Spring框架是如何解决Bean创建过程中的循环依赖问题的

引言

本文主要梳理了Spring框架Bean创建过程中应对循环依赖问题的相关源码。我在手写super-mini-webpackhttps://blog.csdn.net/q1424966670/article/details/135628531?spm=1001.2014.3001.5502的时候也介绍过解决循环依赖的算法：Map+记忆化搜索。可以猜测这段源码也实现了这个算法，所以在看这段源码的时候，我们可以先找到递归点，再去分析调用栈涉及的那些函数，顺便找出其用到的Map数据结构。

本文所用工程https://github.com/Hans774882968/spring-source-code-learn，工程创建方式：用VSCode插件Spring Initializr Java Support，ctrl + shift + P，Spring Initializr: Create a Maven Project。具体看参考链接1https://www.cnblogs.com/miskis/p/9816135.html。

• JDK Version：17

• SpringBoot Version：3.1.2

• Spring Version：6.0.11

三级缓存数据结构简介

三级缓存数据结构定义和操作三级缓存的函数都位于：spring-beans/src/main/java/org/springframework/beans/factory/support/DefaultSingletonBeanRegistry.java

    /** Cache of singleton objects: bean name to bean instance. */
    private final Map singletonObjects = new ConcurrentHashMap<>(256);

    /** Cache of singleton factories: bean name to ObjectFactory. */
    private final Map> singletonFactories = new HashMap<>(16);

    /** Cache of early singleton objects: bean name to bean instance. */
    private final Map earlySingletonObjects = new ConcurrentHashMap<>(16);

缓存先找到第1级，第1级没有才查第2级，依此类推。singletonObjects, earlySingletonObjects, singletonFactories分别是1到3级。第1级缓存是bean名到成品bean的映射；第2级缓存是bean名到半成品bean的映射；第3级缓存是bean名到函数式接口的映射，作用为延迟调用函数。

@FunctionalInterface
public interface ObjectFactory {

    /**
     * Return an instance (possibly shared or independent)
     * of the object managed by this factory.
     * @return the resulting instance
     * @throws BeansException in case of creation errors
     */
    T getObject() throws BeansException;

}

第3级缓存的Value里的函数的调用方式就是调用.getObject()。

为什么需要2级缓存？因为要体现一个分层的思想，半成品bean原则上是不能暴露到外部的。TODO：外部是指？这里我没研究清楚，就先引用了52pojie@特别的你～大佬的解释。“防止线程切换时，其他线程取到半成品bean”，我觉得很有道理。

外部可能是调用spring容器中BEAN的其他参与者？有些应用启动时（如果懒加载将会是使用时）极其复杂，如果多线程调用将会出现问题（2级缓存加锁是防止多线程创建，这里指该BEAN如果没有创建好，而且只有一个容器，那么其他线程就会使用到，认为是完整的BEAN）。

为什么需要第3级缓存？如果所有bean都没有代理对象就不需要第3级缓存。TODO：补充说明。

源码阅读

如何启动调试？

新建一个普通的SpringBoot项目。然后可以尝试通过xml配置文件和注解等方式构造循环引用。后文称为场景1和场景2。

场景1：通过xml配置文件来构造循环引用

入口src\main\java\com\hans\bean_dependency_cycle\HansApplication.java：

package com.hans.bean_dependency_cycle;

import org.springframework.boot.autoconfigure.SpringBootApplication;
import org.springframework.context.support.AbstractApplicationContext;
import org.springframework.context.support.ClassPathXmlApplicationContext;

@SpringBootApplication
public class HansApplication {

    public static void main(String[] args) {
        AbstractApplicationContext ac = new ClassPathXmlApplicationContext("cycle.xml");
        A beanA = ac.getBean(A.class);
        System.out.println(beanA);
        System.out.println(beanA.getB());
        B beanB = ac.getBean(B.class);
        System.out.println(beanB);
        System.out.println(beanB.getA());
        ac.close();
    }

}

src\main\resources\cycle.xml：

A.java和B.java：

// A 和 B 不能都使用 lombok，否则无法打印。这里选择了 B 不使用 lombok
package com.hans.bean_dependency_cycle;

import lombok.Data;

@Data
public class A {
    private B b;

    public String toString(A o) {
        return o.getClass().getName() + "@" +
                Integer.toHexString(System.identityHashCode(o));
    }

}

package com.hans.bean_dependency_cycle;

public class B {
    private A a;

    public A getA() {
        return this.a;
    }

    public void setA(A a) {
        this.a = a;
    }

}

期望输出：

A(b=com.hans.bean_dependency_cycle.B@750e2b97)
com.hans.bean_dependency_cycle.B@750e2b97
com.hans.bean_dependency_cycle.B@750e2b97
A(b=com.hans.bean_dependency_cycle.B@750e2b97)

场景1不需要配置spring.main.allow-circular-references为true也能得到期望输出，TODO：原因未知。

【Optional】顺便补充一下场景1的单测src\test\java\com\hans\bean_dependency_cycleApplicationTests.java：

package com.hans.bean_dependency_cycle;

import org.junit.jupiter.api.Assertions;
import org.junit.jupiter.api.Test;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.boot.test.context.SpringBootTest;
import org.springframework.context.ApplicationContext;
import org.springframework.test.context.ContextConfiguration;

// 自动收集 cycle.xml 的 bean
@ContextConfiguration(locations = { "classpath:cycle.xml" })
@SpringBootTest(classes = HansApplication.class)
class HansApplicationTests {
    @Autowired
    ApplicationContext ac;

    @Test
    void contextLoads() {
        A beanA = ac.getBean(A.class);
        Assertions.assertNotNull(beanA);
        B beanB = ac.getBean(B.class);
        Assertions.assertNotNull(beanB);
        Assertions.assertEquals(beanA, beanB.getA());
        Assertions.assertEquals(beanB, beanA.getB());
    }

}

场景2：通过注解+自动装配属性来构造循环引用：以@Controller为例

入口src\main\java\com\hans\bean_dependency_cycle\AnotherEntry.java：

package com.hans.bean_dependency_cycle;

import org.springframework.boot.SpringApplication;
import org.springframework.boot.autoconfigure.SpringBootApplication;
import org.springframework.boot.autoconfigure.jdbc.DataSourceAutoConfiguration;
import org.springframework.context.ConfigurableApplicationContext;

@SpringBootApplication(exclude = { DataSourceAutoConfiguration.class })
public class AnotherEntry {
    public static void main(String[] args) {
        ConfigurableApplicationContext cac = SpringApplication.run(AnotherEntry.class, args);
        ControllerA beanA = cac.getBean(ControllerA.class);
        System.out.println(beanA);
        System.out.println(beanA.getCb());
        ControllerB beanB = cac.getBean(ControllerB.class);
        System.out.println(beanB);
        System.out.println(beanB.getCa());
    }
}

根据参考链接2https://juejin.cn/post/7096798740593246222，接下来一定要记得修改application.yml：

spring:
  main:
    allow-circular-references: true

否则会报错：

***************************
APPLICATION FAILED TO START
***************************

Description:

The dependencies of some of the beans in the application context form a cycle:

┌─────┐
|  controllerA (field private com.hans.bean_dependency_cycle.ControllerB com.hansn_dependency_cycle.hans.ControllerA.cb)
↑     ↓
|  controllerB (field private com.hans.bean_dependency_cycle.ControllerA com.hansn_dependency_cycle.hans.ControllerB.ca)
└─────┘

Action:

Relying upon circular references is discouraged and they are prohibited by default. Up your application to remove the dependency cycle between beans. As a last resort, it me possible to break the cycle automatically by setting spring.main.allow-circular-refees to true.

