> it can be a different thread than the one that was running when we got the signal.

Er, I hadn't yet had my first cup of coffee today..

Nevertheless, if the signal handler runs in a random thread and you
raise an exception there, wouldn't that be a problem - shouldn't your
exception should go predictably into a particular thread such as the
primordial thread? Is there some trick to set up the call stack in a
special way when you are in a signal handler (is there a special signal
stack altogether)?

Just reading through SRFI 18: Multithreading support. Marc has done an
impressive job. Is there any SRFI specifying Unix signal behavior?