A bug has existed in Spring's MergedAnnotations support since it was
introduced in Spring Framework 5.2. Specifically, if the
MergedAnnotations API is used to search for annotations with "standard
repeatable annotation" support enabled (which is the default), it's
possible to search for a repeatable annotation but not for the
repeatable annotation's container annotation.
The reason is that MergedAnnotationFinder.process(Object, int, Object,
Annotation) does not process the container annotation and instead only
processes the "contained" annotations, which prevents a container
annotation from being included in search results.
In #29685, we fixed a bug that prevented the MergedAnnotations support
from recognizing an annotation as a container if the container
annotation declares attributes other than the required `value`
attribute. As a consequence of that bug fix, since Spring Framework
5.3.25, the MergedAnnotations infrastructure considers such an
annotation a container, and due to the aforementioned bug the container
is no longer processed, which results in a regression in behavior for
annotation searches for such a container annotation.
This commit addresses the original bug as well as the regression by
processing container annotations in addition to the contained
repeatable annotations.
See gh-29685
Closes gh-32731
Prior to this commit, MethodIntrospector failed to properly detect
bridge methods for subsequent invocations of selectMethods() with the
same targetType and MetadataLookup, if such subsequent invocations
occurred after the ApplicationContext had been refreshed.
The reason this occurs is due to the following.
- Class#getDeclaredMethods() always returns "child copies" of the
underlying Method instances -- which means that `equals()` should be
used instead of `==` whenever the compared Method instances can come
from different sources (such as the static caches mentioned below).
- BridgeMethodResolver caches resolved bridge methods in a static cache
-- which is never cleared.
- ReflectionUtils caches declared methods in a static cache
-- which gets cleared when an ApplicationContext is refreshed.
Consequently, if you attempt to load an ApplicationContext twice in the
same ClassLoader, the second attempt uses the existing, populated cache
for bridged methods but a cleared, empty cache for declared methods.
This results in new invocations of Class#getDeclaredMethods(), and
identity checks with `==` then fail to detect equivalent bridge methods.
This commit addresses this by additionally comparing bridge methods
using `equals()` in MethodIntrospector.selectMethods().
Note that the `==` checks remain in place as an optimization for when
`equals()` is unnecessary.
Closes gh-32586
KClass instantiation in CoroutinesUtils is suboptimal, and should be
replaced by KTypes#isSubtypeOf checks using pre-instantiated types for
Flow, Mono and Publisher.
This commit impact on performances is significant since a throughput
increase between 2x and 3x has been measured on basic endpoints.
Closes gh-32390
This commit fixes a performance regression caused by gh-31698,
and more specifically by KClass#isValue invocations which are slow since
they load the whole module to find the class to get the descriptor.
After discussing with the Kotlin team, it has been decided that only
checking for the presence of `@JvmInline` annotation is enough for
Spring use case.
As a consequence, this commit introduces a new
KotlinDetector#isInlineClass method that performs such check, and
BeanUtils, CoroutinesUtils and WebMVC/WebFlux InvocableHandlerMethod
have been refined to leverage it.
Closes gh-32334
Juergen Hoeller
Sam Brannen
Stéphane Nicoll
Sébastien Deleuze
T45K
Maksim Sasnouski