When are C++ macros beneficial?

Viewed 98923

The C preprocessor is justifiably feared and shunned by the C++ community. In-lined functions, consts and templates are usually a safer and superior alternative to a #define.

The following macro:

#define SUCCEEDED(hr) ((HRESULT)(hr) >= 0)  

is in no way superior to the type safe:

inline bool succeeded(int hr) { return hr >= 0; }

But macros do have their place, please list the uses you find for macros that you can't do without the preprocessor.

Please put each use-cases in a seperate answer so it can be voted up and if you know of how to achieve one of the answers without the preprosessor point out how in that answer's comments.

38 Answers

As wrappers for debug functions, to automatically pass things like __FILE__, __LINE__, etc:

#ifdef ( DEBUG )
#define M_DebugLog( msg )  std::cout << __FILE__ << ":" << __LINE__ << ": " << msg
#else
#define M_DebugLog( msg )
#endif

Since C++20 the magic type std::source_location can however be used instead of __LINE__ and __FILE__ to implement an analogue as a normal function (template).

Methods must always be complete, compilable code; macros may be code fragments. Thus you can define a foreach macro:

#define foreach(list, index) for(index = 0; index < list.size(); index++)

And use it as thus:

foreach(cookies, i)
    printf("Cookie: %s", cookies[i]);

Since C++11, this is superseded by the range-based for loop.

Header file guards necessitate macros.

Are there any other areas that necessitate macros? Not many (if any).

Are there any other situations that benefit from macros? YES!!!

One place I use macros is with very repetitive code. For example, when wrapping C++ code to be used with other interfaces (.NET, COM, Python, etc...), I need to catch different types of exceptions. Here's how I do that:

#define HANDLE_EXCEPTIONS \
catch (::mylib::exception& e) { \
    throw gcnew MyDotNetLib::Exception(e); \
} \
catch (::std::exception& e) { \
    throw gcnew MyDotNetLib::Exception(e, __LINE__, __FILE__); \
} \
catch (...) { \
    throw gcnew MyDotNetLib::UnknownException(__LINE__, __FILE__); \
}

I have to put these catches in every wrapper function. Rather than type out the full catch blocks each time, I just type:

void Foo()
{
    try {
        ::mylib::Foo()
    }
    HANDLE_EXCEPTIONS
}

This also makes maintenance easier. If I ever have to add a new exception type, there's only one place I need to add it.

There are other useful examples too: many of which include the __FILE__ and __LINE__ preprocessor macros.

Anyway, macros are very useful when used correctly. Macros are not evil -- their misuse is evil.

Mostly:

  1. Include guards
  2. Conditional compilation
  3. Reporting (predefined macros like __LINE__ and __FILE__)
  4. (rarely) Duplicating repetitive code patterns.
  5. In your competitor's code.

Inside conditional compilation, to overcome issues of differences between compilers:

#ifdef WE_ARE_ON_WIN32
#define close(parm1)            _close (parm1)
#define rmdir(parm1)            _rmdir (parm1)
#define mkdir(parm1, parm2)     _mkdir (parm1)
#define access(parm1, parm2)    _access(parm1, parm2)
#define create(parm1, parm2)    _creat (parm1, parm2)
#define unlink(parm1)           _unlink(parm1)
#endif

When you want to make a string out of an expression, the best example for this is assert (#x turns the value of x to a string).

#define ASSERT_THROW(condition) \
if (!(condition)) \
     throw std::exception(#condition " is false");

String constants are sometimes better defined as macros since you can do more with string literals than with a const char *.

e.g. String literals can be easily concatenated.

#define BASE_HKEY "Software\\Microsoft\\Internet Explorer\\"
// Now we can concat with other literals
RegOpenKey(HKEY_CURRENT_USER, BASE_HKEY "Settings", &settings);
RegOpenKey(HKEY_CURRENT_USER, BASE_HKEY "TypedURLs", &URLs);

If a const char * were used then some sort of string class would have to be used to perform the concatenation at runtime:

const char* BaseHkey = "Software\\Microsoft\\Internet Explorer\\";
RegOpenKey(HKEY_CURRENT_USER, (string(BaseHkey) + "Settings").c_str(), &settings);
RegOpenKey(HKEY_CURRENT_USER, (string(BaseHkey) + "TypedURLs").c_str(), &URLs);

Since C++20 it is however possible to implement a string-like class type that can be used as a non-type template parameter type of a user-defined string literal operator which allows such concatenation operations at compile-time without macros.

