Yes, you are correct and those indexes will work. It's a trade of, composite
indexes 'pre-intersect' (at write time) properties while zigzag merge join
'post-intersects' properties (at read time). I left the ancestor in because
it is probably very 'selective' which has the potential to greatly
:-)
performance is data dependent.
Here is convoluted explanation of performance:
Sx = set of entities where list = :x
smallest_set = min(S1.size(), S2.size(), ...)
It works best when the intersection(S1, S2, S3,...) is large compared to the
smallest_set.
The pathological case is
