BeanCurrentlyInCreationFailureAnalyzer doesn't always correctly detect the cycle's starting point

[wilkinsona]

This example from @cppwfs produces this output:

|  dataSourceInitializerPostProcessor (field private org.springframework.beans.factory.BeanFactory org.springframework.boot.autoconfigure.jdbc.DataSourceInitializerPostProcessor.beanFactory)
↑     ↓
|  testBeanB0
↑     ↓
|  myConfiguration (field private javax.sql.DataSource io.spring.MyConfiguration.dataSource)
↑     ↓
|  testBeanB1
↑     ↓
|  testBeanA
└─────┘

The ┌─────┐ at the top of the cycle diagram is missing.

[wilkinsona]

The problem's caused by a factory bean, the definition for which is registered by an ImportBeanDefinitionRegistrar. The means that there's insufficient type information available to determine the type produced by the factory bean without instantiating it. Instantiation fails because a constructor argument can't be resolved, but some information about this is hidden in exception messages rather than being represented by an exception in the hierarchy. As a result an entry in the cycle is missed. If that entry should have been the start of the cycle, the strange output shown above is produced. It's also possible for the cycle to look correct, but to actually be wrong as it's missing one or more entries somewhere in the middle. It's unclear how to fix this in a robust manner without a change in Spring Framework.