Re: Feature request, mv

[Jim Meyering]

Thanks to Joshua for the report and to Eric for pursuing it.
There is indeed a bug.

Before the fix, mv would unnecessarily unlink the destination symlink:
  $ rm -rf s[12]; ln -s / s1; ln -s /tmp s2; strace -qe unlink /bin/mv -T s1 s2
  unlink("s2") = 0

With the fix, it doesn't call unlink:
  $ rm -rf s[12]; ln -s / s1; ln -s /tmp s2; strace -qe unlink ./mv -T s1 s2
  $

Here's a snap fix to disable the offending condition in move-mode.
That disjunct *is* required to support cp's use of this code.
This might prompt me to add a first test case using strace.
(of course, it'll run only on systems with strace)

This change (or some variant of it) will go on both
the trunk and the b5_9 branch -- just in time for coreutils-5.95.

Index: src/copy.c
===================================================================
RCS file: /fetish/cu/src/copy.c,v
retrieving revision 1.197
diff -u -p -r1.197 copy.c
--- src/copy.c  12 Mar 2006 22:52:46 -0000      1.197
+++ src/copy.c  10 May 2006 13:25:14 -0000
@@ -1230,8 +1230,10 @@ copy_internal (char const *src_name, cha
          else if (! S_ISDIR (dst_sb.st_mode)
                   && (x->unlink_dest_before_opening
                       || (x->preserve_links && 1 < dst_sb.st_nlink)
-                      || (x->dereference == DEREF_NEVER
-                          && S_ISLNK (src_sb.st_mode))))
+                      || (!x->move_mode
+                          && x->dereference == DEREF_NEVER
+                          && S_ISLNK (src_sb.st_mode))
+                      ))
            {
              if (unlink (dst_name) != 0 && errno != ENOENT)
                {