Move reactive wrapper conversion to ReactiveWrapperConverters to minimize touching points with reactive APIs. ReactiveWrapperConverters is used only from reactive repository components and does not initialize with blocking API use. This removes the need of having Project Reactor on the class path for non-reactive use.
- Rename RxJava...Repository to RxJava1...Repository
- Use Completable and Observable instead of Single for results without values/optional values
- Remove reactive paging for now as it does not really fit reactive data streaming
- Expose ReactiveWrappers.isAvailable(ReactiveLibrary) method to query library availability
We now expose reactive interfaces to facilitate reactive repository support in store-specific modules. Spring Data modules are free to implement their reactive support using either RxJava 1 or Project Reactor (Reactive Streams). We expose a set of base interfaces:
* `ReactiveCrudRepository`
* `ReactiveSortingRepository`
* `RxJava1CrudRepository`
* `RxJava1SortingRepository`
Reactive repositories provide a similar feature coverage to blocking repositories. Reactive paging support is limited to a `Mono<Page>`/`Single<Page>`. Data is fetched in a deferred way to provide a paging experience similar to blocking paging.
A store module can choose either Project Reactor or RxJava 1 to implement reactive repository support. Project Reactor and RxJava types are converted in both directions allowing repositories to be composed of Project Reactor and RxJava 1 query methods. Reactive wrapper type conversion handles wrapper type conversion at repository level. Query/implementation method selection uses multi-pass candidate selection to invoke the most appropriate method (exact arguments, convertible wrappers, assignable arguments).
We also provide ReactiveWrappers to expose metadata about reactive types and their value multiplicity.
PageableHandlerMethodArgumentResolver.isFallbackPageable() now correctly guards against the fallback Pageable being null, a situation that's explicitly deemed valid in setFallbackPageable(…).
This commit changes some ArrayList initializations to use an initial capacity and avoids the unnecessary fetch of the return type of a method in Parameter.isDynamicProjectionParameter().
Original pull request: #180.
The special case of an Iterable parameter doesn't have to be handled explicitly anymore so that we can simplify the type match check in DefaultRepositoryInformation.
Related ticket: DATACMNS-912.
Previously, ClassGeneratingPropertyAccessorFactory uses boxed types for primitive type setters. This threw NoSuchMethodExceptions and caused ExceptionInInitializerError as the generated class could not be initialized. We now use the appropriate argument type to look up setters.
Also, we now only generate accessor usage if the property is configured to use property access in the first place.
Original pull request: #178.
QueryExecutionResultHandler now explicitly handles null values for query methods returning Maps by returning an empty Map. This can be the case if a query is supposed to produce a Map (a single result) but doesn't actually yield any results at all.
We now correctly check the bounds of generic methods if we deal with a type variable found as method parameter. Introduced different code paths for type and method level generics so that the latter is explicitly checking all bounds available.
The proxies created for projections now defensively mask getDecoratedClass(), a method declared on DecoratingProxy, an interface that JDK proxies on Spring 4.3 will implement.
The introduced method allows creating new Order instances for a different property ensuring all other aspects (ignore case, null handling) are retained.
Introduced dedicated RevisionSort property that allows to define the sort order independent of the revision property to be used.
Introduced methods to find out whether a Sort is descending or ascending to avoid having to refer to the enum.
We now support exists projections in derived queries by parsing the exists keyword and expose it via PartTree.isExistsProjection().
Original pull request: #171.
Previously, calls to ParameterizedTypeInformation.getMapValueType() caused an infinite recursion when invoked on a type information that was not a map type. This lead to a StackOverflowError.
We now call the super method doGetMapValueType() instead of calling the entry super method getMapValueType() and return by that null in case the map value type is not resolvable.
Original pull request: #176.
We now provide equals(…) and hashCode() methods for GenericPropertyMatcher, PropertySpecifier and PropertySpecifiers so that they can be compared and used in sets.
Original pull request: #175.
If the same generic type was used on recursively nested generics declarations, like this:
class GenericType<T> {}
class Nested extends GenericType<String> {}
class Concrete extends GenericType<Nested> {}
our traversal of the generics discovered T bound to String in Nested and as that is extending GenericType as well, the T detected here is the same instance as the T within the map discovered for Concrete. As we previously added the nested types after we added the original entry, the nested discovery overwrote the more local one.
