This commit provides the necessary infrastructure to let components
contribute statements that are used to fully instantiate a bean
instance.
To ease code generation, a dedicated infrastructure to register bean
definition is provided in the o.s.beans.factory.generator package.
BeanDefinitionRegistrar offers a builder style API that provides a way
to hide how injected elements are resolved at runtime and let
contributors provide code that may throw a checked exception.
BeanInstanceContributor is the interface that components can implement
to contribute to a bean instance setup. DefaultBeanInstanceGenerator
generates, for a particular bean definition, the necessary statements
to instantiate a bean.
Closes gh-28047
Although Java does not allow the definition of recursive annotations,
Kotlin does, and prior to this commit an attempt to synthesize a
merged annotation using the MergedAnnotation API resulted in a
StackOverflowError if there was a recursive cycle in the annotation
definitions.
This commit addresses this issue by tracking which annotations have
already been visited and short circuits the recursive algorithm if a
cycle is detected.
Closes gh-28012
Although the initial report in gh-28015 only covered inconsistencies
for arrays and strings in the toString() implementations for
annotations between the JDK (after Java 9) and Spring, it has since
come to our attention that there was further room for improvement.
This commit therefore addresses the following in toString() output for
synthesized annotations.
- characters are now wrapped in single quotes.
- bytes are now properly formatted as "(byte) 0x##".
- long, float, and double values are now appended with "L", "f", and
"d", respectively. The use of lowercase for "f" and "d" is solely to
align with the choice made by the JDK team.
However, this commit does not address the following issues which we may
choose to address at a later point in time.
- non-ASCII, non-visible, and non-printable characters within a
character or String literal are not escaped.
- formatting for float and double values does not take into account
whether a value is not a number (NaN) or infinite.
Closes gh-28015
This commit adds an API that lets individual components contribute code,
runtime hints, and protected access information. This ease the cases
where code need to be written in a privileged package if necessary and
let contributors provide hints for the code they generate.
Closes gh-28030
This commit repackages the Javapoet library into spring-core so that it
can be used by the AOT engine without requiring a specific version.
Closes gh-27828
Since the introduction of synthesized annotation support in Spring
Framework 4.2 (a.k.a., merged annotations), the toString()
implementation attempted to align with the formatting used by the JDK
itself. However, Class annotation attributes were formatted using
Class#getName in Spring; whereas, the JDK used Class#toString up until
JDK 9.
In addition, JDK 9 introduced new formatting for toString() for
annotations, apparently intended to align with the syntax used in the
source code declaration of the annotation. However, JDK 9+ formats enum
annotation attributes using Enum#toString instead of Enum#name, which
can lead to issues if toString() is overridden in an enum.
This commit updates the formatting used for synthesized annotations by
ensuring that toString() generates a string that is compatible with the
syntax of the originating source code, going beyond the changes made in
JDK 9 by using Enum#name instead of Enum#toString.
Closes gh-28015
This commit provides an API to record the need for reflection,
resources, serialization, and proxies so that the runtime can be
optimized accordingly.
`RuntimeHints` provides an entry point to register the following:
* Reflection hints: individual elements of a type can be defined, as
well as a predefined categories (identified by the `MemberCategory`
enum). A method or constructor hint can refine whether the executable
should only be introspected or also invoked.
* Resource hints: patterns using includes/excludes identify the
resources to include at runtime. Resource bundles are also supported.
* Java Serialization hints: types that use java serialization can be
registered.
* Proxy hints: both interfaces-based (JDK) proxy and class-based proxy
can be defined.
This commit also introduces a `TypeReference` abstraction that permits
to record hints for types that are not available on the classpath, or
not compiled yet (generated code).
Closes gh-27829
Prior to this commit, the USER_DECLARED_METHODS MethodFilter in
ReflectionUtils did not actually filter methods declared in
java.lang.Object as stated in its Javadoc.
Consequently, if ReflectionUtils.doWithMethods was invoked with
USER_DECLARED_METHODS and Object.class as the class to introspect, all
non-bridge non-synthetic methods declared in java.lang.Object were
passed to the supplied MethodCallback, which breaks the contract of
USER_DECLARED_METHODS.
In addition, if USER_DECLARED_METHODS was composed with a custom
MethodFilter using `USER_DECLARED_METHODS.and(<custom MethodFilter>)`,
that composed method filter allowed all non-bridge non-synthetic
methods declared in java.lang.Object to be passed to the supplied
MethodCallback, which also breaks the contract of
USER_DECLARED_METHODS. This behavior resulted in regressions in code
that had previously used USER_DECLARED_METHODS by itself and then began
using USER_DECLARED_METHODS in a composed filter. For example, since
commit c419ea7ba7 ReflectiveAspectJAdvisorFactory has incorrectly
processed methods in java.lang.Object as candidates for advice methods.
This commit fixes this bug and associated regressions by ensuring that
USER_DECLARED_METHODS actually filters methods declared in
java.lang.Object. In addition, ReflectionUtils.doWithMethods now aborts
its search algorithm early if invoked with the USER_DECLARED_METHODS
filter and Object.class as the class to introspect.
Closes gh-27970
Prior to this commit, XmlValidationModeDetector did not properly parse
all categories of comments (described below). When such categories of
comments were encountered XmlValidationModeDetector may have
incorrectly detected that an XML file used a DTD when it used an XSD,
or vice versa.
This commit revises the parsing algorithm in XmlValidationModeDetector
so that multi-line comments and multiple comments on a single line are
properly recognized.
Specifically, with this commit the following categories of comments are
now handled properly.
- Multiple comments on a single line
- Multi-line comment: beginning on one line and then ending on another
line with an additional comment following on that same line.
- Multi-line comment: beginning at the end of XML content on one line
and then spanning multiple lines.
Closes gh-27915
Prior to this commit, our BlockHound integration tests were disabled
after the migration to a JDK 17 baseline since the build now always
runs on JDK 14 or higher.
To re-enable the tests we now supply the deprecated
-XX:+AllowRedefinitionToAddDeleteMethods command-line argument to the JVM
for tests in the Gradle build. Users can also configure this manually
within an IDE to run SpringCoreBlockHoundIntegrationTests.
If that command-line argument is removed from the JVM at some point in
the future, we will need to investigate an alternative solution.
See https://github.com/reactor/BlockHound/issues/33 for details.
In BridgeMethodResolver#isBridgedCandidateFor, candidateMethod is never
not bridged, so it's unnecessary to judge whether candidateMethod and
bridgeMethod are the same.
Closes gh-27862
Stephane Nicoll
Sébastien Deleuze
Sam Brannen
Brian Clozel
Juergen Hoeller
v_vyqyxiong
kuanghc1