When you want to change the program flow (return, break and continue) code in a function behaves differently than code that is actually inlined in the function.

#define ASSERT_RETURN(condition, ret_val) \
if (!(condition)) { \
    assert(false && #condition); \
    return ret_val; }

// should really be in a do { } while(false) but that's another discussion.

The obvious include guards

#ifndef MYHEADER_H
#define MYHEADER_H

...

#endif

Unit test frameworks for C++ like UnitTest++ pretty much revolve around preprocessor macros. A few lines of unit test code expand into a hierarchy of classes that wouldn't be fun at all to type manually. Without something like UnitTest++ and it's preprocessor magic, I don't know how you'd efficiently write unit tests for C++.

Let's say we'll ignore obvious things like header guards.

Sometimes, you want to generate code that needs to be copy/pasted by the precompiler:

#define RAISE_ERROR_STL(p_strMessage)                                          \
do                                                                             \
{                                                                              \
   try                                                                         \
   {                                                                           \
      std::tstringstream strBuffer ;                                           \
      strBuffer << p_strMessage ;                                              \
      strMessage = strBuffer.str() ;                                           \
      raiseSomeAlert(__FILE__, __FUNCSIG__, __LINE__, strBuffer.str().c_str()) \
   }                                                                           \
   catch(...){}                                                                \
   {                                                                           \
   }                                                                           \
}                                                                              \
while(false)

which enables you to code this:

RAISE_ERROR_STL("Hello... The following values " << i << " and " << j << " are wrong") ;

And can generate messages like:

Error Raised:
====================================
File : MyFile.cpp, line 225
Function : MyFunction(int, double)
Message : "Hello... The following values 23 and 12 are wrong"

Note that mixing templates with macros can lead to even better results (i.e. automatically generating the values side-by-side with their variable names)

Other times, you need the __FILE__ and/or the __LINE__ of some code, to generate debug info, for example. The following is a classic for Visual C++:

#define WRNG_PRIVATE_STR2(z) #z
#define WRNG_PRIVATE_STR1(x) WRNG_PRIVATE_STR2(x)
#define WRNG __FILE__ "("WRNG_PRIVATE_STR1(__LINE__)") : ------------ : "

As with the following code:

#pragma message(WRNG "Hello World")

it generates messages like:

C:\my_project\my_cpp_file.cpp (225) : ------------ Hello World

Other times, you need to generate code using the # and ## concatenation operators, like generating getters and setters for a property (this is for quite a limited cases, through).

Other times, you will generate code than won't compile if used through a function, like:

#define MY_TRY      try{
#define MY_CATCH    } catch(...) {
#define MY_END_TRY  }

Which can be used as

MY_TRY
   doSomethingDangerous() ;
MY_CATCH
   tryToRecoverEvenWithoutMeaningfullInfo() ;
   damnThoseMacros() ;
MY_END_TRY

(still, I only saw this kind of code rightly used once)

Last, but not least, the famous boost::foreach !!!

#include <string>
#include <iostream>
#include <boost/foreach.hpp>

int main()
{
    std::string hello( "Hello, world!" );

    BOOST_FOREACH( char ch, hello )
    {
        std::cout << ch;
    }

    return 0;
}

(Note: code copy/pasted from the boost homepage)

Which is (IMHO) way better than std::for_each.

So, macros are always useful because they are outside the normal compiler rules. But I find that most the time I see one, they are effectively remains of C code never translated into proper C++.

You can't perform short-circuiting of function call arguments using a regular function call. For example:

#define andm(a, b) (a) && (b)

bool andf(bool a, bool b) { return a && b; }

andm(x, y) // short circuits the operator so if x is false, y would not be evaluated
andf(x, y) // y will always be evaluated

To fear the C preprocessor is like to fear the incandescent bulbs just because we get fluorescent bulbs. Yes, the former can be {electricity | programmer time} inefficient. Yes, you can get (literally) burned by them. But they can get the job done if you properly handle it.

When you program embedded systems, C uses to be the only option apart form assembler. After programming on desktop with C++ and then switching to smaller, embedded targets, you learn to stop worrying about “inelegancies” of so many bare C features (macros included) and just trying to figure out the best and safe usage you can get from those features.

