Spring Validation实现原理分析

最近要做动态数据的提交处理，即需要分析提交数据字段定义信息后才能明确对应的具体字段类型，进而做数据类型转换和字段有效性校验，然后做业务处理后提交数据库，自己开发一套校验逻辑的话周期太长，因此分析了Spring Validation的实现原理，复用了其底层花样繁多的Validator，在此将分析Spring Validation原理的过程记录下，不深入细节

如何使用Spring Validation

• Spring Bean初始化时校验Bean是否符合JSR-303规范

  1、手动添加BeanValidationPostProcessor Bean

  2、在model类中定义校验规则，如@Max、@Min、@NotEmpty

  3、声明Bean，综合代码如下：

@Bean
public BeanPostProcessor beanValidationPostProcessor() {
    return new BeanValidationPostProcessor();
}

@Bean
public UserModel getUserModel() {
    UserModel userModel = new UserModel();
    userModel.setUsername(null);
    userModel.setPassword("123");
    return userModel;
}

@Data
class UserModel {
    @NotNull(message = "username can not be null")
    @Pattern(regexp = "[a-zA-Z0-9_]{5,10}", message = "username is illegal")
    private String username;
    @Size(min = 5, max = 10, message = "password's length is illegal")
    private String password;
}

4、BeanValidationPostProcessor Bean内部有个boolean类型的属性afterInitialization，默认是false，如果是false，在postProcessBeforeInitialization过程中对bean进行验证，否则在postProcessAfterInitialization过程对bean进行验证

5、此种校验使用了spring的BeanPostProcessor逻辑，可参考 Spring Boot系列之一：如何快速熟悉Spring技术栈

6、校验底层调用了doValidate方法，进一步调用validator.validate，默认validator为HibernateValidator，validation-api包为JAVA规范，Spring默认的规范实现为hibernate-validator包，此hibernate非ORM框架Hibernate

protected void doValidate(Object bean) {
	Assert.state(this.validator != null, "No Validator set");
	Set<ConstraintViolation<Object>> result = this.validator.validate(bean);

7、HibernateValidator默认调用ValidatorFactoryImpl来生成validator，后面展开将ValidatorFactoryImpl

• 支持方法级别的JSR-303规范

  1、手动添加MethodValidationPostProcessor Bean

  2、类上加上@Validated注解（也支持自定义注解，创建MethodValidationPostProcessor Bean时传入）

  3、在方法的参数中加上验证注解，比如@Max、@Min、@NotEmpty、@NotNull等，如

@Component
@Validated
public class BeanForMethodValidation {
    public void validate(@NotEmpty String name, @Min(10) int age) {
        System.out.println("validate, name: " + name + ", age: " + age);
    }
}

4、MethodValidationPostProcessor内部使用aop完成对方法的调用

public void afterPropertiesSet() {
    Pointcut pointcut = new `AnnotationMatchingPointcut`(this.validatedAnnotationType, true);
    this.advisor = new `DefaultPointcutAdvisor`(pointcut, createMethodValidationAdvice(this.validator));
}
protected Advice createMethodValidationAdvice(@Nullable Validator validator) {
	return (validator != null ? new `MethodValidationInterceptor`(validator) : new MethodValidationInterceptor());
}

5、底层同样默认调用ValidatorFactoryImpl来生成validator，由validator完成校验

• 直接编码调用校验逻辑，如

public class Person {
@NotNull(message = "性别不能为空")
private Gender gender;
@Min(10)
private Integer age;
...
}
ValidatorFactory validatorFactory = Validation.buildDefaultValidatorFactory();
Validator validator = validatorFactory.getValidator();
Person person = new Person();
person.setGender(Gender.Man);
validator.validate(person);

同上，默认调用ValidatorFactoryImpl来生成validator，由validator完成具体校验

• valid或validated

  在Spring controller方法参数中使用 valid或validated 注解标注待校验参数

  

  2、可以看到在各种resolver处理请求参数的过程中做了参数校验

  3、底层统一调用了DataBinder的validate方法

  4、DataBinder的作用：Binder that allows for setting property values onto a target object, including support for validation and binding result analysis，也就是binder处理了request提交的字符串形式的参数，将其转换成服务端真正需要的类型，binder提供了对validation的支持，可以存放校验结果

  5、DataBinder的validator默认在ConfigurableWebBindingInitializer中初始化，默认使用OptionalValidatorFactoryBean，该Bean继承了LocalValidatorFactoryBean，LocalValidatorFactoryBean组合了ValidatorFactory、自定义校验属性等各种校验会用到的信息，默认使用ValidatorFactoryImpl来获取validator

