Re: [bug-diffutils] diff: support for --no-dereference option

[Andreas Gruenbacher]

On Sun, 2012-01-08 at 02:16 +0100, Bruno Haible wrote:
> Andreas Gruenbacher wrote:
> > I wonder if comparing symlinks (instead of what they point to) shouldn't
> > become the default once diff supports it.  At the very least, diff
> > should warn when following symlinks though.
> 
> I disagree. --dereference should stay the default, for two reasons:
> 
>   1) Unix philosophy: By default, symlinks are followed. This is a
>      general principle. It starts with open() and fopen() and continues
>      through most command-line utilities.
> 
>   2) What are the use-cases for "diff"?
>        - Compare files coming from two different users.
>        - Comparing the new state of a directory with a backup.
>        - Compating the new state of a partition with a backup.
> 
>      In the first two cases, --dereference is desired. In the third case,
>      --no-dereference is desired.
> 
>      The first two cases are the most frequent ones.
> 
>      In other words, while you personally may be working frequently at
>      the partition level, many users who put a symbolic link want it to be
>      followed automatically.
> 
> See also the discussion at
> <https://lists.gnu.org/archive/html/bug-gnu-utils/2011-10/msg00011.html>
> <https://lists.gnu.org/archive/html/bug-gnu-utils/2011-10/msg00013.html>

I don't find these three use cases very instructive -- I either compare
two files or two directory trees.  In the two files case, I agree with
what Paul Eggert says in the thread you refer to:

> But the default behavior of "diff FILE1 FILE2" should remain
> unchanged: it should follow symlinks, as many programs depend on this
> behavior. Also, POSIX requires it.

In the two directories case, I really don't see when following symlinks
would be desirable.  Here, diff is comparable to a tool like tar, for
example -- in the tar case, I want to archive what's really there; in
the diff case, I want to compare what's really there.

Could you try to explain in more detail?

In the behavior I have in mind, diff would default to following symlink
command-line arguments but not to follow symlinks encountered while
traversing a directory tree; this is what the -H mode is meant to do.

Thanks,
Andreas