On Wed, Dec 1, 2010 at 16:41, Michal Schmidt <mschm...@redhat.com> wrote: > On Wed, 1 Dec 2010 16:31:20 +0100 Kay Sievers wrote: >> On Wed, Dec 1, 2010 at 16:05, Michal Schmidt wrote: >> > I think it's this way in order to simulate a strict path-based >> > watch, as opposed to an inode-based one. >> > For cases when someone recreates the directory hierarchy in a weird >> > way, like: >> > # a watch is set on /foo/bar/baz >> > mv /foo/bar /foo/qux >> > mkdir -p /foo/bar/baz # a completely NEW /foo/bar/baz inode >> >> That should probably be avoided for DirectoryNotEmpty=. > > I guess Lennart wanted it to work even in cases when the directory to > be watched is created later.
That's fine, but these watches maybe should not be established when the final dir already exists? >> Nothing should watch /dev just for that reason. > > If the watch were only for IN_DELETE_SELF|IN_MOVE_SELF, it wouldn't be > too bad. These events are rare. Right. We need IN_ATTRIB only in the last watch, but not for the parents? Kay _______________________________________________ systemd-devel mailing list systemd-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/systemd-devel
