In 2.3, PersistentCollection::match() has been implemented by doing the query directly. But sometimes, the only meaningful information about the matched collection would be its length. In this case, it would be great to handle it in the same way than extra lazy collections are handled: the matched collection would be initialized lazily, and could do the count in an extra lazy way (if the original collection was extra lazy).
This would of course not change anything in the case where the original collection was already initialized.