Ok, it seems like I did not explain it right.

Well, it goes like this:
The program is set up, and, after the work is done, it is shut down.
If an error occurs during these two periods, it just exits without attempting
to clean things up.

When an error occurs after the setup, the shutdown routine is called and the
program then exits.

@strangequark:
Well, if i can't create a directory in /tmp and start a process, it is the users
fault. And the consequences are not that bad-some files in /tmp.

@rocx:
When it fails, it shouts.
I don't like programs that try to repair themselves.
»Windows is searching for a solution, please wait…«
I understand this rule as »if there is a 100% chance the problem can be solved elegantly
by itself, do it, otherwise shout while dying.«
This is what I tried to do.

This keeps the code fairly simple, but does things right.