前情回顾
在前两篇文章中，多次提及AutowireCapableBeanFactory#resolveDependency方法，原因是该方法很重要，在Spring很多场合都涉及该方法的调用，包括但不限于以下场景:
- 解析@Resouce注解的元素(CommonAnnotationBeanPostProcessor#autowireResource)
- 解析@Autowired、@Value注解的元素(AutowiredAnnotationBeanPostProcessor.AutowiredFieldElement#inject、AutowiredAnnotationBeanPostProcessor.AutowiredMethodElement#inject)
- autowire = byType(AbstractAutowireCapableBeanFactory#autowireByType)
- 构造器装配(ConstructorResolver#resolveAutowiredArgument)
了解此方法的底层工作原理，有助于提升对Spring Bean解析的认知能力
案例
本文以一种经典的注入案例进行探讨：注入集合对象
// 接口
public interface BarService {}
// 实现类1
@Service
public class BarServiceImplOne implements BarService {}
// 实现类2
@Service
public class BarServiceImplTwo implements BarService {}
@Service
public class FooService {
// barServices集合有两个元素，分别是BarServiceImplOne、BarServiceImplTwo
@Resource
private List<BarService> barServices;
}
注入集合的姿势Spring官网就有介绍，开发中也比较常用，现在借助该方式来探寻其中的一些细节问题，并介绍resolveDependency在其中起了怎样的作用
源码分析
源码基于Spring 5.1.11.RELEASE
resolveDependency翻译成中文即是解析依赖，其方法签名如下:
Object resolveDependency(DependencyDescriptor descriptor, @Nullable String requestingBeanName,
@Nullable Set<String> autowiredBeanNames, @Nullable TypeConverter typeConverter) throws BeansException;
- descriptor: 依赖描述符，它描述了一个待注入的依赖信息：要么是构造器参数，要么是方法参数，要么是字段，并且提供了非常友好的、一种统一的方式去访问
- requestingBeanName: 声明依赖的Bean。举例，如果A依赖B，则requestingBeanName表示的是A
- autowiredBeanNames: 待装配的Bean名称列表，即解析出来的bean names。使用上，一般是由外部传进来一个空的集合，在方法内部进行Bean的解析，如果符合条件，就将该bean name添加到集合内。潜台词是，可能会有多个符合条件的Bean，其实也很好理解，如果被依赖的类(接口)有多个实现类，且都被Spring管理，就存在多个符合条件的Bean的可能性
- typeConverter: 类型转化器，用于类型转换
由于使用的是@Resouce注解，故直接定位到CommonAnnotationBeanPostProcessor#autowireResource，AutowireCapableBeanFactory探密(2)——传统装配模式与现代注解驱动注入方式一文中也有简单的介绍
protected Object autowireResource(BeanFactory factory, LookupElement element, @Nullable String requestingBeanName)
throws NoSuchBeanDefinitionException {
// ...(省略)
if (factory instanceof AutowireCapableBeanFactory) {
AutowireCapableBeanFactory beanFactory = (AutowireCapableBeanFactory) factory;
// 依赖描述符
DependencyDescriptor descriptor = element.getDependencyDescriptor();
if (this.fallbackToDefaultTypeMatch && element.isDefaultName && !factory.containsBean(name)) {
// 满足条件，进入该分支
// 空集合
autowiredBeanNames = new LinkedHashSet<>();
// 进行Bean解析, requestingBeanName: fooService
resource = beanFactory.resolveDependency(descriptor, requestingBeanName, autowiredBeanNames, null);
if (resource == null) {
throw new NoSuchBeanDefinitionException(element.getLookupType(), "No resolvable resource object");
}
}
// ...(省略)
}
接着调用beanFactory.resolveDependency
// DefaultListableBeanFactory#resolveDependency
public Object resolveDependency(DependencyDescriptor descriptor, @Nullable String requestingBeanName,
@Nullable Set<String> autowiredBeanNames, @Nullable TypeConverter typeConverter) throws BeansException {
// 初始化参数解析器
