On Wed, Nov 10, 2010 at 3:15 PM, Roger Hui <[email protected]> wrote: > The change you propose is non-trivial and has consequences > that may surprise you (and many others, I expect). > > Suppose f is <"_1, and you make the "combining rank" of > f be _1 instead of the current _ . > > So the ranks of ]...@f would be _1 (_1 _1 _1). Suppose the > argument x is i.2 3 4. Since the ranks of ]...@f are _1, > x is split into the cells i.3 4 and 12+i.3 4 and you apply > ]...@f to each cell. [...]
You're right in that combining ranks shouldn't have negative integers, because that too would make (f"f) mean something different from f, such as in the case when (f=:<"_1). I accept that that change would be a bad idea, so I retract that part of my suggestion. On the other hand, are there many surprises or incompatibilities even if you take only the weaker variant of my proposal? That says that if rl is a negative integer and rr is a nonnegative integer, the diadic combining rank of f"(rl,rr) should be (_ _) instead of (_,rr). For example, the combining rank of (]"_ _3 7) would be (_ _ _). The case with left and right swapped would be handled similarly. The combining ranks would still always be nonnegative. As I said, this too would be better than the current state because it would help uphold the f -: f"f identity, which the current semantics does not do for such verbs. Ambrus ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
