Workaround for ament_lint issues. Run CI every night.

[esteve]

No description provided.

[nnmm]

Looks like the same error happens also on humble.

[esteve]

@nnmm @jhdcs It seems to be caused by pydocstyle >= 6.2.0 ament/ament_lint#428, which gets installed by ros-tooling/setup-ros in https://github.com/ros-tooling/setup-ros/blob/master/src/package_manager/pip.ts#L51

Options I can think of:

• disable ament_pep257 for the rclrs_example_msgs package temporarily (https://github.com/ros2-rust/ros2_rust/blob/main/rclrs_example_msgs/CMakeLists.txt#L22)
• manually build https://github.com/ament/ament_lint and install it after running ros-tooling/setup-ros
• manually patch ament_pep257/ament_pep257/main.py to update it for the new pydocstyle version

3 seems easiest, 2 would be the most "correct" one (but every CI run will take longer) and I'd try to avoid 1

[nnmm]

@esteve Have you seen ament/ament_lint#428? I think it has already been fixed, we just need to wait for a Rolling sync: ros-tooling/action-ros-ci#795 (comment)

[nnmm]

I did try using the prerelease/testing APT repo as suggested in the latter link, but it didn't work.

[jhdcs]

My personal vote would be going for the most correct option - however, how much time would this be adding to CI?

[esteve]

My personal vote would be going for the most correct option - however, how much time would this be adding to CI?

This would mean rebuilding ament_lint, which won't take long (it's just a bunch of CMake scripts), and I'm assuming that installing all its development dependencies won't take long either, I don't know for sure, but it's worth a try.

[esteve]

I did try using the prerelease/testing APT repo as suggested in the latter link, but it didn't work.

Yeah, that's weird, there are in fact .debs with newer releases of ament_pep257 for both humble and rolling in the testing repo:

http://packages.ros.org/ros2-testing/ubuntu/pool/main/r/ros-rolling-ament-cmake-pep257/
http://packages.ros.org/ros2-testing/ubuntu/pool/main/r/ros-humble-ament-cmake-pep257/

I don't know why they don't get installed. I've updated setup-ros to v0.6 in case that's why the action was not picking up the testing repository.

[esteve]

setup-ros@v0.6 dropped support for Galactic as it's no longer supported by Open Robotics (support ended in November https://www.ros.org/reps/rep-2000.html#galactic-geochelone-may-2021-november-2022)

[esteve]

Unfortunately, until ament_lint gets tagged for foxy and the latest release includes ament/ament_lint#431, the foxy job will continue failing. I filed ament/ament_lint#434, so hopefully we can use the ros-testing repo with foxy at some point.

[nnmm]

@esteve There is also another error in the rolling build:

Starting >>> rclrs_example_msgs
  --- output: rclrs
  error[E0308]: mismatched types
     --> src/subscription/message_info.rs:120:19
      |
  120 |             data: rmw_message_info.publisher_gid.data,
      |                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ expected an array with a fixed size of 24 elements, found one with 16 elements

[esteve]

@nnmm yeah, saw that, but I'm more interested now in getting humble and foxy passing. Looks like the publisher_gid field got updated to use 16 bytes, instead of 24, in rolling (see ros2/rmw#345)

[esteve]

@nnmm I've redone this PR to just install a non-problematic version of pydocstyle until foxy gets a new version of ament_lint / there's a new sync. I've also added a schedule for running the CI job every night so that we can catch errors in rolling earlier, we didn't see the issue with the publisher GID until there was a PR, so it'd be better to be able to catch these issues earlier. We can add use-ros2-testing: ${{ matrix.ros_distribution == 'rolling' }}, but I'd rather do it once #306 gets merged and we upgrade setup-ros to v0.6

@maspe36 once this PR is merged, can you rebase #306 on top of main? Or rebase #306 on top of this PR now? I minimized the changes so it shouldn't have any conflicts with yours. It should fix the CI failures you see in your PR (which are not caused by your changes, it's still the pydocstyle problem). Thanks.

[nnmm]

LGTM!

[nnmm]

I hope it's fine if I merge your PR this time @esteve, to unblock other PRs.

[esteve]

@nnmm sure! Thanks for merging this PR 🙂