Difference between revisions of "Coding Standards"

From MythTV Official Wiki
Jump to: navigation, search
(Work in progress on disambiguation strings...)
(Translation guidelines: Complete disambiguation info)
Line 191: Line 191:
 
'''Disambiguation string'''
 
'''Disambiguation string'''
  
It obviously could be as simple as that..
+
It obviously couldn't be as simple as that..
  
 
Just kidding...
 
Just kidding...
Line 209: Line 209:
 
</nowiki></pre>
 
</nowiki></pre>
  
All three "Unknown" strings will be in the context "FirstClass" but will have their own translation. The first two will have a disambiguation string, the third one will have none.
+
All three "Unknown" strings will be in the context "FirstClass" but will have their own translation. The first two will have a disambiguation string, the third one will have none. This could be used for example when the gender of the nouns to which these strings refer to could affect the translation ("Unknown" could be written differently in another language depending on whether it refers to a feminine or masculine noun).
 +
 
 +
There is also a poorly documented (it appears to be only documented for the lower level (and which you should never have to use) QTranslator::translate function), the fallback to the string with no disambiguation string if the string with the same context, original string and disambiguation string is not present in the translation file. This could happen when the translation file is not up to date with the application or if one of the parameters passed to tr() (or the other translation functions) is a variable. We will discuss what to do when you have to pass variables to the translation functions later...
 +
 
 +
{{Note box|Never pass a '''null''' disambiguation string as parameter, if you need to provide one pass an empty string ("").}
 +
 
 +
'''But I lack class (or what to do when you are not in a member function)'''
  
 
Adding the tr() function using either the Q_OBJECT or Q_DECLARE_TR_FUNCTIONS macro only works for classes, a different approach must be used when your are not in a member function.
 
Adding the tr() function using either the Q_OBJECT or Q_DECLARE_TR_FUNCTIONS macro only works for classes, a different approach must be used when your are not in a member function.
  
If you have a mix of member functions (ie part of a class) and regular (C-like) functions (which is pretty common in the settings pages for example), you can call the tr() function of the class from your function. The translations done that way will share the same translation context (category) as the class which is usually what you want when those functions are called from the member functions of that class.
+
If you have a mix of member functions (ie part of a class) and regular (C-like) functions (which is pretty common in the settings pages for example), you can call the tr() function of the class from your function. The translations done that way will share the same translation context as the class which is usually what you want when those functions are called from the member functions of that class.
  
 
If you don't have any functions or at least any function you would like to share it's translation context with then you have to resort to using QCoreApplication::translate().
 
If you don't have any functions or at least any function you would like to share it's translation context with then you have to resort to using QCoreApplication::translate().

Revision as of 02:34, 11 August 2013

Design guidelines

Here's some advice to keep in mind while designing or updating code:

