Generic constructor argument (e.g. ObjectProvider) fails to resolve for inner class [SPR-16734]

[spring-projects-issues]

David Kensche opened SPR-16734 and commented

I have a class containing an inner class configuration that inherits from JpaBaseConfiguration. It therefore overrides the protected super constructor. Upon application startup I get a misleading error message stating that there is no ObjectProvider<JtaTransactionManager>. However, the root cause for this is that the inner class itself is listed as first argument in the list of constructor arguments, leading to an off-by-1 error, therefore not satisfying the condition to create the ObjectProvider automatically.

This again resulted for me in hours of searching for dependency conflicts as the internet suggests this is a dependency conflict.

Here is the code example:

@SpringBootApplication
@EnableJms
public class ShowcaseManager {
    ...

    @Configuration
    @EnableTransactionManagement
    public static class JpaConfiguration extends JpaBaseConfiguration {

        protected JpaConfiguration(DataSource dataSource, JpaProperties properties,
                ObjectProvider<JtaTransactionManager> jtaTransactionManager,
                ObjectProvider<TransactionManagerCustomizers> transactionManagerCustomizers) {
            super(dataSource, properties, jtaTransactionManager, transactionManagerCustomizers);
        }
       ...
    }
}

---

Affects: 4.3.16, 5.0.5

Reference URL: https://bugs.java.com/view_bug.do?bug_id=5087240

Issue Links:

• Annotation lookup on parameter in inner class constructor fails when using javac from JDK versions prior to 9 [SPR-16652] #21193 Annotation lookup on parameter in inner class constructor fails when using javac from JDK versions prior to 9

Backported to: 4.3.17

[spring-projects-issues]

Juergen Hoeller commented

I suppose this only happens when you forget to declare the nested class as static? At least I can only reproduce it in such a scenario.

Even there we need to fix it, of course, but we highly recommend to declare nested classes as static unless they have an actual need to access an instance of the containing class.

[spring-projects-issues]

Juergen Hoeller commented

It turns out that this is a variant of the javac bug in #21193, showing through the JDK's core reflection libraries: Constructor.getGenericParameterTypes() is off by one for an inner class since it doesn't expose the enclosing instance, just like Constructor.getParameterAnnotations. I'll use a similar workaround within our MethodParameter implementation.

[spring-projects-issues]

Juergen Hoeller commented

Sam Brannen, you'll love this: Constructor.getGenericParameterTypes() is off by one... but only if generic type information is actually included since it'll fall back to a regular getParameterTypes() call otherwise. As mentioned above, I've worked around it along the lines of our fix for getParameterAnnotations on inner classes. FWIW, this is the same on JDK 8 and 9+, so hasn't been addressed at JDK level recently either, in contrast to the parameter annotation case.

[spring-projects-issues]

Juergen Hoeller commented

The corresponding test unexpectedly fails on the CI server (on Linux? on a different JDK 8 patch level?), still mismatching the parameter between getParameterTypes and getGenericParameterTypes... We'll have to revisit why the new check doesn't kick in there.

[spring-projects-issues]

Juergen Hoeller commented

Turns out I only committed the test, not the actual fix... Pushing to the CI server once again :-)

[spring-projects-issues]

Andrei Ivanov commented

Btw, I've noticed on the jdk8u-dev mailing lists that fixed bugs get back ported from newer versions, so maybe someone can ask for the related ones too?

[spring-projects-issues]

Juergen Hoeller commented

This one hasn't even been addressed in JDK 11 yet, so we'll rather have to push for an actual fix first...

At least the workaround seems as safe as the one we introduced for #21193. No information lost, just having to adapt the index assumptions in the returned array.

[spring-projects-issues]

Sam Brannen commented

Sam Brannen, you'll love this: Constructor.getGenericParameterTypes() is off by one... but only if generic type information is actually included since it'll fall back to a regular getParameterTypes() call otherwise.

That is indeed quite amusing! :-o
 

As mentioned above, I've worked around it along the lines of our fix for getParameterAnnotations on inner classes.

Nice!
 

FWIW, this is the same on JDK 8 and 9+, so hasn't been addressed at JDK level recently either, in contrast to the parameter annotation case.

Hmmm.... that's a pity.