原型Bean加载过程

之前的文章，分析了非懒加载的单例Bean整个加载过程，除了非懒加载的单例Bean之外，Spring中还有一种Bean就是原型（Prototype）的Bean，看一下定义方式：

<pre>1 <?xml version="1.0" encoding="UTF-8"?>
2 <beans xmlns="http://www.springframework.org/schema/beans"
3 xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
4 xsi:schemaLocation="http://www.springframework.org/schema/beans 5 http://www.springframework.org/schema/beans/spring-beans-3.0.xsd">
6
7 <bean id="prototypeBean" class="org.xrq.action.PrototypeBean" scope="prototype" />
8
9 </beans></pre>

原型Bean加载流程总得来说和单例Bean差不多，看一下不同之处，在AbstractBeanFactory的doGetBean的方法的这一步：

<pre> 1 else if (mbd.isPrototype()) { 2 // It's a prototype -> create a new instance.
3 Object prototypeInstance = null;
4 try { 5 beforePrototypeCreation(beanName);
6 prototypeInstance = createBean(beanName, mbd, args); 7 }
8 finally { 9 afterPrototypeCreation(beanName); 10 } 11 bean = getObjectForBeanInstance(prototypeInstance, name, beanName, mbd); 12 }</pre>

第6行createBean是一样的，原型Bean实例化的主要区别就在于第6行，它是直接创建bean的，而单例bean我们再对比一下：

<pre> 1 if (mbd.isSingleton()) { 2 sharedInstance = getSingleton(beanName, new ObjectFactory() { 3 public Object getObject() throws BeansException { 4 try { 5 return createBean(beanName, mbd, args); 6 }
7 catch (BeansException ex) { 8 // Explicitly remove instance from singleton cache: It might have been put there 9 // eagerly by the creation process, to allow for circular reference resolution. 10 // Also remove any beans that received a temporary reference to the bean.
11 destroySingleton(beanName); 12 throw ex; 13 } 14 } 15 }); 16 bean = getObjectForBeanInstance(sharedInstance, name, beanName, mbd); 17 }</pre>

它优先会尝试getSington，即先尝试从singletonObjects中获取一下bean是否存在，如果存在直接返回singletonObjects中的bean对象。

接着，我们看到原型bean创建和单例bean创建的区别还在于第5行和第9行，先看第5行的代码：

<pre> 1 protected void beforePrototypeCreation(String beanName) { 2 Object curVal = this.prototypesCurrentlyInCreation.get();
3 if (curVal == null) {
4 this.prototypesCurrentlyInCreation.set(beanName);
5 }
6 else if (curVal instanceof String) { 7 Set<String> beanNameSet = new HashSet<String>(2);
8 beanNameSet.add((String) curVal);
9 beanNameSet.add(beanName); 10 this.prototypesCurrentlyInCreation.set(beanNameSet); 11 } 12 else { 13 Set<String> beanNameSet = (Set<String>) curVal; 14 beanNameSet.add(beanName); 15 } 16 }</pre>

这段主要是说bean在创建前要把当前beanName设置到ThreadLocal中去，其目的是保证多线程不会同时创建同一个bean。接着看第9行的代码实现，即bean创建之后做了什么：

<pre> 1 protected void afterPrototypeCreation(String beanName) { 2 Object curVal = this.prototypesCurrentlyInCreation.get();
3 if (curVal instanceof String) { 4 this.prototypesCurrentlyInCreation.remove();
5 }
6 else if (curVal instanceof Set) { 7 Set<String> beanNameSet = (Set<String>) curVal;
8 beanNameSet.remove(beanName);
9 if (beanNameSet.isEmpty()) { 10 this.prototypesCurrentlyInCreation.remove(); 11 } 12 } 13 }</pre>

