Add Support for SDF

[Amronos]

Fixes #632
I have modified hardware_interface/src/component_parser.cpp to add support for SDF models.

To test this out the demos being added through ros-controls/gz_ros2_control#427 can be used.

[saikishor]

Can you please add a test with an example SDF here: https://github.com/ros-controls/ros2_control/blob/master/hardware_interface/test/test_component_parser.cpp?

[codecov]

Codecov Report

Attention: Patch coverage is 80.00000% with 2 lines in your changes missing coverage. Please review.

Project coverage is 87.99%. Comparing base (d714e8b) to head (26b7f3b).
Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
hardware_interface/src/component_parser.cpp	66.66%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1763      +/-   ##
==========================================
- Coverage   88.01%   87.99%   -0.02%     
==========================================
  Files         121      121              
  Lines       12403    12411       +8     
  Branches     1108     1108              
==========================================
+ Hits        10916    10921       +5     
- Misses       1083     1085       +2     
- Partials      404      405       +1     

Flag	Coverage Δ
unittests	87.99% <80.00%> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
hardware_interface/test/test_component_parser.cpp	98.83% <100.00%> (+0.01%)	⬆️
hardware_interface/src/component_parser.cpp	94.47% <66.66%> (+0.04%)	⬆️

... and 1 file with indirect coverage changes

[saikishor]

@Amronos The changes look good to me. Can you add that SDF parsing is also supported through robot_description topic in the release_notes.rst?.

Sorry for not pointing out earlier

[saikishor]

LGTM

[azeey]

This is great! Thanks @Amronos

[Amronos]

Are there any updates on when this PR can be reviewed and merged?

[christophfroehlich]

Are there any updates on when this PR can be reviewed and merged?

Maintainers were busy with big PRs being merged the last days. We have so many other (older) PRs waiting for reviews, you can speed up the process by also reviewing them.

[Amronos]

The test successfully_parse_valid_sdf is passing for me when running colcon test, but failing in the GitHub checks. Any idea why this is the case?

[bmagyar]

Do you have a non-standard URDF library installed? The CI errors come from this segment

  // parse full URDF for mimic options
  urdf::Model model;
  if (!model.initString(urdf))
  {
    throw std::runtime_error("Failed to parse URDF file");
}

located here https://github.com/ros-controls/ros2_control/pull/1763/files#diff-ae927ba56161377a082def3ca0417c1052c37949d94e27a7713e3f9f798e790eR853
and the other print comes directly from within the initString() call from here

Seems legit to me but I'm wondering how it worked for you :D

[Amronos]

Do you have a non-standard URDF library installed?

I don't have a non-standard URDF library installed and the test is passing for me.
I did try rebuilding with colcon and testing multiple times. I also tried re-cloning the repo.
I also tested with a regular ROS launch file, in which I passed the SDF file to robot_state_publisher.
Is it possible that something in the CI is outdated?

[christophfroehlich]

Is it possible that something in the CI is outdated?

I don't think so, it fails on debian, RHEL and ubuntu

[Amronos]

ros-controls/gz_ros2_control#427 should help in testing this PR.

[christophfroehlich]

I am confused. Debian and RHEL are compiling fine now, but the test fails on these systems. What could be different here compared to the ubuntu jobs?
on ubuntu it reports

6: Warning [Utils.cc:132] [/sdf/model[@name="my_robot"]/ros2_control[@name="GazeboSimSystem"]:<data-string>:L191]: XML Element[ros2_control], child of element[model], not defined in SDF. Copying[ros2_control] as children of [model].

which is fine I guess?

[azeey]

@christophfroehlich in ros-controls/ros2_debian#11, even though sdformat_urdf was added to the .repos file, looking at the build logs shows that it was not built. Taking a closer look, I see that the set of packages to be built in the docker image is set in
https://github.com/ros-controls/ros2_debian/blob/ecfe2c5d801e0521f21402c3d8c88704df381c18/Dockerfile.debian12#L137-L144

and
https://github.com/ros-controls/ros2_debian/blob/ecfe2c5d801e0521f21402c3d8c88704df381c18/Dockerfile.debian11#L137-L144

but I don't see sdformat_urdf in there.

[christophfroehlich]

@azeey you are totally right, i fixed that and CI is happy now.

still on RHEL

dnf: Failed to detect successful installation of [ros-jazzy-sdformat-urdf]

I was on the wrong track again, because it compiles but behaves differently during runtime.

the gz_tools_vendor are merged but failed to build and sdformat_urdf still deactivated.

[azeey]

the gz_tools_vendor are merged but failed to build and sdformat_urdf still deactivated.