两个普通的Controller：

// src\main\java\com\hans\bean_dependency_cycle\ControllerA.java
package com.hans.bean_dependency_cycle;

import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.web.bind.annotation.RequestMapping;
import org.springframework.web.bind.annotation.RestController;

@RestController
public class ControllerA {
    @Autowired
    private ControllerB cb;

    public ControllerB getCb() {
        return cb;
    }

    @RequestMapping("/controllerA")
    public String index() {
        return "hello controllerA!";
    }
}

// src\main\java\com\hans\bean_dependency_cycle\ControllerB.java
package com.hans.bean_dependency_cycle;

import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.web.bind.annotation.RequestMapping;
import org.springframework.web.bind.annotation.RestController;

@RestController
public class ControllerB {
    @Autowired
    private ControllerA ca;

    public ControllerA getCa() {
        return ca;
    }

    @RequestMapping("/controllerB")
    public String index() {
        return "hello controllerB!";
    }
}

期望输出：

com.hans.bean_dependency_cycle.ControllerA@39f82681
com.hans.bean_dependency_cycle.ControllerB@4bd9e7fd
com.hans.bean_dependency_cycle.ControllerB@4bd9e7fd
com.hans.bean_dependency_cycle.ControllerA@39f82681

【Optional】顺便补充一下场景2的单测src\test\java\com\hans\bean_dependency_cycle\AnotherEntryTests.java

package com.hans.bean_dependency_cycle;

import org.junit.jupiter.api.Assertions;
import org.junit.jupiter.api.Test;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.boot.test.context.SpringBootTest;

@SpringBootTest(classes = AnotherEntry.class)
class AnotherEntryTests {
    @Autowired
    ControllerA ca;

    @Autowired
    ControllerB cb;

    @Test
    void contextLoads() {
        Assertions.assertNotNull(ca);
        Assertions.assertNotNull(cb);
        Assertions.assertEquals(ca, cb.getCa());
        Assertions.assertEquals(cb, ca.getCb());
    }

}

脉络

根据参考链接3，有一个经验：“do”开头的方法名是真正含有大量逻辑的方法。参考链接3Java之父精选的脉络函数如下：

• getBean
• doGetBean
• createBean
• doCreateBean
• createBeanInstance
• populateBean

场景1到递归点为止的调用链：preInstantiateSingletons > getBean > doGetBean > getSingleton函数有多个，其中调用了beforeSingletonCreation的函数调用singletonFactory.getObject()时才真正调用了createBean > createBean > doCreateBean > createBeanInstance, populateBean 都在 doCreateBean 的实现里 > populateBean调用了applyPropertyValues > resolveValueIfNecessary > resolveReference > getBean。

场景2的调用链到populateBean开始和场景1的调用链岔开，两个场景的差异放在后文《Controller和普通Bean解决循环依赖过程的相同点与不同点》分析。

场景1递归到A -> B -> A时的调用链：getBean > doGetBean > 未调用beforeSingletonCreation的getSingleton，操作第2级缓存后离开。

操作三级缓存的函数都位于https://github.com/spring-projects/spring-framework/blob/502997d8e986dcfde1f49b2b2f443a32b5488b13/spring-beans/src/main/java/org/springframework/beans/factory/support/DefaultSingletonBeanRegistry.java：

1. 放入第3级缓存的函数：doCreateBean调用的addSingletonFactory。
2. 放入第2级缓存的函数：首先要知道getSingleton在DefaultSingletonBeanRegistry.java里本质上的实现有两个，一个public Object getSingleton(String beanName, ObjectFactory singletonFactory)调用了beforeSingletonCreation并间接调用了createBean；另一个protected Object getSingleton(String beanName, boolean allowEarlyReference)则是操作了第2级缓存。
3. 放入第1级缓存的函数：addSingleton。在间接调用了createBean函数的getSingleton处调用。

场景1的执行过程

真正意义上的入口：spring-context/src/main/java/org/springframework/context/support/AbstractApplicationContext.javahttps://github.com/spring-projects/spring-framework/blob/3a481a7d7f7ee485e25003b2855b4379ab5cce3a/spring-context/src/main/java/org/springframework/context/support/AbstractApplicationContext.java的beanFactory.preInstantiateSingletons();。这里的beanFactory就有singletonObjects那3级缓存的对象。于是跳到https://github.com/spring-projects/spring-framework/blob/bbde68c49e66c3c531920cb80a55742262507be7/spring-beans/src/main/java/org/springframework/beans/factory/support/DefaultListableBeanFactory.java：

    @Override
    public void preInstantiateSingletons() throws BeansException {
        if (logger.isTraceEnabled()) {
            logger.trace("Pre-instantiating singletons in " + this);
        }

        // Iterate over a copy to allow for init methods which in turn register new bean definitions.
        // While this may not be part of the regular factory bootstrap, it does otherwise work fine.
        List beanNames = new ArrayList<>(this.beanDefinitionNames);

        // 触发所有非延迟加载的（non-lazy）单例 bean 的初始化
        // Trigger initialization of all non-lazy singleton beans...
        for (String beanName : beanNames) {
            // 假如先遍历 A 再遍历 B 那么遍历到 B 的时候，因为循环引用解决的关系，B 已经放到了第1级缓存，所以 doGetBean 的 getSingleton 可以直接从第1级缓存取到值，不用再走一遍 createBean 方法
            // 合并父类 BeanDefinition
            RootBeanDefinition bd = getMergedLocalBeanDefinition(beanName);
            // 非抽象、是单例、非懒加载
            if (!bd.isAbstract() && bd.isSingleton() && !bd.isLazyInit()) {
                // 如果实现了 FactoryBean 接口则是 FactoryBean
                if (isFactoryBean(beanName)) {
                    Object bean = getBean(FACTORY_BEAN_PREFIX + beanName); // 比如：FACTORY_BEAN_PREFIX + beanName = "&A"
                    if (bean instanceof SmartFactoryBean smartFactoryBean && smartFactoryBean.isEagerInit()) {
                        getBean(beanName);
                    }
                }
                else {
                    // 不是 FactoryBean，只是普通 Bean，则走这个分支
                    getBean(beanName);
                }
            }
        }

        // 触发所有 SmartInitializingSingleton 的后初始化回调
        // Trigger post-initialization callback for all applicable beans...
        for (String beanName : beanNames) {
            Object singletonInstance = getSingleton(beanName);
            if (singletonInstance instanceof SmartInitializingSingleton smartSingleton) {
                StartupStep smartInitialize = getApplicationStartup().start("spring.beans.smart-initialize")
                        .tag("beanName", beanName);
                smartSingleton.afterSingletonsInstantiated();
                smartInitialize.end();
            }
        }
    }

顺便看下isFactoryBean的实现https://github.com/spring-projects/spring-framework/blob/4786e2bf53a3f882c10e25d7ff79a18ff47b5e51/spring-beans/src/main/java/org/springframework/beans/factory/support/AbstractBeanFactory.java：

    @Override
    public boolean isFactoryBean(String name) throws NoSuchBeanDefinitionException {
        String beanName = transformedBeanName(name);
        Object beanInstance = getSingleton(beanName, false);
        if (beanInstance != null) {
            return (beanInstance instanceof FactoryBean);
        }
        // No singleton instance found -> check bean definition.
        if (!containsBeanDefinition(beanName) && getParentBeanFactory() instanceof ConfigurableBeanFactory cbf) {
            // No bean definition found in this factory -> delegate to parent.
            return cbf.isFactoryBean(name);
        }
        return isFactoryBean(beanName, getMergedLocalBeanDefinition(beanName));
    }

getBean和isFactoryBean都位于https://github.com/spring-projects/spring-framework/blob/4786e2bf53a3f882c10e25d7ff79a18ff47b5e51/spring-beans/src/main/java/org/springframework/beans/factory/support/AbstractBeanFactory.java，getBean只有1行

    @Override
    public Object getBean(String name) throws BeansException {
        return doGetBean(name, null, null, false);
    }

doGetBean和getBean, isFactoryBean都在AbstractBeanFactory.java：

    /**
     * Return an instance, which may be shared or independent, of the specified bean.
     * @param name the name of the bean to retrieve
     * @param requiredType the required type of the bean to retrieve
     * @param args arguments to use when creating a bean instance using explicit arguments
     * (only applied when creating a new instance as opposed to retrieving an existing one)
     * @param typeCheckOnly whether the instance is obtained for a type check,
     * not for actual use
     * @return an instance of the bean
     * @throws BeansException if the bean could not be created
     */
    @SuppressWarnings("unchecked")
    protected  T doGetBean(
            String name, @Nullable Class requiredType, @Nullable Object[] args, boolean typeCheckOnly)
            throws BeansException {

        String beanName = transformedBeanName(name);
        Object beanInstance;

        // 提前检查单例缓存中是否有手动注册的单例对象，跟循环依赖有关。如果是初次进这里，比如 controllerA 初次进这个方法，肯定是拿不到值的，就会是 null
        // 对于最简单的循环依赖，controllerA -> controllerB -> controllerA 之后，需要进入 getSingleton 了，这里的逻辑就是要从第3级缓存里拿到工厂函数，调用后得到 controllerA 半成品，从而可以直接 return
        // Eagerly check singleton cache for manually registered singletons.
        Object sharedInstance = getSingleton(beanName);
        if (sharedInstance != null && args == null) {
            if (logger.isTraceEnabled()) {
                if (isSingletonCurrentlyInCreation(beanName)) {
                    logger.trace("Returning eagerly cached instance of singleton bean '" + beanName +
                            "' that is not fully initialized yet - a consequence of a circular reference");
                }
                else {
                    logger.trace("Returning cached instance of singleton bean '" + beanName + "'");
                }
            }
            beanInstance = getObjectForBeanInstance(sharedInstance, name, beanName, null);
        }

        else {
            // Fail if we're already creating this bean instance:
            // We're assumably within a circular reference.
            if (isPrototypeCurrentlyInCreation(beanName)) {
                throw new BeanCurrentlyInCreationException(beanName);
            }

