feat: support both qt5 and qt6

[wep21]

rework on #707

[ahcorde]

• Linux
• Linux-aarch64
• Windows

[wep21]

windows script uses Qt5_DIR here.

[wep21]

@ahcorde I appreciate it if you could rerun ci.

[ahcorde]

• Windows

[wep21]

@ahcorde The ci still failed, so I modified to use env.

[ahcorde]

• Windows

[wep21]

@ahcorde I'm sorry for the inconvenience, there was a small mistake in statement, so I fixed it.

[ahcorde]

• Windows

[wep21]

I'm sorry, but I have no idea why the windows ci failed. It may take some time to fix the issue.

[clalancette]

I'm sorry, but I have no idea why the windows ci failed. It may take some time to fix the issue.

One thing you should know about our Qt installation on Windows; it is using a very old version of Qt, version 5.12.10: https://github.com/ros-infrastructure/ros2-cookbooks/blob/8ac2ddf1ceec864211971478bf3d515fbe71298e/cookbooks/ros2_windows/recipes/qt5.rb#L74

That's because after that version, Qt stopped providing downloadable executable files for Windows, so we had no way to auto-install it.

In any case, that very old version may not have the QT CMake target that you are currently expecting. It is worth investigating in that direction.

[MichaelOrlov]

@ahcorde Could you please help to handle this issue. On the latest waffle triage meeting we decided to move it to the backlog with the following notes:

Failure on Windows. Windows setup script hardcoded installation for qt5 only. Qt5 doesn't support the new style find_package(QT NAMES QT6 QT5) https://github.com/ros2/rviz/pull/1187/files#diff-f058f5df521489857a0241a6838777373aa24c99a073c3cd9cf2eee427b167eaR26-R27. A proper way to handle it would be to use conda package manager. It seems a lot of work will need to switch to conda.

[wep21]

I added workaround for windows ci again. I appreciate it if someone could rerun ci.

[MichaelOrlov]

Re-run Windows CI after attempting to fix the build errors.

• Windows

[wep21]

The error says

qt_wrap_cpp called with incorrect number of arguments

so I added the following workaround for windows ci

  function(qt_wrap_cpp out)
    qt5_wrap_cpp(_sources ${ARGN})
    set("${out}" ${_sources} PARENT_SCOPE)
  endfunction()

[ahcorde]

Re-run Windows CI after attempting to fix the build errors.

• Windows

[wep21]

I also updated workaround in rviz_rendering_tests which I forgot to modify.
cc @ahcorde @MichaelOrlov

[ahcorde]

Re-run Windows CI after attempting to fix the build errors.

• Windows

[wep21]

note: need to fix this error

mock_display.obj : error LNK2001: unresolved external symbol "public: virtual struct QMetaObject const * __cdecl rviz_common::MockDisplay::metaObject(void)const " (?metaObject@MockDisplay@rviz_common@@UEBAPEBUQMetaObject@@XZ) [C:\ci\ws\build\rviz_common\rviz_common_display_test.vcxproj]

[srmainwaring]

Great to see Qt6 support for rviz. The package.xml still have a dependency on Qt5:

rviz/rviz2/package.xml

Line 31 in 32eb8b9

<build_depend>qtbase5-dev</build_depend>

Is there a preferred method to optionally include either Qt5 or Qt6 dependencies (or include them both but make them optional)?

[clalancette]

Is there a preferred method to optionally include either Qt5 or Qt6 dependencies (or include them both but make them optional)?

We have conditionals in package.xml, but they can really only select on an environment variable.

In reality, though, there will only be one "supported" configuration at a time. That is, we can leave these as Qt5 dependencies for now, and then when we officially switch over, we can change these to Qt6 dependencies. For users of the other one (that is not officially supported), they can still build from source, they just won't be able to successfully run rosdep.

[wep21]

@ahcorde @clalancette I appreciate it if you could rerun windows ci.

[ahcorde]

• Windows

[ahcorde]

• Linux
• Linux-aarch64
• Windows
• Windows Debug

[clalancette]

Can we make sure to run Windows Debug as well? Otherwise, this is great!

[ahcorde]

Can we make sure to run Windows Debug as well? Otherwise, this is great!

Sure, added to the list

[ahcorde]

• Windows Debug

[wep21]

@clalancette @ahcorde Is it possible to go ahead to merge?

[ahcorde]

@wep21 not really, there are some failing tests on Windows Debug.

I will take a look too

[ahcorde]

I was able to compile rviz2 with qt6, just using the comparison that I mentioned in the comment

[ahcorde]

instead of using in the condition windows why don't use the Qt version instead?

[wep21]

addressed here 3977093

[ahcorde]

• Windows

[wep21]

Building rosidl_generator_py seems to fail.

[ahcorde]

• Windows