Re: bugs in strlen, strstr setlocale -- i386

On Fri, Apr 03, 2009 at 06:05:29PM +0200, Frank Bergmann wrote:
Hi Frank!

> at least 32-Bit Intel-Arch should not segfault due to jecxz:
> > cat dietlibc-0.31/i386/strlen.S
> [...]
> strlen:
>         movl    4(%esp),%ecx
>         xorl    %eax,%eax
>         jecxz   .Lnull
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...

You are right!  I know about this line above.

vladov@picard:4 /tmp$ cat t.c               
#include <string.h>

int main() {
  return strlen((char *)0);
}

vladov@picard:4 /tmp$ diet -Os gcc -Wall t.c
t.c: In function 'main':
t.c:4: warning: null argument where non-null required (argument 1)
vladov@picard:4 /tmp$ ./a.out 
Segmentation fault (core dumped)

vladov@picard:4 /tmp$ diet gcc -Wall -O2 t.c
t.c: In function 'main':
t.c:4: warning: null argument where non-null required (argument 1)
vladov@picard:4 /tmp$ ./a.out 
vladov@picard:4 /tmp$ echo $?
0

Seems that strlen works bad with diet -Os.  At least on i386
with my compiler.  What should strlen(NULL) return?

POSIX says nothing, true?

Nikola

Seems that gcc makes the things wrong with -Os!

diet -Os gcc -S -Wall t.c
t.c: In function 'main':
t.c:4: warning: null argument where non-null required (argument 1)
cat t.s
...
        .text
.globl main
        .type   main, @function
main:
        cld
        pushl   %edi
        xorl    %eax, %eax
        xorl    %edi, %edi
        orl     $-1, %ecx
        repnz
        scasb
        popl    %edi
        notl    %ecx
        leal    -1(%ecx), %eax
        ret
...