ACE Implementation Guidelines

Coding Guidelines

• General
  • Every text file must end with a newline.

  • Use spaces instead of tabs, except in Makefiles. Emacs users can add this to their .emacs:

    (setq-default indent-tabs-mode nil)

    Microsoft Visual C++ users should do the following:

            Choose:  Tools -- Options -- Tabs
            Then Set:  "Tab size" to 8 and "Indent size" to 2, and
            indent using spaces.
            

  • Every program should have a ``usage'' message. It should be printed out if erroneous command line arguments, or a -? command line argument, are provided to the program.

  • The program main function should always be declared with arguments, e.g.,

            int
            main (int argc, char *argv[])
            {
              [...]
    
              return 0;
            }
            

    It should also return 0 on successful termination, and non-zero otherwise.

• Code Documentation
  • Use comments and whitespace (:-) liberally.

  • Insert a CVS/RCS keyword string at the top of every source file, Makefile, config file, etc. For C++ files, it is:

            // $Id$
            

    It is not necessary to fill in the fields of the keyword string, or modify them when you edit a file that already has one. CVS does that automatically when you checkout or update the file.

    To insert that string at the top of a file:

            perl -pi -e 'if (! $o) {printf "// \$Id\$\n\n";}; $o = 1;' file
            

  • Comments, especially in header files, must follow the OSE Tools format requirements. Please see the ``Classinfo Tools'' section of the OSE ``Tools Manual'' for these requirements.

    Please note that the Classinfo tools in OSE have been developed independently since the copies in ACE were made and thus they support new features not supported in the ACE version of the tools. Certain subtle changes were also made in the ACE copies for indicating sections, etc., which makes them different from the OSE documentation. Please see the ACE-doctools page for descriptions of these changes.

• Preprocessor
  • Never #include standard headers directly, except in a few specific ACE files, e.g., OS.h and stdcpp.h. Let those files #include the correct headers. If you do not do this, your code will not compile with the Standard C++ Library.

  • Always follow a preprocessor #endif with a /* */ C-style comment. It should correspond to the condition in the matching #if directive. For example,

            #if defined (ACE_HAS_THREADS)
            # if defined (ACE_HAS_STHREADS)
            #   include /**/ <synch.h>
            #   include /**/ <thread.h>
            #   define ACE_SCOPE_PROCESS P_PID
            #   define ACE_SCOPE_LWP P_LWPID
            #   define ACE_SCOPE_THREAD (ACE_SCOPE_LWP + 1)
            # else
            #   define ACE_SCOPE_PROCESS 0
            #   define ACE_SCOPE_LWP 1
            #   define ACE_SCOPE_THREAD 2
            # endif /* ACE_HAS_STHREADS */
            #endif /* ACE_HAS_THREADS */
            

  • Always insert a /**/ between an #include and filename, as shown in the above example. This avoids dependency problems with Visual C++.

• C++ Syntax and Constructs
  • for loops should look like:

            for (size_t i = 0; i < Options::instance ()->spawn_count (); i++)
              spawn ();
            

    (though I prefer to always wrap the body of the loop in braces, to avoid surprises when other code or debugging statements are added, and to maintain sanity when the body consists of a macro, such as ACE_ASSERT:)

            for (size_t i = 0; i < Options::instance ()->spawn_count (); i++)
              {
                ACE_ASSERT (spawn () == 0);
              }
            

    Similarly, if statements should have a space after the ``if'', and no spaces just after the opening parenthesis and just before the closing parenthesis.

  • Avoid unnecessary parenthesis. We're not writing Lisp :-)

  • Put inline member functions in a .i file. That file is conditionally included by both the .h file, for example:

                class ACE_Export ACE_High_Res_Timer
                {
                  [...]
                };
    
                #if defined (__ACE_INLINE__)
                #include "ace/High_Res_Timer.i"
                #endif /* __ACE_INLINE__ */
                

    and .cpp file:

                #define ACE_BUILD_DLL
                #include "ace/High_Res_Timer.h"
    
                #if !defined (__ACE_INLINE__)
                #include "ace/High_Res_Timer.i"
                #endif /* __ACE_INLINE__ */
    
                ACE_ALLOC_HOOK_DEFINE(ACE_High_Res_Timer)
                

    NOTE: It is very important to ensure than an inline function will not be used before its definition is seen. Therefore, the inline functions in the .i file should be arranged properly. Some compilers, such as g++ with the -Wall option, will issue warnings for violations.

