Hi Corentin, Interesting. Have you thought about following the example of XMonad instead? The analogy could goes as follows: XMonad's configuration file (~/.xmonad/xmonad.hs) <=> Your rules. XMonad's state <=> Your state. Editing the config file <=> Changing the rules. Of course you normally edit the config file while XMonad is running, and in theory XMonad could do funny things to your Emacs, i.e. require you to vote before letting any edit-command through. When you change your configuration file, you send a specific key-sequence to XMonad (Alt-q is the default), and XMonad serializes it's state into a string, compiles the new config file, and then replaces itself with the new instance, handing over it's state. No mutable-state-in-IO trickery necessary. Matthias.