This change simplifies the CacheInterceptor way of dealing with cached
coroutines, thanks to the fact that lower level support for AOP has been
introduced in c8169e5c. This fix is similar to the one applied for
`@Transactional` in gh-33095.
Closes gh-33210
Previous to this change, the transactional aspect would supersed the
user-defined AspectJ aspect, shortcircuiting to calling the original
Kotlin suspending function.
This change simplifies the TransactionAspectSupport way of dealing with
transactional coroutines, thanks to the fact that lower level support
for AOP has been introduced in c8169e5c.
Closes gh-33095
To improve consistency and avoid confusion regarding primitive types
and their wrapper types, this commit ensures that we always use class
literals for primitive types.
For example, instead of using the `Void.TYPE` constant, we now
consistently use `void.class`.
This commit adds support for Kotlin Coroutines to Spring AOP
by leveraging CoroutinesUtils#invokeSuspendingFunction in
AopUtils#invokeJoinpointUsingReflection to convert it to the
equivalent Publisher return value, like in other parts of Spring
Framework.
That allows method interceptors with Reactive support to process
related return values.
CglibAopProxy#processReturnType and JdkDynamicAopProxy#invoke
take care of the conversion from the Publisher return value
to Kotlin Coroutines.
Reactive transactional and HTTP service interface support
have been refined to leverage those new generic capabilities.
Closes gh-22462
This commit refactors some AssertJ assertions into more idiomatic and
readable ones. Using the dedicated assertion instead of a generic one
will produce more meaningful error messages.
For instance, consider collection size:
```
// expected: 5 but was: 2
assertThat(collection.size()).equals(5);
// Expected size: 5 but was: 2 in: [1, 2]
assertThat(collection).hasSize(5);
```
Closes gh-30104
Prior to this commit, the AOT infrastructure would rely on the fact that
native runtime reflection on a type would only consider
methods/fields/constructors that had specific hints contributed. When
listing them through the reflection API on the type, the native image
would only return those for which we had hints contributed.
This behavior will soon change in GraalVM and will better align with the
JVM behavior: when asking for all declared methods on a type in a native
image, we should get all existing methods, not just the ones registered
previously in the native image.
This commit aligns the behavior of the `RuntimeHintsAgent` and removes
the now misleading predicates as a consequence.
Closes gh-29205
Prior to this commit, the `RuntimeHintsAgent` and its testing
infrastructure would assume that calling `MyClass.class.getMethods()`
requires a reflection hint on the class for introspecting public/private
methods.
GraalVM does not require this, in fact this call only returns methods
that have reflection hints in the native image.
This commit refines the agent behavior for `Class.getMethods()`,
`Class.getDeclaredMethods()`, `Class.getFields()` and
`Class.getDeclaredFields()`. With this change, registering at least one
method/field for reflection is enough to match.
During the execution of Java tests, all methods and fields will be
provided, regardless of hints being registered or not. This could cause
false negatives where we're missing reflection hints on methods or
fields.
This risk is mitigated thanks to additional instrumentation on
`Method.getAnnotations()`, `Method.getParameterTypes()` and
`Method.invoke()`. If a method is found reflectively, chances are it
will be used for further reflection.
Closes gh-29091
Prior to this commit, the `RuntimeHintsAgent` would instrument
`Method.invoke` in a way that fails when invoked methods are not public.
This commit ensures that before delegating the invocation call, the
instrumentation makes the method accessible before delegating the call.
Fixes gh-29046
In order to write proper integration tests for the `RuntimeHintsAgent`,
we need to load the java agent on the JVM and instrument test code to
check that invocations are properly recorded.
This commit adds the relevant build configuration to the
integration-tests module and adds reflection tests for the agent.
Closes gh-27981
Prior to this commit, some tests would belong to the PERFORMANCE
`TestGroup`, while they were not testing for performance but rather
performing functional tests that involve long running operations or
timeouts.
This commit moves those tests to the LONG_RUNNING `TestGroup`.
See gh-24830
The AspectJPrecedenceComparator was designed to mimic the precedence
order enforced by the AspectJ compiler with regard to multiple 'after'
methods defined within the same aspect whose pointcuts match the same
joinpoint. Specifically, if an aspect declares multiple @After,
@AfterReturning, or @AfterThrowing advice methods whose pointcuts match
the same joinpoint, such 'after' advice methods should be invoked in
the reverse order in which they are declared in the source code.
