The current dependency relationship in vtk is unclear and misleading.
This MR tries to rewrite them based on header files inclusion of headers
and source files in each module. Corner cases are considered and
modules are sorted in alphabetical order to facilitate future reference.
See details in my gitlab python based script project. In future we can continue
using this script to clean the VTK Dependency easily from
time to time.

This is a fix for:
vtk/vtk#14310

Windows applications that run for a long time report that rendered
objects do not change. This is because the modified time on a Windows
system is 32 bits. This causes overflows that defeat the modified time
mechanism. This patch defines a new type, vtkMTimeType that is 64
unsigned integer regardless of the architecture.

A mechanism to provide backward compatibility is introduced. The
preprocessor define "VTK_HAS_MTIME_TYPE" can be used in applications
that must build against VTK versions that use the "unsigned long" type
for MTime's.

Methodology used to find MTime occurences:
1) Identify files as follows:
   git grep "unsigned long" | grep ime | cut -d":" -f1,1 | sort | uniq
2) Hand edit each of those files replacing "unsigned long" with
   "vtkMTimeType" where appropriate.
3) Temporarily change typedef for vtkMTimeType to "double" to detect
   missing conversions

Added backwards compatible names for older SDKs

Of note: NSOpenGLPFAStereo is deprecated, with no replacement
indicated, so just suppress that warning.

Check return value for null to prevent possible crash.
Also, get ASCII from NSString instead of UTF8.

Find/replace of:

;[ ]*//\s*Not implemented[\.]*
to
 VTK_DELETE_FUNCTION;

To catch a few remaining ones missed by previous greps.

Manually reverted changed files in VPIC and KWSys folders, and a couple other of places.

vtk(.*)\(const vtk\1&\);\s*//\s*Not implemented[\.]*
to
vtk\1(const vtk\1\&) VTK_DELETE_FUNCTION;

vtk(.*)\(const vtk\1 &\);\s*//\s*Not implemented[\.]*
to
vtk\1(const vtk\1 \&) VTK_DELETE_FUNCTION;

vtk(.*)\( const vtk\1 & \);\s*//\s*Not implemented[\.]*
to
vtk\1( const vtk\1 \& ) VTK_DELETE_FUNCTION;

vtk(.*)\( const vtk\1& \);\s*//\s*Not implemented[\.]*
to
vtk\1( const vtk\1\& ) VTK_DELETE_FUNCTION;

vtk(.*) \(const vtk\1&\);\s*//\s*Not implemented[\.]*
to
vtk\1 (const vtk\1\&) VTK_DELETE_FUNCTION;

(operator\s*=.*);\s*//\s*Not\s*implemented[\.]*
to
\1 VTK_DELETE_FUNCTION;

After that, this finds basically nothing:

operator.*implemented

then manually reverted changed files in VPIC and KWSys folders.

The 10.12 SDK made a bunch of enum names more
logical, map old names to new names to continue
supporting old SDKs, and use new names in VTK code.

A loop where the render window is kept around and everything in the pipeline is removed
and added at the next iteration leaks memory, in OpenGL, linux. For the OpenGL2 backend same
loop does not leak.
Two problems were fixed. First ReleaseGraphicsResources was not called when
a renderer was removed from a render window. Second, for a offscreen render window
Mapped was set to 0, which meant that no resources were released.

Kitware Robot authored May 24, 2016 and David Gobbi committed May 25, 2016

Perl was used to remove all the BTX and ETX markers from the code.
The specific command that was run on all "vtk*.h" files was this:
    perl -0777 -i -pe 's/(\n* *\/\/ *[BE]TX *\n+)+/\n\n/g'

This regex replaces each BTX/ETX line, plus any leading or trailing
blank lines, with a single blank line.

Remove 'virtual' keywords and add 'override' (via 'VTK_OVERRIDE' macro) instead.
This is the first actual use of 'override' in the VTK codebase.

Warned by cppcheck

SafeDownCast performs a series of virtual calls and string comparisons,
which is quite slow, especially when used in worker functions.

vtkArrayDownCast will switch between SafeDownCast and the more
efficient FastDownCast (only available for common vtkAbstractArray
subclasses) depending on whether or not FastDownCast is defined for
the specific array type.

