But the point of using doxygen is that it's supposed to be an improvement ...
if it's really 'horrible' then we should stick with autogsdoc and just tweak
the output to look like whatever we want.
A big part of the point of having autogsdoc was to avoid any external
dependencies so any gnustep code could guarantee it was always available to
generate documentation ... if we lose that advantage then we want a big
corresponding advantage somewhere else to compensate. Now if most people are
familiar with doxygen and it's easy to use and produces great output, then that
familiarity/ease of use could be the compensating advantage, but if it's
complicated and 'horrible' then why bother?