Re: Add --dereference option to df

[Bernhard Voelker]

On 05/15/2013 04:39 PM, Pádraig Brady wrote:
> On 05/15/2013 03:32 PM, Ondrej Oprala wrote:

>> Let me explain by example from my system:
>>
>> /dev/sda1 is mounted on /boot
>> /dev/disk/by-uuid/37a5c8d4-78ff-408f-8ae8-70555df51f7f is a symlink to 
>> /dev/sda1
>>
>> #in the following example df prints info about the fs where the symlink file 
>> is.
>> $ df /dev/disk/by-uuid/37a5c8d4-78ff-408f-8ae8-70555df51f7f
>> Filesystem     1K-blocks  Used Available Use% Mounted on
>> devtmpfs         1013888     0   1013888   0% /dev
>>
>> #with -L, information about the symlinked FS is printed.
>> $ df -L /dev/disk/by-uuid/37a5c8d4-78ff-408f-8ae8-70555df51f7f
>> Filesystem     1K-blocks  Used Available Use% Mounted on
>> /dev/sda1        1013888     0   1013888   0% /boot
>>
>> So although the argument is stat-ed(dereferenced) in both cases, different 
>> parts
>> of the information are used to decide what to output if -L is specified 
>> (which is why
>> I find the option name misleading).
> 
> Ah /dev/sda1 (/boot) isn't mounted on my system
> (as I don't want that disk spinning up (long story)).
> That's why I got the results I presented.
> More thinking required...

Isn't that what the findmnt(8) command achieves with
the -S,--source option, i.e. it's not only about dereferencing
per se but rather telling the tool that we're searching for
the source of a mount instead of the target?

Therefore, I think making the changes in get_point() is abusing the
semantics of that function.  Instead, get_disk() should be enhanced
in a similar way as in get_point() via canonicalize_file_name().

Another question is if such behavior could/should be made the default
or if a new option has to be introduced.
POSIX [1] does not explicitly mention symbolic link arguments:

[1] http://pubs.opengroup.org/onlinepubs/9699919799/utilities/df.html

  A pathname of a file within the hierarchy of the desired file system.
  If a file other than a FIFO, a regular file, a directory,
  [XSI] [Option Start]  or a special file representing the device
  containing the file system (for example, /dev/dsk/0s1) [Option End]
  is specified, the results are unspecified.

  If the file operand names a file other than a special file containing
  a file system, df shall write the amount of free space in the file system
  containing the specified file operand.
  [XSI] [Option Start]  Otherwise, df shall write the amount of free space
  in that file system. [Option End]

Well, we could either ask the OpenGroup for clarification.
How do other df's behave? Any opinions?

Have a nice day,
Berny