Macos -latomic from rcutils

[gftabor]

Solution to issue cannot be found in the documentation.

• I checked the documentation.

Issue

Trying to build ros2_control package (specifically here https://github.com/gftabor/pixi_test ) on macOS fails with make -latomic call.

(mqp) gftabor@x86_64-apple-darwin13 pixi_test % colcon build --symlink-install  --cmake-args -DBUILD_TESTING=OFF
dyld[49989]: symbol '_iconv_close' missing from root that overrides /usr/lib/libiconv.2.dylib. Use of that symbol in /usr/lib/libcups.2.dylib is being set to 0xBAD4007.
dyld[49989]: symbol '_iconv_open' missing from root that overrides /usr/lib/libiconv.2.dylib. Use of that symbol in /usr/lib/libcups.2.dylib is being set to 0xBAD4007.
dyld[49989]: symbol '_iconv' missing from root that overrides /usr/lib/libiconv.2.dylib. Use of that symbol in /usr/lib/libcups.2.dylib is being set to 0xBAD4007.
Starting >>> joint_limits
[0.390s] colcon.colcon_ros.prefix_path.ament WARNING The path '/Users/gftabor/Documents/pixi_test/install/rcutils' in the environment variable AMENT_PREFIX_PATH doesn't exist
[0.390s] colcon.colcon_ros.prefix_path.catkin WARNING The path '/Users/gftabor/Documents/pixi_test/install/rcutils' in the environment variable CMAKE_PREFIX_PATH doesn't exist
Starting >>> ros2_control_test_assets
Starting >>> controller_manager_msgs
[0.774s] colcon.colcon_cmake.task.cmake.build WARNING Unable to determine CMake version, building the installtarget instead of invoking 'cmake --install'
Finished <<< ros2_control_test_assets [0.96s]
--- stderr: controller_manager_msgs
ld: library not found for -latomic
x86_64-apple-darwin13.4: error: linker command failed with exit code 1 (use -v to see invocation)
make[2]: *** [CMakeFiles/controller_manager_msgs__rosidl_generator_c.dir/build.make:1340: libcontroller_manager_msgs__rosidl_generator_c.dylib] Error 1
make[1]: *** [CMakeFiles/Makefile2:293: CMakeFiles/controller_manager_msgs__rosidl_generator_c.dir/all] Error 2
make: *** [Makefile:146: all] Error 2
---
Failed   <<< controller_manager_msgs [1.03s, exited with code 2]
Aborted  <<< joint_limits [3.72s]

Summary: 1 package finished [3.85s]
  1 package failed: controller_manager_msgs
  1 package aborted: joint_limits
  2 packages had stderr output: controller_manager_msgs joint_limits
  11 packages not processed
(mqp) gftabor@x86_64-apple-darwin13 pixi_test % 

This seems to be caused by installed version of rcutils

pixi_test/.pixi/envs/default/share/rcutils/cmake/ament_cmake_export_link_flags-extras.cmake

# generated from ament_cmake_export_link_flags/cmake/ament_cmake_export_link_flags-extras.cmake.in

set(_exported_link_flags "-latomic")

# append link_flags to rcutils_LINK_FLAGS
if(NOT "${_exported_link_flags} " STREQUAL " ")
  foreach(_link_flag ${_exported_link_flags})
    list(APPEND rcutils_LINK_FLAGS "${_link_flag}")
  endforeach()
endif()

I am not sure if there is a trick the CI server is doing to ignore this error? Seems like any building from source that links with rcutils would fail.

Installed packages

ros-jazzy-rcutils                                 6.7.5               np2py312hb847498_16     115.70 KiB  conda  https://prefix.dev/robostack-jazzy
ros2-distro-mutex                                 0.14.0              jazzy_16                  2.26 KiB  conda  https://prefix.dev/robostack-jazzy

Environment info

Pixi version: 0.63.1
        TLS backend: rustls
           Platform: osx-arm64
   Virtual packages: __unix=0=0
                   : __osx=15.5=0
                   : __archspec=1=m1
          Cache dir: /Users/gftabor/Library/Caches/rattler/cache
       Auth storage: /Users/gftabor/.rattler/credentials.json
   Config locations: No config files found

Global
------------
            Bin dir: /Users/gftabor/.pixi/bin
    Environment dir: /Users/gftabor/.pixi/envs
       Manifest dir: /Users/gftabor/.pixi/manifests/pixi-global.toml

Workspace
------------
               Name: mqp
            Version: 0.1.0
      Manifest file: /Users/gftabor/Documents/pixi_test/pixi.toml
       Last updated: 14-05-2026 13:06:46

