subclass the original file chooser and move it into the GUI class, so
that we can set its position (the FLTK folks don't want to patch the
standard file chooser, so this is the only solution...)

fix problem with new dynamic menus and plugin dialogs (unfortunately, this
re-adds the (small) memory leak in the GUI)

- the main menu is now scrollable when there are more than 25
buttons and it can handle an arbitrary number of buttons (this removes
the "100 views max in the GUI" limitation)

- the view number is now also displayed in the menu (to make it easier
to find the view when one has many many views)

- added "Remove empty views" menu item

add missing #ifdef HAVE_FLTK

make code compile on Mac again

re-add return statements in functions returning non-void. JF: my compiler
wants this (and he's right)

make depend

- fixed a couple of gcc warnings + added missing o->end() in
  jf's StructuralSolver plugin

- fixed the view option code: FLTK > 1.1.5rc1 breaks our use of
  widget->changed()...

- added background mesh support in the aniso algorithm

little hack to visualize overlapping multi-patch spheres with constant
offsets (use timeStep<0...)

make depend

- Triangle now supports our background meshes

- fixed memory leak in Optimize_Netgen (forgot to free the volume
  vertices) + added missing 2nd order reset

- fixed a couple of typos in the docs

set a parameter for the max number of elements per bucket
(default=100, which seems to be a good memory vs. speed trade-off for
meshes up to 1/2 million elements)

- allow more than 2 elements per bucket for ST!
- add missing tensor simplex case

- fixed seg fault in Plugin(Triangulate)
- added BBox computation in plugins when the coordinates are changed

mention explicitly if a plugin only works for simplices

Generalized Plugin(Triangulate) to handle vector and tensor views

- mesh.algo -> algo2d and algo3d
- replaced most of the remaining radio buttons with Fl_Choices (requires
  less screen real estate)

reset the input lists (do not delete them)

forgot to destroy the evaluators

nbU -> nPointsU

replace CutCircle/CutCurve with CutParametric

- fixed out of bound access in OctreePost for multi-step views

- added dView option to StreamLines plugin to interpolate an
  arbitrary scalar view on the stream lines

- set VectorType to Displacement in Plugin(StreamLines)
- shout when a plugin option/name does not exist

- generalized vertex array class to use reallocatable arrays

- introduced vertex arrays in post for all triangles: this speeds
  things up tremendously for ALL scalar views: 2d scalar views of
  course, but also for 3d isos (tets, hexas, etc.) and all smooth
  normals. The costly computations (compute the isos, get the normals)
  are only done ONCE each time the view is changed. Then we just pass
  the vertex array to OpenGL everytime we need to redraw (e.g. when we
  rotate the scene)

- removed the display list code alltogether: it's completely blown
  away by the new vertex array code!

for the record: with the new drawing code, you can

- rotate 500,000 triangles in real time on a 3-year old linux PC
(tested with a boing 747 surface mesh from CU: 10 times faster than
old code)

- display and rotate in real time 10 transparent+smooth shaded 3D isos
for a 1 million tet mesh (about 1 frame per second with the new code;
took tens of seconds with the old one)

Awesome :-)

Added final bit for 2nd order elements: the middle face nodes for
quadrangular faces (i.e., for quadrangles, hexahedra, prisms and
pyramids).

Small changes inspired by Wednesday's discussions:

- Mac only: Gmsh now publishes its "file types"... That means that
  geo/msh/pos files now have nice icons, and that you can double-click
  on a geo/msh/pos file in the Finder to open it :-)

- removed support for fltk 1.0 (wasn't compiling anymore anyway)

- doubled the timout in GmshServer (2->4s): this seems to help on
  Windows (thanks to jkools@veeco.com)

- added warnings in the docs about the Attractor feature

- increased WB by 1 pixel in the GUI

- replaced more Msg(FATAL) with Msg(ERROR) where it makes sense

missed these bits in my commit yesterday

adapt verbosity test to new default

secured most of the list_pointer accesses

generalized Plugin(Evaluate) to work on any (even non-scalar) views

generalized Plugin(Extract) so that we can also create vector views

better docs