GeezerVison

Three crashes on ordinary Java code, met together in one real class (JabRef's EditAction). One commit each.

1. An enum with a method crashes the verifier

enum Action { UNDO, REDO; String getText() { return name(); } }

NullPointerException: Cannot invoke "String.endsWith(String)" because "qualifiedName" is null
  at ExternalRefinementTypeChecker.visitCtMethod

ExternalRefinementTypeChecker skips classes and records (user code) but not enums, so it visits the enum's methods with no external target. Fix: a no-op visitCtEnum, as for records.

2. A varargs call with more arguments than parameters

void debug(String format, Object... args) {}
logger.debug("{} and {}", a, b);   // IndexOutOfBoundsException: Index 2 out of bounds for length 2

mapInvocation and checkParameters index the parameters by argument position. Fix: both use only the arguments bound one to one to parameters; the ones collected into the varargs array are left out (an array parameter's refinement is not checked per element anyway).

3. A call on the result of a call without a return refinement

holder.get().close();
// Argument Mismatch Error: Argument '#holder_14' and its respective parameter of ghost 'Door.state1'
// types are incompatible: expected Door, got Holder

When get() has no return refinement, its TARGET metadata is still its own receiver (holder), and prepareInvocationTarget used it as the receiver of close(). Fix: a call whose TARGET is not its return value (RETURN_VAR_NAME) gives a fresh receiver of the call's type, in an unknown state, created once per call (Keys.RECEIVER, as for constructor calls). Now: found true but expected open(#fresh).

Tests

• CorrectEnumWithMethod: crashed before.
• CorrectVarargsCall (more, fewer and an explicit array) and ErrorVarargsCall (a refined fixed parameter before the varargs is still checked): crashed before.
• CorrectCallResultReceiver (a call with no precondition on a call result) and ErrorCallResultReceiver (reported as a state error, not an argument mismatch).
• mvn test: 463 tests, 0 failures.

Found on a real fixed bug: JabRef's Edit command calling undo()/redo() without canUndo()/canRedo() (PR JabRef/jabref#9777). We also checked 53 exercises built from real OSS bugs on main (69895fa) and with these changes: the results are identical.

🤖 Generated with Claude Code