  • Use C++. Avoid outdated C constructs like TRUE/FALSE macros, C-style casting etc. One or two C-isms are preferred, like using NULL over the C++ preference of 0 (see this ticket). Myth supports gcc 2.95.x and above.
  • Encapsulate Well Make a Get method const. Return a copy and not a reference. Use Set methods, when defined in the header these entail no efficiency loss and allow you do add debugging code when necessary. Make local functions static and constants static const.
  • Use Qt for non-GUI tasks. Don't design your own helper classes or use platform-dependent code when Qt already offers a class. Strings, containers (lists etc) and mutexes are good examples of Qt's generic functionality. The current development version of MythTV requires Qt 4.8, so check your calls against the Qt 4.8 documentation so you don't break things.
    • We have our own replacement for QProcess called MythSystem. QProcess has had an unfixed deadlocking bug in Qt 4.0 through the time of the writing of this guide (4.7).
    • We have our own logging macro LOG, use this instead of the Qt debugging output macros and cerr.
  • Use MythUI for GUI tasks. We are currently working on a UI layer that's independent of Qt (MythUI), and patches that use native Qt widgets or the old libMyth code will be rejected. Use libmythui/* instead.
  • Stay platform-independent. Use Qt instead of #ifdefs where possible. If you must add something that only works on one platform, add a compile-time flag so that it's only built where it works. See settings.pro for several examples.
  • Avoid dependencies. Dependencies are added very rarely to Myth, so think carefully before you suggest a new item. If you want to link to a library that's small and has unique functionality, you may be able to get it added to the core Myth build. In general, though, avoid external dependencies if at all possible.
  • Use private classes to keep popular header files stable. This design pattern is used in MythContext and others; most of the private data and methods are stored in a private class, defined in the .cpp file. A pointer to this class is stored as a member variable in the public class. This way, the private data can change without changing the header file, which prevents everyone from having to rebuild most of the source tree for a popular header like mythcontext.h.
  • Don't use assert() or abort() Crashing is never good. Use proper error handling in place of assert. This doesn't mean you can't use asserts during development, just that by the time the code is ready for commit there should be no need for them.
  • Don't use C++ exceptions Unlike say Java, C++ doesn't have any way to tell the user of a method what exceptions must be caught, but gcc does let you mark a return value as something that must be examined. To deal with exceptions that can occur during object construction use either two part initialization or a static factory method that returns NULL when it fails.
  • Try to keep style coherent If a class prepends all it's member variables with _ rather than the "m_" don't add a variable with "m_", use the class style unless you are doing a major rewrite and renaming all the existing variables.
  • Make large style changes in a separate patch from functional changes This makes it easier for us to comb through the history looking for the commit that introduced the bug.
  • QString and other Qt container classes are not thread-safe. You can not use them in multiple threads without locking as you might a "const char*" string or a std::string. As of Qt4 assigned copies can be used in another thread, so something like this is now safe "QString MyClass::GetString(void) const { QMutexLocker(&m_lock); return myString; }" You may run across code like this from the Qt3 days, "QString MyClass::GetString(void) const { QMutexLocker(&m_lock); QString tmp = myString; tmp.detach(); return tmp; }" The detach() is no longer required but is a relic of our semi-automated conversion from Qt3 to Qt4.
  • Don't use Qt signals and slots in backend code There are a number of gotchas with signals and slots and we're decided to use them only in UI code as a result of dealing with all the hard to debug crashes that have resulted.
  • Locking order Most classes with more than one lock/mutex/semaphore will have locking order comment. You must follow this locking order to avoid A,B;B,A deadlocks. If you create a class with more than one lock, document your locking order.
  • Unused parameters Functions that have to satisfy a given interface definition sometimes have parameters that are not used. The name of an unused parameter should be put in a comment to document that the parameter is intentionally unused in such a way that using it anyway will not work. If the function does not have to follow such an interface definition the parameter should simply be removed.

For more of an overview of the development process for Myth, see MythPlugin Architecture.

MythTV Style

The rules of thumb on this page are meant to help developers ensure that their code fits in with the conventions used in the MythTV source. The goal is to enhance readability. After your code becomes a part of MythTV many other hands are likely to touch it and they will be following these conventions so if you apply some other convention the code will over time become more difficult to read. Do not apply these conventions to externally maintained code, such as ffmpeg, try your best to follow their conventions and whenever possible get your changes to the external code applied upstream as soon as possible.

Spaces and braces

The one hard and fast rule of Myth development: don't use tab characters.

Indents are accomplished with 4 spaces. Please set your editor to convert tabs to spaces, or at least do the conversion before you submit a patch.

You should put 1 space between keywords like if or for and their parenthesized lists. By contrast, don't do the same for function names; there should be no space between the name and its opening parenthesis. There should be no spacing between parentheses and the items they contain.

Spaces are used between binary operators (meaning + or =).

Braces should go on their own line. The exception to this is the declaration of structs or enums, where the opening brace goes on the same line as the type declaration. One-line if or else bodies do not need braces.

Lines should not exceed 80 characters in width. If you have a long statement, indent the continuations so they line up vertically. Generally, if you indent past the last opening parenthesis of the previous line, you'll be in good shape.

typedef enum my_enum {
    kTestOne,   ///< This represents blah
    kTestTwo,   ///< Widget Blah
    kTestThree, ///< Blah Blah
    kTestSentinel,
} MyEnum;

static inline void run_three_times(const QStringList &strings, const char *buffer)
{
    for (int i = 0; i < 3; i++)
    {
        MyEnum result = my_func(i, strings);
        if (kTestOne == result || kTestTwo == result)
            DoSomething();
        else
            DoTheOpposite();

        result = my_other_func(result, strings);
        if (kTestOne == result || kTestTwo == result)
        {
            if (DoSomething())
                DoMore();
        }
        else
        {
            DoTheOpposite();
        }
    }
}

Note: There are some things above that are not required by the coding standard. Using result == kTestOne is equally valid as kTestOne == result. The reason some developers prefer it is because it avoids introducing a "=" typo bug. If you run across the variation you don't prefer, leave it alone.