It looks like an issue with the build infrastructure. I'm working with @cottsay to get it resolved.

[christophfroehlich]

It looks like an issue with the build infrastructure. I'm working with @cottsay to get it resolved.

https://build.ros2.org/job/Jbin_rhel_el964__sdformat_vendor__rhel_9_x86_64__binary/
https://build.ros2.org/job/Rbin_rhel_el964__sdformat_vendor__rhel_9_x86_64__binary/

it seems this fails because tinyxml was not installed for urdf_parser as part of urdfdom

07:24:44 DEBUG: /usr/include/urdf_parser/urdf_parser.h:44:10: fatal error: tinyxml.h: No such file or directory
07:24:44 DEBUG:    44 | #include <tinyxml.h>
07:24:44 DEBUG:       |          ^~~~~~~~~~~

maybe not related to infrastructure, but only a missing dependency on rhel?

[bmagyar]

Thanks to all involved for pushing this!

[cottsay]

The infrastructure issue was resolved, the build failure you're seeing now is legitimate.

maybe not related to infrastructure, but only a missing dependency on rhel?

Seems like a bug in sdformat or the rosdep rules at first glance. The CMakeLists.txt finds tinyxml2, but tinyxml.h is provided by tinyxml (not 2). There completely separate packages.

[cottsay]

Ohh, wait - the include is referenced from the urdfdom-devel system package.

That's a missing dependency from the system package. I'll try to get it fixed upstream.

[azeey]

The version of liburdfdom-dev in Ubuntu is 4.0.0 whereas in RHEL, it is 1.0.4, so there seems to be a significant difference. I'm not sure what the timeline would be to get the update upstream, so I'll do the the alternative thing and use the urdfdom package from packages.ros.org in sdformat_vendor.

[cottsay]

I'm not sure what the timeline would be to get the update upstream

The same ABI stability we rely on to keep our builds turning over is the source of friction for getting this package updated. It's possible, but intentionally difficult to make such a large jump within a RHEL major.

Here's the bugfix to urdfdom in RHEL 9: https://bodhi.fedoraproject.org/updates/FEDORA-EPEL-2024-2c516e302e

We can work around the problem in the rosdep DB until the update goes stable next week, but it will require re-blooming to take the new rule. If you think that urdfdom 1.0.4 is likely to work and you don't want to wait until next week, I can start that process.

[azeey]

@cottsay and I spoke offline and agreed that using the ROS package urdfdom is the best option. New releases of sdformat_vendor are on their way:

• sdformat_vendor: 0.0.7-1 in 'jazzy/distribution.yaml' [bloom] ros/rosdistro#43363
• sdformat_vendor: 0.2.1-1 in 'rolling/distribution.yaml' [bloom] ros/rosdistro#43362

[saikishor]

@azeey Thank you for following up. Once these PRs are merged, we can trigger the CI.

I see that these are released to Rolling and Jazzy. Out of curiosity, with the Humble version of sdformat_urdf these tests are going to fail or would they work?.

We are making sure when the rolling is compiled from the source, it is compatible on Humble. That's the reason, I'm asking this question.

[azeey]

I see that these are released to Rolling and Jazzy. Out of curiosity, with the Humble version of sdformat_urdf these tests are going to fail or would they work?.

We've changed how Gazebo packages (e.g. sdformat) are released into ROS in Jazzy, that's why these releases are necessary. For Humble, sdformat_urdf should work fine on Ubuntu without any of these changes. I'm not sure about RHEL though. Do you in RHEL for the rolling/humble compatibility?

[saikishor]

We've changed how Gazebo packages (e.g. sdformat) are released into ROS in Jazzy, that's why these releases are necessary. For Humble, sdformat_urdf should work fine on Ubuntu without any of these changes. I'm not sure about RHEL though. Do you in RHEL for the rolling/humble compatibility?

For now, mainly focusing on the Ubuntu part. @christophfroehlich am I right?

[christophfroehlich]

For now, mainly focusing on the Ubuntu part. @christophfroehlich am I right?

We have only Ubuntu CI jobs for that. But I'm personally interested in having this running on Debian (I use that in my main project). But if its compilable from source, then everything fine.

[Amronos]

Can the checks be run again? I think they should pass now.
Thanks a lot everyone for all the help!

[azeey]

RHEL is ✅ 🎉

I'm not sure why the Jazzy and Rolling Binary Build checks are failing. The error messages don't seem related to this PR.

[bmagyar]

Thank you! Those tests are failing because we have a new function recently released in real-time_tools that isn't in main yet. Good thing we have all the variations of CI setup ;) we are good to merge this as soon as it gets a second approval

[saikishor]

LGTM