descriptor.initParameterNameDiscovery(getParameterNameDiscoverer());
// 处理依赖类型为Option类的case，很显然，需要JDK1.8以上才支持，一般不会进入此处
if (Optional.class == descriptor.getDependencyType()) {
return createOptionalDependency(descriptor, requestingBeanName);
}
// 处理类型为ObjectFactory、ObjectProvider，略过
else if (ObjectFactory.class == descriptor.getDependencyType() ||
ObjectProvider.class == descriptor.getDependencyType()) {
return new DependencyObjectProvider(descriptor, requestingBeanName);
}
// 处理类型为JSR-330的javax.inject.Provider，略过
else if (javaxInjectProviderClass == descriptor.getDependencyType()) {
return new Jsr330Factory().createDependencyProvider(descriptor, requestingBeanName);
}
// 99%的情况进入else的分支
else {
// 处理@Lazy 注解的情况，一般特殊需要才会在字段或方法上标注@Lazy，不是本文重点，略过
Object result = getAutowireCandidateResolver().getLazyResolutionProxyIfNecessary(
descriptor, requestingBeanName);
if (result == null) {
// 大部分情况会走下面的case，进行真正的解析
result = doResolveDependency(descriptor, requestingBeanName, autowiredBeanNames, typeConverter);
}
return result;
}
}
虽然有很多条件分支存在，但大部分场景都不会用到，因此只需要关注result = doResolveDependency(descriptor, requestingBeanName, autowiredBeanNames, typeConverter);这一行核心代码即可
有意思的是，一些人认为Spring考虑周全，兼容各种case，360度无死角，正是Java web领域它独领风骚的魅力之所在；与此同时，另一些人认为正是因为考虑太周全，Spring变的越来越臃肿不堪，代码阅读越发困难，需要瘦瘦身，甚至需要一个更轻量级的框架来替代。天下大势，分久必合，合久必分，苍天饶过谁？
整体上看一下doResolveDependency方法核心逻辑
public Object doResolveDependency(DependencyDescriptor descriptor, @Nullable String beanName,
@Nullable Set<String> autowiredBeanNames, @Nullable TypeConverter typeConverter) throws BeansException {
InjectionPoint previousInjectionPoint = ConstructorResolver.setCurrentInjectionPoint(descriptor);
try {
// 只有ShortcutDependencyDescriptor实现了resolveShortcut方法，返回了非空值。目前版本代码只在AutowiredFieldElement、AutowiredMethodElement类中使用到，也即是说，只有解析@Autowired、@Value注解的元素才会用到，目的是为了将解析结果缓存起来，避免重复解析
Object shortcut = descriptor.resolveShortcut(this);
if (shortcut != null) {
return shortcut;
}
// 依赖的类型type: java.util.List
Class<?> type = descriptor.getDependencyType();
// 处理@Value注解，取值注解中的value属性中的值(原样，未经过解析的)，如果descriptor未被@Value标注，则返回null
// 注：从此处可知，@Value注解的优先级较高，只要找到了就处理，不再往下走
Object value = getAutowireCandidateResolver().getSuggestedValue(descriptor);
if (value != null) {
if (value instanceof String) {
// 处理占位符如${}，做占位符的替换(不解析SP EL表达式)
String strVal = resolveEmbeddedValue((String) value);
BeanDefinition bd = (beanName != null && containsBean(beanName) ?
getMergedBeanDefinition(beanName) : null);
// 解析SP EL(如#{})
value = evaluateBeanDefinitionString(strVal, bd);
}
TypeConverter converter = (typeConverter != null ? typeConverter : getTypeConverter());
try {
// 类型转换，把解析出来的结果转成type类型
return converter.convertIfNecessary(value, type, descriptor.getTypeDescriptor());
}
catch (UnsupportedOperationException ex) {
// A custom TypeConverter which does not support TypeDescriptor resolution...
return (descriptor.getField() != null ?
