What is a BeanFactoryPostProcessor? How does it differ from BeanPostProcessor?

A BeanFactoryPostProcessor (BFPP) operates on bean definitions — the metadata — before any bean is instantiated, whereas a BeanPostProcessor operates on bean instances after they're created. Definitions vs instances is the entire distinction. Get that contrast crisp and this question is yours.

The interface

One callback, handed the whole bean factory while it still holds only definitions:

public interface BeanFactoryPostProcessor {
    void postProcessBeanFactory(ConfigurableListableBeanFactory beanFactory);
}

At this point no application bean has been constructed yet. You can inspect and modify bean definitions — change a property value, alter a scope, add or remove a definition — and those changes take effect when Spring later instantiates the beans.

When it runs

Early, during context refresh, before singleton instantiation begins — and therefore before any BeanPostProcessor callback fires (BPPs need instances to exist; BFPPs run before instances exist).

refresh()
  -> invoke BeanFactoryPostProcessors   (edit DEFINITIONS)
  -> instantiate singletons
       -> BeanPostProcessors             (edit INSTANCES)

The canonical example

PropertySourcesPlaceholderConfigurer is a BeanFactoryPostProcessor. It resolves ${...} placeholders in bean definitions before beans are built, so the right literal values are already in place at instantiation. ConfigurationClassPostProcessor (which processes @Configuration / @Bean) is another.

@Component
public class EnforceScopeBfpp implements BeanFactoryPostProcessor {
    public void postProcessBeanFactory(ConfigurableListableBeanFactory bf) {
        BeanDefinition def = bf.getBeanDefinition("reportBuilder");
        def.setScope(BeanDefinition.SCOPE_PROTOTYPE);   // rewrite metadata
    }
}

BFPP vs BPP at a glance