Bug in GCC?

Douglas Warren (dwarren@ic.sunysb.edu)
Tue, 25 Jul 1995 22:46:52 -0400 (EDT)

I believe I have found a serious bug in either GCC or the kernel,
I am unable to trace it further then this, consider two files:
test.c:
main() {
int i;
double test,ANGLE;
srandom(getpid());
for(i=0;i<100;i++) {
test=.5+double_rand();
printf("%d %f\n",i,test);
}
}
test1.c:
double double_rand(void)
{
return (double) random() / 2147483648.0;
}
compiled with:
gcc -fpcc-struct-return -O -c test1.c -o test1.o
gcc -fpcc-struct-return -O -c test.c -o test.o
gcc -o testa test.o test1.o -lm
on the 7th invocation of the loop, it will crash everytime
a sample run is:
0 1059284663.500000
1 1262909464.500000
2 383691163.500000
3 898825246.500000
4 976362629.500000
5 1993317795.500000
Floating point exception
HOWEVER, if the 2 files are removed, and the double_rand() function
is in the same source function as it is called, it works flawlessly.
I have seen this bug in multiple kernels/libaries/versions of GCC, for
atleast a year now, but this is the first time I was able to reliably
reproduce it. The code was tested on GCC 2.5.8, libc 4.7.2, libm 4.6.27,
and Linux 1.2.8. Any replies would be apperiated, as well as any
advise on where to go from here with this problem.
-- 
     Douglas ``Wildcat'' Warren
     dwarren@netusa.net <---- Preferred
     dwarren@ic.sunysb.edu
     root@netusa.net
     dwarren@sunysb.edu


"I hold no enmity against those coerced into evil.   But to those vile 
beings that toy with the hearts and souls of men we, since the time of 
ancient gods, have been your destroyers Now, the 108th generation devil 
hunter, Yokoho is here!  BEWARE!"  -- Yokoho _Devil_Hunter_Yokoho_