Yes, recursive macros are possible in C using a fancy workaround. The end goal is to create a MAP
macro which works like this:
#define PRINT(a) printf(#a": %d", a)
MAP(PRINT, a, b, c) /* Apply PRINT to a, b, and c */
Basic Recursion
First, we need a technique for emitting something that looks like a macro
call, but isn't yet:
#define MAP_OUT
Imagine we have the following macros:
#define A(x) x B MAP_OUT (x)
#define B(x) x A MAP_OUT (x)
Evaluating the macro A (blah)
produces the output text:
blah B (blah)
The preprocessor doesn't see any recursion, since the B (blah)
call is
just plain text at this point, and B
isn't even the name of the current
macro. Feeding this text back into the preprocessor expands the call,
producing the output:
blah blah A (blah)
Evaluating the output a third time expands the A (blah)
macro, carrying
the recursion full-circle. The recursion continues as long as the caller
continues to feed the output text back into the preprocessor.
To perform these repeated evaluations, the following EVAL
macro passes
its arguments down a tree of macro calls:
#define EVAL0(...) __VA_ARGS__
#define EVAL1(...) EVAL0 (EVAL0 (EVAL0 (__VA_ARGS__)))
#define EVAL2(...) EVAL1 (EVAL1 (EVAL1 (__VA_ARGS__)))
#define EVAL3(...) EVAL2 (EVAL2 (EVAL2 (__VA_ARGS__)))
#define EVAL4(...) EVAL3 (EVAL3 (EVAL3 (__VA_ARGS__)))
#define EVAL(...) EVAL4 (EVAL4 (EVAL4 (__VA_ARGS__)))
Each level multiplies the effort of the level before, evaluating the input
365 times in total. In other words, calling EVAL (A (blah))
would
produce 365 copies of the word blah
, followed by a final un-evaluated B (blah)
. This provides the basic framework for recursion, at least within a
certain stack depth.
End Detection
The next challenge is to stop the recursion when it reaches the end of the
list.
The basic idea is to emit the following macro name instead of the normal
recursive macro when the time comes to quit:
#define MAP_END(...)
Evaluating this macro does nothing, which ends the recursion.
To actually select between the two macros, the following MAP_NEXT
macro compares a single list item against the special end-of-list marker
()
. The macro returns MAP_END
if the item matches, or the next
parameter if the item is anything else:
#define MAP_GET_END() 0, MAP_END
#define MAP_NEXT0(item, next, ...) next MAP_OUT
#define MAP_NEXT1(item, next) MAP_NEXT0 (item, next, 0)
#define MAP_NEXT(item, next) MAP_NEXT1 (MAP_GET_END item, next)
This macro works by placing the item next to the MAP_GET_END
macro. If
doing that forms a macro call, everything moves over by a slot in the
MAP_NEXT0
parameter list, changing the output. The MAP_OUT
trick
prevents the preprocessor from evaluating the final result.
Putting it All Together
With these pieces in place, it is now possible to implement useful versions
of the A
and B
macros from the example above:
#define MAP0(f, x, peek, ...) f(x) MAP_NEXT (peek, MAP1) (f, peek, __VA_ARGS__)
#define MAP1(f, x, peek, ...) f(x) MAP_NEXT (peek, MAP0) (f, peek, __VA_ARGS__)
These macros apply the operation f
to the current list item x
. They then
examine the next list item, peek
, to see if they should continue or not.
The final step is to tie everything together in a top-level MAP
macro:
#define MAP(f, ...) EVAL (MAP1 (f, __VA_ARGS__, (), 0))
This macro places a ()
marker on the end of the list, as well as an extra
0
for ANSI compliance (otherwise, the last iteration would have an illegal
0-length list). It then passes the whole thing through EVAL
and
returns the result.
I have uploaded this code as a library on github for your convenience.