For Vi(m) users: the following options can be put into your ~/.vimrc file -

set expandtab
set tabstop=4
set shiftwidth=4

And if it's not too intrusive, this line can appear in the source code, and Vim will apply the correct settings:

/* vim: set expandtab tabstop=4 shiftwidth=4: */

Class and variable names

Class names and enum names are written in StudlyCaps, with an initial capital letter and no underscores. Methods are StudlyCaps. Variable names are all lowercase, or lowerWithCaps. Non-static Member variables in classes should be prefixed with "m_"; other variables should not be preceded by "m_". Constants should be written with a "k" prefix and inner caps. The few global variables use a "g" prefix.

Make your variable names left-to-right specific. Related variables should sort together alphabetically; audioId and audioMutex should stand apart from videoTimeRemaining. The names should be compact, but be as self-explanatory as possible.

enum AudioChannelEnum {
    kChannelLeft      = 0x01,
    kChannelRight     = 0x02,
    kChannelCenter    = 0x04,
    kChannelLeftBack  = 0x08,
    kChannelRightBack = 0x10,
    kChannelBase      = 0x20,
    kChannelAll       = 0xff
} AudioChannels;

class AudioOutput
{
  public:
    AudioOutput(const QString &device_name);

    /// Adjust the volume by a delta, from +100 to -100.
    void AdjustVolume(int change);
    /// \return Device name
    QString GetDeviceName(void) const
    {
        QMutexLocker locker(&m_lock);
        return m_deviceName;
    }
  protected:
    mutable QMutex m_lock; // This is mutable so that 
    /// The absolute volume, from 0-100.
    int m_volume; // protected by m_lock
    QString m_deviceName; // protected by m_lock

    const static int kAudioBuffers;
};

It's a good idea to follow Google's C++ standards when they don't contradict our standards.

Translation guidelines

Important.png Note: This section is still a work in progress. Please do not apply those guidelines or edit this section for now.

Dont's

  • Never use QObject::tr() directly. You can use the tr() function of a class that inherits from QObject but do not directly call the QObject one... When you call QObject::tr() directly it puts all the strings to translate in a QObject "translation context" and this gives no clue to the translator as to what they are actually used for. There is also the possibility of translation clashes this way.
  • Don't build a sentence by concatenating together many words. That might work fine for languages which have a phrase structure similar to English but not for the others.
  • Avoid concatenating together many sentences unless they are separated by a carriage return. It's usually better to regroup all thoses sentences together and call the translation function on that group of sentences.

Do's

If your class inherits from QObject, you must use the Q_OBJECT macro in the class declaration (before you declare anything else). Please note that the class declaration should usually be in a .h header file. There are workarounds around this but it's preferable to avoid using them (it has to do with the way Qt adds, amongst other things, translation support).

If your class doesn't inherit from QObject, you need to call the Q_DECLARE_TR_FUNCTIONS macro to be able to use the tr() function from your class. As with the Q_OBJECT macro, it must be declared before anything else and you must provide it the class name (usually, it could be something else but that's not usually recommended).

e.g.

class AudioOutput
{
    Q_DECLARE_TR_FUNCTIONS(AudioOutput)

  public:
    AudioOutput(const QString &device_name);

<snip>
};

Once you have added the Q_OBJECT or Q_DECLARE_TR_FUNCTIONS macro you should be able to call the tr() function without any class prefix.

Translation context

What are those translation context we mention briefly earlier?

Usually, but it doesn't always have to be, it's the name of the class you call tr() from. It's part of the lookup key used to retrieve the translation from the translation files.

The lookup key used to retrieve the translations is composed of the translation context, the original untranslated string (in our case in US English) and, when provided, a disambiguation string (sometimes still called comments in Qt's documentation but they are way more than that..).

For now, let's focus on strings which do not have disambiguation strings.

