preprocessor concatenation issue ...

Mark Nicholson mnichols@cosinecom.com
Wed Dec 3 17:07:00 GMT 2003


Hi,

	I'm trying to migrate from gcc 2.95.2 to gcc 3.3.2 and I've run 
into an odd incompatibility.  Here is the test code:


typedef struct
{
	struct
	{
		unsigned long	allocs;
		unsigned long	fails;
		unsigned long	frees;
	} sleeplocks;
	struct
	{
		unsigned long	allocs;
		unsigned long	fails;
		unsigned long	frees;
	} rwsleeplocks;
	struct
	{
		unsigned long	allocs;
		unsigned long	fails;
		unsigned long	frees;
	} conditions;
	struct
	{
		unsigned long	allocs;
		unsigned long	fails;
		unsigned long	frees;
	} events;

}	lockstats_t;

lockstats_t lockstats;

#define LOCKSTATS_ALLOCS_INC(_locktype) \
	lockstats. ##_locktype##.allocs++
#define LOCKSTATS_FAILS_INC(_locktype) \
	lockstats.##_locktype##.fails++
#define LOCKSTATS_FREES_INC(_locktype) \
	lockstats.##_locktype##.frees++

int main( void )
{
	LOCKSTATS_FREES_INC(sleeplocks);

	/*
	 *  desire to get:
	 *
	 *   lockstats.sleeplocks.frees++;
	 */

	return( 0 );
}

	I expect to get a main routine with the line:

		lockstats.sleeplocks.frees++;

	in place of the macro.

	When I compile this with GCC 2.95.2, it compiles fine (and works). 
When I use GCC 3.3.2, I get these errors:

tcsh-102: gcc -Wall -o go substitution.c
substitution.c:42:39: pasting "." and "sleeplocks" does not give a valid 
preprocessing token
substitution.c:42:39: pasting "sleeplocks" and "." does not give a valid 
preprocessing token
tcsh-103: 

	Any ideas if this is actually a bug?  Can anyone suggest a 
workaround?

	By the way, when I recompile it with --save-temps, the code it 
spits out is correct (trimmed for brevity):

lockstats_t lockstats;
# 43 "substitution.c"
int main( void )
{
        lockstats.sleeplocks.frees++;







        return( 0 );
}

	Thanks in advance for any wisdom and help!

Cheers!

Mark Nicholson




More information about the Gcc-help mailing list