  • Mutators and accessors should be of this form:

            void object_addr (const ACE_INET_Addr &);
            // Sets <object_addr_> cache from <host> and <port>.
    
            ACE_INET_Addr &object_addr (void);
            // Returns the <ACE_INET_Addr> for this profile.
            

    instead of the ``set_'' and ``get_'' form.

  • Never use delete to deallocate memory that was allocated with malloc. Similarly, never associate free with new. ACE_NEW or ACE_NEW_RETURN should be used to allocate memory, and delete should be used to deallocate it. And be careful to use the correct form, delete or delete [] to correspond to the allocation.

  • Always use ACE_NEW or ACE_NEW_RETURN to allocate memory, because they check for successful allocation and set errno appropriately if it fails.

  • Never compare or assign a pointer value with NULL; use 0 instead. The language allows any pointer to be compared or assigned with 0. The definition of NULL is implementation dependent, so it is difficult to use portably without casting.

  • Never cast a pointer to or from an int. On all currently supported ACE platforms, it is safe to cast a pointer to or from a long.

  • Be very careful when selecting an integer type that must be a certain size, e.g., 4 bytes. long is not 4 bytes on all platforms; it is 8 bytes on many 64-bit machines. ACE_UINT32 is always 4 bytes, and ACE_UINT64 is always 8 bytes.

  • If a class has any virtual functions, and its destructor is declared explicitly in the class, then the destructor should always be virtual as well. And to support compiler activities such as generation of virtual tables and, in some cases, template instantiation, the virtual destructor should not be inline. (Actually, any non-pure virtual function could be made non-inline for this purpose. But, for convenience, if its performance is not critical, it is usually easiest just to make the virtual destructor non-inline.)

  • Constructor initializers must appear in the same order as the data members are declared in the class header. This avoids subtle errors, because initialization takes place in the order of member declaration.

  • Initialization is usually cleaner than assignment, especially in a conditional. So, instead of writing code like this:

            ssize_t n_bytes;
    
            // Send multicast of one byte, enough to wake up server.
            if ((n_bytes = multicast.send ((char *) &reply_port, sizeof reply_port)) == -1)
            

    Write it like this:

            ssize_t n_bytes = multicast.send ((char *) &reply_port, sizeof reply_port)
    
            // Send multicast of one byte, enough to wake up server.
            if (n_bytes == -1)
            

    But, beware if the initialization is of a static variable. A static variable is only initialized the first time its declaration is seen. Of course, we should avoid using static variables at all.

  • It is usually clearer to write conditionals that have both branches without a negated condition. For example,

            if (test)
              {
                // true branch
              }
            else
              {
                // false branch
              }
            

    is preferred over:

            if (! test)
              {
                // false test branch
              }
            else
              {
                // true test branch
              }
            

  • If a cast is necessary, avoid use of ANSI-style casts, e.g., int (foo). Instead, use the ACE_static_cast macro:

            return ACE_static_cast(size_t, this->count_) > that->size_;
            

  • In general, if instances of a class should not be copied, then a private copy constructor and assignment operator should be declared for the class, but not implemented. For example:

            // Disallow copying by not implementing the following . . .
            ACE_Object_Manager (const ACE_Object_Manager &);
            ACE_Object_Manager &operator= (const ACE_Object_Manager &);
            

    If the class is a template class, then the ACE_UNIMPLEMENTED_FUNC macro should be used:

            // = Disallow copying...
            ACE_UNIMPLEMENTED_FUNC (ACE_TSS (const ACE_TSS<TYPE> &))
            ACE_UNIMPLEMENTED_FUNC (void operator= (const ACE_TSS<TYPE> &))
            

    ACE_UNIMPLEMENTED_FUNC can be used with non-template classes as well. Though for consistency and maximum safety, it should be avoided for non-template classes.

• I/O
  • Use ACE_DEBUG for printouts, and ACE_OS::scanf/fprintf () for file I/O. Avoid using iostreams because of implementation differences across platforms.

ACE Usage Guidelines

• Always use ACE_OS (static) member functions instead of bare OS system calls.

• Use the ACE_SYNCH_MUTEX macro, instead of using one of the specific mutexes, such as ACE_Thread_Mutex. This provides portability between threaded and non-threaded platforms.

• Avoid creating a static instance of user-defined (class) type. Instead, either create it as an ACE_Singleton, ACE_TSS_Singleton, or as an ACE_Cleanup object. See the ACE Singleton.h, Object_Manager.h, and Managed_Object.h header files for more information.

  Static instances of built-in types, such as int or any pointer type, are fine.

