Can macros be overloaded by number of arguments?

(Edit: See the end for a ready-made solution.)

To get an overloaded macro, first we need a macro which selects between several implementations. This part doesn't use a variadic macro. Then a variadic macro which generically counts its arguments produces a selector. Plugging the argument count into a dispatcher produces an overloaded macro.

Caveat: This system cannot tell the difference between zero and one arguments because there is no difference between no argument, and a single empty argument. They both look like MACRO().


To select between implementations, use the macro catenation operator with a series of function-like macros.

#define select( selector, ... ) impl ## _ ## selector( __VA_ARGS__ )
#define impl_1() meh
#define impl_2( abc, xyz ) # abc "wizza" xyz()
//etc

// usage: select( 1 ) => impl_1() => meh
//        select( 2, huz, bar ) => impl_2( huzza, bar ) => "huz" "wizza" bar()

Because the ## operator suppresses macro expansion of its arguments, it's better to wrap it in another macro.

#define CAT( A, B ) A ## B
#define SELECT( NAME, NUM ) CAT( NAME ## _, NUM )

To count arguments, use __VA_ARGS__ to shift arguments like so (this is the clever part):

#define GET_COUNT( _1, _2, _3, _4, _5, _6 /* ad nauseam */, COUNT, ... ) COUNT
#define VA_SIZE( ... ) GET_COUNT( __VA_ARGS__, 6, 5, 4, 3, 2, 1 )

Library code:

#define CAT( A, B ) A ## B
#define SELECT( NAME, NUM ) CAT( NAME ## _, NUM )

#define GET_COUNT( _1, _2, _3, _4, _5, _6 /* ad nauseam */, COUNT, ... ) COUNT
#define VA_SIZE( ... ) GET_COUNT( __VA_ARGS__, 6, 5, 4, 3, 2, 1 )

#define VA_SELECT( NAME, ... ) SELECT( NAME, VA_SIZE(__VA_ARGS__) )(__VA_ARGS__)

Usage:

#define MY_OVERLOADED( ... ) VA_SELECT( MY_OVERLOADED, __VA_ARGS__ )
#define MY_OVERLOADED_1( X ) foo< X >
#define MY_OVERLOADED_2( X, Y ) bar< X >( Y )
#define MY_OVERLOADED_3( X, Y, Z ) bang_ ## X< Y >.Z()

Following is an improvement upon Potatoswatter's answer, which can differentiate between zero and one argument.

In a nutshell, when __VA_ARGS__ is empty, EXPAND __VA_ARGS__ () inside VA_SIZE macro becomes EXPAND () and is substituted with 6 commas. So, VA_SIZE... becomes COMPOSE( GET_COUNT, (,,,,,, , 0, 6, 5, 4, 3, 2, 1) ), and that becomes GET_COUNT (,,,,,, , 0, 6, 5, 4, 3, 2, 1) and returns 0.

On the other hand, when __VA_ARGS__ is eg, int, 5, EXPAND __VA_ARGS__ () becomes EXPAND int, 5 (). So, VA_SIZE... becomes COMPOSE( GET_COUNT, (EXPAND int, 5 (), 0, 6, 5, 4, 3, 2, 1) ), which becomes GET_COUNT (EXPAND int, 5 (), 0, 6, 5, 4, 3, 2, 1) and returns 2, as described in Potatoswatter's answer.

I got the EXPAND idea from Jason Dang's answer.

Library code:

#define CAT( A, B ) A ## B
#define SELECT( NAME, NUM ) CAT( NAME ## _, NUM )
#define COMPOSE( NAME, ARGS ) NAME ARGS

#define GET_COUNT( _0, _1, _2, _3, _4, _5, _6 /* ad nauseam */, COUNT, ... ) COUNT
#define EXPAND() ,,,,,, // 6 commas (or 7 empty tokens)
#define VA_SIZE( ... ) COMPOSE( GET_COUNT, (EXPAND __VA_ARGS__ (), 0, 6, 5, 4, 3, 2, 1) )

#define VA_SELECT( NAME, ... ) SELECT( NAME, VA_SIZE(__VA_ARGS__) )(__VA_ARGS__)

Usage:

#define MY_OVERLOADED( ... ) VA_SELECT( MY_OVERLOADED, __VA_ARGS__ )

#define MY_OVERLOADED_0( ) meh()
#define MY_OVERLOADED_1( X ) foo< X >
#define MY_OVERLOADED_2( X, Y ) bar< X >( Y )
#define MY_OVERLOADED_3( X, Y, Z ) bang_ ## X< Y >.Z()

MY_OVERLOADED()                // meh()
MY_OVERLOADED(bool)            // foo< bool >
MY_OVERLOADED(int, 5)          // bar< int >( 5 )
MY_OVERLOADED(me, double, now) // bang_me< double >.now()

I would post this as a comment to Potatoswatter's post, but it's too long and requires a code listing.

Here is a bit of perl code for generating a set of macros which are meant to be overloaded macros.

