I spent many hours debugging a problem that turned out to be caused by
two source files including two header files in a different order.
One of those headers defined _FILE_OFFSET_BITS
to 64, and the other header
file included <sys/types.h>
, which defined off_t
to be either 32 or 64
bits long, depending on the setting of _FILE_OFFSET_BITS
. I've
included below a short example of this situation. This was on x86_32 Linux
(both Debian unstable and CentOS 4.8).
Neither gcc -Wall main.c other.c
, nor Solaris 9 lint, nor splint detects
this situation.
Does anyone know of a software tool that can detect this situation?
#define _FILE_OFFSET_BITS 64
#include <sys/types.h>
#include <stdio.h>
#include "header.h"
int
main(int argc, char **argv) {
struct foo bar = {(off_t) 0, "foo"};
showproc(&bar);
printf("sizeof(off_t) in main.c is %d\n", sizeof(off_t));
return 0;
}
#include <sys/types.h>
#define _FILE_OFFSET_BITS 64
#include <stdio.h>
#include "header.h"
void
showproc(const struct foo *p)
{
if (p->offset == 0) {
if (p->s == NULL)
puts("NULL pointer reference");
else
printf("Structure value is %s\n", p->s);
}
printf("sizeof(off_t) in other.c is %d\n", sizeof(off_t));
}
struct foo {
off_t offset;
const char * s;
};
extern void showproc(const struct foo *);
NULL pointer reference
sizeof(off_t) in other.c is 4
sizeof(off_t) in main.c is 8
I would recommend putting defines that modify headers like that in the makefile instead of in the code. Otherwise you can't be sure which compilation units have one definition or the other, as you've experienced.
Sometimes modifying headers with macros is the intended behavior (e.g. using headers as templates) and sometimes it isn't (like in your case), so it is hard to produce meaningful warnings from a tool, I think.
If you need to define something in a header file, make sure that anything that uses it includes that header. That includes other headers.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With