My guess is that somebody assumed the same semantics for splice as substr implements, which can be used as an lvalue.

So the old behaviour is roughly equivalent to die "That sub should never be called, and is not tested";, while the new one modifies @_.