- added new "spectral" colormap (truncated hsv)
- added GUI button for elliptic mesh + geometry coherence
- added GMSH_MAJOR_VERSION, GMSH_MINOR_VERSION and GMSH_PATCH_VERSION
  in the parser

allow negative line loop number in surface definitions

- fixed the "unix socket name is one character short" problem on
MacOS and AIX. Unfortunately, this required the client code to be
slightly modified as well (and as a consequence, old versions of getdp
will for example not work with new versions of Gmsh on MacOs and
AIX). But it's better to do things "the right way" (TM) for people who
want to connect using different languages (Perl, Python).

- moved the GmshServer stuff in its own class

- added missing rules in parser makefile (so that the parser gets
  rebuilt automatically when one changes Gmsh.y or Gmsh.l)

new "AliasWithOptions" command

- renamed "Duplicate View" into "Alias View" (suggested by Philippe;
  less confusing since "Duplicate" implies that we actually copy the
  data, which is not what we are doing)

- added general "on the fly" tranformation matrix option. This allows
  for very efficient visualization of aliased datasets with
  non-destructive rotations, symmetries, etc.

- removed all the crappy STL code and rewrote it using JF's POLY_rep
  class

- generalized POLY_rep so that we can use the polygonal discretization
  as a surface mesh, and mesh in 3D afterwards. I.e., we can now take
  an input triangulation (a single surface in STL format, multiple
  surfaces in STL format, one or more surfaces defined using the new
  "Discrete Surface" commands), and generate a 3D mesh that uses it. We
  could in theory even mix triangulated and "normal" surfaces in the
  same geometry, but nothing is done at the moment to ensure that the
  mesh at the interfaces would match (if it does, it actually works
  very nicely)

- new STL mesh output format to export a surface mesh as a STL file

- added an option to the GEO output routine to save the surface mesh
  as discrete surfaces associated with the geometrical surfaces

- added STL and Text output formats for post-processing views (the
  text output allows for example to exploit plugin-generated data in
  gnuplot)

- generalized Plugin(Evaluate):

  * can loop automatically over all the timestep and/or components

  * can do operations using data from an external view

     - if the 2 views are based on the same grid, the plugin does the
       evaluation very efficiently

     - if the 2 views are based on differenet grids, the plugin
       automatically interpolates the external view data onto the
       grid of the current view

- added new Rand() function in MathEval

- default colormap is now # 2 (the Matlab "Jet" colormap)

add special token "Today" to return the current date

copyright update

- reworked the keyboard shortcuts with Laurent (our very own Macintosh
  Usability Guidelines Expert)

- new smoothing option for jpeg output

- new small GUI dialogs for ps/pdf/jpeg/gif/msh output (makes the
  'format' menu in the file dialog much shorter)

bulletproof parse_string and skip_until so that we don't crash and simply
report the error + use Msg instead of YY_FATAL_ERROR in YY_INPUT

Fixed all (well, most) string memory leaks in the parser. I've run
automated tests on all the demos and benchmarks, and I did not see
anything suspicious (no crashes, double frees, etc.)

******************************************************************
******************************************************************
*********** Pretty please, please give this version **************
*********** a try, especially on files with lots of **************
*********** loops, includes, arrays of variables,   **************
*********** etc., and report any problems. Thanks!  **************
******************************************************************
******************************************************************

Don't optimize Gmsh.tab.o: on many architectures, g++ -O2 generates
assembly code too large to be addressed correctly. This will hopefully
fix the build problems reported on ppc.

don't need UpdateViewsInGUI in parser anymore

better test for invalid For/EndFor loops

Don't update the views in the GUI directly after each view is created.
Instead, wait until the whole file is read. This speeds up the creation
of the interface dramatically when we have a large number of views in
a single file. (In my code I routinely merge around 1000 views
in a single file to do animations; with this patch, the GUI creation
time goes from 30 seconds to less than 1 second.)

more robust "Delete Empty Views"

my last commit also fixed several problems in For/EndFor loops:

- the body of the loop was executed at least once, even if the test was
  always false (e.g. For (1:-1:1) )
- the body could be executed one extra time when using optional
  step values
- the for loops didn't work with negative step values

- new parser function GetValue("text", default_value) to query a value
  interactively in a script

- added a nice example using this (to compute slices automatically):
  demos/multislice.script

- new "Delete Empty Views" command
- new View.MinX, View.MinY, View.MinZ, View.MaxX, View.MaxY, View.MaxZ options
- added example on how to do simple volume visualization

- fix cast in a couple of error msg
- more fixes in elliptic algorithm (surf->TrsfVertices and surf->Vertices
  were incomplete)

one can now specify a TIME list of values in parsed views to give the
values of the time (or whatever) steps (like in ascii/binary views)

- moved the AdaptiveView stuff in its own header file

- fixed small bug in IsoSimplex (when Raise != 0)

- fixed small memory leak in Plugin/Extract (when one or more
  expressions are invalid)

- Added a bunch of colormaps from:

  Color Theory and Modeling for
  Computer Graphics, Visualization,
  and Multimedia Applications

  Haim Levkowitz
  University of Massachusetts Lowell
  Lowell, Massachusetts, USA

- Increased max size of colormap to 1024

- Small cleanup in parser+plugin

This patch fixes an old design flaw in the post-processing module,
namely that we used a list of Post_View objects instead of a list of
pointers to Post_View objects in CTX.post.list.

This had many annoying consequences, in particular the fact that we
needed to be extra careful every time the list was reallocated (as
pointers to the list elements would become invalid).

I tried very hard to change the code everywhere it should be changed,
but I might have missed something. Please let me know if you see
anything suspicious (like a crash when you duplicate/remove/combine/...
post-processing views).

fixed error checking code to work also with adaptive views.

cleanup (and remove 2 shift/reduce conflicts in parser)

proof of concept for unproject (e.g. to add points with the mouse)

- make the rotation matrix a matrix of doubles (instead of floats)

- store the modelview and the projection matrices in CTX

- new unproject() routine to transform window coordinates to model
  coordinates

- cleaned up all the static variables in the Opengl_Window class

- small fix to reduce normal clamping when zooming

if we have no geometry and no mesh, use the max of all view bounding boxes
(instead of the bounding box of the last view only)

small fixes for last commit:

- only delete view.adaptive if we really delete the view (we might want to
  use the adaptive data in a duplicate view)

- add missing opt_view_global_zoom in update_view()

- don't even try to open StructuralSolver's data files if GMSHPLUGINSHOME
  is not defined

New high order interpolation.

JFR

Trying to fix Patrick's problems when launching getdp on Windows...

When gmsh is started from the command line on Windows, cygwin does
not (for some obscure reason) prepend '/cygdrive/x/' to the path. Since
we check for '/cygdrive/' in SystemCall() in order to sanitize the
path before passing it to CreateProcess(), the system call fails.

I think the right way to deal with this is to actually remove the
path change in SystemCall() (since we only change the leading part),
and use the official cygwin API to change the path of every piece
that will eventually end up in the argument of CreateProcess().
And that's what "FixWindowsPath()" now does...

forgot to commit this

added ability to save views in "parsed" format

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...)