Added outline selection in test and its baseline image.

GPURayCastMapper renders a valid attribute Id during selection.

Disabled depth mask write in gpuRayCastMapper. Cleaned up hardwareSelector.

Style fixes.

GPUVolumeRayCastMapper renders its prop Id when a vtkHardwareSelector is available.

This was already documented in README.md

Simplify another check

Seems like no longer needed OS X workaround.

make ospray pass create the scene graph so user doesn't have to
move ospray specific controls to ivars on ospraynodes
Improve caching
Now we cache what we can but not so much that we
fail time, or dynamic, or actor hiding. Also prevent double
delete on actor hiding and style
no special access just for ospray
Access to scenegraph view node would probably be OK.
make sure we get cell colors right for arbitrary polygons
using a reverse index from triangulated (or cylindered or pointed)
output cell back to input vtk cell. The output cells are homogenous
but the input cells aren't necessarily.
quad 0 -> tri 0 : rindex 0
          tri 1 : rindex 0
tri 1  -> tri 2 : rindex 0
color 0 -> color 0,1
color 1 -> color 2
flesh out point size and line width controls
Now user chosen pointsize and linewidth on mapper matter.
Also, user can designate a point aligned array to vary elements individually
and user can provide a lookuptable to manipulate those values arbitrarily.
In all cases there is an additional scaling control for fine tuning.
start to clean up actor to ospray translation
just use c++ new and remove embree calls to simplify
a single namespace for helpers
more logical flow for readability
Fix image space transformations
Also flesh out the implicits test and remove the related check from
render mesh now that it has its own test.
Fix some subtle bugs in caching.
Moved cache of ospray objects out to composite level. Would be nice to
cache indiviual blocks but the way I did it before was broken.
In process I found and fixed a bug regarding normals.
more cleanup actor to ospray translation and fix line sizing
was using wrong index into point value for line size, also changed
to use average of values at endpoints instead of first value
Also promote large chunks of the conversion process to the scenegraph level
because it is reusable. That and cleanup the ospray actor a lot.
move light vector out to renderer to allow multiple lights
also, just say no to c style casts.
toward consistent API for setting ospray specific state
The plan is for the scenegraph level to be restricted to read only
access to rendering/core state. Since backend specific classes are
hard to get a hold of (created late), options for them go into
information objects on the core classes. There will be setters/getters
on the backend classes though to make it easier to set them in the
case where you have access.
Fix multiple lights and add a test for lights
Make a consistent API for setting options that affect ospray.
fine tune lighting
lights now obey intensity and switch and color range is closer for
positional and directional lights. positional are still much
brighter for some reason so I've kludged in a 1/5 factor to make them
close.
actor colors obey diffuse and specular multipliers now and cell
aligned color maps now repect opacity and specularity. not sure if
I'll make point aligned colors do the same just yet since not
doing so saves space and keeps rgb and val interpolated consistent and
doesn't require changes to ospray to make texture materials respect them.
The key thing I understand now is that ospray does colors with
everything 0..1 except texture maps, which go 0..255.
Interesting to note that Ospray images are whiter than GL counterparts
when ambient occlusion enabled.
SceneGraph is not ready for default on, so take it out of rendering group.
Bunch of compiler warnings.
The collection iterator remove actor in loop change was important.
Don't know yet why implicits don't work on windows, otherwise in pretty good
shape as rest of tests pass. cmake find and ospray version
include need more work.
fix inconsistent background clearing
testdynamicobject was wrong before this, as was GL2 and interactive
camera motion. First would have nonsensical bckground, second would
streak.
Remove implicit SCALE hack and add a light Intensity hack.
Mapper SCALE is not needed since pointsize and linewidth are doubles
not integers in modern paraview. Add a light multiplier, applied to
ALL lights, for ParaView because with 6 double sided lights, the
default is too bright.
guard against actor without mapper and get scalars for multiblock
enable block colors and improve implicit colors and sizes
We now handle block visibility, opacity and color.
Fixed a case of always black implicits for paraview.
Added a new control that by default disables implicit scaling.
improve caching
had some cases where it would not rebuild when it needed to.
One was in paraview when changing colormap variable.
Confine ospray to a renderer instead of a renderwindow
Makes for easier interoperability with GL layers.
Fix lighting and multiple layers
Now that we are using the lights pass (and we should so that
we update camera positions) it exposed a bug where the default
created camera was not actually removed from the Renderer's light
collection. As a result, ospray renderings added duplicate default
lights and the images became washed out.
Also, added back in compositing of ospray layers which used to
happen and the ospraywindownode level. TODO: add a compositeRGB
call to renderwindow to do it nicely.
update ospray to only build MPI test when supported
Removed the dependence on MPI and replaced it with a
conditional. If parallel and MPI are already on then
it will include the MPI based ospray tests and the
required libs and includes
move actor code down to mappers
introduce mapperNodes to do the bulk of the work
also some cleanup to the updating children code
bunch more changes for mapper
more mapper related changes
fixes and resolve restore some missed features
Was a reference leak in parent, use weakpointer to fix
Need another mapper override for OpenGL1
HeaderTest fix
minor Style //---'s per vtk style guide
added back the use ambient color fallback and a handful of other
features that were in actornode only after start of move to mappernode
Fix scalar colors with multiblock data.
Before it would reuse block0's color arrays for block1 and the
result would be wrong.
Fill in per block and from field data color paths
Restore window node, move compositing to pass, cleanup traversal
This also implies that you could render without
even having opengl. Added a test for the windownode entry point.
Moved some of the compositing code into the pass. The idea
being that if you are building with OpenGL then you would
also build ospraypass. Otherwise you would not. The other classes
should not require a OpenGL implementation. Only the pass.
Changed the traversal implementation so that each step gets called
both before and after its decendents. This is implemented with a
prepass argument that is true before the decendents and false after.
Removed the YYYSelf() methods.
Removed most the ORender methods as they were a tad confusing.
Converted some ospray specific subclasses to just implement
Render() where before they had ORender.
Fixed two bugs in the writelayer compositing code.
Remove scenegraph's copies of state
Going with direct access to renderable
fix scenegraph test
miscellaneous fixes
Now specifying implicit scaling array by name rather than pointer.
Made Mapper:ClearColorArrays() not modify the mapper because it
disallows caching and is not necessary.
Restored the use ambient when diffuse is not available behavior
so that paraview point and cylinder representations look OK.
Restores ability to have ospray layers not at layer 0.
Re-enabled RenderPass::ClearLights to fix some GL1 tests.
Fixed header test.

