Spring Framework
  1. Spring Framework
  2. SPR-7341

Allow bean definitions from @Configuration classes to override those defined in XML

    Details

    • Last commented by a User:
      true

      Description

      The ConfigurationClassBeanDefinitionReader hardcodes the behavior that bean definitions from XML always override bean definitions from JavaConfig classes:

      ConfigurationClassBeanDefinitionReader.java
      		// has this already been overridden (e.g. via XML)?
      		if (this.registry.containsBeanDefinition(beanName)) {
      			BeanDefinition existingBeanDef = registry.getBeanDefinition(beanName);
      			// is the existing bean definition one that was created from a configuration class?
      			if (!(existingBeanDef instanceof ConfigurationClassBeanDefinition)) {
      				// no -> then it's an external override, probably XML
      				// overriding is legal, return immediately
      				if (logger.isDebugEnabled()) {
      					logger.debug(String.format("Skipping loading bean definition for %s: a definition for bean " +
      							"'%s' already exists. This is likely due to an override in XML.", method, beanName));
      				}
      				return;
      			}
      		}
      

      I don't see why this decision was made. The normal semantics of using the order in which bean definitions were registered would seem to make sense to apply here as well.

      For example:

      @Configuration
      @ImportResource("foo.xml")
      class FooConfig {
        @Bean bean1() { ... }
      }
      
      @Configuration
      @ImportResource("bar.xml")
      class BarConfig {
        @Bean bean1() { ... }
        @Bean bean2() { ... }
      }
      
      ...
      new AnnotationConfigApplicationContext(FooConfig.class, BarConfig.class)
      

      I would expect the following override order to apply:
      1. beans defined in foo.xml
      2. overridden by beans defined in FooConfig
      3. overridden by beans defined in bar.xml
      4. overridden by beans defined in BarConfig

      Instead it appears the override order is:
      1. beans defined in FooConfig
      2. overridden by beans defined in BarConfig
      3. overridden by beans defined in foo.xml
      4. overridden by beans defined in bar.xml

      This is not intuitive. Specifically, note the interleaving of beans between both Foo and Bar configurations. The clients who use FooConfig and BarConfig don't care whether the beans defined by those Configuration classes come from XML or java. Yet, the override ordering forces them to be aware of this implementation detail.

      I realize changing the behavior outright might present a backwards compatibility issue, so is it possible to add a boolean option to the ApplicationContext to respect the registration ordering regardless of whether the bean definition comes from XML or JavaConfig?

        Activity

        Hide
        Eric Sirianni added a comment -

        Revising... I overlooked the already well-defined override order for imports (imported resources override beans defined in the importer) this would be my expected ordering:
        1. beans defined in FooConfig
        2. overridden by beans defined in foo.xml
        3. overridden by beans defined in BarConfig
        4. overridden by beans defined in bar.xml

        Show
        Eric Sirianni added a comment - Revising... I overlooked the already well-defined override order for imports (imported resources override beans defined in the importer) this would be my expected ordering: 1. beans defined in FooConfig 2. overridden by beans defined in foo.xml 3. overridden by beans defined in BarConfig 4. overridden by beans defined in bar.xml
        Hide
        Christopher Ng added a comment -

        I wish this had a higher priority. Basically java config is second-class at the moment, it can make it difficult to migrate to java config bit by bit, you have to do it all at once.

        btw, I'm not sure the ordering in Eric's comment is correct, afaik the ordering in the original post is the correct one.

        Show
        Christopher Ng added a comment - I wish this had a higher priority. Basically java config is second-class at the moment, it can make it difficult to migrate to java config bit by bit, you have to do it all at once. btw, I'm not sure the ordering in Eric's comment is correct, afaik the ordering in the original post is the correct one.
        Hide
        Abdoul Cisse added a comment - - edited

        This feature is important for Integration testing. Since Beans could be overriden inside Test class to mock behaviour needed like throwing exeptions to assert retry aspects configuration for example. At this moment the only solution is to override beans using XML file but this generates addtional test XML files that we could get rid of if Java Config done inside the test class could override any of the beans defined in application xml imported by @ImportResource.

        I'm not interested declaring the whole context with Java Config.

        Show
        Abdoul Cisse added a comment - - edited This feature is important for Integration testing. Since Beans could be overriden inside Test class to mock behaviour needed like throwing exeptions to assert retry aspects configuration for example. At this moment the only solution is to override beans using XML file but this generates addtional test XML files that we could get rid of if Java Config done inside the test class could override any of the beans defined in application xml imported by @ImportResource. I'm not interested declaring the whole context with Java Config.
        Hide
        Eric Sirianni added a comment -

        Agreed. My use case is similar to Abdoul's. We use JavaConfig extensively for JUnit testing (having to look in XML files for test fixtures is messy in my opinion), and this issue prevents us from injecting mocks in place of real collaborators.

        Show
        Eric Sirianni added a comment - Agreed. My use case is similar to Abdoul's. We use JavaConfig extensively for JUnit testing (having to look in XML files for test fixtures is messy in my opinion), and this issue prevents us from injecting mocks in place of real collaborators.

          People

          • Assignee:
            Chris Beams
            Reporter:
            Eric Sirianni
            Last updater:
            Eric Sirianni
          • Votes:
            15 Vote for this issue
            Watchers:
            15 Start watching this issue

            Dates

            • Created:
              Updated:
              Days since last comment:
              25 weeks, 4 days ago