loop-related regression on Fedora 16

[Jim Meyering]

On Fedora 16 (3.1.5-6.fc16.x86_64), the two t8000*loop.sh tests have
been failing for some time.  I've finally investigated and found it is
due to this:

_disk_sync_part_table returns 0 because "lpn" below is 0 for a loop device:
(indicating non-partitionable device)

    static int
    _disk_sync_part_table (PedDisk* disk)
    {
            PED_ASSERT(disk != NULL);
            PED_ASSERT(disk->dev != NULL);
            int lpn;

            unsigned int part_range = _device_get_partition_range(disk->dev);

            /* lpn = largest partition number. */
            if (ped_disk_get_max_supported_partition_count(disk, &lpn))
                    lpn = PED_MIN(lpn, part_range);
            else
                    lpn = part_range;

            /* Its not possible to support largest_partnum < 0.
             * largest_partnum == 0 would mean does not support partitions.
             * */
            if (lpn < 1)
                    return 0;

That makes parted exit nonzero, but with no diagnostic.  Oops.
That's two bugs.  Failing at all, and failing with no diagnostic.
That function has three other return points that lack a diagnostic,
but they're all for heap allocation failure.

This looks like a regression over parted-3.0, though it may well
be due to changes in the kernel.

A simple test:

    truncate -s10m a && t=$(losetup --show -f a) \
      && { parted -s $t mklabel gpt && echo ok || echo FAIL; }
    losetup -d $t

With parted-3.0, that prints ok.
With parted built from git, it prints FAIL.

I get the same results when using rawhide's 3.2.0-0.rc6.git2.1.fc17.x86_64