In the current code, the running effect can only ensure that the first NamedMatch satisfies not isInterface

@wengyingjian I think the current Junction expression is

(A and not(isInterface())) or (B and not(isInterface())) or ...

Is my understanding right?

I think original request is all things should be not an interface.

First or last are not accurate.

It should use and with all other condition or before all other checks run.

Yes, because I didn't find a place that required modifying the bytecode for interfaces, I think all matching rules should be non-interface based.

If that's the case, adding and (not isInterface) in the for loop is indeed a safer approach.

Additionally, I've noticed that some custom IndirectMatch instances include not interface, while some do not.
I'm not sure if this means there is already a place that requires modifying the interface bytecode, or if it's just to maintain flexibility, allowing subclasses to implement it themselves.

In theory, there is no point to intrument an interface, especially we require to add new field to the target.

I've noticed that some custom IndirectMatch instances include not interface, while some do not.

I think those are not necessary, but as the bug you are fixing, they may be added as it is.

In the current code, the running effect can only ensure that the first NamedMatch satisfies not isInterface

@wengyingjian I think the current Junction expression is

(A and not(isInterface())) or (B and not(isInterface())) or ...

Is my understanding right?

I think the logic error is, that the is not interface could be overridden by the or from the 2nd condition, as it is flat.