When the AspectJPrecedenceComparator was introduced in Spring Framework
2.0, it achieved its goal of mimicking the AspectJ compiler since the
JDK at that time (i.e., Java 5) ensured that an invocation of
Class#geDeclaredMethods() returned an array of methods that matched the
order of declaration in the source code. However, Java 7 removed this
guarantee. Consequently, in Java 7 or higher,
AspectJPrecedenceComparator no longer works as it is documented or as
it was designed when sorting advice methods in a single @Aspect class.
Note, however, that AspectJPrecedenceComparator continues to work as
documented and designed when sorting advice configured via the
<aop:aspect> XML namespace element.
PR gh-24673 highlights a use case where AspectJPrecedenceComparator
fails to assign the highest precedence to an @After advice method
declared last in the source code. Note that an @After advice method
with a precedence higher than @AfterReturning and @AfterThrowing advice
methods in the same aspect will effectively be invoked last due to the
try-finally implementation in AspectJAfterAdvice.invoke() which invokes
proceed() in the try-block and invokeAdviceMethod() in the
finally-block.
Since Spring cannot reliably determine the source code declaration
order of annotated advice methods without using ASM to analyze the byte
code, this commit introduces reliable invocation order for advice
methods declared within a single @Aspect. Specifically, the
getAdvisors(...) method in ReflectiveAspectJAdvisorFactory now hard
codes the declarationOrderInAspect to `0` instead of using the index of
the current advice method. This is necessary since the index no longer
has any correlation to the method declaration order in the source code.
The result is that all advice methods discovered via reflection will
now be sorted only according to the precedence rules defined in the
ReflectiveAspectJAdvisorFactory.METHOD_COMPARATOR. Specifically, advice
methods within a single @Aspect will be sorted in the following order
(with @After advice methods effectively invoked after @AfterReturning
and @AfterThrowing advice methods): @Around, @Before, @After,
@AfterReturning, @AfterThrowing.
The modified assertions in AspectJAutoProxyAdviceOrderIntegrationTests
demonstrate the concrete effects of this change.
Closes gh-25186
Prior to this commit we did not have tests in place to verify the status
quo for the invocation order of all advice types when declared within
a single aspect, either via the <aop:aspect> XML namespace or AspectJ
auto-proxy support.
This commit introduces such tests that demonstrate where such ordering
is broken or suboptimal.
The only test for which the advice invocation order is correct or at
least as expected is the afterAdviceTypes() test method in
ReflectiveAspectJAdvisorFactoryTests, where an AOP proxy is hand crafted
using ReflectiveAspectJAdvisorFactory without the use of Spring's
AspectJPrecedenceComparator.
See gh-25186
Previously, when a project's jar was an input into a test task, a
cache hit required the current build to be using the same JDK as the
one that created the cache entry. This was due to the Created-By
entry in the jar's manifest which will vary if JDKs with different
values for the java.version and java.specification.vendor version are
used.
This commit configures normalization of the runtime classpath to ignore
META-INF/MANIFEST.MF, thereby allowing a cache hit when the tests were
previously run on a different JDK than the one being used now. Typically
this is a different update release being used on a CI agent and a
developer's machine. This change will therefore improve the likelihood
of a cache hit once remote caching has been enabled.
Closes gh-23872
Prior to this commit, a lot of work had been done to prevent improper
use of testing Framework APIs throughout the codebase; however, there
were still some loopholes.
This commit addresses these loopholes by introducing additional
Checkstyle rules (and modifying existing rules) to prevent improper use
of testing framework APIs in production code as well as in test code.
- Checkstyle rules for banned imports have been refactored into
multiple rules specific to JUnit 3, JUnit 4, JUnit Jupiter, and
TestNG.
- Accidental usage of org.junit.Assume has been switched to
org.junit.jupiter.api.Assumptions.
- All test classes now reside under org.springframework packages.
- All test classes (including abstract test classes) now conform to the
`*Tests` naming convention.
- As an added bonus, tests in the renamed
ScenariosForSpringSecurityExpressionTests are now included in the
build.
- Dead JUnit 4 parameterized code has been removed from
DefaultServerWebExchangeCheckNotModifiedTests.
Closes gh-22962
Johnny Lim
Simon Baslé
Juergen Hoeller
Sam Brannen
Stéphane Nicoll
Yanming Zhou
Sam Brannen
Sébastien Deleuze
Juergen Hoeller
Krzysztof Krasoń
Sam Brannen
Brian Clozel
Phillip Webb
Сергей Цыпанов
Qimiao Chen
Sam Brannen
Andy Wilkinson