Background
Spring 3.1 introduced the @ComponentScan annotation, which can accept
an optional array of include and/or exclude @Filter annotations, e.g.
@ComponentScan(
basePackages = "com.acme.app",
includeFilters = { @Filter(MyStereotype.class), ... }
)
@Configuration
public class AppConfig { ... }
@ComponentScan and other annotations related to @Configuration class
processing such as @Import, @ImportResource and the @Enable*
annotations are parsed using reflection in certain code paths, e.g.
when registered directly against AnnotationConfigApplicationContext,
and via ASM in other code paths, e.g. when a @Configuration class is
discovered via an XML bean definition or when included via the
@Import annotation.
The ASM-based approach is designed to avoid premature classloading of
user types and is instrumental in providing tooling support (STS, etc).
Prior to this commit, the ASM-based routines for reading annotation
attributes were unable to recurse into nested annotations, such as in
the @Filter example above. Prior to Spring 3.1 this was not a problem,
because prior to @ComponentScan, there were no cases of nested
annotations in the framework.
This limitation manifested itself in cases where users encounter
the ASM-based annotation parsing code paths AND declare
@ComponentScan annotations with explicit nested @Filter annotations.
In these cases, the 'includeFilters' and 'excludeFilters' attributes
are simply empty where they should be populated, causing the framework
to ignore the filter directives and provide incorrect results from
component scanning.
The purpose of this change then, is to introduce the capability on the
ASM side to recurse into nested annotations and annotation arrays. The
challenge in doing so is that the nested annotations themselves cannot
be realized as annotation instances, so must be represented as a
nested Map (or, as described below, the new AnnotationAttributes type).
Furthermore, the reflection-based annotation parsing must also be
updated to treat nested annotations in a similar fashion; even though
the reflection-based approach has no problem accessing nested
annotations (it just works out of the box), for substitutability
against the AnnotationMetadata SPI, both ASM- and reflection-based
implementations should return the same results in any case. Therefore,
the reflection-based StandardAnnotationMetadata has also been updated
with an optional 'nestedAnnotationsAsMap' constructor argument that is
false by default to preserve compatibility in the rare case that
StandardAnnotationMetadata is being used outside the core framework.
Within the framework, all uses of StandardAnnotationMetadata have been
updated to set this new flag to true, meaning that nested annotation
results will be consistent regardless the parsing approach used.
Spr9031Tests corners this bug and demonstrates that nested @Filter
annotations can be parsed and read in both the ASM- and
reflection-based paths.
Major changes
- AnnotationAttributes has been introduced as a concrete
LinkedHashMap<String, Object> to be used anywhere annotation
attributes are accessed, providing error reporting on attribute
lookup and convenient type-safe access to common annotation types
such as String, String[], boolean, int, and nested annotation and
annotation arrays, with the latter two also returned as
AnnotationAttributes instances.
- AnnotationUtils#getAnnotationAttributes methods now return
AnnotationAttributes instances, even though for binary compatibility
the signatures of these methods have been preserved as returning
Map<String, Object>.
- AnnotationAttributes#forMap provides a convenient mechanism for
adapting any Map<String, Object> into an AnnotationAttributes
instance. In the case that the Map is already actually of
type AnnotationAttributes, it is simply casted and returned.
Otherwise, the map is supplied to the AnnotationAttributes(Map)
constructor and wrapped in common collections style.
- The protected MetadataUtils#attributesFor(Metadata, Class) provides
further convenience in the many locations throughout the
.context.annotation packagage that depend on annotation attribute
introspection.
- ASM-based core.type.classreading package reworked
Specifically, AnnotationAttributesReadingVisitor has been enhanced to
support recursive reading of annotations and annotation arrays, for
example in @ComponentScan's nested array of @Filter annotations,
ensuring that nested AnnotationAttributes objects are populated as
described above.
AnnotationAttributesReadingVisitor has also been refactored for
clarity, being broken up into several additional ASM
AnnotationVisitor implementations. Given that all types are
package-private here, these changes represent no risk to binary
compatibility.
