Noetic exotica_pinocchio failing to build - cc1plus killed

[sloretz]

I seem to remember this being a known issue, but I couldn't find where it was talked about. The exotica_pinocchio job often fails to build on the Noetic buildfarm. If I remember correctly if the job runs on a machine by itself then it will pass, otherwise it runs out of memory.

https://build.ros.org/job/Nbin_uF64__exotica_pinocchio_dynamics_solver__ubuntu_focal_amd64__binary/82

00:10:50.344 c++: fatal error: Killed signal terminated program cc1plus

Past work at reducing the memory usage in: #729

[wxmerkt]

Hi @sloretz, Indeed I've been seeing these failures quite a lot and was hoping the jobs will eventually pass. There is little more we can do - I split the compilation units down even further (e.g. in 8fb184c0c8226ae5777b4a837ffe909df2237a12) and it didn't help.

One thing that helps - and something I might set as a patch in the next release - is to switch the compiler to clang. I reckon by adding a <buildtool_depend>clang</buildtool_depend> and setting it in the CMakeLists as a patch that should work - or would there be additional steps required?

[sloretz]

I didn't find any packages using clang to build themselves in Melodic, Noetic, or Rolling, so this package would be the first. As long as the libraries and executables output from clang are ABI compatible with gcc then I would expect it to work with only those two steps.

[wxmerkt]

I've tested building only one package with clang and the rest with gcc and they are ABI compatible (the Pinocchio package is used as a plugin via pluginlib in the other packages). I've just released a new version to Melodic & Noetic and hope this resolves our memory problems (beyond clang there is nothing further we can do):

• https://github.com/ros/rosdistro/pull/30317
• https://github.com/ros/rosdistro/pull/30318

[wxmerkt]

Closing as this should be resolved. We still have failures on armhf due to the derivatives of the dynamics in Pinocchio - but we cannot split the compile units any further.