the file they are in. This applies mostly

    to files which collect a bunch of non-method utility functions.


    Broken parameter documentation: Documenting parameters to methods can be done two ways: you can document none of them, or you can document all of them for a given method. There is no middle ground (that doesn't generate gobs of errors that you should fix).

    If you document all of your parameters (which is a good thing to do, and generates things in a nicely formatted fashion), then your method apidox consists first of a general description of the method and then a bunch of @param tags which describe each individual parameter. The @param tag is followed by the name of the parameter -- watch out for spelling and case-sensitivity! -- and then the description. The description can span multiple lines.

    /** Calculate the root-mean-square distance from the
     * origin to the given integral coordinate.
     *
     * @param y y-coordinate, which is on an axis perpendicular to the
     *        x-axis, yet still in the same plane.
     * @param x x-coordinate
     */
    

    When you document all the parameters like this, Doxygen will complain if you misspell parameter names, or forget some, or mention parameters that are not there. This will force you to update the documentation if you change the method signature. And that's a good thing.

    Additional pitfalls are putting the type of the parameter in the list, like @param int x, which makes "int" the name of the parameter. Another pitfall is that @param should come after the general description. Once the @param list starts, nothing can stop it except for other list-style Doxygen tags like @since, @author or @return. So write @param after the story and before, say, @return.

    If you document none of the parameters, you do not use the @param tag at all. You can talk about your parameters by writing @p before the name of a parameter (or anything else, really, but it only makes sense in front of a name of a parameter). Like this:

    /** Calculate the root-mean-square distance from the origin to
     * the given integral coordinate which is given as a pair @p x,
     * @p y. If @p nonFree is true, use ESR instead of RMS in the
     * computation.
     */
    double distance(int x, int y, bool nonFree=false);
    

    Misuse of tags in running text: This boils down to the warning above about where to put @param: not every Doxygen tag can go anywhere. Some start lists or basically end the general description part of a description, so you need to avoid using them in running text.

    Chosing between Doxygen errors and compiler warnings: If you are going to document your parameters, you need to name them. If you are defining a stub method, this can lead to compiler warnings.

    Accidental tags: It can be easy to accidentally use Doxygen tags in running text -- email addresses, backslash escapes, those are the easy ones. Watch the Doxygen logs and escape that at sign with a backslash when needed (i.e. write \@ to get an at sign). It's probably a good idea to avoid the backslash style of apidox entirely; at the same time, if you happen to write \moose, Doxygen will complain that that is an invalid tag.

    Accidental HTML: If you use < and > in your apidox, these may confuse Doxygen -- especially if you write things like <service> which look like HTML tags. This is a common way of writing some kind of element that may be replaced in a method call or string or something, so it crops up all the time. You know, you write some thing like this:

    /** This method connects to a database. The connection string is
     * composed of a username, password, host and database name as
     * follows:
     * <username>:<password>@<host>//<database>
     *
     * @return true if the connection succeeded.
     */
    

    This bit of dox is terribly broken, because the whole sample connection string will be interpreted as (bad) HTML and ignored. There are two solutions:

      • Write \< instead of just < to escape the < and make Doxygen output it normally. Doxygen is smart enough to turn this into &lt; in HTML output. This has only a minimal impact on readability of the dox in the header files themselves.
      • Use the HTML formatting that Doxygen makes available, and write <i>foo</i> instead of <foo>. This produces nicer output with italics instead of plain text, so it is easier to spot what is the replaceable part of the text. The downside is that it has a larger visible impact on the apidox in the headers.
    Superfluous @ref
    It can be tempting -- certainly if you've written dox for other projects -- to use #method or @ref method to force a reference to another method somewhere. Relax, it's not needed and usually causes Doxygen warnings to boot. Just name the method normally, make apidox and watch the references appear naturally.

    KDE Specific Tags

    @bc: This tag indicated binary compatibility, much like @since does. The argument after @bc is the scope of binary compatibility (for instance, KDE4). Classes that are not marked with @bc may, in some modules, be flagged as incompatible so that they can be avoided.

    Example:

    /**
     * This class emulates a Moose.
     * @bc KDE4.2 Compatibility is expected to break 
     *     with next-generation mooses.
     */
    

    @libs: Use this tag to indicate what libraries should be linked in for the given class. Although this is usually the name of the directory the class lives in, this is not always the case.

    Example:

    /**
     * This class emulates a Moose.
     * @libs
     *     @a -lkdeui or use \$(LIB_KDEUI) in the KDE build framework.
     */
    

    Code Examples in APIDOX

    It can be useful to put example code in the APIDOX for a class. Very useful, in fact, for the reader who is wondering how exactly to use the class in a straightforward way. Most classes in kdelibs/kdecore have example code.

    One way to write example code is to use @code and @endcode around blocks of example code in the Doxygen comments themselves, like this:

    /**
     * This class represents a Moose.
     * The correct way of generating Meese is to use the factory:
     *
     * @code
     * Moose::Factory outlet = Moose::factory();
     * Moose *m = outlet.spawn();
     * @endcode
     */
    

    This is how most of the examples in kdelibs are done, actually. It works reasonably well, you can pare the example down to something really minimal and leave out all the boilerplate.

    An important drawback of this approach to writing example code is that the code is never checked to see if it actually works. The code is also so terse, usually, that it's hard to expand into a complete example that actually compiles and runs.

    To solve this problem, you can write real code that actually compiles (as part of the test suite for the code that you have anyway) and that illustrates exactly how to use the class under consideration in an actual program. Better yet, you do not need to include all the code in the APIDOX, just the interesting bits. The way to do this is to put the code in a file in the tests/ subdirectory of your code. Then use Doxygen's @include tag to include the code from that file into the API documentation.

    Warning
    Do not use @example! It does something totally different from what you might expect (it doesn't insert example code, for one thing).


    Links

    How to build apidox

    List of supported doxygen tags

    This is a list of all available documentation tags of doxygen, from the official site. Note that the list uses a backslash in front of a tag, while this page always used the at sign. Both are allowed, but for KDE you should use the at sign to stay consistent within KDE.