chore: disable colcon-powershell

[bjsowa]

This is a hotfix for the CI failing due to ModuleNotFoundError: No module named 'ament_package' error (see here for example). The workflow has been failing for about a month now and I suspect it is due to some recent change to the Github runner images. I couldn't find any other solution than running the job in a container.

The other test is failing due to recent changes in rclpy that changed some executor behavior (which I authored 😅). It will be fixed in another PR later.

[MatthijsBurgh]

In my mind it is very unlikely an update in a runner image causes the CI to break. But here is the diff between the image releases, link

You can check it to see whether you see something interesting.

[bjsowa]

In my mind it is very unlikely an update in a runner image causes the CI to break. But here is the diff between the image releases, link

You can check it to see whether you see something interesting.

I tried looking for something interesting there but couldn't find anything. I'm suspicious of runner image for two reasons:

• Running the CI in ubuntu:noble container image or using ubuntu-slim runner (see here) does not result in the error.
• I have a private repo where the CI builds the whole ROS (including ament_package) in a Nix devshell which freezes all the dependencies and it started suffering from the same error around similar time.

[MatthijsBurgh]

Can we print the filesystem tree or similar just before building? I am just looking for a way to debug this issue.

[bjsowa]

@MatthijsBurgh Github actions seem to be down for now

[MatthijsBurgh]

Yes Indeed 😢

[bjsowa]

@MatthijsBurgh I have no more ideas for things I can check for now.

[sea-bass]

honestly? I had CI fail on one of my repos and deleting all the github actions caches and rerunning solved it for me. Try that?

EDIT: oh, there are no caches here...

[bjsowa]

honestly? I had CI fail on one of my repos and deleting all the github actions caches and rerunning solved it for me. Try that?

EDIT: oh, there are no caches here...

There is only cache from pre-commit job. It shouldn't affect anything but I cleared it anyway and it did not change anything.

[bjsowa]

@MatthijsBurgh Any idea how to debug this further?

[MatthijsBurgh]

Preferably we can print the state midway through the ros-ci-action. Are there any hooks we can define?

[bjsowa]

Preferably we can print the state midway through the ros-ci-action. Are there any hooks we can define?

We could just try running the commands manually, instead of using the action

[MatthijsBurgh]

We can try checking the colcon logs.

See here how to store the logs as artifacts

[MatthijsBurgh]

@bjsowa when running on the runner the logs show the following PYTHONPATH

PYTHONPATH=/home/runner/work/rosbridge_suite/rosbridge_suite/ros_ws/install/rosbridge_test_msgs\lib/python3.12/site-packages;${PYTHONPATH}

The backslash and the semi-colon show that some part of the tooling, either colcon/ament or the action-ros-ci, thinks we are on Windows. I think this is something we should investigate.

[MatthijsBurgh]

@bjsowa I don't know what changed in december causing the issues. But the issue is colcon-powershell not being skipped. It wasn't skipped, because the PSModulePath is set in the runner. See the check For me it is hard to say whether that check is okay or too much of a shortcut.

But by overriding it in the CI config, we make it work for now.

Might be fixed by colcon/colcon-powershell#42

[bjsowa]

Thank you @MatthijsBurgh for finding the culprit! Should we remove the other changes or leave it as it is?

[MatthijsBurgh]

I kept these changes as I think both changes (bash as default shell; upload log artifacts) are useful. So you can merge if you agree.

Changes