至此，所有的线索都指向了ValidatorFactoryImpl，下面分析下该类

public Validator `getValidator`() {
	return `createValidator`(
		constraintValidatorManager.getDefaultConstraintValidatorFactory(),
		valueExtractorManager,
		validatorFactoryScopedContext,
		methodValidationConfiguration
	);
}
Validator `createValidator`(ConstraintValidatorFactory constraintValidatorFactory,
	ValueExtractorManager valueExtractorManager,
	ValidatorFactoryScopedContext validatorFactoryScopedContext,
	MethodValidationConfiguration methodValidationConfiguration) {
	
	BeanMetaDataManager beanMetaDataManager = beanMetaDataManagers.computeIfAbsent(
		new BeanMetaDataManagerKey( validatorFactoryScopedContext.getParameterNameProvider(), valueExtractorManager, methodValidationConfiguration ),
		key -> new BeanMetaDataManager(
			`constraintHelper`,
			executableHelper,
			typeResolutionHelper,
			validatorFactoryScopedContext.getParameterNameProvider(),
			valueExtractorManager,
			validationOrderGenerator,
			buildDataProviders(),
			methodValidationConfiguration
		)
	 );
   
        return `new ValidatorImpl`(
			constraintValidatorFactory,
			beanMetaDataManager,
			valueExtractorManager,
			constraintValidatorManager,
			validationOrderGenerator,
			validatorFactoryScopedContext
	);
}
public final <T> Set<ConstraintViolation<T>> validate(T object, Class<?>... groups) {
	Contracts.assertNotNull( object, MESSAGES.validatedObjectMustNotBeNull() );
	sanityCheckGroups( groups );

	ValidationContext<T> validationContext = `getValidationContextBuilder().forValidate( object )`;

	if ( !validationContext.getRootBeanMetaData().hasConstraints() ) {
		return Collections.emptySet();
	}

	ValidationOrder validationOrder = determineGroupValidationOrder( groups );
	ValueContext<?, Object> valueContext = `ValueContext.getLocalExecutionContext`(
			validatorScopedContext.getParameterNameProvider(),
			object,
			validationContext.getRootBeanMetaData(),
			PathImpl.createRootPath()
	);

	return validateInContext( validationContext, valueContext, validationOrder );
}

1、getValidator->createValidator->ValidatorImpl->validate

在执行过程中封装了beanMetaDataManager、validationContext、valueContext等内容，都是校验时会用到的上下文信息，如待校验bean的所有校验项（含父类和接口）、property、method parameter的校验信息，从ValidatorFactoryScopedContext继承过来的validator通用的各种工具类（如message、script等的处理）等，内容比较复杂

2、分组（group）校验忽略，来到默认分组处理validateConstraintsForDefaultGroup->validateConstraintsForSingleDefaultGroupElement->validateMetaConstraint(注：metaConstraints维护了该bean类型及其父类、接口的所有校验，需要遍历调用validateMetaConstraint)

3、继续调用MetaConstraint的doValidateConstraint方法，根据不同的annotation type走不同的ConstraintTree

public static <U extends Annotation> ConstraintTree<U> of(ConstraintDescriptorImpl<U> composingDescriptor, Type validatedValueType) {
	if ( composingDescriptor.getComposingConstraintImpls().isEmpty() ) {
		return new SimpleConstraintTree<>( composingDescriptor, validatedValueType );
	}
	else {
		return new ComposingConstraintTree<>( composingDescriptor, validatedValueType );
	}
}

4、具体哪些走simple，哪些走composing暂且不管，因为二者都调用了ConstraintTree的'getInitializedConstraintValidator'方法，该步用来获取校验annotation（如DecimalMax、NotEmpty等）对应的validator并初始化validator 5、 ConstraintHelper 类维护了所有builtin的validator，并根据校验annotation（如DecimalMax）分类，validator的描述类中维护了该validator的泛型模板（如BigDecimal），如下：

putConstraints( tmpConstraints, DecimalMax.class,  Arrays.asList(
	DecimalMaxValidatorForBigDecimal.class,
	DecimalMaxValidatorForBigInteger.class,
	DecimalMaxValidatorForDouble.class,
	DecimalMaxValidatorForFloat.class,
	DecimalMaxValidatorForLong.class,
	DecimalMaxValidatorForNumber.class,
	DecimalMaxValidatorForCharSequence.class,
	DecimalMaxValidatorForMonetaryAmount.class
) );

在获取具体bean类的validator时，先根据annotation获取所有的validator，对应方法是ConstraintManager.findMatchingValidatorDescriptor，然后根据被校验对象的类型获取唯一的validator

6、然后根据上下文信息initializeValidator，进而调用validator的isValid方法校验