GitLab

Fixes #19033. The box widget needed to faithfully follow the state of
UseReferenceBounds even when the ReferenceBounds were overridden. By
keeping PlaceWidget bounds and ReferenceBounds separate, we can support
this use-case.

Also added a test for this use-case.

Box implicit function was specified using a transform relative the data
bounds which were not exposed in the UI. This made it extremely tricky
to specify the box using absolute position/orientation/scale parameters.
Fixed that.

By default, now the box is always placed using absolute values i.e.
relative to a unit box anchored at origin. Added a mode that enables
user to provide an explicit bounding box to use for reference instead.
This will make it possible to support legacy state files.

Adding a new mode to vtkSMBoundsDomain where each the range is magnitude
for the bounds in each direction separately.

Fixes #18928. The issue was arising due to a bug in vtkSMViewLayoutProxy
where it did not consider separator width when positioning views in the
layout. Fixed that.

Improve support for stereo screenshots. vtkSMViewProxy now explicitly
handles the stereo left-right compositing after rendering each eye
separately. This is required to support stereo captures that need
tiling. Fixes #18974.

fixes #19017.

Avoid setting default values for filename prefix etc. for the Exodus
reader. This was causing issues like the one reported in #18962 which
resulted since incorrect values were getting saved for these properties.
Fixed by removing the logic to init these values by default. Users can
still manually specify them if needed.

Corresponds to d9bd44a5. Here we fix the catalyst extensions of
the writer proxy definitions to have them show up again in the
exporter GUI.

This remove usages of GetDomain(name) and replaces them by FindDomain<class>()
or FindDomain("class") in python.

This also transform usages of FindDomain("class") into FindDomain<class>()

This also add support for StringListDomain in python

  molecules can now be colored using RGB values from 3-components array.

vtkPVRenderView now sets up server and client stereo types used in a
fashion that takes into consideration which processes are the one
showing the results to the user. In tile display/cave mode, it's the
server side processes while in others it's the client. This makes it
possible to support active stereo with remote rendering without
requiring active stereo on the server side, for example.

fixes #18996.

vtkSMViewLayoutProxy was not pushing its custom state to the server side
when XML state was loaded. Fixed that.

fixes #19006.

vtkPVView was calling Window::Initialize to ensure that the shared
render window is initialized, but that doesn't seem to be adequate for
certain implementations. Reworking code to call `Render` on the shared
window before the very first render to overcome the issue.

ref #18938.

Layout proxy was not getting its initial state correctly in
collaboration mode. This was happening because unlike earlier, we now
create vtkSIProxy has objects on server side cause it to pull real
state for the proxy. The vtkSIProxy::Pull was not restoring user_data.
Fixed that.

Since we no longer automatically create layouts when a view is created,
we need to explicitly call `AssignViewToLayout` to assign a view to a
layout.

To fix #18964, pqMultiViewWidget no longer
automatically grabs a newly created view. Instead association of a view
with a layout is now an explicit action. This makes it possible to
faithfully reproduce Python traces when played back in the Qt GUI.

Since all views are now created 'detached', the APIs to create detached
views is being removed since it's no longer applicable.

Python state support was also imporved to correctly record view layouts.
Previously, layouts were simply not saved in Python state files.

SplitViewTrace and TraceMultiViews tests now do baseline compares to
validate the fixes.

Fixes #18964.

For now, we don't support comparative views on tile displays and cave
displays. We will add support for that later. These changes ensure that
comparative views otherwise work as expected.

Reported as #18985.

the test can now compare plots too. updated acordingly.

Previously in batch mode, the `vtkPVView::SetPPI` failed to propagate
the DPI correctly to the underlying render window thus ending up using
the default DPI instead. Since now the correct PPI is passed through,
the baselines needed to be updated.

The data server is never involved in actual rendering. Let's not set
rendering messages to it. The `asserts` added to
`vtkPVRenderView::Render` helped catch this unnecessary communication in
client-dataserver-renderserver mode.

vtkViewLayout support rendering a single view in CAVE mode. The active
view is rendered on the cave display. This commit adds logic to ensure
the windows are sized properly.

Since vtkIceTCompositePass renders to full window irrespective of
renderer's viewport, need to make sure that the vtkRenderer's viewport
is updated to full size as well, otherwise internal passes like depth
peeling can have issues when rendering on tile displays.

Fixes #18978

Updating TileDisplaySplitView to test translucent geomtry. Also added
support to render larger tiles for better test image comparison.

vtkIceTCompositePass unnecessarily maintained separate color buffers for
each eye. Since vtkRenderPass is full executed for each eye, there's
nothing we need to do really. Even in client-server, each eye will be
individually synchronized by vtkSynchronizedRenderer (or subclass).

This commit cleans up the code that manages remote rendering.
This fixes several issues: #15205, #17611, #18940, #18152, #18938,
and #18937. #17815 is also addressed since its no longer applicable.

ParaView no longer creates a single render window to use among all
vtkPVView instances on non-client processes like pvbatch or pvserver,
pvrenderserver. This shared render window was the root cause of several
headaches and necessitated classes like vtkPVSynchronizedRenderWindows
and extra burden on views to re-direct several calls like creating
windows, resizing windows to vtkPVSynchronizedRenderWindows. This is no
longer the case. Non-client processes still have a single viewable
window per proceses (vtkPVProcessWindow). However, this is not the
window that each view renders into. Instead, each view creates its own
vtkRenderWindow subclass. However, it's a FBO-based render window which
shares the context with the process specific render window
(created in vtkPVProcessWindow). This makes it possible to simply blit
back the rendering results from each view to the user-viewable window
when needed (e.g. tile display, cave, etc.).

`vtkPVSynchronizedRenderWindows` is now obsolete and removed with
appropriate convenience API now moved over to vtkPVView. This also
removes the need to initialize each view with a unique ID and hence
`vtkPVView::Initialze()` API is no longer needed and hence removed.

vtkIceTCompositePass was fixed to address several issues including tile
display aspect ratios (#18152, #17611). vtkPVIceTCompositePass is no
longer needed and hence removed which further simplifies the code.

Cleaning up the logic to generate images for debugging to be more
robust.