- Reflection-based StandardAnnotationMetadata updated
As described above, the 'nestedAnnotationsAsMap' constructor argument
has been added, and all framework-internal uses of this class have
been updated to set this flag to true.
Issue: SPR-7979, SPR-8719, SPR-9031
A number of users reported issues with comparing method identity vs
equivalence when discovering JavaBeans property methods in
ExtendedBeanInfo.
This commit updates the implementation to consistently use '.equals()'
instead of '=='.
Issue: SPR-8079, SPR-8347
Prior to this change, roughly 5% (~300 out of 6000+) of files under the
source tree had CRLF line endings as opposed to the majority which have
LF endings.
This change normalizes these files to LF for consistency going forward.
Command used:
$ git ls-files | xargs file | grep CRLF | cut -d":" -f1 | xargs dos2unix
Issue: SPR-5608
Prior to this change, an assumption was made in
AbstractAutowireCapableBeanFactory that any factory-method would have
zero parameters. This may not be the case in @Bean methods.
We now look for the factory-method by name in a more flexible fashion
that accomodates the possibility of method parameters.
There remains at least one edge cases here where things could still fail,
for example a @Configuration class could have two FactoryBean-returning
methods of the same name, but each with different generic FactoryBean
types and different parameter lists. In this case, the implementation
may infer and return the wrong object type, as it currently returns
the first match for the given factory-method name. The complexity cost
of ensuring that this never happens is not likely worth the trouble
given the very low likelihood of such an arrangement.
Issue: SPR-8762
Allowing beans of primitive type to be looked up via getBean(Class), or
to be injected using @Autowired or @Injected or @Resource. Prior to
these changes, an attempt to lookup or inject a bean of, for example,
type boolean would fail because all spring beans are Objects, regardless
of initial type due to the way that ObjectFactory works.
Now these attempts to lookup or inject primitive types work, thanks to
simple changes in AbstractBeanFactory using ClassUtils#isAssignable
methods instead of the built-in Class#isAssignableFrom. The former takes
into account primitives and their object wrapper types, whereas the
latter does not.
The need to declare, look up or inject primitive-typed beans is probably
low -- how often does one need a bean of type boolean or int after all?.
Prior to the introduction of @Bean methods in Spring 3.0, it was not
possible in practice to register primitive beans, so this issue never
came up. Now that one can declare primitive-typed beans, it does make
sense that we properly support by-type lookup and injection without
forcing the user to work with object wrappers.
Issue: SPR-8874
Certain edge cases around return type covariance can trigger an
IntrospectionException when trying to create a new PropertyDescriptor;
particularly around the addition of write methods with parameter types
that do not match read method return types.
These type mismatch exceptions are raised during normal Introspector
operations as well (i.e. without ExtendedBeanInfo in the mix), but
the Introspector intentionally supresses them. In covariance cases,
there is often already a suitable write method present, e.g. discovered
in a supertype or superinterface, that, with the benefit of bridge
methods works just fine in practice at runtime. That is to say, in
these suppression cases, the rejection of the write method is 'OK' in
that there is already a write method present that can handle a call.
ExtendedBeanInfo now mirrors this suppression behavior, but does issue
a WARN-level log message to let the user know.
An important effect of this change is that ExtendedBeanInfo now modifies
the delegate BeanInfo object, whereas previously it did not. In practice
this probably matters very little, but it is a design change worth
noting. The reason for this change was to avoid the need to create new
PropertyDescriptors wherever possible. It was discovered that by updating
existing PDs, one can avoid these IntrospectionExceptions a greater
percentage of the time.
Issue: SPR-8806
Anywhere the value of a destroy method may be expressed, specifying
the value "(inferred)" now indicates that the container should attempt
to automatically discover a destroy method. This functionality is
currently limited to detecting public, no-arg methods named 'close';
this is particularly useful for commonly used types such as Hibernate
SessionFactory most JDBC DataSource implementations, JMS connection
factories, and so forth.
