Bug 1478864

Description of problem:

The following code snippet spits a warning with gcc-7.1.1-3.fc26 and -Wall on
ppc64le AND x86_64, while we would expect the compiler to optimize the loop
away since the array is empty.

int foo(void)
{
    char empty_array[] = { };
    int i, ret = 0;

    for (i = 0; i < (int) (sizeof(empty_array) / sizeof(empty_array[0])); i++) {
        ret = empty_array[i];
    }

    return ret;
}

Version-Release number of selected component (if applicable):

gcc-7.1.1-3.fc26

How reproducible:

Always.

Steps to Reproduce:

1. cat>foo.c<<EOF
int foo(void)
{
    char empty_array[] = { };
    int i, ret = 0;

    for (i = 0; i < (int) (sizeof(empty_array) / sizeof(empty_array[0])); i++) {
        ret = empty_array[i];
    }

    return ret;
}
EOF
2. gcc -c -Wall foo.c
3.

Actual results:

foo.c: In function ‘foo’:
foo.c:7:26: warning: ‘empty_array[i]’ may be used uninitialized in this function [-Wmaybe-uninitialized]
         ret = empty_array[i];

Expected results:

No warning.

Additional info:

This doesn't happen on fedora24 (gcc-6.3.1-1.fc24) or fedora25 (gcc-6.4.1-1.fc25).

This code pattern is used in QEMU, and causes the build to fail because we pass -Wall -Werror to gcc by default.