Let recursive mapping calls of collection elements use the value type instead of the collections component type. Refactored MappingMongoConverter to make collection handling more maintainable. Added code to always add custom type information if the actual value being stored differs from the declared one.
Moved some of the DBRef discovering code into implementations of MongoPersistentProperty. Renamed MongoPersistentProperty.getKey() to ….getFieldName().
Several changes to how objects are initialized inside the template:
1. In one is not specified, a MappingMongoConverter is created and set as the default.
2. A special ApplicationEventPublisher implementation is installed by default to handle creating indexes when the template isn't used inside a Spring application context.
3. If a Spring application context is available, it will be set as the template's application context and eventPublisher, with the index creator being registered as an event listener if one isn't already present.
The tests had to be changed in a couple places to accurately reflect how mapping contexts and converters are now handled.
When annotation a repository method with e.g. @Query("{ 'firstname' : ?0, 'lastname' : ?1 }") all placeholders get now replaced correctly. Added unit tests and fixed broken logging in StringBasedQueryCreator,
Removed Criteria.whereId(…). Updated SimpleMongoRepository to use the new method and use more core template methods to prevent objects from being marshalled to find out whether a particular object exists.
Introduced writeSimpleInternal(…) method that automatically stores the name of an enum instead of the enum itself. Changed quite a few places to rather use MongoPersistentProperty.getKey() over getName().
Reding values into a map assumed map keys to be String values. We know leverage the conversion service to convert the key to the type we discover from the generics property information.
User repository interfaces are not required to implement MongoRepository anymore. Added missing override for save(Iterable<? extends T> entities) to return a List<T>.
The document being persisted now gets a _class attribute to carry the actual type. That field key will be made configurable by a subsequent commit and its value should be interpreted as type hint to a client and might be interpreted using a type mapper at a later stage as well (see DATADOC-63). For now it carries the fully-qualified Java class name.
On reads MappingMongoConverter will consider this field when choosing a type to bind the data to if - and only if - the type stored in there is a subtype of the actually requested one. So if we have a document carrying Person type information and you query for Contact you would get back a Person object. If you query for any other type not extending Contact you would get this custom type instead.
Added unit tests and an integration tests covering the Contact/Person scenario.
Don't use ConversionService for simple type keys as we unregistered ObjectToStringConverter. This in turn causes Number, Boolean, Locale and the like not being convertible to String anymore as for those types only the Type -> String converter is registered but no corresponding converter back. Opened a ticket for this against Core Spring (SPR-8306).
Fixed Criteria being able to create correct query for ….is(null). MongoQueryCreator now uses ….ne(…) instead of ….not().is(…) as this doesn't generate a correct query (see DATADOC-129).
Oliver Gierke
Thomas Risberg
J. Brisbin
Mark Pollack