converter.convertIfNecessary(value, type, descriptor.getField()) :
converter.convertIfNecessary(value, type, descriptor.getMethodParameter()));
}
}
// 本文重点，解析"集合类"Bean，如果依赖的类型不是"集合类"，则返回null
// 注："集合类"是口语描述，目的是方便记忆，实际上，还支持数组类型和Map类型
/**
* 1. array
* 2. Collection及其子类
* 3. Map
*/
Object multipleBeans = resolveMultipleBeans(descriptor, beanName, autowiredBeanNames, typeConverter);
if (multipleBeans != null) {
return multipleBeans;
}
// 代码走到此处，说明依赖的是非"集合类"，
// 查找所有类型为type的实例，存放在matchingBeans <beanName, bean>
Map<String, Object> matchingBeans = findAutowireCandidates(beanName, type, descriptor);
if (matchingBeans.isEmpty()) {
if (isRequired(descriptor)) {
// 如果IoC容器中找不到符合条件的Bean，且依赖项标识为required，则抛出NoSuchBeanDefinitionException异常
raiseNoMatchingBeanFound(type, descriptor.getResolvableType(), descriptor);
}
return null;
}
String autowiredBeanName;
Object instanceCandidate;
// 如果找到多个元素，Spring要按一定的机制进行挑选，如果不满足规则可能需要抛出异常
if (matchingBeans.size() > 1) {
// 按以下顺序，找到符合条件的就直接返回
// 1. 挑选出被标识为primary的bean
// 2. 挑选标识了@Priority，且先级级最高的bean。可以不标识，一旦标识，不允许同一优先级的存在
// 3. fallback，依赖的名称与matchingBeans中任意一Key匹配
autowiredBeanName = determineAutowireCandidate(matchingBeans, descriptor);
if (autowiredBeanName == null) {
if (isRequired(descriptor) || !indicatesMultipleBeans(type)) {
// 非集合类，找到了多个符合条件的Bean，抛出异常
return descriptor.resolveNotUnique(descriptor.getResolvableType(), matchingBeans);
}
else {
// In case of an optional Collection/Map, silently ignore a non-unique case:
// possibly it was meant to be an empty collection of multiple regular beans
// (before 4.3 in particular when we didn't even look for collection beans).
return null;
}
}
instanceCandidate = matchingBeans.get(autowiredBeanName);
}
else {
// We have exactly one match.
// 找到匹配的唯一元素，直接取出来
Map.Entry<String, Object> entry = matchingBeans.entrySet().iterator().next();
autowiredBeanName = entry.getKey();
instanceCandidate = entry.getValue();
}
if (autowiredBeanNames != null) {
// 将待装配的Bean名称放入autowiredBeanNames集合里
autowiredBeanNames.add(autowiredBeanName);
}
if (instanceCandidate instanceof Class) {
instanceCandidate = descriptor.resolveCandidate(autowiredBeanName, type, this);
}
Object result = instanceCandidate;
if (result instanceof NullBean) {
if (isRequired(descriptor)) {
raiseNoMatchingBeanFound(type, descriptor.getResolvableType(), descriptor);
}
result = null;
}
// 类型校验，确保类型与解析出来的Bean实例能够匹配
if (!ClassUtils.isAssignableValue(type, result)) {
throw new BeanNotOfRequiredTypeException(autowiredBeanName, type, instanceCandidate.getClass());
}
return result;
}
finally {
ConstructorResolver.setCurrentInjectionPoint(previousInjectionPoint);
}
}
本文案例是注入集合类对象，因此把关注点放到Object multipleBeans = resolveMultipleBeans(descriptor, beanName, autowiredBeanNames, typeConverter);
private Object resolveMultipleBeans(DependencyDescriptor descriptor, @Nullable String beanName,
@Nullable Set<String> autowiredBeanNames, @Nullable TypeConverter typeConverter) {
// 依赖的类型
final Class<?> type = descriptor.getDependencyType();
if (descriptor instanceof StreamDependencyDescriptor) {
// ...(特殊的用法，省略)
}
// 如果是数组
else if (type.isArray()) {
// 与下边的分支逻辑类似
Class<?> componentType = type.getComponentType();
ResolvableType resolvableType = descriptor.getResolvableType();
Class<?> resolvedArrayType = resolvableType.resolve(type);
if (resolvedArrayType != type) {
componentType = resolvableType.getComponentType().resolve();
}
if (componentType == null) {
return null;
}
Map<String, Object> matchingBeans = findAutowireCandidates(beanName, componentType,
new MultiElementDescriptor(descriptor));
if (matchingBeans.isEmpty()) {
return null;
}
if (autowiredBeanNames != null) {
autowiredBeanNames.addAll(matchingBeans.keySet());
}
TypeConverter converter = (typeConverter != null ? typeConverter : getTypeConverter());
Object result = converter.convertIfNecessary(matchingBeans.values(), resolvedArrayType);
if (result instanceof Object[]) {
Comparator<Object> comparator = adaptDependencyComparator(matchingBeans);
