Support new target export template introduced with CMake 3.24

[roehling]

The latest CMake 3.24 has tweaked its export template, so the regular expression in ament_cmake_export_targets-extra.cmake no longer matches. This PR relaxes the regex to match both the old and the new template.

[ooeygui]

Thank you for submitting a fix! I've confirmed that this does build. I would love to see this pulled into humble as well.

[clalancette]

Thanks for the change, this looks reasonable to me. I'm going to go ahead and run full CI on this.

[clalancette]

CI:

• Linux
• Linux-aarch64
• Windows

[clalancette]

The flaky test on Windows is well known, so going ahead and merging this (and backporting to stable distributions).

[clalancette]

@Mergifyio backport humble galactic foxy

[mergify]

backport humble galactic foxy

✅ Backports have been created

• #397 Support new target export template introduced with CMake 3.24 (backport #395) has been created for branch humble
• #398 Support new target export template introduced with CMake 3.24 (backport #395) has been created for branch galactic
• #399 Support new target export template introduced with CMake 3.24 (backport #395) has been created for branch foxy

[ijnek]

May I request for a release in all distros please @clalancette ?

[ijnek]

I can't quite understand the reason why using custom messages are failing on Foxy, Galactic and Humble even they aren't using CMake > 3.24, but there are many recent issues (Failed to find exported target names in) that this PR has probably fixed.

Humble:
moveit/moveit2#1611
moveit/moveit2#1570

Foxy:
dusty-nv/jetson-containers#181
https://answers.ros.org/question/406709/sudden-custom_msgs-related-error-in-ros2-build-failed-in-ci/
colcon/colcon-core#525

This started happening in one of my galactic CIs: https://github.com/ros-sports/soccer_interfaces/actions/runs/3333641352/jobs/5515776865 , and I can confirm that the changes in this PR does fix the failure (when i build ament_cmake from source in CI).

I have been trying to make a minimal reproducible example (https://github.com/ijnek/reproduce_ament_cmake_395) but I haven't been able to reproduce it yet, or reproduce the issue locally on my computer in a container.

Does anyone know what the changes were that broke the galactic builds?
@clalancette Could we get these changes released in all distros please?

[clalancette]

@clalancette Could we get these changes released in all distros please?

I mean, we can, but we really should figure out why it is happening in all distributions. By default, it shouldn't happen anywhere but on Windows (which ends up rolling forwards all the time). I suspect somewhere in setup-ros or action-ci-ros is installing a custom version of CMake, and it shouldn't do that.

[ijnek]

From the logs, the CMake version being installed does seem to be 3.16.3.

2022-10-27T00:47:57.1119909Z Setting up cmake (3.16.3-1ubuntu1) ...

I understand that this is something for me to figure out, but I'm sort of stuck on this one. So if anyone else out there is experiencing something similar, I'd like to know!

[quarkytale]

I was trying to build moveit2 on humble and it requires CMake >= 3.22. So I manually upgraded to 3.24, stumbling upon this issue of "Failed to find exported target names".

[ijnek]

My issue was caused by Github's action runner manually installing CMake > 3.24, which I didn't notice due to cmake 3.16.3 also being installed side-by-side. I've written up an explanation on ROS Answers for this specific case and how to prevent being affected by such problems by using containers in Github Actions rather than running builds directly on the Github Action Runners.

[clalancette]

My issue was caused by Github's action runner manually installing CMake > 3.24, which I didn't notice due to cmake 3.16.3 also being installed side-by-side. I've written up an explanation on ROS Answers for this specific case and how to prevent being affected by such problems by using containers in Github Actions rather than running builds directly on the Github Action Runners.

All right, thanks for tracking this down. The good news is that we'll be releasing this fix into Humble in the next patch release, soon-ish. Then we'll be able to handle CMake 3.24 on both Rolling and Humble.

[ijnek]

Then we'll be able to handle CMake 3.24 on both Rolling and Humble.

Foxy too?

[clalancette]

It doesn't look like we've backported it there, so no, not at the moment. Have we seen instances of the problem on Foxy?

[ijnek]

I've been reproducing it on galactic, and foxy (although I haven't checked) would very likely have the same problem too.

It seems like you have backported it to galactic and foxy.

[clalancette]

I've been reproducing it on galactic, and foxy (although I haven't checked) would very likely have the same problem too.

It seems like you have backported it to galactic and foxy.

Oh, I looked too quickly. In the case of Foxy, that was backported and released back in September, so it should not be a problem anymore there.

We backported and merged it to Galactic, but have not yet released it. Since this month will be the final patch release of Galactic, it should make it in there too.

[ijnek]

Great, thanks @clalancette !