This special value is captured as the constant
AbstractBeanDefinition#INFER_METHOD, which in turn serves as the default
value of the @Bean#destroyMethod attribute.
For example in the following case
@Bean
public BasicDataSource dataSource() { ... }
the container will automatically detect BasicDataSource#close and invoke
it when the enclosing ApplicationContext is closed. This is exactly
equivalent to
@Bean(destroyMethod="(inferred)")
public BasicDataSource dataSource() { ... }
A user may override this inference-by-default convention simply by
specifying a different method
@Bean(destroyMethod="myClose")
public MyBasicDataSource dataSource() { ... }
or, in the case of a bean that has an otherwise inferrable 'close'
method, but the user wishes to disable handling it entirely, an empty
string may be specified
@Bean(destroyMethod="")
public MyBasicDataSource dataSource() { ... }
The special destroy method name "(inferred)" may also be specified in
an XML context, e.g.
<bean destroy-method="(inferred)">
or
<beans default-destroy-method="(inferred)">
Note that "(inferred)" is the default value for @Bean#destroyMethod,
but NOT for the destroy-method and default-destroy-method attributes
in the spring-beans XML schema.
The principal reason for introducing this feature is to avoid forcing
@Configuration class users to type destroyMethod="close" every time a
closeable bean is configured. This kind of boilerplate is easily
forgotten, and this simple convention means the right thing is done
by default, while allowing the user full control over customization or
disablement in special cases.
Issue: SPR-8751
Add BridgeMethodResolver#isJava6VisibilityBridgeMethodPair to
distinguish between (a) bridge methods introduced in Java 6 to
compensate for inheriting public methods from non-public superclasses
and (b) bridge methods that have existed since Java 5 to accommodate
return type covariance and generic parameters.
In the former case, annotations should be looked up from the original
bridged method (SPR-7900). In the latter, the annotation should be
looked up against the bridge method itself (SPR-8660).
As noted in the Javadoc for the new method, see
http://stas-blogspot.blogspot.com/2010/03/java-bridge-methods-explained.html
for a useful description of the various types of bridge methods, as
well as http://bugs.sun.com/view_bug.do?bug_id=6342411, the bug fixed in
Java 6 resulting in the introduction of 'visibility bridge methods'.
Issue: SPR-8660, SPR-7900
Previously, #containsBean Javadoc advertised that a true return value
indicates that a call to #getBean for the same name would succeed.
This is actually not the case, and has never been. The semantics
of #containsBean have always been to indicate whether a bean definition
with the given name is definied or a singleton instance with the given
name has been registered.
The Javadoc now reflects this accurately.
Issue: SPR-8690
Commit http://bit.ly/nXumTs ensured that component methods and fields
marked with 'common annotations' such as @Resource, @PostConstruct and
@PreDestroy are invoked/assigned once and only once, even if multiple
instances of the CommonAnnotationBeanPostProcessor are processing the
same bean factory.
The implementation works against the InjectionMetadata API, adding and
removing these members from sets that track whether they are already
'externally managed', i.e. that another CABPP has already handled them,
thus avoiding redundant processing.
Prior to this change, the #remove operations against these sets were
not synchronized. In a single-threaded context this is fine thanks to
logic in AbstractAutowireCapableBeanFactory#doCreateBean that checks to
see whether a given bean definition has already been post processed.
However, as reported by SPR-8598, certain cases involving multiple
threads and annotated prototype-scoped beans can cause concurrent
modification exceptions during the #remove operation (ostensibly because
another thread is attempting to do the same removal at the same time,
though this has yet to be reproduced in isolation).
Now the sets originally introduced by the commit above are decorated
with Collections#synchronizedSet and any iterations over those sets
are synchronized properly. This change should have low performance
impact as such processing happens at container startup time (save for
non-singleton lookups at runtime), and there should be little
contention in any case.
