Re: autoconf bug

[Joerg Schilling]

>From address@hidden Thu Jul 10 08:58:49 2003

>(CC'ing to bug-autoconf)

>Joerg Schilling <address@hidden> writes:

>> find_stack_direction() 
>> { 
>>         static char *addr = 0; 
>>         auto char dummy; 
>>         if (addr == 0) { 
>>                 addr = &dummy; 
>>                 return find_stack_direction (); 
>>         } else { 
>>                 return (&dummy > addr) ? 1 : -1; 
>>         } 
>> } 
>> 
>> found in acspecific.m4 is not working correctly because
>> GCC optimizes this code in a way that &dummy may be == addr.

>Thanks for the bug report.  If I understand you correctly, it's a GCC
>bug, as that behavior doesn't conform to the C standard.  We should
>file a bug report.  Which version of GCC was it, and on what platform?

Why would you call this a GCC bug?

GCC just detects that 'dummy' is not (except for an address) used and does
not reserve stack space for this object.

In the usual case, when ./configure is called without parameters/environ,
then GCC does not optimize and the code above gives the expected results
except for the fact that is is lazy code anyway and ignores the possibility
that &dummy may be == addr.

If you call env CC=gcc -O' ./configure

then the test does not work and while code:

int
stack_direction(lp)
        long    *lp;
{
        auto long       *dummy[4];
        int             i;

        for (i=0; i < 4; i++)
                dummy[i] = lp;

        if (lp == 0) {
                return (stack_direction((long *)dummy));
        } else {
                if ((long *)dummy == lp)
                        return (0);
                return (((long *)dummy > lp) ? 1 : -1);
        }
}

still works.

BTW: the tests have been made on Solaris x86.

>Hmm, but we can't simply modify acspecific.m4; we also need to modify
>alloca.c, so that it is consistent with acspecific.m4.

Then you should also fuile a bug against GCC

>But I'm puzzled as to why this matters, if you're using GCC.
>GCC has builtin alloca, so shouldn't this issue be moot for GCC?

????

I was just working on a portable software signal system that needs to
know the direction in which the stack grows... so I was looking for a
usable test mechanism.

After I run my tests on all plartforms that I have in our reseach
institute, I found that HP-PA uses a stack that grows to higher addresses.

Jörg

-- 
 EMail:address@hidden (home) Jörg Schilling D-13353 Berlin
       address@hidden           (uni)  If you don't have iso-8859-1
       address@hidden   (work) chars I am J"org Schilling
 URL:  http://www.fokus.fraunhofer.de/usr/schilling 
ftp://ftp.berlios.de/pub/schily