Spring源碼剖析4：其他方式獲取Bean的過程分析

<!-- more -->

原型Bean加載過程

以前的文章，分析了非懶加載的單例Bean整個加載過程，除了非懶加載的單例Bean以外，Spring中還有一種Bean就是原型（Prototype）的Bean，看一下定義方式：

<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.or...d">
6
7 <bean id="prototypeBean" class="org.xrq.action.PrototypeBean" scope="prototype" />
8
9 </beans></pre>

原型Bean加載流程總得來講和單例Bean差很少，看一下不一樣之處，在AbstractBeanFactory的doGetBean的方法的這一步：

<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是同樣的，原型Bean實例化的主要區別就在於第6行，它是直接建立bean的，而單例bean咱們再對比一下：

<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>

它優先會嘗試getSington，即先嚐試從singletonObjects中獲取一下bean是否存在，若是存在直接返回singletonObjects中的bean對象。

接着，咱們看到原型bean建立和單例bean建立的區別還在於第5行和第9行，先看第5行的代碼：

<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>

這段主要是說bean在建立前要把當前beanName設置到ThreadLocal中去，其目的是保證多線程不會同時建立同一個bean。接着看第9行的代碼實現，即bean建立以後作了什麼：

<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，就是把當前bean移除一下，這樣其它線程就能夠建立bean了。第11行的代碼不看了，意思是若是bean是FactoryBean的實現類的話，調用getObject()方法獲取真正的對象。

byName源碼實現

Spring有爲開發者提供Autowire（自動裝配）的功能，自動裝配最經常使用的就是byName和byType這兩種屬性。因爲自動裝配是爲了解決對象注入致使的<property>過多的問題，所以很容易找到byName與byType的Spring源碼實現應該在屬性注入這一塊，定位到屬性注入的代碼AbstractAutowireCapableBeanFactory的populateBean方法，直接截取重點：

<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行判斷是否byName形式，是就執行byName自動裝配代碼；第11行~第13行判斷是否byType形式，是就執行byType自動裝配代碼。那麼首先看一下第7行的byName代碼實現：

<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>

篇幅問題，代碼不一層層跟了，邏輯梳理一下：

• 第4行，找到Bean中不是簡單屬性的屬性，這句話有點繞，意思就是找到屬性是對象類型的屬性，但也不是全部的對象類型都會被找到，好比CharSequence類型、Number類型、Date類型、URL類型、URI類型、Locale類型、Class類型就會忽略，具體可見BeanUtils的isSimpleProperty方法
• 第5行~第7行，遍歷全部被找到的屬性，若是bean定義中包含了屬性名，那麼先實例化該屬性名對應的bean
• 第9行registerDependentBean，註冊一下當前bean的依賴bean，用於在某個bean被銷燬前先將其依賴的bean銷燬

其他代碼都是一些打日誌的，沒什麼好說的。

byType源碼實現

上面說了byName的源碼實現，接下來看一下byType源碼實現：

<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>

前面同樣，到第10行都是找到Bean中屬性是對象類型的屬性。

接着就是遍歷一下PropertyName，獲取PropertyName對應的屬性描述，注意一下16行的判斷及其對應的註釋：不要嘗試自動裝配Object類型，這沒有任何意義，即便從技術角度看它是一個非簡單的對象屬性。

第18行~第20行跳過（沒有太明白是幹什麼的），byType實現的源碼主要在第21行的方法resolveDependency中，這個方法是AbstractAutowireCapableBeanFactory類的實現類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>

這裏判斷一下要自動裝配的屬性是ObjectFactory.class仍是javaxInjectProviderClass仍是其餘的，咱們裝配的是其餘的，看一下12行的代碼實現：

<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>

第四行結果是null不看了，爲了簡化代碼Array裝配、Collection裝配、Map裝配的代碼都略去了，重點看一下普通屬性的裝配。首先是第25行獲取一下自動裝配的候選者：