if (comparator != null) {
Arrays.sort((Object[]) result, comparator);
}
}
return result;
}
// 如果依赖的类型是Collection及其子接口(不能是具体实现类)
else if (Collection.class.isAssignableFrom(type) && type.isInterface()) {
// 获取集合元素的泛型信息，即集合元素类型。如果没有泛型信息，即获取不了元素类型，则返回null
Class<?> elementType = descriptor.getResolvableType().asCollection().resolveGeneric();
if (elementType == null) {
return null;
}
// 查找所有类型为type的实例，存放在matchingBeans <beanName, bean>
Map<String, Object> matchingBeans = findAutowireCandidates(beanName, elementType,
new MultiElementDescriptor(descriptor));
if (matchingBeans.isEmpty()) {
return null;
}
// 将待装配的Bean名称放入autowiredBeanNames集合里
if (autowiredBeanNames != null) {
autowiredBeanNames.addAll(matchingBeans.keySet());
}
TypeConverter converter = (typeConverter != null ? typeConverter : getTypeConverter());
// 将解析出来的结果转换成目标类型type的元素(List)
Object result = converter.convertIfNecessary(matchingBeans.values(), type);
if (result instanceof List) {
// 如果待注入对象为List实例，就再按AnnotationAwareOrderComparator排个序
// 可用PriorityOrdered、Ordered、@Order、@Priority定义顺序
Comparator<Object> comparator = adaptDependencyComparator(matchingBeans);
if (comparator != null) {
((List<?>) result).sort(comparator);
}
}
return result;
}
// 如果依赖的是Map类型(如: Map<String, BarService>)
else if (Map.class == type) {
ResolvableType mapType = descriptor.getResolvableType().asMap();
Class<?> keyType = mapType.resolveGeneric(0);
if (String.class != keyType) {
// Map的Key必须为String类型，表示Bean的名称
return null;
}
Class<?> valueType = mapType.resolveGeneric(1);
// 同样的，value类型的泛型信息必须指定，否则为null
if (valueType == null) {
return null;
}
Map<String, Object> matchingBeans = findAutowireCandidates(beanName, valueType,
new MultiElementDescriptor(descriptor));
if (matchingBeans.isEmpty()) {
return null;
}
if (autowiredBeanNames != null) {
autowiredBeanNames.addAll(matchingBeans.keySet());
}
// map无序，直接返回
return matchingBeans;
}
else {
return null;
}
}
Array与Collection及其子接口的处理逻辑相似，都是找到matchingBeans，并通过TypeConverter转换成目标类型，再经过AnnotationAwareOrderComparator排序，如此，返回的结集便是带有顺序的Array或Collection
Map的处理是找到matchingBeans，但不排序，此处需要注意的是，Key必须为String类型，表示Bean的名称。在本案例中，可写成依赖的属性是: Map<String, BarService>
在doResolveDependency与resolveMultipleBeans方法中多次出现findAutowireCandidates的调用，它的作用是根据requiredType在IoC中找到匹配的Bean实例，并组装成Map<BeanName, BeanInstance>返回，源码如下:
protected Map<String, Object> findAutowireCandidates(
@Nullable String beanName, Class<?> requiredType, DependencyDescriptor descriptor) {
// 从IoC中拿到所有类型为requiredType的bean name，本质上调用的是ListableBeanFactory#getBeanNamesForType方法进行获取
String[] candidateNames = BeanFactoryUtils.beanNamesForTypeIncludingAncestors(
this, requiredType, true, descriptor.isEager());
Map<String, Object> result = new LinkedHashMap<>(candidateNames.length);
// 处理特殊的依赖，如BeanFactory、ApplicationContext等，这些类的实例并不在狭义的IoC容器中，而是保存在resolvableDependencies
// 只能通过遍历resolvableDependencies与requiredType进行比较，满足条件返返回
for (Map.Entry<Class<?>, Object> classObjectEntry : this.resolvableDependencies.entrySet()) {
Class<?> autowiringType = classObjectEntry.getKey();
if (autowiringType.isAssignableFrom(requiredType)) {
Object autowiringValue = classObjectEntry.getValue();
autowiringValue = AutowireUtils.resolveAutowiringValue(autowiringValue, requiredType);
if (requiredType.isInstance(autowiringValue)) {
result.put(ObjectUtils.identityToString(autowiringValue), autowiringValue);
break;
}
}
}
// 本案例中candidateNames有两个元素[barServiceImplOne、barServiceImplTwo]
for (String candidate : candidateNames) {
// 非自引用 且 candidate对应的BeanDefinition是autowireCandidate，则表明符合条件，添加到CandidateEntry中
// 注：autowireCandidate是AbstractBeanDefinition的一个属性，默认值为true，即所有的Bean默认都支持autowire
if (!isSelfReference(beanName, candidate) && isAutowireCandidate(candidate, descriptor)) {
addCandidateEntry(result, candidate, descriptor, requiredType);
}
}
if (result.isEmpty()) {
boolean multiple = indicatesMultipleBeans(requiredType);
// Consider fallback matches if the first pass failed to find anything...