Alexander Stepanov says:

When we program in C++ we should not be ashamed of its C heritage, but make full use of it. The only problems with C++, and even the only problems with C, arise when they themselves are not consistent with their own logic.

We use the __FILE__ and __LINE__ macros for diagnostic purposes in information rich exception throwing, catching and logging, together with automated log file scanners in our QA infrastructure.

For instance, a throwing macro OUR_OWN_THROW might be used with exception type and constructor parameters for that exception, including a textual description. Like this:

OUR_OWN_THROW(InvalidOperationException, (L"Uninitialized foo!"));

This macro will of course throw the InvalidOperationException exception with the description as constructor parameter, but it'll also write a message to a log file consisting of the file name and line number where the throw occured and its textual description. The thrown exception will get an id, which also gets logged. If the exception is ever caught somewhere else in the code, it will be marked as such and the log file will then indicate that that specific exception has been handled and that it's therefore not likely the cause of any crash that might be logged later on. Unhandled exceptions can be easily picked up by our automated QA infrastructure.

One common use is for detecting the compile environment, for cross-platform development you can write one set of code for linux, say, and another for windows when no cross platform library already exists for your purposes.

So, in a rough example a cross-platform mutex can have

void lock()
{
    #ifdef WIN32
    EnterCriticalSection(...)
    #endif
    #ifdef POSIX
    pthread_mutex_lock(...)
    #endif
}

For functions, they are useful when you want to explicitly ignore type safety. Such as the many examples above and below for doing ASSERT. Of course, like a lot of C/C++ features you can shoot yourself in the foot, but the language gives you the tools and lets you decide what to do.

I occasionally use macros so I can define information in one place, but use it in different ways in different parts of the code. It's only slightly evil :)

For example, in "field_list.h":

/*
 * List of fields, names and values.
 */
FIELD(EXAMPLE1, "first example", 10)
FIELD(EXAMPLE2, "second example", 96)
FIELD(ANOTHER, "more stuff", 32)
...
#undef FIELD

Then for a public enum it can be defined to just use the name:

#define FIELD(name, desc, value) FIELD_ ## name,

typedef field_ {

#include "field_list.h"

    FIELD_MAX

} field_en;

And in a private init function, all the fields can be used to populate a table with the data:

#define FIELD(name, desc, value) \
    table[FIELD_ ## name].desc = desc; \
    table[FIELD_ ## name].value = value;

#include "field_list.h"

Something like

void debugAssert(bool val, const char* file, int lineNumber);
#define assert(x) debugAssert(x,__FILE__,__LINE__);

So that you can just for example have

assert(n == true);

and get the source file name and line number of the problem printed out to your log if n is false.

If you use a normal function call such as

void assert(bool val);

instead of the macro, all you can get is your assert function's line number printed to the log, which would be less useful.

#define ARRAY_SIZE(arr) (sizeof arr / sizeof arr[0])

Unlike the 'preferred' template solution discussed in a current thread, you can use it as a constant expression:

char src[23];
int dest[ARRAY_SIZE(src)];

When you are making a decision at compile time over Compiler/OS/Hardware specific behavior.

It allows you to make your interface to Comppiler/OS/Hardware specific features.

#if defined(MY_OS1) && defined(MY_HARDWARE1)
#define   MY_ACTION(a,b,c)      doSothing_OS1HW1(a,b,c);}
#elif define(MY_OS1) && defined(MY_HARDWARE2)
#define   MY_ACTION(a,b,c)      doSomthing_OS1HW2(a,b,c);}
#elif define(MY_SUPER_OS)
          /* On this hardware it is a null operation */
#define   MY_ACTION(a,b,c)
#else
#error  "PLEASE DEFINE MY_ACTION() for this Compiler/OS/HArdware configuration"
#endif

You can use #defines to help with debugging and unit test scenarios. For example, create special logging variants of the memory functions and create a special memlog_preinclude.h:

#define malloc memlog_malloc
#define calloc memlog calloc
#define free memlog_free

Compile you code using:

gcc -Imemlog_preinclude.h ...

An link in your memlog.o to the final image. You now control malloc, etc, perhaps for logging purposes, or to simulate allocation failures for unit tests.

Compilers can refuse your request to inline.

Macros will always have their place.

Something I find useful is #define DEBUG for debug tracing -- you can leave it 1 while debugging a problem (or even leave it on during the whole development cycle) then turn it off when it is time to ship.

