This fixes:
   VTK_OPENGL_HAS_OSMESA on
   VTK_USE_X             on
   caused link errors in Rendering/OpenGL as some of the cxx
   files contained code ifdef using VTK_OPENGL_HAS_OSMESA instead of
   VTK_USE_OSMESA

In the past, one could link against both MESA and OSMESA and use OSMESA only
for off-screen rendering. In recent versions of MESA this is not the case anymore.

Use vtkOpenGL.cmake to decide to use (LINK_PRIVATE) OpenGL or Offscreen Mesa.
OpenGL is now LINK_PRIVATE, so additional libraries and tests need to include OpenGL.
Include the following in your CMakeLists.txt:

include(vtkOpenGL)
vtk_opengl_link(${module})

for every ${module} that uses OpenGL. This takes care of include directories and
linking of proper libraries.

Minimum SDK and deployment is now 10.6.

Convert double-quote includes to angle brackets:

 git grep -l '#include "sstream"' | xargs sed -i 's/"sstream"/<sstream>/'

We no longer need this compatibility layer for the compilers we support.
Use the following commands to switch to standard header and namespace:

 git grep -l vtksys/ios/ | xargs sed -i 's|vtksys/ios/||'
 git grep -l vtksys_ios | xargs sed -i 's|vtksys_ios|std|g'

We no longer need this compatibility layer for the compilers we support.
Use the following commands to switch to standard header and namespace:

 git grep -l vtksys/stl/ | xargs sed -i 's|vtksys/stl/||'
 git grep -l vtksys_stl | xargs sed -i 's|vtksys_stl|std|g'

In the early days of VTK, support for stl was not portable. vtksys_stl
and vtksys_ios provided a portable implementation of the stl. Now, all
of the VTK supported compilers have portable stl implementations.

This patch:
  1) Replaces the vtksys_ios:: with std::.
  2) Replaces the vtksys_stl:: with std::.
  3) Removes "using" statements for stl

Remove all code marked with VTK_LEGACY...
from before vtk-6.1 (2014-01-21)

Auto-detecting the screen DPI (which the win32 backend does)
interferes with testing after we make 2D text rendering
DPI-aware. Since only one backend supported automatic DPI
detection, I moved this out into a user method so that folks
who depend on this behavior can still have it, but we can
test consistently.

Previously, we had defined GLX_GLXEXT_LEGACY as a workaround for
missing typedefs on newer versions of opengl. However, this broke
multisampling (http://www.vtk.org/Bug/view.php?id=15433).

Adding the typedefs explicitly in the few places we directly include
"GL/glx.h" fixes both issues.

This is not supported by the new OpenGL2 backend, and doesn't support
new text features, like MathText, FontConfig, or custom font files.

It only implements vtkTextMapper, which now will use a texture-based
implementation, similar to vtkTextActor and friends.

Fix in vtkWeb code instead.

This reverts commit 39fedcd6.

Also removed some dead code as a consequence,
like setting the deallocated pointer to null
then to a new address.

Also removed setting some ivars to null in dtors.

NSTrackingArea long ago replaced older APIs
used before this.  We still work back to 10.5
even after this change.  Code affects mouse cursor
shape changes.

I accidently added a memory leak when adding partial ARC support.
I forgot that one can build as mixed GC+MRR, in which case the
explicit retain/release calls are needed, so added them back.

This was a leftover from debugging that I missed earlier...

Found with clang -Wunused-member-function or
-Wunreachable-code.

Fixes clang -Wmissing-variable-declarations warnings

The path where atlbase.h can be found can theb be passed for VS express.
http://www.vtk.org/Wiki/VTK/3DConnexion_Devices_Support

Change-Id: Ic6b6251d436b32debaf7d4d04a8e017b58afe1a2

Change-Id: Idecb2674d9e546fe46080fa59a3420c6d77ab8be

From KillTimer(HWND hWnd, UINT_PTR uIDEvent) documentation:
uIDEvent [in]: The timer to be destroyed. If the window handle passed to
SetTimer is valid, this parameter must be the same as the nIDEvent value
passed to SetTimer. If the application calls SetTimer with hWnd set to
NULL, this parameter must be the timer identifier returned by SetTimer.

Fixes #15334

Change-Id: Ifd88c790f4955dcd352004e2f5cba22a900e02de

Change-Id: I5adee5e4e29ae03aa4363550a11a044d447ae74b

Fixes cppcheck warning, should be a little faster

Change-Id: I24ebe2085e2f1c771d2a83fba6ebfb8761cfb9ab

If the while loop executes 0 times, an undefined
negative bit shift would occur.

Also eliminate ugly cast with simpler constant declaration.

Change-Id: I91237652bf25e0de86b445b77d1760d53d0813b4

These implementations ignore the MultiSamples ivar, and should fall back
to a platform-specific solution when MS is needed.