            // Check if bean definition exists in this factory.
            BeanFactory parentBeanFactory = getParentBeanFactory();
            if (parentBeanFactory != null && !containsBeanDefinition(beanName)) {
                // Not found -> check parent.
                String nameToLookup = originalBeanName(name);
                if (parentBeanFactory instanceof AbstractBeanFactory abf) {
                    return abf.doGetBean(nameToLookup, requiredType, args, typeCheckOnly);
                }
                else if (args != null) {
                    // Delegation to parent with explicit args.
                    return (T) parentBeanFactory.getBean(nameToLookup, args);
                }
                else if (requiredType != null) {
                    // No args -> delegate to standard getBean method.
                    return parentBeanFactory.getBean(nameToLookup, requiredType);
                }
                else {
                    return (T) parentBeanFactory.getBean(nameToLookup);
                }
            }

            if (!typeCheckOnly) {
                markBeanAsCreated(beanName);
            }

            StartupStep beanCreation = this.applicationStartup.start("spring.beans.instantiate")
                    .tag("beanName", name);
            try {
                if (requiredType != null) {
                    beanCreation.tag("beanType", requiredType::toString);
                }
                RootBeanDefinition mbd = getMergedLocalBeanDefinition(beanName);
                checkMergedBeanDefinition(mbd, beanName, args);

                // Guarantee initialization of beans that the current bean depends on.
                String[] dependsOn = mbd.getDependsOn();
                if (dependsOn != null) {
                    for (String dep : dependsOn) {
                        if (isDependent(beanName, dep)) {
                            throw new BeanCreationException(mbd.getResourceDescription(), beanName,
                                    "Circular depends-on relationship between '" + beanName + "' and '" + dep + "'");
                        }
                        registerDependentBean(dep, beanName);
                        try {
                            getBean(dep);
                        }
                        catch (NoSuchBeanDefinitionException ex) {
                            throw new BeanCreationException(mbd.getResourceDescription(), beanName,
                                    "'" + beanName + "' depends on missing bean '" + dep + "'", ex);
                        }
                    }
                }

                // Create bean instance.
                if (mbd.isSingleton()) {
                    // 返回 beanName 的原始单例对象。如果尚未注册，则使用 singletonFactory 创建并注册一个对象
                    sharedInstance = getSingleton(beanName, () -> {
                        try {
                            // 为给定的合并后的 BeanDefinition 和参数创建一个 bean 实例
                            // createBean 真实执行时机是调用了 beforeSingletonCreation 方法的 getSingleton 方法执行 singletonObject = singletonFactory.getObject() 时
                            // 首次运行到 beanName = "controllerA" 时 args = null
                            return createBean(beanName, mbd, args);
                        }
                        catch (BeansException ex) {
                            // Explicitly remove instance from singleton cache: It might have been put there
                            // eagerly by the creation process, to allow for circular reference resolution.
                            // Also remove any beans that received a temporary reference to the bean.
                            // 显式地从单例缓存中删除bean实例：因为这个实例可能是由创建过程急切地放在那里的，放在那里的目的是允许循环引用解析。
                            // 还要删除所有被这个bean临时引用的所有bean。如果找到相应的一次性bean实例，则委托给 destroyBean
                            destroySingleton(beanName);
                            throw ex;
                        }
                    });
                    beanInstance = getObjectForBeanInstance(sharedInstance, name, beanName, mbd);
                }

                else if (mbd.isPrototype()) {
                    // It's a prototype -> create a new instance.
                    Object prototypeInstance = null;
                    try {
                        beforePrototypeCreation(beanName);
                        prototypeInstance = createBean(beanName, mbd, args);
                    }
                    finally {
                        afterPrototypeCreation(beanName);
                    }
                    beanInstance = getObjectForBeanInstance(prototypeInstance, name, beanName, mbd);
                }

                else {
                    String scopeName = mbd.getScope();
                    if (!StringUtils.hasLength(scopeName)) {
                        throw new IllegalStateException("No scope name defined for bean '" + beanName + "'");
                    }
                    Scope scope = this.scopes.get(scopeName);
                    if (scope == null) {
                        throw new IllegalStateException("No Scope registered for scope name '" + scopeName + "'");
                    }
                    try {
                        Object scopedInstance = scope.get(beanName, () -> {
                            beforePrototypeCreation(beanName);
                            try {
                                return createBean(beanName, mbd, args);
                            }
                            finally {
                                afterPrototypeCreation(beanName);
                            }
                        });
                        beanInstance = getObjectForBeanInstance(scopedInstance, name, beanName, mbd);
                    }
                    catch (IllegalStateException ex) {
                        throw new ScopeNotActiveException(beanName, scopeName, ex);
                    }
                }
            }
            catch (BeansException ex) {
                beanCreation.tag("exception", ex.getClass().toString());
                beanCreation.tag("message", String.valueOf(ex.getMessage()));
                cleanupAfterBeanCreationFailure(beanName);
                throw ex;
            }
            finally {
                beanCreation.end();
            }
        }

        return adaptBeanInstance(name, beanInstance, requiredType);
    }

我们先只关注createBean，所以需要关注getSingleton。注意：

getSingleton的函数在DefaultSingletonBeanRegistry.java里本质上的实现有两个，一个调用了beforeSingletonCreation并间接调用了createBean；另一个则是操作了第2级缓存。

所以目前我们需要关注的是调用了beforeSingletonCreation的getSingleton方法。路径：https://github.com/spring-projects/spring-framework/blob/502997d8e986dcfde1f49b2b2f443a32b5488b13/spring-beans/src/main/java/org/springframework/beans/factory/support/DefaultSingletonBeanRegistry.java

    /**
     * Return the (raw) singleton object registered under the given name,
     * creating and registering a new one if none registered yet.
     * @param beanName the name of the bean
     * @param singletonFactory the ObjectFactory to lazily create the singleton
     * with, if necessary
     * @return the registered singleton object
     */
    public Object getSingleton(String beanName, ObjectFactory singletonFactory) {
        Assert.notNull(beanName, "Bean name must not be null");
        synchronized (this.singletonObjects) {
            Object singletonObject = this.singletonObjects.get(beanName);
            if (singletonObject == null) {
                if (this.singletonsCurrentlyInDestruction) {
                    throw new BeanCreationNotAllowedException(beanName,
                            "Singleton bean creation not allowed while singletons of this factory are in destruction " +
                            "(Do not request a bean from a BeanFactory in a destroy method implementation!)");
                }
                if (logger.isDebugEnabled()) {
                    logger.debug("Creating shared instance of singleton bean '" + beanName + "'");
                }
                // 创建单例前的回调，默认实现为：将单例注册为当前正在创建中，实现只有3行，可以看下。
                beforeSingletonCreation(beanName);
                // flag 表示是否生成了新的单例对象
                boolean newSingleton = false;
                // flag 表示是否有抑制异常的记录，true表示没有
                boolean recordSuppressedExceptions = (this.suppressedExceptions == null);
                if (recordSuppressedExceptions) {
                    // 若没有抑制异常记录，则对抑制异常列表进行初始化
                    this.suppressedExceptions = new LinkedHashSet<>();
                }
                try {
                    // 从单例工厂获取对象。注意 singletonFactory 是本方法第二个参数，之前也介绍了
                    // ObjectFactory 对象通过调 getObject 来正式执行函数，所以 createBean 在此时才真正执行
                    singletonObject = singletonFactory.getObject();
                    // 获取后，就已经生成了新的单例对象，标记为 true
                    newSingleton = true;
                }
                catch (IllegalStateException ex) {
                    // Has the singleton object implicitly appeared in the meantime ->
                    // if yes, proceed with it since the exception indicates that state.
                    singletonObject = this.singletonObjects.get(beanName);
                    if (singletonObject == null) {
                        throw ex;
                    }
                }
                catch (BeanCreationException ex) {
                    if (recordSuppressedExceptions) {
                        for (Exception suppressedException : this.suppressedExceptions) {
                            ex.addRelatedCause(suppressedException);
                        }
                    }
                    throw ex;
                }
                finally {
                    if (recordSuppressedExceptions) {
                        this.suppressedExceptions = null;
                    }
                    afterSingletonCreation(beanName);
                }
                if (newSingleton) {
                    // 操作第1级缓存
                    addSingleton(beanName, singletonObject);
                }
            }
            return singletonObject;
        }
    }

