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

[Paul Eggert]

On 01/07/12 18:22, Bruno Haible wrote:
> I claim that this use-case is more frequent than the one which does not
> follow symlinks. Proof: GNU diff users were happy with the existing behaviour
> for 20 years.

I'm not sure that this is much of a proof.  We've had many complaints
over the years.  I know I've not been happy with the existing behavior,
in many practical cases.

No matter what the default is, I imagine that occasionally one will
do a diff -r, be unhappy because the default wasn't the way one wanted it,
and then rerun with diff -rL or diff -rP (or whatever) to fix it.
So the issues are more: which behavior is more likely to be desired in
practice?  And, if they're roughly equally desired, or if we don't know which
is more desired, then which behavior is more likely to cause problems
in practice?  I'm not sure how to answer the first question, but for the
second question it seems that following symlinks is more likely to cause
problems than not following them, because of the problem of loops.

On the other hand, there's the issue of compatibility -- POSIX requires that
symlinks be followed all the time, and GNU diff has behaved
that way for many years.  Changing this is not something we'd want to
do lightly.