zig cc exe OFFSETOF macro crash

[leonardo-ourmachinery]

zig 0.9

This code just crashes if compiled with zig cc -g

#include <stddef.h>
typedef struct foo
{
    int a;
}foo;

#define OFFSETOF(TYPE, ELEMENT) ((size_t)&(((TYPE *)0)->ELEMENT))

int main(int argc, char** argv)
{
    int offset = OFFSETOF(foo, a);
    (void) offset;
    
    return 0;
}

the crash happens on the OFFSETOF line:
Unhandled exception at 0x00007FF6955A100F in a.exe: 0xC000001D: Illegal Instruction.

The problem manifested in a dll compiled with zig cc in a slightly different way, but the crash is exactly the same so I suppose the root cause of the problem is the same:

I tried to isolate the problem and arrived at the small program that I posted above, hope it helps.

[floooh]

My guess is that the OFFSETOF() macro is undefined behaviour because of the null-pointer dereference. Compilers often inject an 'ud2' instruction when they detect UB. zig cc activates various clang sanitizers (see the command line that's dumped with zig cc -v), and maybe one of those sanitizers detects the nullptr dereference as UB and injects the illegal instruction.

FWIW, when replacing the custom OFFSETOF() macro with the standard offsetof() it works fine (offsetof() resolves to __builtin_offsetof() in clang, and it also works in GCC and MSVC, so IMHO it's no longer necessary to use a custom macro).

Also see this wikipedia snippet about potential undefined behavior (https://en.wikipedia.org/wiki/Offsetof):

"While this implementation works correctly in many compilers, it has generated some debate if this is undefined behavior according to the C standard,[2] since it appears to involve a dereference of a null pointer"

[leonardo-ourmachinery]

The actual problem was misaligned memory, I'm so sorry for opening the issue. (At least I learned that the OFFSETOF macro is UB :) )
The fact is that I didn't know (And I still don't) how to make the sanitizer print meaningful errors at runtime instead of just making the program crash, and so I just had to guess... but that's another story.

Thank you for the support and sorry for the useless issue.

Leonardo

[Vexu]

The fact is that I didn't know (And I still don't) how to make the sanitizer print meaningful errors at runtime instead of just making the program crash, and so I just had to guess... but that's another story.

That is an issue with Zig that is tracked in #5163.

Also no need to apologize for opening an issue but I do recommend asking in one of the communities if you're unsure whether something is a bug.