在调用singletonObject = singletonFactory.getObject();以间接调用createBean后，会调用addSingleton，将bean加入第1级缓存，这标志着bean变为成品。接下来我们看下addSingleton的代码，它和getSingleton定义于同一个文件。

    /**
     * Add the given singleton object to the singleton cache of this factory.
     * 
To be called for eager registration of singletons.
     * @param beanName the name of the bean
     * @param singletonObject the singleton object
     */
    protected void addSingleton(String beanName, Object singletonObject) {
        // addSingleton 在调用了 createBean 的 getSingleton 方法中被调用，标志着 bean 变为成品对象
        synchronized (this.singletonObjects) {
            // 第1级缓存添加，第2、3级缓存移除
            this.singletonObjects.put(beanName, singletonObject);
            this.singletonFactories.remove(beanName);
            this.earlySingletonObjects.remove(beanName);
            // 添加到已注册的单例集合。 registeredSingletons 为 Set
            this.registeredSingletons.add(beanName);
        }
    }

接下来看createBean。https://github.com/spring-projects/spring-framework/blob/4786e2bf53a3f882c10e25d7ff79a18ff47b5e51/spring-beans/src/main/java/org/springframework/beans/factory/support/AbstractAutowireCapableBeanFactory.java

    /**
     * Central method of this class: creates a bean instance,
     * populates the bean instance, applies post-processors, etc.
     * @see #doCreateBean
     */
    @Override
    protected Object createBean(String beanName, RootBeanDefinition mbd, @Nullable Object[] args)
            throws BeanCreationException {
        // 首次看源码，直接看 doCreateBean 调用处
        if (logger.isTraceEnabled()) {
            logger.trace("Creating instance of bean '" + beanName + "'");
        }
        RootBeanDefinition mbdToUse = mbd;

        // Make sure bean class is actually resolved at this point, and
        // clone the bean definition in case of a dynamically resolved Class
        // which cannot be stored in the shared merged bean definition.
        Class resolvedClass = resolveBeanClass(mbd, beanName);
        if (resolvedClass != null && !mbd.hasBeanClass() && mbd.getBeanClassName() != null) {
            mbdToUse = new RootBeanDefinition(mbd);
            mbdToUse.setBeanClass(resolvedClass);
        }

        // Prepare method overrides.
        try {
            mbdToUse.prepareMethodOverrides();
        }
        catch (BeanDefinitionValidationException ex) {
            throw new BeanDefinitionStoreException(mbdToUse.getResourceDescription(),
                    beanName, "Validation of method overrides failed", ex);
        }

        try {
            // Give BeanPostProcessors a chance to return a proxy instead of the target bean instance.
            Object bean = resolveBeforeInstantiation(beanName, mbdToUse);
            if (bean != null) {
                return bean;
            }
        }
        catch (Throwable ex) {
            throw new BeanCreationException(mbdToUse.getResourceDescription(), beanName,
                    "BeanPostProcessor before instantiation of bean failed", ex);
        }

        try {
            // 实际创建 bean
            Object beanInstance = doCreateBean(beanName, mbdToUse, args);
            if (logger.isTraceEnabled()) {
                logger.trace("Finished creating instance of bean '" + beanName + "'");
            }
            return beanInstance;
        }
        catch (BeanCreationException | ImplicitlyAppearedSingletonException ex) {
            // A previously detected exception with proper bean creation context already,
            // or illegal singleton state to be communicated up to DefaultSingletonBeanRegistry.
            throw ex;
        }
        catch (Throwable ex) {
            throw new BeanCreationException(
                    mbdToUse.getResourceDescription(), beanName, "Unexpected exception during bean creation", ex);
        }
    }

createBean目前唯一值得关注的点就是调用了doCreateBean。doCreateBean做了几件值得本文关注的事：

• 调用了createBeanInstance完成bean的实例化。
• 调用了addSingletonFactory，即加入了第3级缓存。
• 调用了populateBean完成bean的属性赋值操作。

doCreateBean和createBean都定义在https://github.com/spring-projects/spring-framework/blob/4786e2bf53a3f882c10e25d7ff79a18ff47b5e51/spring-beans/src/main/java/org/springframework/beans/factory/support/AbstractAutowireCapableBeanFactory.java。

    /**
     * Actually create the specified bean. Pre-creation processing has already happened
     * at this point, e.g. checking {@code postProcessBeforeInstantiation} callbacks.
     * 
Differentiates between default bean instantiation, use of a
     * factory method, and autowiring a constructor.
     * @param beanName the name of the bean
     * @param mbd the merged bean definition for the bean
     * @param args explicit arguments to use for constructor or factory method invocation
     * @return a new instance of the bean
     * @throws BeanCreationException if the bean could not be created
     * @see #instantiateBean
     * @see #instantiateUsingFactoryMethod
     * @see #autowireConstructor
     */
    protected Object doCreateBean(String beanName, RootBeanDefinition mbd, @Nullable Object[] args)
            throws BeanCreationException {

        // Instantiate the bean.
        // instanceWrapper 持有创建出的 bean 对象
        BeanWrapper instanceWrapper = null;
        // 获取 factoryBean 实例缓存
        if (mbd.isSingleton()) {
            // 如果是单例对象，从 factoryBean 实例缓存中移除当前 bean 的信息
            // 首次 controllerA 进来会进行移除操作，并且会调用 createBeanInstance
            instanceWrapper = this.factoryBeanInstanceCache.remove(beanName);
        }
        // 没有就创建实例
        if (instanceWrapper == null) {
            // 根据执行 bean 使用对应的策略创建新的工厂实例，如：工厂方法、构造函数主动注入、简单初始化
            // 第一次读源码时不需要点进去看 createBeanInstance 。下一个主干方法 populateBean 是在本函数下文调用的
            instanceWrapper = createBeanInstance(beanName, mbd, args);
        }
        // 从包装类中获取原始 bean 。首次执行 controllerA 的时候, bean 就是普通的 controllerA 对象 ControllerA@100{cb=null}
        Object bean = instanceWrapper.getWrappedInstance();
        // 获取具体 bean 对象的 Class 属性
        Class beanType = instanceWrapper.getWrappedClass();
        if (beanType != NullBean.class) {
            // 不等于 NullBean 类型时就修改目标类型
            mbd.resolvedTargetType = beanType;
        }

        // Allow post-processors to modify the merged bean definition.
        synchronized (mbd.postProcessingLock) {
            if (!mbd.postProcessed) {
                try {
                    applyMergedBeanDefinitionPostProcessors(mbd, beanType, beanName);
                }
                catch (Throwable ex) {
                    throw new BeanCreationException(mbd.getResourceDescription(), beanName,
                            "Post-processing of merged bean definition failed", ex);
                }
                mbd.markAsPostProcessed();
            }
        }

        // Eagerly cache singletons to be able to resolve circular references
        // even when triggered by lifecycle interfaces like BeanFactoryAware.
        // 判断当前 bean 是否需要提前曝光，条件为：是单例 && 允许循环依赖 spring.main.allow-circular-references 配置为 true && 当前 bean 正在创建中
        boolean earlySingletonExposure = (mbd.isSingleton() && this.allowCircularReferences &&
                isSingletonCurrentlyInCreation(beanName));
        if (earlySingletonExposure) {
            if (logger.isTraceEnabled()) {
                logger.trace("Eagerly caching bean '" + beanName +
                        "' to allow for resolving potential circular references");
            }
            // 为避免后期循环依赖，可以在 bean 初始化完成前将创建实例的 ObjectFactory 加入工厂
            // 注意，这个方法会操作三级缓存的数据结构，尤其是第3级缓存。在 controllerA -> controllerB -> controllerA 的时候未调用 createBean 的 getSingleton 方法会真正调用这个匿名函数，从而调用 getEarlyBeanReference 进而操作第2级缓存
            addSingletonFactory(beanName, () -> getEarlyBeanReference(beanName, mbd, bean));
        }

