Steven D'Aprano wrote:
I don't think there is a suitable solution for all of these issues. That would mean that set.universal() can't be a real set object, it has to be some sort of not-quite-a-set object that doesn't provide the entire set interface.
That doesn't strike me as all that different from handling NaN and Inf values in floating point mathematics though. Would set.universal() need to be a special object that required special handling in some set algorithms? Yes, it would, but that doesn't make it an invalid idea. In fact, treating it like a NaN from the decimal module is probably the answer to most of those questions: if a valid answer isn't defined, raise an exception. The semantics would need to be spelled out clearly however, as would a position on whether or not to add support for complementary sets (i.e. sets that are defined as "set.universal() ^ items_not_in_set"). Also the "universal" or "complementary" character of the set would need to be embodied in mutable state on the set object so that in-place operations such as "s |= set.universal()" can work correctly. Cheers, Nick. P.S. I'm not necessarily +1, or even +0, on the idea until it is fleshed out further. I just wanted to point out that there is precedent for including the numerical equivalents of the universal set. -- Nick Coghlan | ncoghlan@gmail.com | Brisbane, Australia ---------------------------------------------------------------