So the first time you care is the first time f(x, y) returns nonzero. Now you can find out what kind of function f is by calling f(y, x). If f(y, x) returns zero, f is <. Otherwise, it's a 3-way comparison.
Guido> Right. There's no flaw in this logic, but I'd hate to have to Guido> explain it over and over... I don't want people to believe Guido> that Python can somehow magically sniff the difference between Guido> two functions; they might expect it in other contexts. I can understand your reluctance -- I was just pointing out that it's possible. However, I'm slightly dubious about the x.sort(lt=f) vs x.sort(cmp=f) technique because it doesn't generalize terribly well. If I want to write a function that takes a comparison function as an argument, and eventualy passes that function to sort, what do I do? Something like this? def myfun(foo, bar, lt=None, cmp=None): # ... x.sort(lt=lt, cmp=cmp) # ... and assume that sort will use None as its defaults also? Or must I write if lt==None: x.sort(cmp=cmp) else: x.sort(lt=lt) Either way it's inconvenient. So I wonder if it might be better, as a way of allowing sort to take two different types of comparison functions, to distinguish between them by making them different types. -- Andrew Koenig, ark@research.att.com, http://www.research.att.com/info/ark