        // Initialize the bean instance.
        // controllerA 首次执行到这里时，exposedObject = ControllerA@100{cb=null}
        Object exposedObject = bean;
        try {
            // populateBean 是主干方法，给刚刚实例化的 bean （半成品）填充属性
            populateBean(beanName, mbd, instanceWrapper);
            // controllerA -> controllerB -> controllerA 之后，getSingleton 返回 controllerA 后回到这里， controllerB 的 ca 属性就有值了
            // 随后递归返回到这句话且调用栈只有 controllerA 的时候，发现两者都有值了
            exposedObject = initializeBean(beanName, exposedObject, mbd);
        }
        catch (Throwable ex) {
            if (ex instanceof BeanCreationException bce && beanName.equals(bce.getBeanName())) {
                throw bce;
            }
            else {
                throw new BeanCreationException(mbd.getResourceDescription(), beanName, ex.getMessage(), ex);
            }
        }

        if (earlySingletonExposure) {
            Object earlySingletonReference = getSingleton(beanName, false);
            if (earlySingletonReference != null) {
                if (exposedObject == bean) {
                    exposedObject = earlySingletonReference;
                }
                else if (!this.allowRawInjectionDespiteWrapping && hasDependentBean(beanName)) {
                    String[] dependentBeans = getDependentBeans(beanName);
                    Set actualDependentBeans = new LinkedHashSet<>(dependentBeans.length);
                    for (String dependentBean : dependentBeans) {
                        if (!removeSingletonIfCreatedForTypeCheckOnly(dependentBean)) {
                            actualDependentBeans.add(dependentBean);
                        }
                    }
                    if (!actualDependentBeans.isEmpty()) {
                        throw new BeanCurrentlyInCreationException(beanName,
                                "Bean with name '" + beanName + "' has been injected into other beans [" +
                                StringUtils.collectionToCommaDelimitedString(actualDependentBeans) +
                                "] in its raw version as part of a circular reference, but has eventually been " +
                                "wrapped. This means that said other beans do not use the final version of the " +
                                "bean. This is often the result of over-eager type matching - consider using " +
                                "'getBeanNamesForType' with the 'allowEagerInit' flag turned off, for example.");
                    }
                }
            }
        }

        // Register bean as disposable.
        try {
            registerDisposableBeanIfNecessary(beanName, bean, mbd);
        }
        catch (BeanDefinitionValidationException ex) {
            throw new BeanCreationException(
                    mbd.getResourceDescription(), beanName, "Invalid destruction signature", ex);
        }

        return exposedObject;
    }

createBeanInstance的细节与本文主题无关，不关注。接下来我们看下addSingletonFactory的实现。addSingletonFactory位于https://github.com/spring-projects/spring-framework/blob/502997d8e986dcfde1f49b2b2f443a32b5488b13/spring-beans/src/main/java/org/springframework/beans/factory/support/DefaultSingletonBeanRegistry.java，主要动作是操作第3级缓存。

    /**
     * Add the given singleton factory for building the specified singleton
     * if necessary.
     * 
To be called for eager registration of singletons, e.g. to be able to
     * resolve circular references.
     * @param beanName the name of the bean
     * @param singletonFactory the factory for the singleton object
     */
    protected void addSingletonFactory(String beanName, ObjectFactory singletonFactory) {
        Assert.notNull(singletonFactory, "Singleton factory must not be null");
        synchronized (this.singletonObjects) {
            // controllerA 首次执行到这里，第1级缓存肯定是查不到的
            if (!this.singletonObjects.containsKey(beanName)) {
                // 第3级缓存放入。再回顾一下， singletonFactory存的是 beanName 到一个延迟执行的函数的映射
                // controllerA 首次执行到这里的时候， singletonFactory = () -> getEarlyBeanReference(beanName, mbd, bean)
                this.singletonFactories.put(beanName, singletonFactory);
                // 从早期单例对象的高速缓存（即第2级缓存）移除当前 beanName 对应的缓存对象
                this.earlySingletonObjects.remove(beanName);
                // 添加到已注册的单例集合里，和三级缓存无关。值得注意的是， A 首次加入第3级缓存时，就是首次加入已注册的单例集合
                this.registeredSingletons.add(beanName);
            }
        }
    }

接下来关注位于https://github.com/spring-projects/spring-framework/blob/4786e2bf53a3f882c10e25d7ff79a18ff47b5e51/spring-beans/src/main/java/org/springframework/beans/factory/support/AbstractAutowireCapableBeanFactory.java的populateBean。populateBean方法有一个功能是给bean的属性赋值，包含了递归点。在创建Bean的源码中，递归点指的是递归调用getBean方法。

    /**
     * Populate the bean instance in the given BeanWrapper with the property values
     * from the bean definition.
     * @param beanName the name of the bean
     * @param mbd the bean definition for the bean
     * @param bw the BeanWrapper with bean instance
     */
    protected void populateBean(String beanName, RootBeanDefinition mbd, @Nullable BeanWrapper bw) {
        if (bw == null) {
            if (mbd.hasPropertyValues()) {
                throw new BeanCreationException(
                        mbd.getResourceDescription(), beanName, "Cannot apply property values to null instance");
            }
            else {
                // Skip property population phase for null instance.
                return;
            }
        }

        if (bw.getWrappedClass().isRecord()) {
            if (mbd.hasPropertyValues()) {
                throw new BeanCreationException(
                        mbd.getResourceDescription(), beanName, "Cannot apply property values to a record");
            }
            else {
                // Skip property population phase for records since they are immutable.
                return;
            }
        }

        // Give any InstantiationAwareBeanPostProcessors the opportunity to modify the
        // state of the bean before properties are set. This can be used, for example,
        // to support styles of field injection.
        if (!mbd.isSynthetic() && hasInstantiationAwareBeanPostProcessors()) {
            for (InstantiationAwareBeanPostProcessor bp : getBeanPostProcessorCache().instantiationAware) {
                if (!bp.postProcessAfterInstantiation(bw.getWrappedInstance(), beanName)) {
                    return;
                }
            }
        }

        PropertyValues pvs = (mbd.hasPropertyValues() ? mbd.getPropertyValues() : null);

        int resolvedAutowireMode = mbd.getResolvedAutowireMode();
        if (resolvedAutowireMode == AUTOWIRE_BY_NAME || resolvedAutowireMode == AUTOWIRE_BY_TYPE) {
            MutablePropertyValues newPvs = new MutablePropertyValues(pvs);
            // Add property values based on autowire by name if applicable.
            if (resolvedAutowireMode == AUTOWIRE_BY_NAME) {
                autowireByName(beanName, mbd, bw, newPvs);
            }
            // Add property values based on autowire by type if applicable.
            if (resolvedAutowireMode == AUTOWIRE_BY_TYPE) {
                autowireByType(beanName, mbd, bw, newPvs);
            }
            pvs = newPvs;
        }
        if (hasInstantiationAwareBeanPostProcessors()) {
            if (pvs == null) {
                pvs = mbd.getPropertyValues();
            }
            for (InstantiationAwareBeanPostProcessor bp : getBeanPostProcessorCache().instantiationAware) {
                // 对于 Controller Bean 自动装配属性产生循环依赖的场景，遍历到 AutowiredAnnotationBeanPostProcessor 时，这句话包含递归点
                PropertyValues pvsToUse = bp.postProcessProperties(pvs, bw.getWrappedInstance(), beanName);
                if (pvsToUse == null) {
                    return;
                }
                pvs = pvsToUse;
            }
        }

        boolean needsDepCheck = (mbd.getDependencyCheck() != AbstractBeanDefinition.DEPENDENCY_CHECK_NONE);
        if (needsDepCheck) {
            PropertyDescriptor[] filteredPds = filterPropertyDescriptorsForDependencyCheck(bw, mbd.allowCaching);
            checkDependencies(beanName, mbd, filteredPds, pvs);
        }

        if (pvs != null) {
            // 对于普通 Bean 的场景，这句话包含递归点
            applyPropertyValues(beanName, mbd, bw, pvs);
        }
    }

applyPropertyValues也位于https://github.com/spring-projects/spring-framework/blob/4786e2bf53a3f882c10e25d7ff79a18ff47b5e51/spring-beans/src/main/java/org/springframework/beans/factory/support/AbstractAutowireCapableBeanFactory.java，主要功能为完成bean初始化。其中，Object resolvedValue = valueResolver.resolveValueIfNecessary(pv, originalValue)包含了递归点，通过动调可以验证bw.setPropertyValues(new MutablePropertyValues(deepCopy))真正完成了属性赋值工作。

    /**
     * Apply the given property values, resolving any runtime references
     * to other beans in this bean factory. Must use deep copy, so we
     * don't permanently modify this property.
     * @param beanName the bean name passed for better exception information
     * @param mbd the merged bean definition
     * @param bw the BeanWrapper wrapping the target object
     * @param pvs the new property values
     */
    protected void applyPropertyValues(String beanName, BeanDefinition mbd, BeanWrapper bw, PropertyValues pvs) {
        // applyPropertyValues 是真正完成赋值操作的函数
        // 如果 pvs 没有 PropertyValue，就直接结束
        if (pvs.isEmpty()) {
            return;
        }

