Fix dependencies in tf2_ros.

[clalancette]

This will allow overlays on it to work properly.

Signed-off-by: Chris Lalancette clalancette@openrobotics.org
Signed-off-by: Dirk Thomas dirk-thomas@users.noreply.github.com

This should fix #268 .

[jacobperron]

LGTM

The testing failing in the PR job is an existing flake.

[dirk-thomas]

Can you clarify what the diff between this PR and the previous branch clalancette/overlay-include-problem is (and why there was a need to create a new separate one)?

[clalancette]

Can you clarify what the diff between this PR and the previous branch clalancette/overlay-include-problem is (and why there was a need to create a new separate one)?

Two main reasons:

1. The clalancette/overlay-include-problem branch has some nonsense code in it that we don't want to merge; it is just there to show the problem.
2. The changes on that branch to all of the cpp/hpp files to add include headers as appropriate are technically more correct, but not really necessary. Given how late we are in Foxy release cycle, this was the minimum amount of change we needed to solve this particular problem. I'll follow up with the include header changes in a separate PR that we can merge later.

[clalancette]

Here's CI:

• Linux
• Linux-aarch64
• macOS
• Windows

[clalancette]

The compilation warnings on macOS are expected; they show up on nightlies as well.

The compilation warnings on Windows are also expected; they show up on nightlies as well.

What is not expected is the test failures on Windows. The tests that failed looked like they immediately segfaulted on startup, and we do not see that on the nightlies. That needs more investigation before we merge this.

[jacobperron]

I can reproduce the RViz test failures locally on Windows. Looks like a bunch of "Entry Point Not Found" errors, for example if I run:

C:/Users/jacob/dev/ws/ros/geometry2/build/rviz_common/Release/frame_manager_test.exe

I get a dialog box containing the following error:

The procedure entry point ?hitButton@QPushButton@@MEBA_NAEBVQPoint@@@Z could not be located in the dynamic link library C:\Users\jacob\dev\ws\ros\geometry2\build\rviz_common\Release\rviz_common.dll

I don't really know why this change is causing the tests in RViz to fail. We also see the same test failures happening in Windows CI for #270 (which is an orthogonal change). Maybe it's a difference in our CI job and the nightly?

[clalancette]

I've now rebased this onto the latest. I think that the rviz failures are unrelated, as we are seeing them on all rviz PRs that go through CI (but not on nightlies). So those will have to be investigated separately.

I'm going to run CI one more time on this change, just to make sure everything else is good. Given that there have been no substantial changes since the approvals, I'm going to take those as still valid, and will merge assuming CI comes back OK.

CI:

• Linux
• Linux-aarch64
• macOS
• Windows