You can #define constants on the compiler command line using the -D or /D option. This is often useful when cross-compiling the same software for multiple platforms because you can have your makefiles control what constants are defined for each platform.

In my last job, I was working on a virus scanner. To make thing easier for me to debug, I had lots of logging stuck all over the place, but in a high demand app like that, the expense of a function call is just too expensive. So, I came up with this little Macro, that still allowed me to enable the debug logging on a release version at a customers site, without the cost of a function call would check the debug flag and just return without logging anything, or if enabled, would do the logging... The macro was defined as follows:

#define dbgmsg(_FORMAT, ...)  if((debugmsg_flag  & 0x00000001) || (debugmsg_flag & 0x80000000))     { log_dbgmsg(_FORMAT, __VA_ARGS__);  }

Because of the VA_ARGS in the log functions, this was a good case for a macro like this.

Before that, I used a macro in a high security application that needed to tell the user that they didn't have the correct access, and it would tell them what flag they needed.

The Macro(s) defined as:

#define SECURITY_CHECK(lRequiredSecRoles) if(!DoSecurityCheck(lRequiredSecRoles, #lRequiredSecRoles, true)) return
#define SECURITY_CHECK_QUIET(lRequiredSecRoles) (DoSecurityCheck(lRequiredSecRoles, #lRequiredSecRoles, false))

Then, we could just sprinkle the checks all over the UI, and it would tell you which roles were allowed to perform the action you tried to do, if you didn't already have that role. The reason for two of them was to return a value in some places, and return from a void function in others...

SECURITY_CHECK(ROLE_BUSINESS_INFORMATION_STEWARD | ROLE_WORKER_ADMINISTRATOR);

LRESULT CAddPerson1::OnWizardNext() 
{
   if(m_Role.GetItemData(m_Role.GetCurSel()) == parent->ROLE_EMPLOYEE) {
      SECURITY_CHECK(ROLE_WORKER_ADMINISTRATOR | ROLE_BUSINESS_INFORMATION_STEWARD ) -1;
   } else if(m_Role.GetItemData(m_Role.GetCurSel()) == parent->ROLE_CONTINGENT) {
      SECURITY_CHECK(ROLE_CONTINGENT_WORKER_ADMINISTRATOR | ROLE_BUSINESS_INFORMATION_STEWARD | ROLE_WORKER_ADMINISTRATOR) -1;
   }
...

Anyways, that's how I've used them, and I'm not sure how this could have been helped with templates... Other than that, I try to avoid them, unless REALLY necessary.

If you have a list of fields that get used for a bunch of things, e.g. defining a structure, serializing that structure to/from some binary format, doing database inserts, etc, then you can (recursively!) use the preprocessor to avoid ever repeating your field list.

This is admittedly hideous. But maybe sometimes better than updating a long list of fields in multiple places? I've used this technique exactly once, and it was quite helpful that one time.

Of course the same general idea is used extensively in languages with proper reflection -- just instrospect the class and operate on each field in turn. Doing it in the C preprocessor is fragile, illegible, and not always portable. So I mention it with some trepidation. Nonetheless, here it is...

(EDIT: I see now that this is similar to what @Andrew Johnson said on 9/18; however the idea of recursively including the same file takes the idea a bit further.)

// file foo.h, defines class Foo and various members on it without ever repeating the
// list of fields.

#if defined( FIELD_LIST )
   // here's the actual list of fields in the class.  If FIELD_LIST is defined, we're at
   // the 3rd level of inclusion and somebody wants to actually use the field list.  In order
   // to do so, they will have defined the macros STRING and INT before including us.
   STRING( fooString )
   INT( barInt )   
#else // defined( FIELD_LIST )

#if !defined(FOO_H)
#define FOO_H

#define DEFINE_STRUCT
// recursively include this same file to define class Foo
#include "foo.h"
#undef DEFINE_STRUCT

#define DEFINE_CLEAR
// recursively include this same file to define method Foo::clear
#include "foo.h"
#undef DEFINE_CLEAR

// etc ... many more interesting examples like serialization

#else // defined(FOO_H)
// from here on, we know that FOO_H was defined, in other words we're at the second level of
// recursive inclusion, and the file is being used to make some particular
// use of the field list, for example defining the class or a single method of it

