The Makefiles generated and executed on the parallel workers are invalid when they are copied back to the client, as all the paths in the Makefiles are relative to the temporary build folders used by the workers. In previous releases, the Makefiles were regenerated on the client during the merge phase, but this is expensive and unnecessary for the parallel build to succeed.
When "GenCodeOnly" is on, where the intention is to compile the code at a later point, the Makefiles are still regenerated. However, when "GenCodeOnly" is off, the Makefiles are discarded which gives a performance boost depending on the size/structure of the model reference hierarchy.
One possible workaround is to regenerate the Makefiles with the "rtwrebuild" command:
Another workaround is to call "rtwbuild" twice, once with "GenCodeOnly" on, and once with "GenCodeOnly" off:
rtwbuild('model_name'); % GenCodeOnly='on';
rtwbuild('model_name', 'generateCodeOnly', false);
