At the moment the cfo_supervisor.sh script doesn't respond correctly to ctrl-c - the child script continues to run while the parent zig and build processes become zombies. This happened after we moved the call to unshare from a command in the cfo_supervisor script to a syscall in the script binary.

After quite a big of debugging I have not solved the problem and think we should revert temporarily.

The unshare code on master is not quite correct yet - after forking the child should be resetting sigprocmask before execing. But that fix does not fix the ctrl-c problem.

I still do not understand what is going wrong but I think it has to do with delivery of SIGINT to the foreground process group, and this is affected by the exact hierarchy of processes in play, how they set their signal masks and how they wait on their children.

The bug only occurs when we are unsharing from a subprocess of the zig build system and the build program. If I e.g. use our unshare code to build an unshare bin, that bin seems to behave the same as stock unshare. But when the process hierarchy has the zig build system in between, like bash > zig > build > script something is going wrong with SIGINT. I am still debugging this but it seems to be pretty deep and I don't see a fix on the horizon.