(j3.2006) Tom explained the flaw in my induction

[Van Snyder]

Thanks to Tom for explaining the flaw in my induction.  A case I had
overlooked does allow to invoke a deferred binding, without knowing it's
a deferred binding, if an object of abstract type is allowed to invoke
nondeferred bindings.

The people who want to access an abstract parent component would like to
specify the parent component instead of all of its components in a
structure constructor.  In order to allow that, either a tiny hole in
C911 would be necessary, or if C911 were moved to C1535a, (at least) two
constraints would be needed to prohibit a nonpolymorphic abstract object
from being an actual argument associated with a polymorphic dummy
argument or a pointer target in a polymorphic pointer assignment.

I told them I couldn't think of another place where this would be
useful, or even possible, and they couldn't either.  Maybe it's not
worth the effort.