Roughly in order of importance.  Feel free to reorder.


For current release
===================

Empty - release?!


For future release
===================

* Clean up API
   This is described in various points below.

* Use attributes intead of getters/setters.
   Getters and setters are not very Python like, they come from Java
   and are no real benefit.  The rule is only to use functions if the
   `properties' way of doing it is non-trivial.

   Mention the public attributes in the class docstrings, or dig out
   an other way of doing this in the PEP's.

* Use node indexes internally.
   Since node index is supported now we could rewrite this module to
   use that exclusively.  This would save space and be more
   traditional.  The core.py wrapper could then still behave the same
   for a while.

* Solve the qh_version problem with qhull3.1.
   Is there no way of figuring out from the header files if it's 3.1?

* Create a flag that will return Numeric sequences.
   In case the users is working with Numeric too it is a waste for the
   data to be converted back and forth.

* Remove option to update the set?
   Is there any point in being able to update the set?  Any advantage
   above instantiating a new Triangulation object?

* Qhull error output.
   Shoud this go to stderr (as it is now) or should we catch it
   somehow different.

* Allow flags to call qhull to be set in Python?
   This is more likely to be converging of the question wheather to
   support more the just Delaunay tiangulation of Qhull.

* Solve the issue demonstrated with TestPredictedOutputs.testSquare()
   This might have to move to the core module (and it's tests) instead
   of in _qhull.  This could also end up in a large rewrite of the
   delny function (and core module), i.e. accept user flags to qhull
   etc.  Basically lots of questions.

* The number of input points seems to be limited by the length of `int'.
   Is this really true?  Should this be mentioned as "Known issues" in
   the "BUGS" file?