        MutablePropertyValues mpvs = null;
        List original;

        if (pvs instanceof MutablePropertyValues _mpvs) {
            mpvs = _mpvs;
            if (mpvs.isConverted()) {
                // Shortcut: use the pre-converted values as-is.
                try {
                    bw.setPropertyValues(mpvs);
                    return;
                }
                catch (BeansException ex) {
                    throw new BeanCreationException(
                            mbd.getResourceDescription(), beanName, "Error setting property values", ex);
                }
            }
            // 获取 mpvs 的 PropertyValue 列表
            original = mpvs.getPropertyValueList();
        }
        else {
            // 获取 pvs 的 PropertyValue 对象数组并转为列表
            original = Arrays.asList(pvs.getPropertyValues());
        }

        // 用户自定义的类型转换器，默认转换器为 bean 的包装类对象
        TypeConverter converter = getCustomTypeConverter();
        if (converter == null) {
            converter = bw;
        }
        BeanDefinitionValueResolver valueResolver = new BeanDefinitionValueResolver(this, beanName, mbd, converter);

        // Create a deep copy, resolving any references for values.
        List deepCopy = new ArrayList<>(original.size());
        // resolveNecessary flag 含义为是否还需要解析
        boolean resolveNecessary = false;
        for (PropertyValue pv : original) {
            // 属性已解析过则加入 deepCopy
            if (pv.isConverted()) {
                deepCopy.add(pv);
            }
            else {
                String propertyName = pv.getName();
                // 获取未经类型转换的值
                Object originalValue = pv.getValue();
                if (originalValue == AutowiredPropertyMarker.INSTANCE) {
                    Method writeMethod = bw.getPropertyDescriptor(propertyName).getWriteMethod();
                    // 如果 setter 方法为空
                    if (writeMethod == null) {
                        // 异常：自动装配标记属性没有写方法
                        throw new IllegalArgumentException("Autowire marker for property without write method: " + pv);
                    }
                    originalValue = new DependencyDescriptor(new MethodParameter(writeMethod, 0), true);
                }
                // 交由 valueResolver 根据 pv 解析出 originalValue 所封装的对象。注意：这个函数包含递归点
                Object resolvedValue = valueResolver.resolveValueIfNecessary(pv, originalValue);
                Object convertedValue = resolvedValue;
                // flag 含义为是否可转换： propertyName 是 bw 中的可写属性 && propertyName 不是索引属性或嵌套属性
                boolean convertible = bw.isWritableProperty(propertyName) &&
                        !PropertyAccessorUtils.isNestedOrIndexedProperty(propertyName);
                if (convertible) {
                    // 可转换则将 resolvedValue 转换为指定的目标属性对象
                    convertedValue = convertForProperty(resolvedValue, propertyName, bw, converter);
                }
                // Possibly store converted value in merged bean definition,
                // in order to avoid re-conversion for every created bean instance.
                if (resolvedValue == originalValue) {
                    if (convertible) {
                        pv.setConvertedValue(convertedValue);
                    }
                    deepCopy.add(pv);
                }
                else if (convertible && originalValue instanceof TypedStringValue typedStringValue &&
                        !typedStringValue.isDynamic() &&
                        !(convertedValue instanceof Collection || ObjectUtils.isArray(convertedValue))) {
                    pv.setConvertedValue(convertedValue);
                    deepCopy.add(pv);
                }
                else {
                    resolveNecessary = true;
                    deepCopy.add(new PropertyValue(pv, convertedValue));
                }
            }
        }
        if (mpvs != null && !resolveNecessary) {
            mpvs.setConverted();
        }

        // Set our (possibly massaged) deep copy.
        try {
            // 完成属性赋值工作。咱们做个简单的实验，动调执行此句前后各点击调用栈看一次 populateBean 调用处的下一句的 exposedObject 或者 bean 变量，这就证实了 bean 的属性赋值确实是在这句话完成的
            bw.setPropertyValues(new MutablePropertyValues(deepCopy));
        }
        catch (BeansException ex) {
            throw new BeanCreationException(mbd.getResourceDescription(), beanName, ex.getMessage(), ex);
        }
    }

resolveValueIfNecessary位于https://github.com/spring-projects/spring-framework/blob/30d6ec3398ce41add7bc44d360b8fb86ac0264b1/spring-beans/src/main/java/org/springframework/beans/factory/support/BeanDefinitionValueResolver.java。其实现也很长，但目前只需要关注return resolveReference(argName, ref)，因为resolveReference包含递归点。

    /**
     * Given a PropertyValue, return a value, resolving any references to other
     * beans in the factory if necessary. The value could be:
     * 
A BeanDefinition, which leads to the creation of a corresponding
     * new bean instance. Singleton flags and names of such "inner beans"
     * are always ignored: Inner beans are anonymous prototypes.
     * 
A RuntimeBeanReference, which must be resolved.
     * 
A ManagedList. This is a special collection that may contain
     * RuntimeBeanReferences or Collections that will need to be resolved.
     * 
A ManagedSet. May also contain RuntimeBeanReferences or
     * Collections that will need to be resolved.
     * 
A ManagedMap. In this case the value may be a RuntimeBeanReference
     * or Collection that will need to be resolved.
     * 
An ordinary object or {@code null}, in which case it's left alone.
     * @param argName the name of the argument that the value is defined for
     * @param value the value object to resolve
     * @return the resolved object
     */
    @Nullable
    public Object resolveValueIfNecessary(Object argName, @Nullable Object value) {
        // We must check each value to see whether it requires a runtime reference
        // to another bean to be resolved.
        // RuntimeBeanReference：当属性值对象是工厂中另一个 bean 的引用时，使用不可变的占位符类，在运行时进行解析
        if (value instanceof RuntimeBeanReference ref) {
            // 解析出对应 ref 所封装的 Bean 的元信息的 Bean 对象。Bean 的元信息：Bean 名，Bean类型。注意，这个函数里包含递归点
            return resolveReference(argName, ref);
        }
        else if (value instanceof RuntimeBeanNameReference ref) {
            String refName = ref.getBeanName();
            refName = String.valueOf(doEvaluate(refName));
            if (!this.beanFactory.containsBean(refName)) {
                throw new BeanDefinitionStoreException(
                        "Invalid bean name '" + refName + "' in bean reference for " + argName);
            }
            return refName;
        }
        else if (value instanceof BeanDefinitionHolder bdHolder) {
            // Resolve BeanDefinitionHolder: contains BeanDefinition with name and aliases.
            return resolveInnerBean(bdHolder.getBeanName(), bdHolder.getBeanDefinition(),
                    (name, mbd) -> resolveInnerBeanValue(argName, name, mbd));
        }
        else if (value instanceof BeanDefinition bd) {
            return resolveInnerBean(null, bd,
                    (name, mbd) -> resolveInnerBeanValue(argName, name, mbd));
        }
        else if (value instanceof DependencyDescriptor dependencyDescriptor) {
            Set autowiredBeanNames = new LinkedHashSet<>(2);
            Object result = this.beanFactory.resolveDependency(
                    dependencyDescriptor, this.beanName, autowiredBeanNames, this.typeConverter);
            for (String autowiredBeanName : autowiredBeanNames) {
                if (this.beanFactory.containsBean(autowiredBeanName)) {
                    this.beanFactory.registerDependentBean(autowiredBeanName, this.beanName);
                }
            }
            return result;
        }
        else if (value instanceof ManagedArray managedArray) {
            // May need to resolve contained runtime references.
            Class elementType = managedArray.resolvedElementType;
            if (elementType == null) {
                String elementTypeName = managedArray.getElementTypeName();
                if (StringUtils.hasText(elementTypeName)) {
                    try {
                        elementType = ClassUtils.forName(elementTypeName, this.beanFactory.getBeanClassLoader());
                        managedArray.resolvedElementType = elementType;
                    }
                    catch (Throwable ex) {
                        // Improve the message by showing the context.
                        throw new BeanCreationException(
                                this.beanDefinition.getResourceDescription(), this.beanName,
                                "Error resolving array type for " + argName, ex);
                    }
                }
                else {
                    elementType = Object.class;
                }
            }
            return resolveManagedArray(argName, (List) value, elementType);
        }
        else if (value instanceof ManagedList managedList) {
            // May need to resolve contained runtime references.
            return resolveManagedList(argName, managedList);
        }
        else if (value instanceof ManagedSet managedSet) {
            // May need to resolve contained runtime references.
            return resolveManagedSet(argName, managedSet);
        }
        else if (value instanceof ManagedMap managedMap) {
            // May need to resolve contained runtime references.
            return resolveManagedMap(argName, managedMap);
        }
        else if (value instanceof ManagedProperties original) {
            // Properties original = managedProperties;
            Properties copy = new Properties();
            original.forEach((propKey, propValue) -> {
                if (propKey instanceof TypedStringValue typedStringValue) {
                    propKey = evaluate(typedStringValue);
                }
                if (propValue instanceof TypedStringValue typedStringValue) {
                    propValue = evaluate(typedStringValue);
                }
                if (propKey == null || propValue == null) {
                    throw new BeanCreationException(
                            this.beanDefinition.getResourceDescription(), this.beanName,
                            "Error converting Properties key/value pair for " + argName + ": resolved to null");
                }
                copy.put(propKey, propValue);
            });
            return copy;
        }
        else if (value instanceof TypedStringValue typedStringValue) {
            // Convert value to target type here.
            Object valueObject = evaluate(typedStringValue);
            try {
                Class resolvedTargetType = resolveTargetType(typedStringValue);
                if (resolvedTargetType != null) {
                    return this.typeConverter.convertIfNecessary(valueObject, resolvedTargetType);
                }
                else {
                    return valueObject;
                }
            }
            catch (Throwable ex) {
                // Improve the message by showing the context.
                throw new BeanCreationException(
                        this.beanDefinition.getResourceDescription(), this.beanName,
                        "Error converting typed String value for " + argName, ex);
            }
        }
        else if (value instanceof NullBean) {
            return null;
        }
        else {
            return evaluate(value);
        }
    }

