If the type lookup from the store source returns a raw generic type (e.g. resolving the value of a generic property against a value of that generic type - i.e. not a more concrete type binding the generic information) in the context of a generic property, we previously did not apply the generics information of the contextual instance to that very raw type.
We now expose a TypeInformation.specialize(ClassTypeInformation) which applies the current generics context to the given raw type and basically creates a synthetic parameterized TypeInformation instance.
DefaultTypeMapper applies this specialization by default now with ClassTypeInformation simply returning the given type as is so that we don't create any resolution overhead in case no generics are involved in the first place.
The ProjectingMethodInterceptor now uses a default ConversionService instance to try to convert a potentially not matching result of the target invocation into the type the projection requires before the attempt to create nested projection.
The assignability check of method return types now explicitly detects wrapper types supported in QueryExceutionConverters and automatically unwraps those before the actual check.
We now correctly inspect the sort parameter values and also fall back to the default if a single, empty parameter value is given.
Related ticket: DATACMNS-408.
Related pull request: #138.
Linked back to the release train names from the section on dependency management with Boot for clarity. Added additional line break to avoid the first line in the repositories.adoc being interpreted as text.
If standard Querydsl StringPaths are used without any reference to a query class tree, we previously dropped the first segment of the path expression.
Moved to a completely different algorithm to determine the path by just using the Path's toString() representation and the root removed if necessary.
If a keyword was contained in a complex property path (like In in SomeInfo), the property path was invalidly cut of in the middle. We now simply use the length of the keyword to extract the actual property from the candidate. This should be faster anyway as it doesn't require another lookup of the pattern in the source candidate.
Related pull request: #136.
The JavaDoc of PersistentPropertyAccessor now explicitly hints to PersistentEntity.getPropertyAccessor(…) to make obvious how instances of it can be obtained.
PointFormatter error message now states it's expecting comma separated values. Previously it erroneously reported expecting a semicolon.
Original pull request: #134.
To prevent classes from being garbage collected due to Class<?> references within DefaultParameterNameDiscoverer we no keep all ParameterNameDiscoverer instances in a non-static field.
Refreshed the copy of AnnotationTransactionAttributeSource to pull in the JTA 2.1 support introduced in the Spring Framework class. Opened up a ticket [0] to improve Spring Framework to eventually be able to ditch the copies to prevent such scenarios in the future (the missing feature introduced that is).
We now handle primitive and non-primitive arrays in ProxyProjectionFactory. WE basically skip element projections for the former and correctly implement exactly that for the latter.
When default transactions get disabled, we now still configure a TransactionInterceptor to make sure explicitly defined transactions are still applied. Previously the interceptor was not registered at all if default transactions were configured to be turned off and thus even explicitly declared transaction configuration was not considered.
Related tickets: DATACMNS-656.
Extended Version value object to correctly dump a non-numeric last segment of a version string to parse. Incorrect source version strings now raise an IllegalArgumentException better explaining the cause.
Related tickets: DATACMNS-496.
ProxyProjectionFactory now implements ResourceLoaderAware to be able to use the class loader used by the framework. SpringDataWebConfiguration now forwards the Application context as ResourceLoader through ProxyingHandlerMethodArgumentResolver into the target factory.
We now implement EnvironmentAware and ResourceLoaderAware to respect the current environment setup when scanning for types. We use the ResourceLoader's ClassLoader if configured but indicate to fall back to the default one if none configured.
We now support collection executions in query derivation in combination with wrapper types like (Future, Optional) that wrap a collection, e.g. Future<List<User>>. Previously we used the "raw" return type which tricked our detection for Collection / Page / Slice executions.
Original pull request: #130.
We now leniently add the TransactionProxy marker interface to the repository proxy to opt-out of any further transaction handling potentially applied through @EnableTransactionManagement or its XML equivalent.
The interface will be introduced in Spring 4.1.7 / 4.2 RC2 so we're reflectively adding it if present. Might be something we can think of making less indirect for Fowler once 4.1.7 has been released.
Deprecate ObjectInstantiatorClassGenerator and provide a replacement
ClassGeneratingEntityInstantiator which doesn't retain a reference to
the context class loader.
Original pull request: #128.
On a last page the total handed into a PageImpl constructor might not necessarily fit the content as there's the change of an insertion or deletion between the calculated count and the retrieval of the content. We now leniently mitigate those differences if the page created is the last page of the result.
Related tickets: DATAJPA-728, DATACMNS-615.
We're now using ClassUtils.getDefaultClassloader() to make sure we adhere to any class loading tricks the framework might impose for domain type instantiation.
