The "by-element" assignment with index expressions (index vectors) looks as if an implicit loop is executed over the index values. In fact it seems that the evaluation is en bloc, i.e. first, the complete rhs is evaluated and then assigned to the range of elements adressed by the index expression on the lhs.
For the user, there is no obvious reason why this equation should fail. The above assumption on how the assignment is executed may be an explanation. I guess that this procedure has performance benefits.
However, the proposal is that in case of failing parallel handling, serial handling should be used, using linear indexing over the lhs index objects. This would change the behaviour from "fail" to "slower" and would be less expensive to explain. Serial execution isn' t a warrant for success but it would cover iterations with backward references to the assignment target.