IllegalAccessException when auto-mapping Records (JEP-359)

[GeorgeSalu]

I want MyBatis to Support for JDK 16 record types

[ksmith97]

This actually does work if you follow the directions for dealing with immutable objects and supply a constructor binding in a custom result map. Unfortunately, it is a lot more verbose than would be preferred. Maybe the code could check if the result type is a record and because result objects are always consistent in constructors it could automatically bind it together properly. Unfortunately, this probably relies on the isRecord method that does not exist before JDK 16. If the code allows for it maybe an optional adapter could be written that would allow for this support to be added for those of us on newer JDKs. If it does not allow for it currently maybe a new feature request could be submitted to allow for the code to allow something to hook in at the relevant location to supply this functionality?

[kazuki43zoo]

@harawata

I will try to support the Record type by following commit(not includes test).
The Record type cannot set a value to the final field, then I changed to skip property mapping for prevent mapping error.

kazuki43zoo@f8a5eda

WDYT?

[harawata]

@kazuki43zoo ,
Is that a workaround for the behavior I mentioned in the p.s. of this comment?
If so, I would like to consider correcting that behavior (#2207 is the same issue if I understand correctly).
If it's not related, please let me know and I'll take another look.
Thanks!

[kazuki43zoo]

@harawata Thanks for your reaction!!
I think that the root cause is the same with #2207 (= MyBatis try to write value to final field). But It is different behavior with standard type and record type on JDK 16.

standard type:

public class MyEntity {
   private final Integer id;
   private final String name;
   public MyEntity(Integer id, String name) {
     this.id = id;
     this.name = name;
   }
   // ...
}

recored type:

public record MyEntity(Integer id, String name) {
}

The standard type work but it write value to final field(Field#trustedFinal=false) at twice by constructor mapping and property(field) auto mapping. The recored type occurred an Exception because can't write a final field(Field#trustedFinal=true) on record type.

I would like to consider correcting that behavior (#2207 is the same issue if I understand correctly).

I agree with this!!

[harawata]

Thank you, @kazuki43zoo for the information!

I didn't know about the trustedFinal. There is no public method to check if a Field is 'trusted' or not, it seems. 😕

Anyway, my first impression is that this is an issue that should be addressed in org.apache.ibatis.reflection.Reflector (e.g. skip setter detection if it's a Record) [1].
Skipping particular processes may look simpler, but it gets harder to maintain in the long run.
Will there be any difficulty with this approach?

[1] There is another outstanding feature request slightly related to this topic. It might be a good time to add support for the prefix-less getter/setter methods to Reflector. This can be a separate discussion, though.

[cmosher01]

FYI, I had record types working fine with JDK 14 in preview mode. But when I switched to JDK 16 (non-preview), I started running into the final field update error: java.lang.IllegalAccessException: Can not set final int field.....

[GeorgeSalu]

a next version with Record support would be great. @kazuki43zoo

[bzhou]

Pull request #2477 sent

[bzhou]

Anyway, my first impression is that this is an issue that should be addressed in org.apache.ibatis.reflection.Reflector (e.g. skip setter detection if it's a Record) [1]. Skipping particular processes may look simpler, but it gets harder to maintain in the long run. Will there be any difficulty with this approach?

Agreed about the approach should be addressing Reflector.

The reason it is trying to write to final field, is that it treat java record like any class, so it tries to

• find all getters, there won't be any normally;
• find all setters, of course there won't be any because records are immutable;
• find all "fields", it finds all record components, but does not understand they are read-only.

Dealing with java record like this actually is a problematic bug.

My patch

• detects a class is a record by checking the name of its super class (so does not require java 16 and backward compatible)
• in that case simply finds all accessors (methods with no parameter) and add them as getters

That's it. You can see easily that in the case of java record, it might even be slightly faster :-)