resolveReference和resolveValueIfNecessary都位于https://github.com/spring-projects/spring-framework/blob/30d6ec3398ce41add7bc44d360b8fb86ac0264b1/spring-beans/src/main/java/org/springframework/beans/factory/support/BeanDefinitionValueResolver.java。其实现也很长，但目前只需要关注bean = this.beanFactory.getBean(resolvedName)，因为这就是递归点。

    /**
     * Resolve a reference to another bean in the factory.
     */
    @Nullable
    private Object resolveReference(Object argName, RuntimeBeanReference ref) {
        try {
            Object bean;
            Class beanType = ref.getBeanType();
            // 如果引用来自父工厂
            if (ref.isToParent()) {
                // 获取父工厂
                BeanFactory parent = this.beanFactory.getParentBeanFactory();
                if (parent == null) {
                    // 没有父工厂则报错：在父工厂中无法解析对 Bean 的引用，因为父工厂就不存在
                    throw new BeanCreationException(
                            this.beanDefinition.getResourceDescription(), this.beanName,
                            "Cannot resolve reference to bean " + ref +
                                    " in parent factory: no parent factory available");
                }
                if (beanType != null) {
                    bean = parent.getBean(beanType);
                }
                else {
                    bean = parent.getBean(String.valueOf(doEvaluate(ref.getBeanName())));
                }
            }
            else {
                String resolvedName;
                if (beanType != null) {
                    // 解析与 beanType 唯一匹配的 bean 实例，包括其 bean 名
                    NamedBeanHolder namedBean = this.beanFactory.resolveNamedBean(beanType);
                    // 让 bean 引用 namedBean 所封装的 bean 对象
                    bean = namedBean.getBeanInstance();
                    resolvedName = namedBean.getBeanName();
                }
                else {
                    resolvedName = String.valueOf(doEvaluate(ref.getBeanName()));
                    // 获取 ref 所包装的 Bean 名对应的 Bean 对象
                    // 注意，这就是递归点了
                    bean = this.beanFactory.getBean(resolvedName);
                }
                // 注册依赖关系到 Bean 工厂
                this.beanFactory.registerDependentBean(resolvedName, this.beanName);
            }
            if (bean instanceof NullBean) {
                bean = null;
            }
            return bean;
        }
        catch (BeansException ex) {
            throw new BeanCreationException(
                    this.beanDefinition.getResourceDescription(), this.beanName,
                    "Cannot resolve reference to bean '" + ref.getBeanName() + "' while setting " + argName, ex);
        }
    }

B创建过程调用栈是完全一样的，接下来我们假设现在走到了A -> B -> A，回到了doGetBean的Object sharedInstance = getSingleton(beanName);处。此时我们需要关注其实现了，因为这个函数要调用第3级缓存的函数，获取半成品bean，并放入第2级缓存。

    /**
     * Return the (raw) singleton object registered under the given name.
     * 
Checks already instantiated singletons and also allows for an early
     * reference to a currently created singleton (resolving a circular reference).
     * @param beanName the name of the bean to look for
     * @param allowEarlyReference whether early references should be created or not
     * @return the registered singleton object, or {@code null} if none found
     */
    @Nullable
    protected Object getSingleton(String beanName, boolean allowEarlyReference) {
        // Quick check for existing instance without full singleton lock
        // 该函数主要是调用放入第3级缓存的 getEarlyBeanReference 并放入第2级缓存
        // 首先从第1级缓存获取 bean
        Object singletonObject = this.singletonObjects.get(beanName);
        // 第1级缓存没有，并且已标记为创建中
        if (singletonObject == null && isSingletonCurrentlyInCreation(beanName)) {
            // 从第2级缓存获取 bean.因为controllerA -> controllerB -> controllerA 的时候是创建中的 bean,只放到了第3级缓存,所以是查不到的
            singletonObject = this.earlySingletonObjects.get(beanName);
            // controllerA -> controllerB -> controllerA 的时候进来, allowEarlyReference 肯定是 true
            if (singletonObject == null && allowEarlyReference) {
                synchronized (this.singletonObjects) {
                    // Consistent creation of early reference within full singleton lock
                    // 这段做二次确认的代码让我联想到线程安全的单例模式的写法
                    // 从第1级缓存取
                    singletonObject = this.singletonObjects.get(beanName);
                    if (singletonObject == null) {
                        // 从第2级缓存取
                        singletonObject = this.earlySingletonObjects.get(beanName);
                        if (singletonObject == null) {
                            // 从第3级缓存取。如果是 controllerA 初次进来，因为 not in creation 所以不会进这里，就算进了这里，因为第3级缓存取不到所以还是会直接 return。如果是 controllerA -> controllerB -> controllerA 则会操作第2级缓存
                            ObjectFactory singletonFactory = this.singletonFactories.get(beanName);
                            if (singletonFactory != null) {
                                // 回顾一下，DefaultSingletonBeanRegistry.java 的 addSingletonFactory 函数在操作第3级缓存的时候，放入的匿名函数就是：
                                // singletonFactory = () -> getEarlyBeanReference(beanName, mbd, bean) 所以 getEarlyBeanReference 返回值会在此被放入第2级缓存
                                singletonObject = singletonFactory.getObject();
                                // 为什么是放第2级缓存？因为 getBean 的递归还没返回
                                // 放入第2级缓存后，第3级缓存的就可以移除了
                                this.earlySingletonObjects.put(beanName, singletonObject);
                                this.singletonFactories.remove(beanName);
                            }
                        }
                    }
                }
            }
        }
        return singletonObject;
    }

再回忆一遍，什么时候加入第1级缓存？调用了beforeSingletonCreation的getSingleton方法在间接调用createBean后，会调用addSingleton方法，将成品bean加入第1级缓存。

A -> B -> A的递归返回后，A, B两个单例bean都已经是成品，beanName遍历到B的时候，进入getSingleton就能命中第1级缓存了，不用再走一遍createBean方法。

        for (String beanName : beanNames) {
            // 假如先遍历 A 再遍历 B 那么遍历到 B 的时候，因为循环引用解决的关系，B 已经放到了第1级缓存，所以 doGetBean 的 getSingleton 可以直接从第1级缓存取到值，不用再走一遍 createBean 方法
            // 合并父类 BeanDefinition
            RootBeanDefinition bd = getMergedLocalBeanDefinition(beanName);
            // 非抽象、是单例、非懒加载
            if (!bd.isAbstract() && bd.isSingleton() && !bd.isLazyInit()) {
                // 如果实现了 FactoryBean 接口则是 FactoryBean
                if (isFactoryBean(beanName)) {
                    Object bean = getBean(FACTORY_BEAN_PREFIX + beanName); // 比如：FACTORY_BEAN_PREFIX + beanName = "&A"
                    if (bean instanceof SmartFactoryBean smartFactoryBean && smartFactoryBean.isEagerInit()) {
                        getBean(beanName);
                    }
                }
                else {
                    // 不是 FactoryBean，只是普通 Bean，则走这个分支
                    getBean(beanName);
                }
            }
        }

