This commit applies the same kind of processing for
Void.class element type in the ParameterizedTypeReference
variant of WebTestClient.ResponseSpec#returnResult than
in the Class variant.
Closes gh-33389
Prior to this commit if the return type is a CompletableFuture but a
cache get returns null, the execution falls through to the
reactiveCachingHandler.
This commit ensures that evaluation instead continues onto the next
cache (if any).
Closes gh-33371
This commit revises QualifierAnnotationAutowireCandidateResolver to
reinstate "qualifier" support for the legacy JSR-330
@javax.inject.Named annotation.
See gh-31090
Closes gh-33345
When ScheduledAnnotationReactiveSupport adds the Observation to the
context, Reactor opens a scope through the Context Propagation API.
This happens before TrackingSubscriber would start the Observation and
opening a scope without starting an Observation is invalid.
This change moves the Observation start before the scope opening.
Closes gh-33349
Before UnknownContentTypeException is thrown, it is added to the current
Observation but since UnknownContentTypeException is a
RestClientException, the exception is added again in one of the
catch blocks later on.
Closes gh-33347
ClientHttpResponse implements Closeable and the close method of
DefaultConvertibleClientHttpResponse also stops the current Observation.
Before this change exiting the try-with-resource block stopped the
Observation since it called close on ClientHttpResponse.
After this, there were multiple error and stop calls on the Observation
in the catch blocks after the Observation was already stopped
which is invalid.
This change reorders the flow by stopping the Observation in the
finally block (closing ClientHttpResponse) and not stopping
the Observation in any of the catch blocks.
Closes gh-33346
Prior to this commit, the Gradle build output the following warning
multiple times.
OpenJDK 64-Bit Server VM warning: Sharing is only supported for boot loader classes because bootstrap classpath has been appended
Since we don't need CDS enabled for our tests, I've added `-Xshare:off`
as a JVM argument for our tests to disable CDS.
(cherry picked from commit 27985b1439)
Since the me.champeau.mrjar stores Java 21 main and test sources in
non-standard locations, and since we can only use a single JDK within
Eclipse, this commit updates `ide.gradle` to remove `java21` classpath
entries in Eclipse projects.
This commit implements modifyContextPath in ServletRequestPath and
apply the same logic of concatenating the servlet path with the
context path.
Closes gh-33251
This changes the criteria for simplified task rejection logging from
to be that the executor is in the shutdown phase, not fully terminated.
See gh-33334
Closes gh-33336
SimpleEvaluationContext.forReadOnlyDataBinding() documents that it
creates a SimpleEvaluationContext for read-only access to public
properties; however, prior to this commit write access was not disabled
for indexed structures when using the assignment operator, the
increment operator, or the decrement operator.
In order to better align with the documented contract for
forReadOnlyDataBinding(), this commit makes it possible to disable
assignment in general in order to enforce read-only semantics for
SpEL's SimpleEvaluationContext when created via the
forReadOnlyDataBinding() factory method. Specifically:
- This commit introduces a new isAssignmentEnabled() "default" method
in the EvaluationContext API, which returns true by default.
- SimpleEvaluationContext overrides isAssignmentEnabled(), returning
false if the context was created via the forReadOnlyDataBinding()
factory method.
- The Assign, OpDec, and OpInc AST nodes -- representing the assignment
(=), increment (++), and decrement (--) operators, respectively --
now throw a SpelEvaluationException if assignment is disabled for the
current EvaluationContext.
Closes gh-33319
The changes made in conjunction with gh-33013 resulted in a regression
for varargs support in SpEL expressions. Specifically, before gh-33013
one could supply a list -- for example, an "inline list" -- as the
varargs array when invoking a varargs constructor, method, or function
within a SpEL expression. However, after gh-33013 an inline list (or
collection in general) is no longer converted to an array for varargs
invocations. Instead, the list is supplied as a single argument of the
resulting varargs array which breaks applications that depend on the
previous behavior.
Although it was never intended that one could supply a collection as
the set of varargs, we concede that this is a regression in existing
behavior, and this commit therefore restores support for supplying a
java.util.List as the varargs "array".
In addition, this commit introduces the same "list to array" conversion
support for MethodHandle-based functions that accept varargs.
Note, however, that this commit does not restore support for converting
arbitrary single objects to an array for varargs invocations if the
single object is already an instance of the varargs array's component
type.
See gh-33013
Closes gh-33315
Prior to this commit, the Indexer in the Spring Expression Language
(SpEL) silently ignored a failure to set a property via the indexed
property syntax (['<property name>'] = <new value>) – for example, if
property write access was disabled in the EvaluationContext.
This commit addresses this issue by properly throwing a
SpelEvaluationException in PropertyIndexingValueRef.setValue(Object) if
the property could not be set.
Closes gh-33310
Prior to this commit, SpEL incorrectly wrapped a primitive array in an
Object[] array when invoking Object[] varargs methods and constructors.
This commit addresses this by updating the convertArguments(...) method
in ReflectionHelper as follows when the user supplies the varargs
already packaged in a primitive array.
- When deciding whether to convert a single element passed as varargs,
we now check if the argument is an array that is assignable to the
varargs array type.
- When converting an array supplied as the varargs, we convert that
array to the varargs array type.
Closes gh-33317
Prior to this commit, the ConversionService failed to convert a primitive
array (such as int[]) to an Object[] due to an error in the logic in
ArrayToArrayConverter.
This commit addresses this by augmenting the "can bypass conversion"
check in ArrayToArrayConverter to ensure that the supplied source object
is an instance of the target type (i.e., that the source array can be
cast to the target type array without conversion).
Closes gh-33212