<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>

代碼邏輯整理一下：

• 首先獲取候選者bean名稱，經過DefaultListableBeanFactory的getBeanNamesForType方法，即找一下全部的Bean定義中指定Type的實現類或者子類
• 接着第7行~第16行的判斷要自動裝配的類型是否是要自動裝配的糾正類型，這個在【Spring源碼分析】非懶加載的單例Bean初始化先後的一些操做一文講PrepareBeanFactory方法的時候有講過，若是要自動裝配的類型是糾正類型，好比是一個ResourceLoader，那麼就會爲該類型生成一個代理實例，具體能夠看一下第10行的AutowireUtils.resolveAutowiringValue方法的實現
• 正常來講都是執行的第17行~第21行的代碼，逐個判斷查找一下beanName對應的BeanDefinition，判斷一下是否是自動裝配候選者，默認都是的，若是<bean>的autowire-candidate屬性設置爲false就不是

這樣，拿到全部待裝配對象的實現類或者子類的候選者，組成一個Map，Key爲beanName，Value爲具體的Bean。接着回看獲取Bean以後的邏輯：

<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>

整理一下邏輯：

• 若是拿到的Map是空的且屬性必須注入，拋異常
• 若是拿到的Map中有多個候選對象，判斷其中是否有<bean>中屬性配置爲"primary=true"的，有就拿執行第13行~第15行的代碼，沒有就第8行的方法返回null，拋異常，這個異常的描述相信Spring用的比較多的應該比較熟悉
• 若是拿到的Map中只有一個候選對象，直接拿到那個

經過這樣一整個流程，實現了byType自動裝配，byType自動裝配流程比較長，中間細節比較多，還須要多看看才能弄明白。

最後注意一點，即全部待注入的PropertyName-->PropertyValue映射拿到以後都只是放在MutablePropertyValues中，最後由AbstractPropertyAccessor類的setPropertyValues方法遍歷並進行逐一注入。

經過FactoryBean獲取Bean實例源碼實現

咱們知道能夠經過實現FactoryBean接口，重寫getObject()方法實現個性化定製Bean的過程，這部分咱們就來看一下Spring源碼是如何實現經過FactoryBean獲取Bean實例的。代碼直接定位到AbstractBeanFactory的doGetBean方法建立單例Bean這部分：

<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首先是個Bean且被實例化出來成爲一個對象以後才能調用getObject()方法，所以仍是會執行第3行~第16行的代碼，這段代碼以前分析過了就不說了。以後執行第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行判斷一下是否beanName以"&"開頭而且不是FactoryBean的實現類，不知足則拋異常，由於beanName以"&"開頭是FactoryBean的實現類bean定義的一個特徵。

接着判斷第12行~第14行，若是：

• bean不是FactoryBean的實現類
• beanName以"&"開頭

這兩種狀況，都直接把生成的bean對象返回出去，不會執行餘下的流程。

最後流程走到第16行~第30行，最終調用getObject()方法實現個性化定製bean，先執行第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行的代碼與第12行~第13行的代碼最終都是同樣的，調用了以下一段：

<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行的代碼，都同樣，最終調用getObject()方法獲取對象。回過頭去看以前的getObjectFromFactoryBean方法，雖然if...else...邏輯最終都是調用了以上的方法，可是區別在於：

• 若是FactoryBean接口實現類的isSington方法返回的是true，那麼每次調用getObject方法的時候會優先嚐試從FactoryBean對象緩存中取目標對象，有就直接拿，沒有就建立並放入FactoryBean對象緩存，這樣保證了每次單例的FactoryBean調用getObject()方法後最終拿到的目標對象必定是單例的，即在內存中都是同一份
• 若是FactoryBean接口實現類的isSington方法返回的是false，那麼每次調用getObject方法的時候都會新建立一個目標對象

微信公衆號【黃小斜】做者是螞蟻金服 JAVA 工程師，專一於 JAVA 後端技術棧：SpringBoot、SSM全家桶、MySQL、分佈式、中間件、微服務，同時也懂點投資理財，堅持學習和寫做，相信終身學習的力量！關注公衆號後回覆」架構師「便可領取 Java基礎、進階、項目和架構師等免費學習資料，更有數據庫、分佈式、微服務等熱門技術學習視頻，內容豐富，兼顧原理和實踐，另外也將贈送做者原創的Java學習指南、Java程序員面試指南等乾貨資源