Re: tests/du/slink.sh: what's the rationale?

[Pádraig Brady]

On 01/19/2013 10:33 PM, Bernhard Voelker wrote:
> Does anyone remember the reason for adding this test?
>
> tests/du/slink.sh:2:
>    # Ensure that the size of a long-named-symlink is > 0.
>
> I'm asking because I'm experiencing failures on OpenSUSE's Build Service.
> It uses XEN for building packages ...
>
>    booting XEN kernel ...
>    xm create -c /var/tmp/build.xen.conf-xEZDA7Af3 name=build_8 \
>       disk=phy:/dev/dm-14,hda1,w disk=phy:/dev/dm-15,hda2,w \
>       extra=quiet init=/.build/build elevator=noop panic=1 console=ttyS0
>
> and the file system is an ordinary ext4:
>
>    + df -Th .
>    Filesystem     Type  Size  Used Avail Use% Mounted on
>    /dev/hda1      ext4   14G  1.2G   13G   9% /
>
>    + stat .
>    File: '.'
>    Size: 4096           Blocks: 8          IO Block: 4096   directory
>    Device: 301h/769d    Inode: 141009      Links: 2
>    Access: (0700/drwx------)  Uid: (  399/  abuild)   Gid: (  399/  abuild)
>    Access: 2013-01-19 13:26:39.280020256 +0000
>    Modify: 2013-01-19 13:26:39.420020256 +0000
>    Change: 2013-01-19 13:26:39.420020256 +0000
>    Birth: -
>
>    + stat -f .
>    File: "."
>    ID: c0e637f6ae90cc28 Namelen: 255     Type: ext2/ext3
>    Block size: 4096       Fundamental block size: 4096
>    Blocks: Total: 3571656    Free: 3274091    Available: 3274091
>    Inodes: Total: 897600     Free: 855410
>
> However, du reports them all with 0 size:
>
>    + grep '^[1-9]' out
>    + fail=1
>    + cat out
>    0    1
>    0    15
>    0    16
>    0    31
>    0    32
>    0    59
>    0    60
>    0    63
>    0    64
>    0    127
>    0    128
>    0    255
>    0    256
>    0    511
>    0    512
>    0    1024
>
> Here's an strace output for the last symlink "1024":
>
>    newfstatat(AT_FDCWD, "1024", {st_dev=makedev(3, 1), st_ino=141025,
>      st_mode=S_IFLNK|0777, st_nlink=1, st_uid=399, st_gid=399,
>      st_blksize=4096, st_blocks=0, st_size=1024,
>      st_atime=2013/01/19-13:26:39, st_mtime=2013/01/19-13:26:39,
>      st_ctime=2013/01/19-13:26:39}, AT_SYMLINK_NOFOLLOW) = 0
>
>
> The test du/slink has been added back in 2002
>    http://git.sv.gnu.org/cgit/coreutils.git/commit/?id=92a2cc65
> but I didn't find a reason for it - neither in the surrounding
> commits nor in today's du.c or du.c of that time.
>
> Is tests/du/slink.sh obsolete nowadays?
>
> Have a nice day,
> Berny

I guess the rationale is to verify the symlink is not being resolved,
though you're right that symlink allocation is dependent on the
file system. I suppose one could use a loopback ext2 file system
like in cp/cp-a-selinux.sh.

cheers,
Pádraig.