Move method from AggregationCommandPreparer and AggregationResultPostProcessor to BatchAggregationLoader. Extract field names to constants. Tiny renames to variables. Add unit test for aggregation response without cursor use.
Original pull request: #521.
We now send aggregation commands along a cursor batch size for compatibility with MongoDB 3.6 that no longer supports aggregations without cursor. We consume the whole cursor before returning and converting results and omit the 16MB aggregation result limit. For MongoDB versions not supporting aggregation cursors we return results directly.
Original pull request: #521.
We now make sure that we convert empty sources to the corresponding target type. This prevents entity instantiation from failing due to incorrect argument types when invoking the constructor.
Original pull request: #520.
In BasicMongoPersistentEntity.verify() we now properly call the super method to make sure the comparators that honor the @Field's order value are initialized properly.
We now rely on PersistentProperty paths to detect cycles between types. Cycles are detected when building up the path object and traversing PersistentProperty stops after the cycle was hit for the second time to generated indexes for at least one hierarchy level.
Previously, we used String-based property dot paths and compared whether paths to a particular property was already found by a substring search which caused false positives if a property was reachable via multiple paths.
Original Pull Request: #500
We now allow to remove the type restriction inferred by the QBE mapping via an ignored path expression on the ExampleMatcher. This allows to create untyped QBE expressions returning all entities matching the query without limiting the result to types assignable to the probe itself.
Original pull request: #496.
On bulk resolution of a DBRef array we now map the resulting documents back to their ids to make sure that reoccurring identifiers are mapped to the corresponding documents.