Promoting RenderPass and thus FrameBufferObject and RenderState to
Rendering Core allows other modules to define RenderPasses that are
are independent of RenderingBackend choice.

Kitware Robot authored Mar 08, 2016 and Dan Lipsa committed Mar 08, 2016

Code extracted from:

    https://gitlab.kitware.com/third-party/proj.git

at commit 41bdf05361007f8c5186f3df9944d86bb273ed13 (for/vtk).

vtkDataArray subclasses now use

ComponentValue --> TypedComponent
TupleValue --> TypedValue

for their type-specific methods.

# Conflicts:
#	Rendering/Annotation/vtkScalarBarActor.cxx

Just added some parens to clarify precedence.

Reduce opengl2 dependencies in two ways
first move a class that renderingcore depends on from
filterExtraction to FiltersGeneral. FiltersExtraction pulls
in a lot of other modules where FiltersGeneral does not.
FiltersExtraction already depends on filtersGeneral so it
should be backwards compatible.

The second change is a new feature to allow a back implementation
to be required or not. In this case gl2ps was originally required
by OpenGl2 now it is optional. This required adding an option to
the module code to specify when a backend implementation is required.

#include <vtk([a-zA-Z0-9]+)\.h>
to
#include "vtk\1\.h"

but only in files named vtk.*\.cxx
excluding ThirdParty folder.

Find/replace of:

with C++ counterparts.  Changed only vtk*.cxx files.
Didn’t touch .h or .c or any 3rd party.

This function used to presume a window style used to compute the whole window
size needed for a requested client frame size. The obtained client size of the
window what finally not the one requested - thus the frame buffer was not of
the size obtained by calling GetSize(), leading to problem when capturing
image from the back buffer for instance.

The current window style is now directly obtained using the related Win32 API
function with the window hanle, ensuring a correct window rect computation.