Environments
------------
        Environment: default
           Features: default
           Channels: https://prefix.dev/conda-forge, https://prefix.dev/robostack-jazzy
   Dependency count: 37
       Dependencies: compilers, cmake, pkg-config, make, ninja, pip, ros-jazzy-desktop, python, setuptools, colcon-common-extensions, rosdep, numpy, ros-jazzy-ament-cmake-google-benchmark, ros-jazzy-performance-test-fixture, ros-jazzy-mimick-vendor, ros-jazzy-ament-cmake-vendor-package, ros-jazzy-ament-cmake-clang-format, ros-jazzy-test-msgs, ros-jazzy-navigation2, ros-jazzy-ros-gz-sim, ros-jazzy-ackermann-msgs, filelock, tinyxml2, ros-jazzy-realtime-tools, ros-jazzy-control-msgs, ros-jazzy-pal-statistics, ros-jazzy-rsl, ros-jazzy-generate-parameter-library, ros-jazzy-xacro, ros2-distro-mutex, ros-jazzy-ros2-control, ros-jazzy-joint-limits, ros-jazzy-ros2-control-cmake, ros-jazzy-control-toolbox, ros-jazzy-ros-gz-bridge, ros-jazzy-gz-ros2-control-demos, ros-jazzy-hardware-interface-testing
   Target platforms: linux-64, osx-64, win-64
    Prefix location: /Users/gftabor/Documents/pixi_test/.pixi/envs/default

[traversaro]

xref: ros2/rcutils#516

[traversaro]

I have no idea why this works in CI, but it seems that anyhow the if in https://github.com/ros2/rcutils/pull/516/changes#diff-1e7de1ae2d059d21e1dd75d5812d5a34b0222cef273b7c3a2af62eb747f9d20aR101 should also have AND NOT APPLE or similar?

[gftabor]

So it seems like building rcutils from source works just fine. Presumably 'HAVE_LIBATOMICS' environment variable in that commit is resolving correctly that the mac doesn't have libatomics?

The installed version from the CI build must have passed the current check. We could maybe make the check more aggressive, or figure out why the CI thinks it has access to libatomics?

Also this dependency is really annoying. Its not listed anywhere colcon can find, so building whole workspace doesnt work correctly. In parallel it builds joint_limits with the robostack installed rcutils, while also building rcutils instead of realizing to do them sequentially. I had to build only rcutils, source the workspace and then I could build all the ros2_control packages that depend on it.

[Tobias-Fischer]

I have no idea why this works in CI, but it seems that anyhow the if in https://github.com/ros2/rcutils/pull/516/changes#diff-1e7de1ae2d059d21e1dd75d5812d5a34b0222cef273b7c3a2af62eb747f9d20aR101 should also have AND NOT APPLE or similar?

I'd be happy to go with that approach!

@gftabor do you want to create a PR with a patch and bumping the build number for rcutils?

[gftabor]

I have no idea why this works in CI, but it seems that anyhow the if in https://github.com/ros2/rcutils/pull/516/changes#diff-1e7de1ae2d059d21e1dd75d5812d5a34b0222cef273b7c3a2af62eb747f9d20aR101 should also have AND NOT APPLE or similar?

I'd be happy to go with that approach!

@gftabor do you want to create a PR with a patch and bumping the build number for rcutils?

To clarify the current code works fine when building everything from source locally.
The current check seems to only do something incorrect on the robostack CI.

[gftabor]

O I just reread your comment. I thought you were saying PR to rcutils. Ya I'll make a patch and PR here

[traversaro]

I think also a PR upstream would be welcome, but yes definitely that is not blocking. fyi @idesign0 as I suspect this will also affect macOS with Homebrew.

[idesign0]

Thanks for the heads up @traversaro !

Source build seems fine on my side as well — CI is building rcutils from source and everything passes cleanly (controller_manager and friends all build fine). I'll inspect the flag propagation issue more closely once I have Mac access. AND NOT APPLE upstream sounds like the right fix. :)

[traversaro]

The job that created the package is https://github.com/RoboStack/ros-jazzy/actions/runs/22041123150/job/63701387268, but no low is available. What I suspect happened is that by chance in the CI runner some kind of gcc installation was available, so libatomic was available, but the fact that libatomic is available does not mean that it should be linked, especially on macOS where it is not particularly useful, so I think ros2/rcutils#565 make total sense anyhow.

[traversaro]

I think also a PR upstream would be welcome, but yes definitely that is not blocking.

Actually I think that upstream the issue or possible issue was already solved by https://github.com/ros2/rcutils/pull/528/changes that did not export the -latomic flag. On the other hand, if atomic flag is not useful on macOS, it is better never to link it at all, so ros2/rcutils#565 still make sense to me.

[gftabor]

The new build works and the upstream PR was merged as well

#218