* Refinement for gateway Mono processing
* Introduce a `MessagingGatewaySupport.MonoReplyChannel` instead of
`FutureReplyChannel` for better on demand handling and reusing a
`Mono` returned from the target handler
* Refactor `GatewayProxyFactoryBean` to identify a target return type
(including generics for `Function`) during initialization
* Handle a `Mono` return type via
`MessagingGatewaySupport.sendAndReceiveMessageReactive()`
* Some `@Nullable` in the `GatewayProxyFactoryBean` and `ExpressionUtils`
* Add `MonoFunction` test-case into the `FunctionsTests.kt`
* * Deprecate `GatewayProxyFactoryBean.setServiceInterface()` in favor of
ctor initialization
* Fix `GatewayProxyFactoryBean.setServiceInterface()` usage in tests
The method reference together with a `IntegrationFlows.frm(Supplier<?>)`
do the trick to avoid a method name resolution and also gives us a code
navigation in the IDE
Fixes https://github.com/spring-projects/spring-integration/issues/3061
Without a `@DirtiesContext` we keep not only a ctx in the cache, but
also an embedded DB instance, which is shared between contexts.
When ctx cache clean up happens, several ctxes would like to destroy
their associated embedded DB which, essentially, is the same in-memory
instance
* The `@DirtiesContext` give us a chance to destroy embedded DB for
the particular ctx in isolation
* Improve `channel.adoc`: polishing, new channels, java config
* Make `LoadBalancingStrategy` in the `DirectChannel` ctor as
`@Nullable` to reflect the real logic behind
* More Dos and polishing according PR comments
* Mention WebFlux and RSocket endpoints in the `endpoint-summary.adoc`
table
* More reactive streams docs
* `SourcePollingChannelAdapter` polishing
* Wrap multi-value publisher into `Mono.just()` in the `AbstractMessageProducingHandler`
instead of emitting just only a first item for the standard reply
Doc polishing
* Hide `toReactivePublisher()` from lambdas
When an `IntegrationFlow` is declared via lambda, it is impossible
to materialize it as a `Publisher` since we lose a Lambda context and
finish with a `NoSuchBeanException` when we would like to inject that
`Publisher`
* Make `IntegrationFlowDefinition.toReactivePublisher()` as `protected`
and expose it on the `IntegrationFlowBuilder` level
* Use `Channels` as a singleton: no reason in overhead for its instances
* Some code style polishing for `IntegrationFlowDefinition`
* Restore `Channels.super()` call
* Fix some typos in `webflux.adoc` and `dsl.adoc`
* Make `FunctionsTests.kt` more Kotlin friendly
* Improve RSocket components and tests for them,
especially rely on the default `RSocketStrategies` in the
target SF RSocket components
* Rework `ServerRSocketConnector.setClientRSocketKeyStrategy` to use the whole `MessageHeaders` for consultation.
It turns out that just destination is not enough since it can be used from different clients
* Make the key based on the provided `setupData` by default
* Include all the `MessageHeaders` into the `RSocketConnectedEvent`
* Improve Docs according review feedback and Docs in SF
Doc polishing
Fixes https://github.com/spring-projects/spring-integration/issues/3052
Starting with version `5.1`, a `LambdaMessageProcessor` is based on the
`MessageConverter` instead of plain `ConversionService`.
(A `ConfigurableCompositeMessageConverter` is used from the application
context.)
However a type conversion based on the `@IntegrationConverter` is lost
in the `GenericMessageConverter` because it doesn't use an
`integrationConversionService` from the application context
* Change `ConfigurableCompositeMessageConverter` to configure a
`GenericMessageConverter` with an `integrationConversionService`
if any
* Fix `MessagingMethodInvokerHelper` to populate a `BeanFactory` into
its internal `ConfigurableCompositeMessageConverter`
* Ensure in the `LambdaMessageProcessorTests` that
`@IntegrationConverter` is applied for ``LambdaMessageProcessor` as well
**Cherry-pick to 5.1.x**
Related to: https://build.spring.io/browse/INT-MASTERSPRING40-826/
If we don't close application context on the embedded (S)FTP server,
we my still have some session opened in other tests, which would cause
some unexpected behavior like extra `SessionClosedEvent` in the
test for FTP events
Fixes https://github.com/spring-projects/spring-integration/issues/3047
* Improve `GatewayProxyFactoryBean` to determine the return type of the
method call from the interface generic types, when the `serviceInterface`
is a `java.util.function.Function`
* Propagate `MethodArgsHolder` as a `rootObject` for SpEL evaluations
* Deprecate `#gatewayMethod` and `#args` evaluation context variables
in favor of `MethodArgsHolder` as root object.
They will be removed in the future release and a single
`EvaluationContext` will be used for all the gateway expressions
* Introduce an
`IntegrationFlows.from(Class<?> serviceInterface, Consumer<GatewayProxySpec> endpointConfigurer)`
to allow to configure any valid gateway proxy options similar to what
we have with the `<gateway>` and `@MessagingGateway`.
This way we are very close to consistency between different approaches
* * Remove `default` prefix from `GatewayProxySpec` options
* Document the change
* Use convenient `Consumer` API to avoid external iteration for
`setupMetadata` in the `ClientRSocketConnector`
* Add `routeVars` and `metadata` support into the `RSocketOutboundGateway`
* Cover new options in the XML and Java DSL configurations
Related to https://github.com/spring-projects/spring-framework/issues/23137
The metadata in Spring Messaging for RSockets now supports any
arbitrary objects for setup payload, including composition.
* Switch the `ClientRSocketConnector` to fully delegate to the
`RSocketRequester.Builder` inheriting possible metadata encoding/decoding
in the target `RSocketRequester` implementation
* Turn off a default `dataMimeType` from the `MimeTypeUtils.TEXT_PLAIN`
to the `null` by default relying on the encoder/decoder logic in the
target RSocket wrappers
* Expose more delegating options in the `ClientRSocketConnector`, like
`setupRouteVars`, `setupMetadata`
Fixes https://github.com/spring-projects/spring-integration/issues/3041
Some Redis clients/servers don't allow to perform an `INFO` command,
therefore we are not able to determine if we can perform `UNLINK` or
not.
* Deprecate `RedisUtils.isUnlinkAvailable()` as not reliable source of
through; use trial with fallback algorithm in the target logic around
`UNLINK` command calls.
**Cherry-pick to 5.1.x**
