If a subquery is used to drive an index, this change avoids making an extra
copy of the result of that subquery. But, such situations are probably
sufficiently unusual that the added complexity of this enhancement is not
worth the performance gain. So I'm going to park this check-in on a branch.
If we later find a use case to justify it, we can merge it to trunk then.
This is the "further change" that was promised by the prior check-in comment.
FossilOrigin-Name:
4488cb883e155a7466946d25569ab2816c4a8de9d4c7d540f5053ea5334e7ae5