install dereferences symlinks given as sources (and I think that's good). The "-C" option, however, uses lstat rather than stat in need_copy and returns true if !S_ISREG.
So the (dereferenced) file will always be copied despite "-C". Even worse, since "-p" cannot be combined with "-C", the installed file will always get a new timestamp which may trigger rebuilds (e.g. if the file is a header) even if nothing was changed, e.g.: % touch a % ln -s a b % install -C b c % ls --time-style=+%T -l total 0 -rw------- 1 frank frank 0 23:46:51 a lrwxrwxrwx 1 frank frank 1 23:46:54 b -> a -rwxr-xr-x 1 frank frank 0 23:46:58 c % install -C b c % ls --time-style=+%T -l total 0 -rw------- 1 frank frank 0 23:46:51 a lrwxrwxrwx 1 frank frank 1 23:46:54 b -> a -rwxr-xr-x 1 frank frank 0 23:47:06 c % install -C b c % ls --time-style=+%T -l total 0 -rw------- 1 frank frank 0 23:46:51 a lrwxrwxrwx 1 frank frank 1 23:46:54 b -> a -rwxr-xr-x 1 frank frank 0 23:47:08 c % I asked Kamil Dudka who implemented "-C" originally and he pointed out that requiring both the files to be regular files was suggested by Jim Meyering in the review: https://lists.gnu.org/archive/html/bug-coreutils/2009-02/msg00106.html However, I don't see a real reason there, only his question: "Have you considered requiring that both files be `regular', too?" I don't know if install didn't dereference symlinks back then, but now that it does, it seems consistent to dereference them in the check as well, i.e. use stat rather than lstat.
