Hey Ludo’,

I've attached an updated patch. I couldn't think of any unwanted 
consequences, so I took your idea of making register-services 
handle most of the details of replacement. With my patch, 
something like
>   herd eval root '(register-services (load "a.scm") (load 
>   "b.scm"))'
will deal with a conflict by either replacing the old service (if 
it's not running), arranging for the old service to be replaced 
when it's stopped, or raising an error. This seems like a sensible 
way for things to function.

>> At the very least we need to control the inherent race 
>> condition [...]
>
> Indeed.

Despite my desire to deal with the race condition, I haven't done 
anything about it in this patch. The modification of %services 
that was done in register-services was already racy, and I don't 
think this patch will make it worse. If it hasn't been a problem 
up until now, then I don't think this will make it a problem.

Carlo