DependencyDescriptor fallbackDescriptor = descriptor.forFallbackMatch();
for (String candidate : candidateNames) {
if (!isSelfReference(beanName, candidate) && isAutowireCandidate(candidate, fallbackDescriptor) &&
(!multiple || getAutowireCandidateResolver().hasQualifier(descriptor))) {
addCandidateEntry(result, candidate, descriptor, requiredType);
}
}
if (result.isEmpty() && !multiple) {
// Consider self references as a final pass...
// but in the case of a dependency collection, not the very same bean itself.
for (String candidate : candidateNames) {
if (isSelfReference(beanName, candidate) &&
(!(descriptor instanceof MultiElementDescriptor) || !beanName.equals(candidate)) &&
isAutowireCandidate(candidate, fallbackDescriptor)) {
addCandidateEntry(result, candidate, descriptor, requiredType);
}
}
}
}
return result;
}
private void addCandidateEntry(Map<String, Object> candidates, String candidateName,
DependencyDescriptor descriptor, Class<?> requiredType) {
if (descriptor instanceof MultiElementDescriptor) {
// 下面的一行代码本质上是: beanFactory.getBean(beanName), 即根据beanName上IoC容器中查找
Object beanInstance = descriptor.resolveCandidate(candidateName, requiredType, this);
if (!(beanInstance instanceof NullBean)) {
// 找到，添加进CandidateEntry中
candidates.put(candidateName, beanInstance);
}
}
else if (containsSingleton(candidateName) || (descriptor instanceof StreamDependencyDescriptor &&
((StreamDependencyDescriptor) descriptor).isOrdered())) {
// 同上
Object beanInstance = descriptor.resolveCandidate(candidateName, requiredType, this);
candidates.put(candidateName, (beanInstance instanceof NullBean ? null : beanInstance));
}
else {
candidates.put(candidateName, getType(candidateName));
}
}
需要说明一下的是，findAutowireCandidates方法中出现了resolvableDependencies(Map<Class<?>, Object>)属性，它定义在DefaultListableBeanFactory，其作用是存放Spring内部一些特殊的Bean，比如BeanFactory、ResourceLoader、ApplicationContext、ServletRequest等；而一般普通的Bean存放在DefaultSingletonBeanRegistry.singletonObjects(Map<String, Object>)属性中，该属性就是狭义上的IoC容器
总结
AutowireCapableBeanFactory#resolveDependency，本质上是根据descriptor(依赖描述符)到Spring中找到符合描述的Bean(们)并返回。既可以解析由@Resouce标注的依赖信息，也可以解析由@Autowired、@Value标注的依赖信息；既可以解析单一依赖元素，也可以解析多个依赖("集合类")元素，当是集合类元素时，如果是Array或者是Collection，还可以根据PriorityOrdered、Ordered、@Order、@Priority定义注入的元素顺序；既在狭义的IoC容器(singletonObjects)中寻找，也在特殊的容器(resolvableDependencies)中寻找。
总之，该方法的能力非常强大，涉及的面也非常地广泛，因此，本文仅分享了其中一些与注入集合对象案例相关的细节。受限于作者的表达功力，本文并不足以描述它的全貌，还有诸多细节未能展开进行讲解，例如：
- @Lazy是如何解析的？
- @Value的占位符如何解析？SP EL表达式又如何解析？
- determineAutowireCandidate的细节是如何展开的？
- 类型转换又是如何进行的？
- 该方法还支持哪些骚操作？
导读:
网友评论