We now make sure the detected nested type variable mappings don't overwrite already existing ones.
Changed the implementation of both the JSON Path and XPath based projecting HttpMessageConverters to make sure the cached decisions don't depend on the media type. Reuse the evaluation of the media type in AbstractHttpMessageConverter.
Previously TransactionalRepositoryFactoryBeanSupport.setBeanFactory(…) didn't invoke the method of the super class preventing the BeanFactory to be propagated to the repository factory resulting in the ProjectionFactory implementation created being unaware of the BeanFactory and thus no bean references being available in projection interfaces.
When a recursive decent operator is used in a JSON Path expression there's no way to get out of the array mode again to indicate one is interested in a particular element (e.g. "the first one no matter where in the document") [0]. We now work around this by always letting the parser return lists, so that the mapping library finally kicking in can be equipped with the correct target type to create.
Unfortunately this causes arrays to be double-wrapped for definite paths so that we basically have to adapt the type handed to Jackson in another round to the unwrap the mapping result in turn [1].
[0] https://github.com/jayway/JsonPath/issues/248
[1] https://github.com/jayway/JsonPath/issues/249
Projection types annotated with @ProjectedPayload can now be used as parameters for @RequestBody annotated Spring MVC controller method parameters.
Accessor methods will be translated into JSON path property expressions which can be customized by using the @JsonPath annotation. The methods are allowed to return simple types, nested projection interfaces or complex classes which will will be mapped using a Jackson ObjectMapper.
PageableExecutionSupport assumes that data queries are cheaper than count queries so we can apply some optimizations. We determine the total from the pageable and the results in which we don't hit the page size bounds (i.e. results are less than a full page without offset or results are greater 0 and less than a full page with offset). In all other cases we issue an additional count query.
Improved the dot-path translation of existing Querydsl Path instances by not relying on the toString() representation of the metadata but manually traversing the elements up to the root. Extracted the translation method into QueryDslUtils.
Furthermore, we now automatically insert an CollectionPathBase.any() step for Path instances of that type when reifying the paths from request attributes.
Introduced dedicated factory methods for ExampleMatcher so that it exposes whether the predicates built from Example instances have to be fulfilled all or if it's sufficient that any of them matches.
Turned the previously anonymous inner class that applies result processing to a JDK 8 Stream into a dedicated inner class to avoid JDK types being loaded on older JDKs. The previous way was causing issues due to the JVM trying to resolve the Function interface when loading the ResultProcessor class.
Original pull request: #163.
We now explicitly report the given object's type and the one expected in case of a mismatch when BasicPersistentEntity.getPropertyAccessor(…) is invoked.
We now also cache a failed property lookup so that we can return null on a subsequent call for the same invalid property name right away and don't have to reattempt the lookup unnecessarily.
The lookup for already available PersistentEntity instances now acquires a read lock to shield the execution from partially equipped PersistentEntity instances currently in creation from another thread.
Original pull request: #157.
ToEntityConverter.matches(…) now doesn't throw an exception anymore as it might be picked up by classpath scanning an the exception previously thrown causing the entire conversion attempt to abort.
We're now rather inspecting the ToEntityConverter in case the target type is a repository managed one and the ToIdConverter otherwise.
ResourceProcessor supports Streams now and adds a map(…) step to apply projections.
Also, the ProjectingConverter falls back to use a default ConversionService instance to make use of the object mapping functionality implemented in ObjectToObjectConverter, i.e. the ability to just map an object returned from the store to a DTO in case that latter exposes a constructor or factory methods taking the source instance as parameter.
Related tickets: DATAJPA-903.
Simplified conditional expressions in DefaultRepositoryInformation. Reordered test method to reflect natural order of adding.
Original pull request: #162.
Add additional check for detecting customized method where method contains generics. Overriding methods without generics already works, this patch simply makes the behavior consistent.
Original pull request: #162.
We're now correctly stating that we're preferring the delete method that takes the entity as argument over the one taking the id to match what's actually implemented.
The translation of the list of actual type parameters into an array now uses the loop index correctly. Previously, it always used the first type parameter.
