We now use the native id within ObjectPath for checking if a DBref has already been resolved. This is required as MongoDB Java driver 3 generation changed ObjectId.equals(…) which now performs a type check.
Original pull request: #334.
Related pull request: #288.
The presence of ObjectToObjectConverter in a DefaultConversionService causes the guard trying to register converters for ObjectIds in AbstractMongoConverter to not trigger the registration. This in turn caused ObjectId conversions to be executed via reflection instead of the straight forward method calls and thus a drop in performance for such operations.
We no unconditionally register the converters to make sure they really get applied.
Related tickets: SPR-13703.
We now also treat references as source of a potential index. This enforces index creation for Objects like:
@Document
class WithDbRef {
@Indexed
@DBRef
ReferencedObject reference;
}
Combining @TextIndexed or @GeoSpatialIndexed with a DBRef will lead to a MappingException.
Original pull request: #329.
We now allow registration of ConverterFactory within CustomConversions by inspecting the generic type arguments for determining the conversion source and target types.
Original pull request: #330.
Triggering data access exception translation could lead to NullPointerException in cases where PersistenceExceptionTranslator returned null because the original exception couldn't be translated and the result was directly used from a throw clause.
This is now fixed by consistently the potentiallyConvertRuntimeException(…) method, which was made static to be able to refer to it from nested static classes.
Refactored Scanner usage to actually close the Scanner instance to prevent a resource leak.
We now retain position parameters in paths used in queries when mapping the field name. This allows to map "list.1.name" to the name property of the first element in the list.
The change also fixes a glitch in mapping java.util.Map like structures having numeric keys.
Original pull request: #314.
We now consider the potential generic type argument of collection elements.
Prior to this change an index within List<GenericWrapper<ConcreteWithIndex>> would not have been resolved.
Original pull request: #312.
If a repository query method returned a primitive, that primitive was exposed as domain type which e.g. caused deleteBy…(…) methods to fail that returned a void.
We now shortcut the MongoEntityMetadata lookup in MongoQueryMethod to use the repository's domain type if a primitive or wrapper is returned.
We now push the aliased field name down the aggregation pipeline for projections including operations. This allows to reference them in a later stage. Prior to this change the field reference was potentially resolved to the target field of the operation which did not result in an error but lead to false results.
Original pull request: #311.
We now reject configuration using MongoClient along with UserCredentials in SimpleMongoDbFactory. This move favors the native authentication mechanism provided via MongoCredential.
<mongo:mongo-client id="mongo-client-with-credentials" credentials="jon:warg@snow?uri.authMechanism=PLAIN" />
Original pull request: #309.
We now allow grouping credentials by enclosing them in single quotes like this:
credentials='CN=myName,OU=myOrgUnit,O=myOrg,L=myLocality,ST=myState,C=myCountry?uri.authMechanism=MONGODB-X509'
We also changed the required argument checks to be more authentication mechanism specific which means the pattern is now username[:password@database][?options].
Original pull request: #310.
UpdateMapper.getMappedObjectForField(…) invokes the very same method of the super class but handed in an already mapped value so that value conversion was invoked twice.
This was especially problematic in cases a dedicated converter had been registered for an object that is already a Mongo-storable one (e.g. an enum-to-string converter and back) without indicating which of the tow converter is the reading or the writing one. This basically caused the source value converted back and forth during the update mapping creating the impression the value wasn't converted at all.
This is now fixed by removing the superfluous mapping.
Removed the creation of a BasicMongoPersistentEntity in favor of always handing ClassTypeInformation.OBJECT into the converter in case not entity can be found.
This makes sure type information is written for updates on properties of type Object (which essentially leads to no PersistentEntity being available).
Original pull request: #301.
We now use property type information when mapping fields affected by an update in case we do not have proper entity information within the context. This allows more precise type resolution required for determining the need to write type hints for a given property.
Original pull request: #301.
We now quote the original criteria before actually wrapping it inside of an regular expression for case insensitive search. This happens not only to case insensitive is, startsWith, endsWith criteria but also to those using like. In that case we quote the part between leading and trailing wildcard if required.
Original pull request: #301.
We now eagerly check DBRef properties for invalid definitions such as final class or array. In that case we throw a MappingException when verify is called.
Previously we tried to apply the ignore case settings found in the PartTree to the root PropertyPath we handle in MongoQueryCreator.create(). This is now changed to work on the leaf property of the PropertyPath.
We now inspect the actual field type during update mapping and provide a type hint accordingly. Simple, non interface and non abstract types will no longer be decorated with the _class attribute. We now honor positional parameters when trying to map paths to properties. This allows more decent type mapping since we have now access to the meta model which allows us to check if presence of type hint (aka _class) is required.
We now add a special type hint indicating nested types to the converter. This allows more fine grained removal of _class property without the need to break the contract of MongoWriter.convertToMongoType(…).
Original pull request: #292.
The problem stems from the inconsistent handling of type hints such as MongoTemplate.save(…) does not add a type hint, but findAndModify(…) does. The same values are then treated differently by MongoDB, depending on whether they have a type hint or not. To verify this behavior, you can manually add the (superfluous) type hint to the saved object - findAndModify will then work as expected.
Additional tests demonstrate that findAndModify(…) removes type hints from complex documents in collections that are either nested in another collection or in a document, or doesn't add them in the first place.
Original pull requests: #290, #291.
Related pull request: #292.
CLA: 119820150506013701 (Stefan Ganzer)
We now skip authentication via an explicit AuthDB when requesting a DB via a MongoClient instance.
Related ticket: DATACMNS-1218
Original pull request: #296.
We now use TypeInformation in IndexResolver to lookup the root PersistentEntity for resolving @Indexed properties to ensure that we retrieve the same PersistentEntity that was stored. Previously we used the Class to lookup up the PersistentEntity which yielded a partially processed result.
Original pull request: #293.
