thus suppressing Ogre runtime warning

Also consider QEvent::Show in SplitterHandle's event filter.

Newer versions of libMesa cause issues when rendering materials with glsl150 scripts.
Thus, downgrade to glsl120 scripts.

* Remove EXPORTs for ScrewVisual, ScrewDisplay

The interactive_markers package auto-completes scales to 1.0 if they are zero.
Hence, we should simply ignore those values.

With OGRE's cmake files prior to version 1.11, OGRE_INCLUDE_DIRS
doesn't comprise include dirs of components. Similarly for OGRE_LIBRARIES.
While we don't find cmake files on Ubuntu (but use pkgconfig instead),
on other systems, e.g. Arch Linux, the build would fail.

Given an empty vector, size()-1 would yield a huge number to delete items at...
Fixup for a22c731
Fixes #1583.

As we see more OpenGL issues, this should facilitate issue handling on github.

PointStampedDisplay didn't display initial points (before the circular buffer was fully filled) because setting the point radius didn't update the (initially zero) scale.

Mesa-based rendering issues showed up e.g. on AMD GPU as well. The new approach should handle any Mesa 20-based system.

* Fix creation of properties
* Fix unresolved symbol processMessagePrivate

They are programmatically set to inform the user.

… is off

Occasionally, the RenderPanel received an additional resize event with a wrong size of 640x480.
Further, floating and docked are two mutually exclusive properties of a QDockWidget. Thus, fixed default args.

Co-authored-by: Robert Haschke <[email protected]>

- trailing whitespace
- end of file
- yaml: single doc

- enable full (and slow) clang-tidy check via manual workflow triggering only
- make CLANG_TIDY non-pendantic (there are too many open issues)
- add prerelease workflow
- auto-format .clang-format
- drop caching of target_ws and docker image
  - the docker image continuously accumulates layers (comprising environment at least)
  - caching the target_ws missed to rebuild sometimes

VisualizationManager uses [start|stop]Update() to pause ROS updates during dialog box interaction (#631).
However, asynchronous updates via threaded_queue_ were still performed, leading to a memory leak (#1621),
because messages were partially processed by the thread (i.e. kept in memory), but not finally processed
(and cleaned up) by the synchronous Display::update() call.
Pausing threaded processing too resolves that issue.

... since ros/geometry2#402 is released

PointCloudCommon was defining its own callback queue and spinner
to allow for asynchronous processing of point cloud data.
However, as most PointCloud displays just inherit MessageFilterDisplay,
which is using the default update_nh_, eventually the default synchronous
callback queue was used.
(Although the callback queue was configured in the constructor to point
to PointCloudCommon's private queue, Display::initialize() was resetting
the queue back to the default.)

... to allow rqt_rviz notice config file changes

... allowing rqt_rviz finding the action easily by name

Partially reverts #631. Fixes #1606.

All sensitive members of PointCloudCommon are mutex-protected anyway.

* Smoothly update the scene node's position + orientation to keep history of arrows/axes 
  even if fixed_frame is different from parent frame_id of Odometry.msg
* Handle large float values in Odometry message (e.g. when transforming GPS to UTM coordinates)
  * Introduce ref_pose
  * Use double precision to calculate the distance between the current and last position

In order to correctly display the path/history of Odometry messages when the
message's frame_id is moving relative to rviz' fixed frame, we need to update
the pose of the corresponding scene node in each update() cycle.
This is a partial backport of #1631.

Update the PCL's scene node in each update() cycle to ensure a smooth motion when the message's frame_id moves w.r.t. rviz' fixed frame. So far, the pose was only computed once when the message arrived.

… is off

Backward port from Noetic

This reverts commit 227678f.

This fixes a segfault reported in #1658

For a small number of total points, we can place several lines into
a single Ogre::BillboardChain. However, if we have more than MAX_ELEMENTS
points, we need to distribute a line across several BillBoardChains.

We need to wait for TF2 0.6.7 to be actually released.
This reverts commit 25d2c81.

…)"

This reverts commit a6e58df.

This reverts commit 7bdbeea.

This reverts commit 5b27a0c.

Fixes #1674

Co-authored-by: Robert Haschke <[email protected]>