The first proposed final draft and the draft of jsr13 from bug 4851776 include a requirement that
"If the result of an operation has the value zero, its scale is 0."
This is not consistent with the existing behavior of exact arithmetic methods, nor is it consistent with the behavior of other decimal arithmetic packages, including the forthcoming decimal section of IEEE 754R.
The working implemenation enforces the zero canonicallization in some situations but not others.
The zero canonicalization requirement should be removed and the usual scale rules apply to zeros too.