很好理解，就是把当前bean移除一下，这样其它线程就可以创建bean了。第11行的代码不看了，意思是如果bean是FactoryBean的实现类的话，调用getObject()方法获取真正的对象。

byName源码实现

Spring有为开发者提供Autowire（自动装配）的功能，自动装配最常用的就是byName和byType这两种属性。由于自动装配是为了解决对象注入导致的<property>过多的问题，因此很容易找到byName与byType的Spring源码实现应该在属性注入这一块，定位到属性注入的代码AbstractAutowireCapableBeanFactory的populateBean方法，直接截取重点：

<pre> 1 if (mbd.getResolvedAutowireMode() == RootBeanDefinition.AUTOWIRE_BY_NAME ||
2 mbd.getResolvedAutowireMode() == RootBeanDefinition.AUTOWIRE_BY_TYPE) { 3 MutablePropertyValues newPvs = new MutablePropertyValues(pvs); 4
5 // Add property values based on autowire by name if applicable.
6 if (mbd.getResolvedAutowireMode() == RootBeanDefinition.AUTOWIRE_BY_NAME) { 7 autowireByName(beanName, mbd, bw, newPvs);
8 }
9
10 // Add property values based on autowire by type if applicable.
11 if (mbd.getResolvedAutowireMode() == RootBeanDefinition.AUTOWIRE_BY_TYPE) { 12 autowireByType(beanName, mbd, bw, newPvs); 13 } 14
15 pvs = newPvs; 16 }</pre>

看到第6行_{第8行判断是否byName形式，是就执行byName自动装配代码；第11行}第13行判断是否byType形式，是就执行byType自动装配代码。那么首先看一下第7行的byName代码实现：

<pre> 1 protected void autowireByName( 2 String beanName, AbstractBeanDefinition mbd, BeanWrapper bw, MutablePropertyValues pvs) {
3
4 String[] propertyNames = unsatisfiedNonSimpleProperties(mbd, bw); 5 for (String propertyName : propertyNames) { 6 if (containsBean(propertyName)) { 7 Object bean = getBean(propertyName); 8 pvs.add(propertyName, bean);
9 registerDependentBean(propertyName, beanName); 10 if (logger.isDebugEnabled()) { 11 logger.debug("Added autowiring by name from bean name '" + beanName +
12 "' via property '" + propertyName + "' to bean named '" + propertyName + "'"); 13 } 14 } 15 else { 16 if (logger.isTraceEnabled()) { 17 logger.trace("Not autowiring property '" + propertyName + "' of bean '" + beanName +
18 "' by name: no matching bean found"); 19 } 20 } 21 } 22 }</pre>

篇幅问题，代码不一层层跟了，逻辑梳理一下：

• 第4行，找到Bean中不是简单属性的属性，这句话有点绕，意思就是找到属性是对象类型的属性，但也不是所有的对象类型都会被找到，比如CharSequence类型、Number类型、Date类型、URL类型、URI类型、Locale类型、Class类型就会忽略，具体可见BeanUtils的isSimpleProperty方法
• 第5行~第7行，遍历所有被找到的属性，如果bean定义中包含了属性名，那么先实例化该属性名对应的bean
• 第9行registerDependentBean，注册一下当前bean的依赖bean，用于在某个bean被销毁前先将其依赖的bean销毁

