The trouble with symbolic links

I view the *at function family as the right thing; it was wrong for 'open' et al. to take an implicit "root" parameter (the Hurd file system interface is very much like that: every file system operation takes a capability to the root directory on which it operates, starting with 'file_name_lookup', which is like 'openat').

In some cases, such as the Stow/Guix/Nix examples, symlinks could be replaced by a unionfs. Maybe a way forward?