no warnings with -Waddress

[llvmbot]

Bugzilla Link	43647
Version	trunk
OS	Linux
Reporter	LLVM Bugzilla Contributor
CC	@zygoloid

Extended Description

clang-10 falls to emit the warning for the following:

$ cat s.c
int main()
{
int *a = (void *) 0;
int b = (&a) == ((void *) 0);
return b;
}

$ clang-10 -Weverything -pedantic s.c

clang version 10.0.0 (https://github.com/llvm/llvm-project.git 49c4e58)
Target: x86_64-unknown-linux-gnu
Thread model: posix

[-Waddress] is enabled by default in Clang, but do not issue the warning. When compiled with GCC, warnings such as:

$ gcc -Wall s.c
s.c: In function ‘main’:
s.c:4:16: warning：the comparison will always evaluate as ‘false’ for the address of ‘a’ will never be NULL [-Waddress]
4 | int b = (&a) == ((void *) 0);
| ^~

[Krishna-13-cyber]

Hi!,
I am trying to solve this issue by targeting def warn_null_pointer_compare which does not have variable at present, I suppose due to which it does not catch the address of the pointer to variable.
Please do let me know if its a right approach and on the correct track.
CC: @tbaederr @AaronBallman @cjdb @Quuxplusone

[AaronBallman]

Please do let me know if its a right approach and on the correct track.

I think it's actually a bit easier than that, even. It looks like Sema::DiagnoseAlwaysNonNullPointer() is not skipping the paren expression node. When you remove the parens from (&a), we diagnose correctly: https://godbolt.org/z/EWTYWqP9Y

I'd recommend looking through that function to see where we're skipping over nodes we don't care about and ensure we skip over paren AST nodes as well.

[Krishna-13-cyber]

Thanks a lot @AaronBallman !
I will work on the suggested approach.

[Krishna-13-cyber]

I have updated with a patch https://reviews.llvm.org/D149000
Thanks!