其余代码都是一些打日志的，没什么好说的。

byType源码实现

上面说了byName的源码实现，接下来看一下byType源码实现：

<pre> 1 protected void autowireByType( 2 String beanName, AbstractBeanDefinition mbd, BeanWrapper bw, MutablePropertyValues pvs) {
3
4 TypeConverter converter = getCustomTypeConverter(); 5 if (converter == null) {
6 converter = bw; 7 }
8
9 Set<String> autowiredBeanNames = new LinkedHashSet<String>(4); 10 String[] propertyNames = unsatisfiedNonSimpleProperties(mbd, bw); 11 for (String propertyName : propertyNames) { 12 try { 13 PropertyDescriptor pd = bw.getPropertyDescriptor(propertyName); 14 // Don't try autowiring by type for type Object: never makes sense, 15 // even if it technically is a unsatisfied, non-simple property.
16 if (!Object.class.equals(pd.getPropertyType())) { 17 MethodParameter methodParam = BeanUtils.getWriteMethodParameter(pd); 18 // Do not allow eager init for type matching in case of a prioritized post-processor.
19 boolean eager = !PriorityOrdered.class.isAssignableFrom(bw.getWrappedClass()); 20 DependencyDescriptor desc = new AutowireByTypeDependencyDescriptor(methodParam, eager); 21 Object autowiredArgument = resolveDependency(desc, beanName, autowiredBeanNames, converter); 22 if (autowiredArgument != null) { 23 pvs.add(propertyName, autowiredArgument); 24 } 25 for (String autowiredBeanName : autowiredBeanNames) { 26 registerDependentBean(autowiredBeanName, beanName); 27 if (logger.isDebugEnabled()) { 28 logger.debug("Autowiring by type from bean name '" + beanName + "' via property '" +
29 propertyName + "' to bean named '" + autowiredBeanName + "'"); 30 } 31 } 32 autowiredBeanNames.clear(); 33 } 34 } 35 catch (BeansException ex) { 36 throw new UnsatisfiedDependencyException(mbd.getResourceDescription(), beanName, propertyName, ex); 37 } 38 } 39 }</pre>

前面一样，到第10行都是找到Bean中属性是对象类型的属性。

接着就是遍历一下PropertyName，获取PropertyName对应的属性描述，注意一下16行的判断及其对应的注释：不要尝试自动装配Object类型，这没有任何意义，即使从技术角度看它是一个非简单的对象属性。

第18行~第20行跳过（没有太明白是干什么的），byType实现的源码主要在第21行的方法resolveDependency中，这个方法是AbstractAutowireCapableBeanFactory类的实现类DefaultListableBeanFactory中的方法：

<pre> 1 public Object resolveDependency(DependencyDescriptor descriptor, String beanName, 2 Set<String> autowiredBeanNames, TypeConverter typeConverter) throws BeansException { 3
4 descriptor.initParameterNameDiscovery(getParameterNameDiscoverer());
5 if (descriptor.getDependencyType().equals(ObjectFactory.class)) {
6 return new DependencyObjectFactory(descriptor, beanName); 7 }
8 else if (descriptor.getDependencyType().equals(javaxInjectProviderClass)) { 9 return new DependencyProviderFactory().createDependencyProvider(descriptor, beanName); 10 } 11 else { 12 return doResolveDependency(descriptor, descriptor.getDependencyType(), beanName, autowiredBeanNames, typeConverter); 13 } 14 }</pre>

这里判断一下要自动装配的属性是ObjectFactory.class还是javaxInjectProviderClass还是其他的，我们装配的是其他的，看一下12行的代码实现：

<pre> 1 protected Object doResolveDependency(DependencyDescriptor descriptor, Class<?> type, String beanName, 2 Set<String> autowiredBeanNames, TypeConverter typeConverter) throws BeansException { 3
4 Object value = getAutowireCandidateResolver().getSuggestedValue(descriptor); 5 if (value != null) {
6 if (value instanceof String) { 7 String strVal = resolveEmbeddedValue((String) value); 8 BeanDefinition bd = (beanName != null && containsBean(beanName) ? getMergedBeanDefinition(beanName) : null);
9 value = evaluateBeanDefinitionString(strVal, bd); 10 } 11 TypeConverter converter = (typeConverter != null ? typeConverter : getTypeConverter()); 12 return converter.convertIfNecessary(value, type); 13 } 14
15 if (type.isArray()) { 16 ... 17 } 18 else if (Collection.class.isAssignableFrom(type) && type.isInterface()) { 19 ... 20 } 21 else if (Map.class.isAssignableFrom(type) && type.isInterface()) { 22 ... 23 } 24 else { 25 Map<String, Object> matchingBeans = findAutowireCandidates(beanName, type, descriptor); 26 if (matchingBeans.isEmpty()) { 27 if (descriptor.isRequired()) { 28 raiseNoSuchBeanDefinitionException(type, "", descriptor); 29 } 30 return null; 31 } 32 if (matchingBeans.size() > 1) { 33 String primaryBeanName = determinePrimaryCandidate(matchingBeans, descriptor); 34 if (primaryBeanName == null) { 35 throw new NoSuchBeanDefinitionException(type, "expected single matching bean but found " +
36 matchingBeans.size() + ": " + matchingBeans.keySet()); 37 } 38 if (autowiredBeanNames != null) { 39 autowiredBeanNames.add(primaryBeanName); 40 } 41 return matchingBeans.get(primaryBeanName); 42 } 43 // We have exactly one match.
44 Map.Entry<String, Object> entry = matchingBeans.entrySet().iterator().next(); 45 if (autowiredBeanNames != null) { 46 autowiredBeanNames.add(entry.getKey()); 47 } 48 return entry.getValue(); 49 } 50 }</pre>

