Index creation now double checks the MappingContext a MappingContextEvent originates from before actually creating indexes. This avoids invalid indexes being created in a multi-database scenario.
AbstractMappingConfiguration now uses the package of the class extending it to enable entity scanning for that package. To disable entity scanning entirely override the method to return null or an empty String.
Reading an empty collection always returned a HashSet assuming the returned value would be converted into the assigned properties value later on. However the method should rather return the correct type already which we do now by invoking the potential conversion.
$ne expressions are now only being tried to be converted into an ObjectId in case they follow an id property. Previously they tried in every case which might have led to Strings being converted into ObjectIds that accidentally were valid ObjectIds but didn't represent an id at all.
So far the algorithm to inspect whether an id property has to be set after a save(…) operation has used the plain BeanWrapper.getProperty(PersistentProperty property) method. This caused problems in case the getter of the id field returned something completely different (to be precise: a complex type not convertible out of the box).
We now inspect the id field only to retrieve the value.
When reading BasicDBLists we now make sure the resulting collection is converted into the actual target type eventually. It might be an array and thus need an additional round of massaging before being returned as value.
As we try to massage the value of the id property into an ObjectId if possible we need to do so as well when mapping the Querydsl query. Adapted SpringDataMongoDbSerializer accordingly.
MongoDbUtils now correctly returns DB instances for others than the first one bound. So far the lookup for an alternate database resulted in the first one bound to be returned. Polished log statements a bit.
ServerAddressPropertyEditor now only eventually fails if none of the configured addresses can be parsed correctly. Strengthened the parsing implementation to not fail for host-only parsing or accidental double commas.
Cleaned up integration tests for replica set configuration.
The type of the value returned for the total field of the timing map in map-reduce results has changed from Integer to Long as of MongoDB version 2.1.0 apparently. Changed MapReduceResults to accommodate either Integer or Long types.
So far URL instances were treated as entities and serialized as nested document. As there was no custom converter registered to re-instantiate the objects and URL does not contain a no-arg constructor, reading the instances back in resulted in an ugly exception in ReflectionEntityInstantiator. We now register a custom Converter to serialize URL instances as their plain toString() representation. This causes the reading working out of the box as the StringToObjectConverter registered by default uses the constructor taking a String on URL accidentally. To make sure this still works we added an explicit StringToURLConverter to implement symmetric conversions.
The raw result for a map-reduce operation might contain a complex element containing the output collection in case the original request configured an output database as option. Adapted the parsing of the output collection to accommodate both scenarios (plain String value as well as DBObject wrapper).
So far we have resolved DBRef values eagerly without inspecting the actual property type the would have to be assigned eventually. Now we simply skip the recursive resolve process if the property type (or component type, map value type) is DBRef actually.
The debug output now uses the already mapped query object when concatenating the log string. Improved applying the id after save operations by inspecting whether the object already has the id set before trying to set it. This could have caused problems in case you use a complex id and don't provide a custom converter as it can be serialized out of the box. Fixed minor glitch in MappingMongoConverter which was not really a bug as another path through the code has covered the scenario later on. Introduced SerializationUtils class that provides a method to safely serialize objects to pseudo JSON. Pseudo in the sense that it simply renders a complex object as { $java : object.toString() }. This is useful for debug output before the DBObject was mapped into Mongo-native types.
So far we've read empty collections and populated the property value of the Java object being created with Collections.emptySet(). This returns an unmodifiable Set so that further modifications fail with an UnsupportedOperationException. We now simply use new HashSet<Object>().
Upgraded to JUnit 4.10. Move to junit-dep dependency to allow cleaning up hamcrest dependencies. Use hamcrest-library ins tread of hamcrest-all. Upgrade to Mockito 1.9.0, use mockito-core instead of mockito-all. Upgrade to Spring Data Core 1.3.1.BUILD-SNAPSHOT. Upgraded to Querydsl 2.6.0. Upgraded compiler and Surefire plugin.
Expose the failOnWarnings property to be able to override it through a command line argument. Necessary as the Sonar build uses Clover which instruments source code and thus creates type dependencies, Bundlor complains about not being declared in template.mf.
Using List as return type for paginating methods didn't work for query methods currently. Fixed by inspecting the Pageable parameter potentially handed into them and restricting the result set accordingly.
