[Looks like unlink_file_dir should be calling CVS_LSTAT not CVS_STAT
(two calls, stat call in unlink_file_dir and isdir call in
deep_remove_dir).

-kingdon]

Date: Wed, 16 Sep 1998 09:33:34 +0200
From: Joerg Faschingbauer <jfasch@hyperwave.com>

To: info-cvs@gnu.org
Subject: cvs release -d and symbolic links

Hi,

I am relatively new to cvs, so I haven't a grasp if this is a desired
behavior or not (probably not). I am using cvs 1.9, any architecture.

During the build process in my working copy symbolic links are created
with the purpose to have the compiled object files reside on a local
disk, where the working copy's file system is NFS mounted.

(Anyway.)

When the time comes to release my copy I use cvs release -d to have it
rm -r my tree. During the recursion, when it gets to unlink the link,
a stat() is done in isdir(), stating that it is a directory (which is
where the link points to). Ok, the directory where the link points to
is rm -r'd, but the link itself remains in the directory. When this
recursion level is left, the directory is considered empty, rmdir() is
called, and the error EEXIST is reported.

I think this is not the desired behavior, but I'm not sure what is
desired.

- should the directory where the link points to be removed, and,
  afterwards, the link as well?

- should only the link be removed?

Any opinions?

Joerg
____________________________________________________________________________
Joerg Faschingbauer                         Hyperwave Research & Development
                                                            Schloegelgasse 9 
mailto:jfasch@hyperwave.com                                      A-8010 Graz
____________________________________________________________________________