第四行结果是null不看了，为了简化代码Array装配、Collection装配、Map装配的代码都略去了，重点看一下普通属性的装配。首先是第25行获取一下自动装配的候选者：

<pre> 1 protected Map<String, Object> findAutowireCandidates( 2 String beanName, Class requiredType, DependencyDescriptor descriptor) {
3
4 String[] candidateNames = BeanFactoryUtils.beanNamesForTypeIncludingAncestors( 5 this, requiredType, true, descriptor.isEager());
6 Map<String, Object> result = new LinkedHashMap<String, Object>(candidateNames.length);
7 for (Class autowiringType : this.resolvableDependencies.keySet()) {
8 if (autowiringType.isAssignableFrom(requiredType)) { 9 Object autowiringValue = this.resolvableDependencies.get(autowiringType); 10 autowiringValue = AutowireUtils.resolveAutowiringValue(autowiringValue, requiredType); 11 if (requiredType.isInstance(autowiringValue)) { 12 result.put(ObjectUtils.identityToString(autowiringValue), autowiringValue); 13 break; 14 } 15 } 16 } 17 for (String candidateName : candidateNames) { 18 if (!candidateName.equals(beanName) && isAutowireCandidate(candidateName, descriptor)) { 19 result.put(candidateName, getBean(candidateName)); 20 } 21 } 22 return result; 23 }</pre>

代码逻辑整理一下：

• 首先获取候选者bean名称，通过DefaultListableBeanFactory的getBeanNamesForType方法，即找一下所有的Bean定义中指定Type的实现类或者子类
• 接着第7行~第16行的判断要自动装配的类型是不是要自动装配的纠正类型，这个在【Spring源码分析】非懒加载的单例Bean初始化前后的一些操作一文讲PrepareBeanFactory方法的时候有讲过，如果要自动装配的类型是纠正类型，比如是一个ResourceLoader，那么就会为该类型生成一个代理实例，具体可以看一下第10行的AutowireUtils.resolveAutowiringValue方法的实现
• 正常来说都是执行的第17行~第21行的代码，逐个判断查找一下beanName对应的BeanDefinition，判断一下是不是自动装配候选者，默认都是的，如果<bean>的autowire-candidate属性设置为false就不是

这样，拿到所有待装配对象的实现类或者子类的候选者，组成一个Map，Key为beanName，Value为具体的Bean。接着回看获取Bean之后的逻辑：

<pre> 1 Map<String, Object> matchingBeans = findAutowireCandidates(beanName, type, descriptor); 2 if (matchingBeans.isEmpty()) { 3 if (descriptor.isRequired()) { 4 raiseNoSuchBeanDefinitionException(type, "", descriptor);
5 }
6 return null;
7 }
8 if (matchingBeans.size() > 1) {
9 String primaryBeanName = determinePrimaryCandidate(matchingBeans, descriptor); 10 if (primaryBeanName == null) { 11 throw new NoSuchBeanDefinitionException(type, "expected single matching bean but found " +
12 matchingBeans.size() + ": " + matchingBeans.keySet()); 13 } 14 if (autowiredBeanNames != null) { 15 autowiredBeanNames.add(primaryBeanName); 16 } 17 return matchingBeans.get(primaryBeanName); 18 } 19 // We have exactly one match.
20 Map.Entry<String, Object> entry = matchingBeans.entrySet().iterator().next(); 21 if (autowiredBeanNames != null) { 22 autowiredBeanNames.add(entry.getKey()); 23 } 24 ... 25 }</pre>