  Construction of static instance of a user-defined type should never spawn threads. Because order of construction of statics across files is not defined by the language, it is usually assumed that only one thread exists during static construction. This allows statics suchs as locks to be safely created. We do not want to violate this assumption.

• Do not use run-time type identification (RTTI). Some platforms do not support it.

• Do not use exception handling. Some platforms do not support it. And, it imposes an execution speed penalty.

• Because ACE does not use exception handling, dealing with failures requires a bit of care. This is especially true in constructors. Consider the following approach:

        ACE_NEW_RETURN (this->name_space_, LOCAL_NAME_SPACE, -1);
  
        if (ACE_LOG_MSG->op_status () != 0)
        ....
        

  This snip of code is from ACE_Naming_Context. All failed constructors in ACE (should) call ACE_ERROR. This sets the thread specific op_status, which can be checked by the caller. This mechanism allows the caller to check for a failed constructor without the requiring the constructor to throw exceptions.

• Avoid using the C++ Standard Template Library (STL) in our applications. Some platforms do not support it yet.

ACE Design Rules

Initialization on first use is preferred over using static initializations, especially when creating OO class libraries. I.e., use Double Check Locking.

This is a well known problem among OO C++ class designers. See Designing and Coding Reusable C++ by Caroll and Ellis, for example.

If you are parameterizing a searchable container template class with readers/writer locks, do not use sentinels or any state that requires updating during a a search.

....

In multi-threaded objects, all methods should be re-entrant..

Framework users should always use wrappers instead of directly making system calls.

....

Always guard against multiple inclusion of header files..

Templates may require source. Hence, source files may need to be guarded against multiple inclusion too.

Template classes should not be mixed with non-template classes..

Guard against quirky compiler implementations through #define abstractions (whereever possible)..

Use consistent error-handling..

Abstract away different representations of handles / file descriptors..

Avoid the use of global functions..

If functions are required, place them within a namespace or nested in a class..

Avoid polluting the namespace..

Don't use the inline keyword directly..

Separate inline code into their own implementation files. I.e. do not write code in header files..

(Internationalization) Do not use the char type directly. Abstract it for wide characters. Thus, do not assume your character abstraction is a single byte..

Don't use conditional compilation based on compiler/OS/hardware platform, but upon available features..

Centralize portability #ifdefs in a single place to ease portability maintainence..

Either check or return all return values (from framework calls)..

Apply strategy pattern to factor out common sources of variation in a component..

Use traits to merge multiple template arguments..

Allow customization of memory allocation.

Instrument framework methods to allow for run-time tracing.

Avoid long chains of dereference operators.

Use open() methods rather than constructors.

Instrument a dump() method in every object.

Qualify references to base class methods and data. I.e., anything outside the scope of the class.

Guard against multiple open() and close() calls, for idempotency.

The state of an object does not need to be guarded in destructors.

Base classes should not store information about derived classes.

Don't make static objects whose correctness depends on constructors being called. (That is, if 0 initialization is not sufficient.)

Guarantee singleton destruction. E.g., provide hooks that allow singletons to be deleted before the process exits.

All uses of the placement new operator must be accompanied by a corresponding explicit destructor call.

In classes that require locking, have public methods acquire the locks and call protected/private methods that do not acquire the locks.

Try to minimize the number of template parameters passed to a base class. Try to have a base class that is not a template, from which specialized classes can be derived that may have template parameters.

Define wrappers around clusters of functionality that are semantically the same but may have accidental incompatibilities. E.g., semaphores, readers/writer locks, mutex + condition variables, which can be used by threads, processes, that may or may not reside on the same machine.

Define generic components that can be parameterized by the wrappers defined above (in the previous point).

Separate creation from use. E.g., acceptor/connector and service handler.

In searchable containers, separate traversal logic from item operations.

...

CVS Usage Guidelines

• Always make sure that a change builds and executes correctly on at least one platform before checking it into the CVS repository.

Software Engineering Guidelines

• Advise: Keep other developers informed of problems and progress.

• Authorize: We have contractual obligations to not unilaterally change interfaces. If you need to change or remove an interface, get an OK.

• Minimize risk: Test all changes. Solicit review of changes.

• Revise only when necessary: Every change has risk, so avoid making any change unless there is a good reason for it.

• Normalize: Factor out commonality. For example, maintain a data value in only one place.

• Synthesize: Build stubs and scaffolding early to simulate the complete system. Maintain a checked-in version of the system that cleanly builds and tests at all times.