Let's take the following example

void FirstClass::init(void)
{
    setLabel(tr("A String"));
}

void SecondClass::init(void)
{
    setLabel(tr("A String"));
}

This will, under normal circumstances (there are ways around it if the class doesn't inherit from QObject), create two different strings, one in the "FirstClass" context the other in the "SecondClass" context and both will need to be translated separately. The translation editor will notice these are identical and offer the translator to use the translation of the first one filled in the fill in the translation of the second one).

If I want to look up the translation of translation of either one of those strings from another class we will have to specify their context (either "FirstClass" or "SecondClass" in this case) either in by prefixing tr with the class name or by specifying their context to a function that allows use to specify the translation context such as QCoreApplication::translate().

Disambiguation string

It obviously couldn't be as simple as that..

Just kidding...

When the translation context and the original untranslated string is not enough to differentiate one string from the other, the disambiguation string comes to the rescue. It gets added to the lookup key and must be provided to retrieve that translation. It used to be the old way of commenting translation and is still documented in quite a few places in Qt's documentation (and even the translation file format calls the field in which they are stored comment) but don't be fooled, it becomes part of the lookup key and must be provided to lookup a translation which used them.

This is usually how they are used:

void FirstClass::init(void)
{
    setLabel(tr("A String"));
    setCategory(tr("Unknown", "Category"));
    setType(tr("Unknown", "Type"));
    setUnknown(tr(Unknown"));
}

All three "Unknown" strings will be in the context "FirstClass" but will have their own translation. The first two will have a disambiguation string, the third one will have none. This could be used for example when the gender of the nouns to which these strings refer to could affect the translation ("Unknown" could be written differently in another language depending on whether it refers to a feminine or masculine noun).

There is also a poorly documented (it appears to be only documented for the lower level (and which you should never have to use) QTranslator::translate function), the fallback to the string with no disambiguation string if the string with the same context, original string and disambiguation string is not present in the translation file. This could happen when the translation file is not up to date with the application or if one of the parameters passed to tr() (or the other translation functions) is a variable. We will discuss what to do when you have to pass variables to the translation functions later...

{{Note box|Never pass a null disambiguation string as parameter, if you need to provide one pass an empty string ("").}

But I lack class (or what to do when you are not in a member function)

Adding the tr() function using either the Q_OBJECT or Q_DECLARE_TR_FUNCTIONS macro only works for classes, a different approach must be used when your are not in a member function.

If you have a mix of member functions (ie part of a class) and regular (C-like) functions (which is pretty common in the settings pages for example), you can call the tr() function of the class from your function. The translations done that way will share the same translation context as the class which is usually what you want when those functions are called from the member functions of that class.

If you don't have any functions or at least any function you would like to share it's translation context with then you have to resort to using QCoreApplication::translate().

There are many ways to call it (more info will be added on them as I update this section) but in its simply way it must be called like so

QCoreApplication::translate("(My new translation context)", "The text I want translated")

qApp->translate("(My new translation context)", "The text I want translated")

Both of these examples are equivalent, the first time we directly call translate from QCoreApplication and the second time we call it from QApplication which inherits it from QCoreApplication. qApp is actually a macro that return a pointer to the unique application object. QApplication is meant to be used with GUI applications so it's usually safer to call it from QCoreApplication.

In MythTV, any context which is not a class name should be put in parenthesis. It makes those "virtual" contexts easier to spot and avoid name clashes with contexts which are actual class names.


Important.png Note: If you add any additional source code folders which contains strings which should be translated, please contact knightr on IRC or nriendeau (at) mythtv.org.

Comments

Comments should be compact; you should spend your time making your code readable and your variable names self-explanatory. Comments should give necessary overviews of functionality, or document strange cases, ranges, etc.

Description of variables' use and purpose should accompany the declaration. For class variables, a rough outline of how they fit into the algorithm should be given and any assumptions or notes on valid ranges should be given.

Doxygen comments are encouraged in the code. But keep them brief and don't use them when they don't add information not already obvious from the variable or function name.

MythWeb Style

MythWeb is written in PHP and javascript so a different style makes sense.

Comment Style

Outdented, meaning one indent level less then the code it's talking about, and above the code block, not below.

White Space Style

Soft Tab style (spaces, not tabs, and 4 spaces to a tabstop)

Classes

* Autoload
* Multiton pattern