整理一下逻辑：

• 如果拿到的Map是空的且属性必须注入，抛异常
• 如果拿到的Map中有多个候选对象，判断其中是否有<bean>中属性配置为"primary=true"的，有就拿执行第13行~第15行的代码，没有就第8行的方法返回null，抛异常，这个异常的描述相信Spring用的比较多的应该比较熟悉
• 如果拿到的Map中只有一个候选对象，直接拿到那个

通过这样一整个流程，实现了byType自动装配，byType自动装配流程比较长，中间细节比较多，还需要多看看才能弄明白。

最后注意一点，即所有待注入的PropertyName-->PropertyValue映射拿到之后都只是放在MutablePropertyValues中，最后由AbstractPropertyAccessor类的setPropertyValues方法遍历并进行逐一注入。

通过FactoryBean获取Bean实例源码实现

我们知道可以通过实现FactoryBean接口，重写getObject()方法实现个性化定制Bean的过程，这部分我们就来看一下Spring源码是如何实现通过FactoryBean获取Bean实例的。代码直接定位到AbstractBeanFactory的doGetBean方法创建单例Bean这部分：

<pre> 1 // Create bean instance.
2 if (mbd.isSingleton()) { 3 sharedInstance = getSingleton(beanName, new ObjectFactory() { 4 public Object getObject() throws BeansException { 5 try { 6 return createBean(beanName, mbd, args); 7 }
8 catch (BeansException ex) { 9 // Explicitly remove instance from singleton cache: It might have been put there 10 // eagerly by the creation process, to allow for circular reference resolution. 11 // Also remove any beans that received a temporary reference to the bean.
12 destroySingleton(beanName); 13 throw ex; 14 } 15 } 16 }); 17 bean = getObjectForBeanInstance(sharedInstance, name, beanName, mbd); 18 }</pre>

FactoryBean首先是个Bean且被实例化出来成为一个对象之后才能调用getObject()方法，因此还是会执行第3行~第16行的代码，这段代码之前分析过了就不说了。之后执行第17行的方法：

<pre> 1 protected Object getObjectForBeanInstance( 2 Object beanInstance, String name, String beanName, RootBeanDefinition mbd) {
3
4 // Don't let calling code try to dereference the factory if the bean isn't a factory.
5 if (BeanFactoryUtils.isFactoryDereference(name) && !(beanInstance instanceof FactoryBean)) { 6 throw new BeanIsNotAFactoryException(transformedBeanName(name), beanInstance.getClass()); 7 }
8
9 // Now we have the bean instance, which may be a normal bean or a FactoryBean. 10 // If it's a FactoryBean, we use it to create a bean instance, unless the 11 // caller actually wants a reference to the factory.
12 if (!(beanInstance instanceof FactoryBean) || BeanFactoryUtils.isFactoryDereference(name)) { 13 return beanInstance; 14 } 15
16 Object object = null; 17 if (mbd == null) { 18 object = getCachedObjectForFactoryBean(beanName); 19 } 20 if (object == null) { 21 // Return bean instance from factory.
22 FactoryBean factory = (FactoryBean) beanInstance; 23 // Caches object obtained from FactoryBean if it is a singleton.
24 if (mbd == null && containsBeanDefinition(beanName)) { 25 mbd = getMergedLocalBeanDefinition(beanName); 26 } 27 boolean synthetic = (mbd != null && mbd.isSynthetic()); 28 object = getObjectFromFactoryBean(factory, beanName, !synthetic); 29 } 30 return object; 31 }</pre>

首先第5行~第7行判断一下是否beanName以"&"开头并且不是FactoryBean的实现类，不满足则抛异常，因为beanName以"&"开头是FactoryBean的实现类bean定义的一个特征。