#if defined( DEFINE_STRUCT )
#define STRING(a)  std::string a;
#define INT(a)     long a;
   class Foo
   {
      public:
#define FIELD_LIST
// recursively include the same file (for the third time!) to get fields
// This is going to translate into:
//    std::string fooString;
//    int barInt;
#include "foo.h"
#endif

      void clear();
   };
#undef STRING
#undef INT
#endif // defined(DEFINE_STRUCT)


#if defined( DEFINE_ZERO )
#define STRING(a) a = "";
#define INT(a) a = 0;
#define FIELD_LIST
   void Foo::clear()
   {
// recursively include the same file (for the third time!) to get fields.
// This is going to translate into:
//    fooString="";
//    barInt=0;
#include "foo.h"
#undef STRING
#undef int
   }
#endif // defined( DEFINE_ZERO )

// etc...


#endif // end else clause for defined( FOO_H )

#endif // end else clause for defined( FIELD_LIST )

Yet another foreach macros. T: type, c: container, i: iterator

#define foreach(T, c, i) for(T::iterator i=(c).begin(); i!=(c).end(); ++i)
#define foreach_const(T, c, i) for(T::const_iterator i=(c).begin(); i!=(c).end(); ++i)

Usage (concept showing, not real):

void MultiplyEveryElementInList(std::list<int>& ints, int mul)
{
    foreach(std::list<int>, ints, i)
        (*i) *= mul;
}

int GetSumOfList(const std::list<int>& ints)
{
    int ret = 0;
    foreach_const(std::list<int>, ints, i)
        ret += *i;
    return ret;
}

Better implementations available: Google "BOOST_FOREACH"

Good articles available: Conditional Love: FOREACH Redux (Eric Niebler) http://www.artima.com/cppsource/foreach.html

Often times I end up with code like:

int SomeAPICallbackMethod(long a, long b, SomeCrazyClass c, long d, string e, string f, long double yx) { ... }
int AnotherCallback(long a, long b, SomeCrazyClass c, long d, string e, string f, long double yx) { ... }
int YetAnotherCallback(long a, long b, SomeCrazyClass c, long d, string e, string f, long double yx) { ... }

In some cases I'll use the following to make my life easier:

#define APIARGS long a, long b, SomeCrazyClass c, long d, string e, string f, long double yx
int SomeAPICallbackMethod(APIARGS) { ... } 

It comes with the caveat of really hiding the variable names, which can be an issue in larger systems, so this isn't always the right thing to do, only sometimes.

I think this trick is a clever use of the preprocessor that can't be emulated with a function :

#define COMMENT COMMENT_SLASH(/)
#define COMMENT_SLASH(s) /##s

#if defined _DEBUG
#define DEBUG_ONLY
#else
#define DEBUG_ONLY COMMENT
#endif

Then you can use it like this:

cout <<"Hello, World!" <<endl;
DEBUG_ONLY cout <<"This is outputed only in debug mode" <<endl;

You can also define a RELEASE_ONLY macro.

You need a macros for resource identifiers in Visual Studio as the resource compiler only understands them (i.e., it doesn't work with const or enum).

Can you implement this as an inline function?

#define my_free(x) do { free(x); x = NULL; } while (0)

You can enable additional logging in a debug build and disable it for a release build without the overhead of a Boolean check. So, instead of:

void Log::trace(const char *pszMsg) {
    if (!bDebugBuild) {
        return;
    }
    // Do the logging
}

...

log.trace("Inside MyFunction");

You can have:

#ifdef _DEBUG
#define LOG_TRACE log.trace
#else
#define LOG_TRACE void
#endif

...

LOG_TRACE("Inside MyFunction");

When _DEBUG is not defined, this will not generate any code at all. Your program will run faster and the text for the trace logging won't be compiled into your executable.

Macros are useful for simulating the syntax of switch statements:

switch(x) {
case val1: do_stuff(); break;
case val2: do_other_stuff();
case val3: yet_more_stuff();
default:   something_else();
}

for non-integral value types. In this question:

Using strings in switch statements - where do we stand with C++17?

you'll find answers suggesting some approaches involving lambdas, but unfortunately, it's macros that get us the closest:

SWITCH(x)
CASE val1  do_stuff(); break;
CASE val2  do_other_stuff();
CASE val3  yet_more_stuff();
DEFAULT    something_else();
END
Related