TODO

- eliminate minor code duplication: we are running dlsym() for a second
time inside do_fopen_or_freopen_or_open() (after we have already done so in
init()).

- rethink support for *at() functions.

- make *at() wrappers honor the preference set in in_case_of_failure;
currently, they all fail if they can't derive the correct path using
make_abs_path_from_dirfd_relpath().

- Reimplement libtrash using partition-specific trash cans. This has always
been the number one feature request. It would put an end to the lengthy file
copy operations which occur whenever GLOBAL_PROTECTION is set to YES and a
large file outside of the partition on which the user's home directory
resides is deleted. This is a major change, and I am afraid it will have to
wait until I have some spare time.

Currently I plan to have a "Trash" directory beneath each mount point, and
populate it with user-specific trash cans which will only be
readable/removeable by their owners. All the different trash cans belonging
to a user (one on each mounted partition) will be connected through symbolic
links to points inside a "Trash" directory in that user's home dir,
something which will be done by a small script which runs when the system
boots. That same script will also create the "Trash" directories beneath
each mount point (because, as a rule, root privileges will be necessary to
do that).

Some people wrote that symbolic links are too "fragile": eg, "what if mount
points are changed? Then the links in the users' trash directories will
become dangling symlinks". However, others pointed out (convincingly, IMHO)
that in a production system mount points won't be changing on regular basis,
and that after a reboot everything will once again be working correctly.

Feedback on this issue is welcome; however, code contributions are even more
so... :-)

- Allow the user to define IGNORE_USERS: a list of usernames for which
libtrash is disabled. This was suggested by Frederic Connes and it would be
useful, eg, for daemons.