All of this sounds pretty reasonable, but my goal was to document the simplest way I saw it work back in the day. You can keep the process alive if copyover fa

All of this sounds pretty reasonable, but my goal was to document the simplest way I saw it work back in the day. You can keep the process alive if copyover fails by writing to a temporary file and designing the server so that you get a "rescue console" if it fails to come up. Then you could inspect the dumped file, see what's going wrong, and exec() into another binary to try again. You could even use sqlite as your state file, which would let you interactively explore and edit it during debugging. A custom allocator sounds like an extremely interesting approach. I'm still not sure whether I'd want to rely on struct layouts not changing, but if you use it diligently you at least make it easier to find everything you've allocated.