Take a slice of the metadata if not using composite metadata to allow
reading it multiple times. For composite metadata this is not an isuse
as it is ensured by the underlying RSocket Java API.
TransactionOperator.as(Mono) now no longer short-cuts via a Flux.next() but provides an implementation via Mono.usingWhen(…).
The short-cut previously issued a cancellation signal to the transactional Mono causing the transaction cleanup to happen without a handle for synchronization.
Using Mono.usingWhen(…) initiates transaction cleanup when the Mono completes eliminating the need for cancellation of the transactional Publisher.
This change does not fully fix gh-23304 but it softens its impact because TransactionalOperator.transactional(Mono) avoids cancellation.
This commit reverts b2704e1db6 and
configures the ReactorResourceFactory not to use global resources,
thereby allowing all tests in WebClientDataBufferAllocatingTests to
execute within approximately 2 seconds again on Mac OS.
Prior to this commit, the name of the Thread registered via
ConfigurableApplicationContext#registerShutdownHook() was the generic,
default thread name ("Thread-#"). That made it difficult to discern
which executing thread was the Spring ApplicationContext shutdown hook.
This commit improves diagnostics by setting the thread name of the
ApplicationContext shutdown hook to "SpringContextShutdownHook".
Closes gh-23670
When index == arrayLength, the array index is also out of bounds.
For this scenario, a SpelEvaluationException should be thrown instead
of ArrayIndexOutOfBoundsException.
Closes gh-23658
While resolving the regression raised in gh-23571, it came to our
attention that not all of our ClassFilter and MethodMatcher
implementations were properly cacheable with CGLIB generated proxies
due to missing (or improper) equals() and hashCode() implementations.
Although such deficiencies may not manifest themselves as bugs in Core
Spring's default arrangements, these might cause issues in custom
arrangements in user applications.
This commit addresses this by ensuring that ClassFilter and
MethodMatcher implementations properly implement equals() and
hashCode(). In addition, missing toString() implementations have been
added to improve diagnostics for logging and debugging.
Closes gh-23659
This commit renames the `metadataExtractors` method to something that is
closer to the actual intent: addming new metadata extractors against the
registry given as a parameter of the `Consumer`.
The method is renamed to `metadataExtractorRegistry`.
This commit makes sure we narrow down dependencies of the Spring
Framework to only Maven central. Optional dependencies that are not
available on Maven Central are now served from a more specific
repository (`libs-spring-framework-build`).
Closes gh-23124
Prior to this commit, Spring Framework would use `Schedulers.elastic()`
in places where we needed to process blocking tasks in a reactive
environment.
With reactor/reactor-core#1804, a new `Schedulers.boundedElastic()`
scheduler is available and achieves the same goal with added security;
it guarantees that resources are bounded.
This commit uses that new scheduler in the standard websocket client,
since the underlying API is blocking for the connection phase and we
need to schedule that off a web server thread.
Closes gh-23661
See gh-23665
This commit makes it possible to send requests without
requiring to call data(Mono.empty()). It introduces a
dedicated MetadataSpec interface and merge ResponseSpec
into RequestSpec for more flexibility.
Closes gh-23649
It seems that publishing a Gradle platform as a Maven POM writes all
dependencies with their scope information by default.
We don't want that when publishing the Spring Framework BOM, as it
forces the scope on projects depending on the BOM, unless they've
specified the scope when they added the dependency to their build.
Typically, developers could get spring-test as a compile dependency
without this change.
This commit removes the scope information from the published BOM.
Fixes gh-23660
Spring Framework 5.2 M1 introduced a memory leak for applications using
@Async methods. Specifically, in a large test suite with multiple
ApplicationContexts that were closed (e.g., via @DirtiesContext,
@MockBean, or context cache eviction), the JVM process could run out of
memory.
Underlying cause: Due to a missing equals() implementation in Spring's
new AnnotationCandidateClassFilter, CGLIB's static cache of generated
classes indirectly retained references to BeanFactory instances for the
closed ApplicationContexts for the duration of the test suite.
This commit fixes this regression by introducing a proper equals()
implementation in AnnotationCandidateClassFilter. This commit also
introduces corresponding hashCode() and toString() implementations.
Closes gh-23571
Prior to this commit, concurrent invocations of
DefaultListableBeanFactory.removeBeanDefinition() could result in a
NullPointerException.
This commit fixes this by adding an appropriate not-null check in
resetBeanDefinition().
Closes gh-23542
Prior to this commit, customizing the extraction of RSocket metadata
from frames would require developers to override the default
`MetadataExtractor` while configuring `RSocketStrategies`.
This touches on many infrastructure parts, whereas the goal is just to
configure an extra metadata entry extractor using already configured
codecs.
This commit adds a way to register metadata entry extractors on the
`RSocketStrategies` builder with a `Consumer`-based API.
Closes gh-23645
This commit removes the duplicated CRON_DISABLED constant value from
@Scheuled and simply refers to the ScheduledTaskRegistrar.CRON_DISABLED
constant.
This avoids a potential package cycle by ensuring that the `annotation`
package depends on the `config` package but not the other way around.
See gh-23568
This commit makes sure that DefaultResourceLoader consistently use
getProtocolResolvers() to access additional protocol resolvers. This
allows subclasses to define how the list is provided.
Closes gh-23564
Previously, tests in ConcurrentWebSocketSessionDecoratorTests that use
the BlockingSession would fail intermittently. This appears to have
been due to a race condition in sendBlockingMessage where the call
to getSentMessageLatch() that stores a latch in nextMessageLatch on
the main thread may happen after the call to sendMessage that counts
down the latch if it is non-null occurs on the executor's thread.
This commit updates sendBlockingMessage to call getSentMessageLatch()
(and therefore store the latch) before it sumbmits the task to the
executor. This ensures that the latch will be available when the
exeuctor's thread attempts to retrieve and decrement it.
BlockingSession's AtomicReference fields have also been made final to
eliminate the possibility of any visibility problems across threads.
Closes gh-23642
Arjen Poutsma
Rossen Stoyanchev
Sam Brannen
Juergen Hoeller
Mark Paluch
i321222
Brian Clozel
Stephane Nicoll
Sebastien Deleuze
Andy Wilkinson
Kwon Young Jae
Phillip Webb