Why limit yourself to three worlds? Perhaps warnings should be considered to be like some kind of row in a relational table, and you can apply any selection criterion on any attribute of the data. Or consider a warning to be an object that decides whether or not to say anything based on the selection criteria it can see in its various outward lexical scopes. The actual selection criteria do not need to be physically replicated as long as you can follow the chain of criteria outward somehow. Bitmasks are merely a useful optimization for certain selection criteria.
in reply to Re: future warnings
in thread future warnings