Investigate on the issue and the solution proposed about exceptions raised

[masesdevelopers]

Looking at masesgroup/KEFCore#448 and analyzing the code:
https://github.com/masesgroup/KNet/blob/024d22b6e15a7fd2725b4b83d563cb742d9dfa90/src/jvm/knet/src/main/java/org/mases/knet/generated/org/apache/kafka/streams/processor/TimestampExtractor.java#L73-L77

• the first issue is related to a missing assignment of the return value in the JNetEventResult: the default null is returned back and the unboxing conversion fails raising java.lang.NullPointerException: Cannot invoke "java.lang.Long.longValue()" because "retVal" is null
• the second issue is related to a similar condition since the function getHasOverride probably returns the default value which is false

The JNetEventResult instance is created locally and the extract method invokes the synchronized method https://github.com/masesgroup/KNet/blob/024d22b6e15a7fd2725b4b83d563cb742d9dfa90/src/jvm/knet/src/main/java/org/mases/knet/generated/org/apache/kafka/streams/processor/TimestampExtractor.java#L45-L48

The code of JNetEventResult is very simple:

public class JNetEventResult {
   boolean _hasOverride = false;
   Object _returnData;

   public boolean getHasOverride() {
      return _hasOverride;
   }

   public void setHasOverride(boolean hasOverride) {
      _hasOverride = hasOverride;
   }

   public Object getReturnData() {
      return _returnData;
   }

   public void setReturnData(Object retData) {
      _returnData = retData;
   }
   
   public void setReturnData(boolean hasOverride, Object retData) {
      _hasOverride = hasOverride;
      _returnData = retData;
   }
}

but for the second issue seems clear that the default value is returned from getHasOverride.
Maybe it can be an option to replace the _hasOverride field with a type that accept to be a three state or verify if the value was returned adding an extra field updated when setReturnData or setHasOverride are invoked: if the optional boolean field is still false the getHasOverride can raise an internal exception since no one updated its internal value.
Shall be checked if the method setReturnData with one parameter is in use: otherwise the one with two parameter can change its name to avoid method matching confusion, leaving the value of _hasOverride to its default value.
Anyway an update on JNet shall be made to verify the previous assumptions.

Originally posted by @masesdevelopers in #1058

[masesdevelopers]

An issue was opened in JCOReflector (masesgroup/JCOReflector#253) with a similar condition: the JVM was not able to manage the returned value from an invocation to the CLR when there is an heavy invocation loop: probably masesgroup/KNet#1058 and masesgroup/KEFCore#448 fallback in a similar condition when Apache Kafka Streams invokes the callback of TimestampExtractor into .NET.
In masesgroup/JCOReflector#253 there was an heavy invocation loop solved with a little waiting time which helps to avoid useless CPU waste too.
In the case of masesgroup/KEFCore#448, the objects are returned to the JVM using the setReturnData of JNetEventResult before the control is returned back to the JVM from the invocation to the Extract method in CLR: the expectation is that setReturnData is able to pass the objects into the JVM when the callaback method is still in the CLR context, so when the code exit from CLR and returns to the JVM the objects lives there and can be used.

Following the tip in masesgroup/KNet#1058 a branch will be opened to create a JNet test package can used to make some tests in the environment of masesgroup/KEFCore#448.

The Java code will be changed in this way to intercept if setReturnData with two parameters was never invoked or it was invoked the one with one parameter:

public class JNetEventResult {
   volatile boolean _hasOverrideSet = false;
   volatile boolean _hasOverride = false;
   volatile Object _returnData;

   public boolean getHasOverride() {
      if (!_hasOverrideSet) throw new IllegalStateException(".NET side never changed the value of HasOverride property.");
      return _hasOverride;
   }

   public void setHasOverride(boolean hasOverride) {
      _hasOverrideSet = true;
      _hasOverride = hasOverride;
   }

   public Object getReturnData() {
      if (!_hasOverrideSet) throw new IllegalStateException(".NET side never changed the value of HasOverride property.");
      return _returnData;
   }

   public void setReturnData(Object retData) {
      throw new IllegalStateException(".NET side shall never invokes this method.");
      //_returnData = retData;
   }
   
   public void setReturnData(boolean hasOverride, Object retData) {
      _hasOverrideSet = true;
      _hasOverride = hasOverride;
      _returnData = retData;
   }
}

the previous code check only the second condition reported in masesgroup/KNet#1058.

[masesdevelopers]

Currently the tests done with artifact of the branch with modification described in #856 (comment) does not solves the issue, however in masesgroup/KNet#1058 (comment) is described an applied solution that mitigate the problem.

[mariomastrodicasa]

This behavior belongs to the JCOBridge layer on which JNet is built, rather than to JNet-specific logic.

The standard/public line is evolved when a case requires critical fixes or API changes to preserve ordinary usability. For this class of issue, however, the complete bridge-layer solution is currently available only in the HPA Edition.

So, in the current public stack, this is not something that is fully solved at JNet level; the complete solution is available in HPA.