Bad gcc -O optimization cause core dump. What to do?

[Maxime Henrion]

Ricardo Nabinger Sanchez wrote:
> On Tue, 13 Mar 2007 15:11:07 +0300
> Andrey Chernov <ache at freebsd.org> wrote:
> 
> > cc -O -S a.c
> > 	.file	"a.c"
> > 	.text
> > 	.p2align 2,,3
> > .globl main
> > 	.type	main, @function
> > main:
> > 	pushl	%ebp
> > 	movl	%esp, %ebp
> > 	subl	$8, %esp
> > 	andl	$-16, %esp
> > 	subl	$28, %esp
> > 	pushl	$0
> > 	call	puts
> > 	leave
> > 	ret
> > 	.size	main, .-main
> > 	.ident	"GCC: (GNU) 3.4.6 [FreeBSD] 20060825"
> 
> Confirmed on FreeBSD-6.1 RELEASE:
> 
>         .file   "bla.c"
>         .text
>         .p2align 2,,3
> .globl main
>         .type   main, @function
> main:
>         pushl   %ebp
>         movl    %esp, %ebp
>         subl    $8, %esp
>         andl    $-16, %esp
>         subl    $28, %esp
>         pushl   $0
>         call    puts
>         leave
>         ret
>         .size   main, .-main
>         .ident  "GCC: (GNU) 3.4.4 [FreeBSD] 20050518"
> 
> > It calls "puts(NULL)" with core dump.
> > It means "printf("%s\n", NULL)" is overoptimized.
> > BTW, things like "printf("1%s\n", NULL)" are not overoptimized.
> > Any ideas? Is it right or needs to be fixed?
> 
> Given that this is not what the user asked (replacing printf with puts), I
> consider this a bug.  GCC made its assumption, and it was incorrect--it's not
> user's fault.

GCC can do whatever it wants here, even printing "foobar42", because the
C standard says that passing a NULL pointer to a %s format will yield
undefined behaviour.  It *is* user's fault to have passed NULL to
printf() in the first place.

So, while we could argue that GCC's behaviour here is useless, annoying,
etc, this just can't be called a bug in GCC.  As a side note, these
"optimizations" are in place since a *long* time now.

Cheers,
Maxime