Enhanced sort join produces duplicate results

In a plan where an enhanced sort join chooses to create an index on a child that should be SORT_DISTINCT, the distinct processing does not occur.

For example this would occur with a query such as:

select a.e1, b.e2 from pm1.g1 as a, (select e1, e2 from pm2.g2 union select e1, e2 from pm2.g2) as b where a.e1 = b.e1

where the source does not support union.