$ perl -le 'map{
        $arity = $_; map {
                $ar = 2 + $arity + $_; $arm = $ar - 1; $arlist = join("", map{"A$_, "} 1..$arity); $warlist = "WHAT, $arlist";
                @li = map {"_$_"} 0..$_; $lis = join(", ", @li); $lim = pop @li; $lims = join(", ", @li);
                print "#define FEI_${arity}A_$ar($warlist$lis) FEI_${arity}A_$arm($warlist$lims) WHAT($_, $arlist$lim)"
        } 1..3; print ""
} 0..4'

Here is the output of the script:

#define FEI_0A_3(WHAT, _0, _1) FEI_0A_2(WHAT, _0) WHAT(1, _1)
#define FEI_0A_4(WHAT, _0, _1, _2) FEI_0A_3(WHAT, _0, _1) WHAT(2, _2)
#define FEI_0A_5(WHAT, _0, _1, _2, _3) FEI_0A_4(WHAT, _0, _1, _2) WHAT(3, _3)

#define FEI_1A_4(WHAT, A1, _0, _1) FEI_1A_3(WHAT, A1, _0) WHAT(1, A1, _1)
#define FEI_1A_5(WHAT, A1, _0, _1, _2) FEI_1A_4(WHAT, A1, _0, _1) WHAT(2, A1, _2)
#define FEI_1A_6(WHAT, A1, _0, _1, _2, _3) FEI_1A_5(WHAT, A1, _0, _1, _2) WHAT(3, A1, _3)

#define FEI_2A_5(WHAT, A1, A2, _0, _1) FEI_2A_4(WHAT, A1, A2, _0) WHAT(1, A1, A2, _1)
#define FEI_2A_6(WHAT, A1, A2, _0, _1, _2) FEI_2A_5(WHAT, A1, A2, _0, _1) WHAT(2, A1, A2, _2)
#define FEI_2A_7(WHAT, A1, A2, _0, _1, _2, _3) FEI_2A_6(WHAT, A1, A2, _0, _1, _2) WHAT(3, A1, A2, _3)

#define FEI_3A_6(WHAT, A1, A2, A3, _0, _1) FEI_3A_5(WHAT, A1, A2, A3, _0) WHAT(1, A1, A2, A3, _1)
#define FEI_3A_7(WHAT, A1, A2, A3, _0, _1, _2) FEI_3A_6(WHAT, A1, A2, A3, _0, _1) WHAT(2, A1, A2, A3, _2)
#define FEI_3A_8(WHAT, A1, A2, A3, _0, _1, _2, _3) FEI_3A_7(WHAT, A1, A2, A3, _0, _1, _2) WHAT(3, A1, A2, A3, _3)

#define FEI_4A_7(WHAT, A1, A2, A3, A4, _0, _1) FEI_4A_6(WHAT, A1, A2, A3, A4, _0) WHAT(1, A1, A2, A3, A4, _1)
#define FEI_4A_8(WHAT, A1, A2, A3, A4, _0, _1, _2) FEI_4A_7(WHAT, A1, A2, A3, A4, _0, _1) WHAT(2, A1, A2, A3, A4, _2)
#define FEI_4A_9(WHAT, A1, A2, A3, A4, _0, _1, _2, _3) FEI_4A_8(WHAT, A1, A2, A3, A4, _0, _1, _2) WHAT(3, A1, A2, A3, A4, _3)

These are the (regularly structured sections of) groups of macro overloads that are used for generating FOR_EACH (a.k.a. FE) macros which can dispatch a WHAT macro optionally with an arbitrary number of constant arguments (A1, A2...) in addition to an arbitrary number of arguments in a list, along with an index in the proper ordering (a naive implementation without using something like SELECT for overloading would yield reversed indices).

As an example, the remaining section (the nonregular "base case" portion of the second block) looks like the following:

#define FE_INDEXED_1ARG(...) VA_SELECT(FEI_1A, __VA_ARGS__)
#define FEI_1A_3(WHAT, A1, _0) WHAT(0, A1, _0)

The utility of this can perhaps be brought into question (I built it because I saw a use for it...), and neither does this answer the OP's question directly (in fact, it sort of does the opposite -- a foreach construct does the same thing to all variadic arguments...), but I just thought that the technique is pretty interesting (as well as utterly horrifying in some ways) and allows for quite some expressive power using the preprocessor and it will be possible to generate very efficient machine code in this manner. I think it also serves as a poignant example of why I personally think the C preprocessor still has room for improvement.

By which I mean that the C preprocessor is an absolute abomination and we should probably scrap it and start from scratch :)


Though it's already answered, I have prepared a very short version of it. Hope it may help.

Implementation

// Variable Argument Macro (VA_MACRO) upto 6 arguments
#define NUM_ARGS_(_1, _2, _3, _4, _5, _6, TOTAL, ...) TOTAL
#define NUM_ARGS(...) NUM_ARGS_(__VA_ARGS__, 6, 5, 4, 3, 2, 1)

#define CONCATE_(X, Y) X##Y  // Fixed the double '_' from previous code
#define CONCATE(MACRO, NUMBER) CONCATE_(MACRO, NUMBER)
#define VA_MACRO(MACRO, ...) CONCATE(MACRO, NUM_ARGS(__VA_ARGS__))(__VA_ARGS__)

Customization

// This is how user may define own set of variadic macros
#define MY_MACRO(...) VA_MACRO(MY_MACRO, __VA_ARGS__)
#define MY_MACRO1(_1) "One"
#define MY_MACRO2(_1, _2) "Two"
#define MY_MACRO3(_1, _2, _3) "Three"

Usage

// While using those, user needs to use only the main macro
int main ()
{
  auto one = MY_MACRO(1);
  auto two = MY_MACRO(1, 2); 
  auto three = MY_MACRO(1, 2, 3); 
}