Change-Id: Ifb45196dbb49f78d2c44a5e2862e708a0a758d2d

Calling XSync(...) is not enough. This causes a timing issue
went SetSize(...) is called following be a Render(...). In
Render(...) the window is queried for the window size which
will not have be resize yet so we get the old size! ( seen
particularly in vtkWeb ). Add loop to wait for ConfigNotify
event with the new size.

Change-Id: I72d34cae16215b6887467496df23d7a8e0bdcd37

Change-Id: Ib877b5c85111b360a79b7395f25ec52f1e3a6799

Fixed inverted ifdef for NDEBUG (doh!).

Change-Id: I864b3a2265c322a5d2ff75b5654430b3f52ef586

Change-Id: I4ba032c598bf63e4ae7b9193dec938df603d3139

Change-Id: If71032baf636db53e46cbf8f60f69d66b8903bb4

Used the following command:

find . \( -iname *.h -and -type f \) -exec perl -pi -w -e 's/__vtk(.*)_h/vtk$1_h/g;' {} \;

Fixes many, but not all, clang -Wreserved-id-macro warnings.

Change-Id: I694ff053cb204c70e37b261296f496064d91fc07

Used the following command:

find . \( -iname *.in -and -type f \) -exec perl -pi -w -e 's/__vtk(.*)_h/vtk$1_h/g;' {} \;

Fixes many, but not all, clang -Wreserved-id-macro warnings.

Change-Id: I86db91ad530c203347f2b0dfd29c31f843c3f2b7

Now only perform observation at all if it was VTK that created the NSWindow/NSView.
This avoids receiving many NSView frame change messages when not needed.

Some renaming, new asserts, fix minor type mismatch.
Things are more symmetrical now.

Change-Id: I337afef3a07fe9e33abdfc6e7e714502262bcc91

OpenGL error checks have been shown to affect performance dramtically
when rendering large datasets. That being the case, this change adds a
new flag which disables these checks in Release builds, by default.
Users can still use the advanced variable
VTK_REPORT_OPENGL_ERRORS_IN_RELEASE_BUILDS to enable checks in release builds as well.

Change-Id: Ib7a46292fcbc9dd166dbbd6d9217ee71f98f14df

This change makes it possible to map vtkStringArrays to colors using
indexed lookup mode. The vtkStringArrays may be associated with
points, cells, or neither (field data). A slew of changes were
required to give this capability:

- Changed signature of vtkScalarsToColors::MapScalars(...) and
  vtkDiscretizableColorTransferFunction::MapScalars(...)  to take a
  vtkAbstractArray instead of a vtkDataArray. This enables mapping
  non-vtkDataArray arrays, such as vtkStringArrays. These changes are
  backwards compatible because vtkAbstractArray is the parent class of
  vtkDataArray.

- Changed vtkLookupTable::MapScalarsThroughTable2(...) to handle
  vtkStringArrays.

- Changed vtkColorTransferFunction::MapScalarsThroughTable2(...) to handle
  vtkStringArrays.

- Added vtkAbstractMapper::GetAbstractScalars(...) that returns a data
  array as a vtkAbstractArray - needed to retrieve vtkStringArray
  scalars.

- Added some error reporting when unhandled array type is encountered
  in vtkScalarsToColors::MapScalarsThroughTable2.

- Enabled use of vtkStringArrays for coloring in the surface mappers.

- Added some tests for mapping vtkStringArrays with
  vtkColorTransferFunction and vtkDiscretizableColorTransferFunction.

- Added some tests for coloring objects by vtkStringArrays in cell data
  and point data.

- Added logic to vtkMapper and vtkPolyDataMapper2D that generates a
  reasonable default lookup table when a vtkStringArray or other
  non-numeric array is selected for coloring by scalars. This requires
  making the vtkRenderingCore module depend on vtkCommonColor. Added
  tests for this feature.

Change-Id: I91e2e451217139a7daabd7f4a63de0b8ff707ad4

Some level of support was previously available for coloring by field
data, but this was restricted to drawing triangle strips. This patch
adds another mode to coloring by field data that enables coloring the
entire data set by a particular tuple in the field data. The field
data tuple gets mapped through the color map, and that color is used
to color the entire data set. The default behavior for triangle strips
is retained by setting this tuple index to -1, the default value.

The motivation for this change is to make it possible to color blocks
in a multi-block dataset according to a value in that block's field
data array, e.g., an initial condition, block ID, etc. Vector values
(arrays with more than one component) are fully supported.

Several other improvements were also made:

- Added test for existing capability of coloring by field data.

- Added test for coloring by tuple of field data.

- Modified vtkMapper::GetScalarModeAsString() to return
  string for field data scalar mode.

- Fixed vtkDataSetSurfaceFilter::UnstructuredGridExecute(...) and
  vtkUnstructuredGridGeometryFilter::RequestData(...)
  to pass field data through to output

Change-Id: I0d97e84ec77e7ea01ca1b1350acce73cb6c4c15b