Issue: SPR-8598
Prior to this change, a parsing exception would be thrown if a
<description> element was placed within a <map><entry/></map>
element. The beans XSD has always permitted this arrangement,
but BeanDefinitionParserDelegate did not respect it. The latter now
simply ignores any <description> element, rather than failing.
Issue: SPR-8563
For the particular use case detailed in SPR-8514, with this change we
now attempt to determine the object type of a FactoryBean through its
generic type parameter if possible.
For (a contrived) example:
@Configuration
public MyConfig {
@Bean
public FactoryBean<String> fb() {
return new StringFactoryBean("foo");
}
}
The implementation will now look at the <String> generic parameter
instead of attempting to instantiate the FactoryBean in order to call
its #getObjectType() method.
This is important in order to avoid the autowiring lifecycle issues
detailed in SPR-8514. For example, prior to this change, the following
code would fail:
@Configuration
public MyConfig {
@Autowired Foo foo;
@Bean
public FactoryBean<String> fb() {
Assert.notNull(foo);
return new StringFactoryBean("foo");
}
}
The reason for this failure is that in order to perform autowiring,
the container must first determine the object type of all configured
FactoryBeans. Clearly a chicken-and-egg issue, now fixed by this
change.
And lest this be thought of as an obscure bug, keep in mind the use case
of our own JPA support: in order to configure and return a
LocalContainerEntityManagerFactoryBean from a @Bean method, one will
need access to a DataSource, etc -- resources that are likely to
be @Autowired across @Configuration classes for modularity purposes.
Note that while the examples above feature methods with return
types dealing directly with the FactoryBean interface, of course
the implementation deals with subclasses/subinterfaces of FactoryBean
equally as well. See ConfigurationWithFactoryBeanAndAutowiringTests
for complete examples.
There is at least a slight risk here, in that the signature of a
FactoryBean-returing @Bean method may advertise a generic type for the
FactoryBean less specific than the actual object returned (or than
advertised by #getObjectType for that matter). This could mean that an
autowiring target may be missed, that we end up with a kind of
autowiring 'false negative' where FactoryBeans are concerned. This is
probably a less common scenario than the need to work with an autowired
field within a FactoryBean-returning @Bean method, and also has a clear
workaround of making the generic return type more specific.
Issue: SPR-8514
Prior to this change, (Context)SingletonBeanFactoryLocator used
BeanFactoryUtils#beanOfType(ListableBeanFactory, Class) to locate the
bean of type BeanFactory.
The more modern approach is to use BeanFactory#getBean(Class), which
removes a dependency on ListableBeanFactory interface while at the same
time opening the implementation up to respecting autowiring exclusions,
primary metadata, etc.
Issue: SPR-8489
Prior to this change, @Autowired injection against an instance of a
subclass having different visibility than its superclass would fail
if the @Autowired method is declared only in the superclass. This is due
to an apparent change in the rules around bridge method generation
between Java 5 and Java 6, and possibly even varying across compiler
vendors.
Now, BridgeMethodResolver is used consistently when detecting
@Autowired, @Inject and @Resource metadata to bypass these bridge
methods if they exist.
Issue: SPR-7900
Prior to this change, AbstractAutowireCapableBeanFactory did not support
a default ParameterNameDiscoverer. This meant that attempting to use
<constructor-arg name=".."> syntax would fail (with a fairly obscure
exception) as that feature depends on a ParameterNameDiscoverer to
introspect the constructor arguments.
This lack of a default was originally intended to avoid a dependency on
ASM, but now that (a) .asm is a built-in module and (b) .beans has a
non-optional compile-time dependency on .asm, there is no reason not to
provide this default.
The net effect is that in a number of locations throughout the
framework, namely in GenericApplicationContext and
AbstractRefreshableApplicationContext, it is no longer necessary to
explicitly call AACBF#setParameterNameDiscoverer. This also means that
using a naked BeanFactory (likely for testing scenarios) is that much
easier.
Issue: SPR-8184
Chris Beams
Juergen Hoeller
Sam Brannen