经过上面的分析，我们来看更简单的情况：如果没有循环依赖，比如只有A依赖B，对三级缓存数据结构的操作是怎样的？梳理出调用栈如下：A doGetBean -> A doCreateBean -> A addSingletonFactory -> A applyPropertyValues -> B doGetBean -> B doCreateBean -> B addSingletonFactory -> B applyPropertyValues不走到递归点 -> B回到getSingleton调用addSingleton -> A addSingleton。可见如果没有循环依赖，就不会操作第2级缓存，但仍然会操作第1级缓存。第3级缓存虽然被操作但没有影响。

至此，场景1的递归点和三级缓存的操作时机都已经清楚了。

@Controller+自动装配属性和普通Bean解决循环依赖过程的相同点与不同点

场景2的Controller使用了@Autowired注解来构造循环依赖。动调可知，这个场景并不是在populateBean的applyPropertyValues(beanName, mbd, bw, pvs);完成修改的，而是在populateBean的：

            for (InstantiationAwareBeanPostProcessor bp : getBeanPostProcessorCache().instantiationAware) {
                PropertyValues pvsToUse = bp.postProcessProperties(pvs, bw.getWrappedInstance(), beanName);
                if (pvsToUse == null) {
                    return;
                }
                pvs = pvsToUse;
            }

这个循环里完成属性赋值的。动调发现getBeanPostProcessorCache().instantiationAware有一个元素是AutowiredAnnotationBeanPostProcessor@133（所有元素分别是ConfigurationClassPostProcessor$ImportAwareBeanPostProcessor@129, InfrastructureAdvisorAutoProxyCreator@130, PersistenceExceptionTranslationPostProcessor@131, CommonAnnotationBeanPostProcessor@132, AutowiredAnnotationBeanPostProcessor@133），遍历到这个元素时执行的操作完成了自动装配属性的赋值。那我们跟进去看下。

AutowiredAnnotationBeanPostProcessor.postProcessProperties位于https://github.com/spring-projects/spring-framework/blob/6183f0684684912802021556dce916ba26228c26/spring-beans/src/main/java/org/springframework/beans/factory/annotation/AutowiredAnnotationBeanPostProcessor.java

    @Override
    public PropertyValues postProcessProperties(PropertyValues pvs, Object bean, String beanName) {
        InjectionMetadata metadata = findAutowiringMetadata(beanName, bean.getClass(), pvs);
        try {
            metadata.inject(bean, beanName, pvs);
        }
        catch (BeanCreationException ex) {
            throw ex;
        }
        catch (Throwable ex) {
            throw new BeanCreationException(beanName, "Injection of autowired dependencies failed", ex);
        }
        return pvs;
    }

显然是在metadata.inject处完成自动装配的。InjectionMetadata.inject位于https://github.com/spring-projects/spring-framework/blob/2f33e77ab49f136d83b6ebf5eeb72d200fe23c0b/spring-beans/src/main/java/org/springframework/beans/factory/annotation/InjectionMetadata.java

    public void inject(Object target, @Nullable String beanName, @Nullable PropertyValues pvs) throws Throwable {
        Collection checkedElements = this.checkedElements;
        Collection elementsToIterate =
                (checkedElements != null ? checkedElements : this.injectedElements);
        if (!elementsToIterate.isEmpty()) {
            for (InjectedElement element : elementsToIterate) {
                element.inject(target, beanName, pvs);
            }
        }
    }

element.inject最终跳入的是https://github.com/spring-projects/spring-framework/blob/6183f0684684912802021556dce916ba26228c26/spring-beans/src/main/java/org/springframework/beans/factory/annotation/AutowiredAnnotationBeanPostProcessor.java的protected void inject(Object bean, @Nullable String beanName, @Nullable PropertyValues pvs) throws Throwable。AutowiredFieldElement, AutowiredMethodElement都有inject方法，显然这个case里我们调用的是AutowiredFieldElement的inject方法。

        @Override
        protected void inject(Object bean, @Nullable String beanName, @Nullable PropertyValues pvs) throws Throwable {
            Field field = (Field) this.member;
            Object value;
            if (this.cached) {
                try {
                    value = resolveCachedArgument(beanName, this.cachedFieldValue);
                }
                catch (BeansException ex) {
                    // Unexpected target bean mismatch for cached argument -> re-resolve
                    this.cached = false;
                    logger.debug("Failed to resolve cached argument", ex);
                    value = resolveFieldValue(field, bean, beanName);
                }
            }
            else {
                value = resolveFieldValue(field, bean, beanName);
            }
            if (value != null) {
                ReflectionUtils.makeAccessible(field);
                field.set(bean, value);
            }
        }

显然field.set(bean, value);最终完成了属性的自动装配。值得注意的是，动调看到resolveFieldValue获取到的cb是已经装配好的，这里一定存在递归调用。我们用一个简单的动态调试技巧来找到递归点：在执行到value = resolveFieldValue(field, bean, beanName)时，给doGetBean函数下一个临时的断点。得到的调用栈如下：

1. AutowiredAnnotationBeanPostProcessor.AutowiredFieldElement.inject()
2. AutowiredAnnotationBeanPostProcessor.AutowiredFieldElement.resolveFieldValue()的value = beanFactory.resolveDependency(desc, beanName, autowiredBeanNames, typeConverter);
3. DefaultListableBeanFactory.java的resolveDependency方法的result = doResolveDependency(descriptor, requestingBeanName, autowiredBeanNames, typeConverter);
4. DefaultListableBeanFactory.java的doResolveDependency方法的instanceCandidate = descriptor.resolveCandidate(autowiredBeanName, type, this);
5. https://github.com/spring-projects/spring-framework/blob/f1fe16e3cda66b164f77489f82287116477197bc/spring-beans/src/main/java/org/springframework/beans/factory/config/DependencyDescriptor.java的resolveCandidate方法的return beanFactory.getBean(beanName);，这就是递归点了，beanName = controllerB。

总而言之，Controller的bean自动装配属性的场景和普通的bean的递归点不一样，但对三级缓存的操作逻辑是完全一致的。谜底已揭晓~

一些扩展结论

【1】为什么Spring不能解决构造器的循环依赖？

在doCreateBean调用createBeanInstance时，一二三级缓存都没有Bean的相关信息，在实例化之后才调用addSingletonFactory放入第3级缓存中，因此当getBean的时候缓存不会命中，所以会抛出循环依赖的异常。

【2】为什么多实例Bean不能解决循环依赖？

多实例Bean是每次创建都会调用doGetBean方法，mbd.isSingleton()是false，不走sharedInstance = getSingleton(beanName, () -> {这个分支，所以也不会使用三级缓存，不能解决循环依赖。

总结

1. 根据常识猜测Spring创建Bean过程解决循环依赖的算法也是Map+记忆化搜索。所以我们可以先找到递归点，再去分析调用栈涉及的那些函数，顺便找出其用到的Map数据结构。
2. 在不了解Spring框架的情况下可以用一个动态调试技巧快速找到递归点：在执行到某条顺序靠后的语句时，给顺序靠前的语句下一个断点，若下断成功，则说明找到了递归点。剩下的工作就是关注调用栈涉及的函数。

参考资料

1. VSCode搭建Java开发环境（SpringBoot项目创建、运行、调试）：https://www.cnblogs.com/miskis/p/9816135.htmlhttps://www.cnblogs.com/miskis/p/9816135.html
2. 痛快！SpringBoot终于禁掉了循环依赖！痛快！SpringBoot终于禁掉了循环依赖！ - 掘金Spring的Bean管理，一直是整个体系中津津乐道的东西。尤其是Bean的循环依赖，更是很多面试官最喜欢考察的2B知识点之一。 但事实上，项目中存在Bean的循环依赖，是代码质量低下的表现。多数人寄https://juejin.cn/post/7096798740593246222
3. 循环依赖的详细debug过程_哔哩哔哩_bilibili循环依赖的详细debug过程是2023金三银四Java面试突击班！短期跳槽、长期系统学习、面试题、，马士兵系统课程都在这里！的第82集视频，该合集共计153集，视频收藏或关注UP主，及时了解更多相关视频内容。https://www.bilibili.com/video/BV1J14y1A7eE/?p=82 震惊！这位零基础前端同学在学习Java的第7天就可以研究Spring框架源码了！这是怎么做到的呢？噢！原来是站在了Java之父士兵的肩膀上！