接着判断第12行~第14行，如果：

• bean不是FactoryBean的实现类
• beanName以"&"开头

这两种情况，都直接把生成的bean对象返回出去，不会执行余下的流程。

最后流程走到第16行~第30行，最终调用getObject()方法实现个性化定制bean，先执行第28行的方法：

<pre> 1 protected Object getObjectFromFactoryBean(FactoryBean factory, String beanName, boolean shouldPostProcess) { 2 if (factory.isSingleton() && containsSingleton(beanName)) { 3 synchronized (getSingletonMutex()) { 4 Object object = this.factoryBeanObjectCache.get(beanName);
5 if (object == null) {
6 object = doGetObjectFromFactoryBean(factory, beanName, shouldPostProcess); 7 this.factoryBeanObjectCache.put(beanName, (object != null ? object : NULL_OBJECT)); 8 }
9 return (object != NULL_OBJECT ? object : null); 10 } 11 } 12 else { 13 return doGetObjectFromFactoryBean(factory, beanName, shouldPostProcess); 14 } 15 }</pre>

第1行_{第11行的代码与第12行}第13行的代码最终都是一样的，调用了如下一段：

<pre> 1 private Object doGetObjectFromFactoryBean( 2 final FactoryBean factory, final String beanName, final boolean shouldPostProcess) 3 throws BeanCreationException { 4
5 Object object;
6 try { 7 if (System.getSecurityManager() != null) {
8 AccessControlContext acc = getAccessControlContext(); 9 try { 10 object = AccessController.doPrivileged(new PrivilegedExceptionAction<Object>() { 11 public Object run() throws Exception { 12 return factory.getObject(); 13 } 14 }, acc); 15 } 16 catch (PrivilegedActionException pae) { 17 throw pae.getException(); 18 } 19 } 20 else { 21 object = factory.getObject(); 22 } 23 } 24 catch (FactoryBeanNotInitializedException ex) { 25 throw new BeanCurrentlyInCreationException(beanName, ex.toString()); 26 } 27 catch (Throwable ex) { 28 throw new BeanCreationException(beanName, "FactoryBean threw exception on object creation", ex); 29 } 30
31 // Do not accept a null value for a FactoryBean that's not fully 32 // initialized yet: Many FactoryBeans just return null then.
33 if (object == null && isSingletonCurrentlyInCreation(beanName)) { 34 throw new BeanCurrentlyInCreationException( 35 beanName, "FactoryBean which is currently in creation returned null from getObject"); 36 } 37
38 if (object != null && shouldPostProcess) { 39 try { 40 object = postProcessObjectFromFactoryBean(object, beanName); 41 } 42 catch (Throwable ex) { 43 throw new BeanCreationException(beanName, "Post-processing of the FactoryBean's object failed", ex); 44 } 45 } 46
47 return object; 48 }</pre>

第12行和第21行的代码，都一样，最终调用getObject()方法获取对象。回过头去看之前的getObjectFromFactoryBean方法，虽然if...else...逻辑最终都是调用了以上的方法，但是区别在于：

• 如果FactoryBean接口实现类的isSington方法返回的是true，那么每次调用getObject方法的时候会优先尝试从FactoryBean对象缓存中取目标对象，有就直接拿，没有就创建并放入FactoryBean对象缓存，这样保证了每次单例的FactoryBean调用getObject()方法后最终拿到的目标对象一定是单例的，即在内存中都是同一份
• 如果FactoryBean接口实现类的isSington方法返回的是false，那么每次调用getObject方法的时候都会新创建一个目标对象

微信公众号【黄小斜】作者是蚂蚁金服 JAVA 工程师，专注于 JAVA
后端技术栈：SpringBoot、SSM全家桶、MySQL、分布式、中间件、微服务，同时也懂点投资理财，坚持学习和写作，相信终身学习的力量！关注公众号后回复”架构师“即可领取
Java基础、进阶、项目和架构师等免费学习资料，更有数据库、分布式、微服务等热门技术学习视频，内容丰富，兼顾原理和实践，另外也将赠送作者原创的Java学习指南、Java程序员面试指南等干货资源