Only clean when the annotated type mirror is type variable

[Ao-senXiong]

Fixes #786

[wmdietl]

Can you add the test case from the issue?
Can you make that test case self-contained, so that it doesn't depend on the CF classes?

[wmdietl]

Can you provide a reason why this new error is desirable? Why is this change in behavior an improvement?

[wmdietl]

Also note that a wilcard jtreg test fails.

[Ao-senXiong]

Can you provide a reason why this new error is desirable? Why is this change in behavior an improvement?

Because nn is @Nonnull and method1() is @Nullable. So there should be an assignment incompatiable error.

[wmdietl]

But method1 is in the stub file without annotations.
Shouldn't the logic be that, if something appears in the stub-file, those annotations are used. If there are no annotations in the stub-file, the usual defaults are used.
That's at least what the current behavior seems to be. This makes the stub-files readable and understandable without looking at what the corresponding source says.

[Ao-senXiong]

Can you add the test case from the issue? Can you make that test case self-contained, so that it doesn't depend on the CF classes?

Simplified version. Also updated in the issue.

import javax.lang.model.element.TypeElement;

public class Test {
    void test(TypeElement element) {
        element.asType();
    }
}

[Ao-senXiong]

Interesting. This crash does not happen under java 17 and happens under Java 8 and Java 11.

[wmdietl]

Simplified version. Also updated in the issue.

It would be nicer if you just had a class in the test case instead of depending on TypeElement#asType. Now to understand the test, one has to look at the annotated JDK and try to see what annotations are on that method.

[wmdietl]

Simplified version. Also updated in the issue.

It would be nicer if you just had a class in the test case instead of depending on TypeElement#asType. Now to understand the test, one has to look at the annotated JDK and try to see what annotations are on that method.

You only updated the test in the issue. You should add test cases to this PR to ensure the issue is actually fixed.

[Ao-senXiong]

Simplified version. Also updated in the issue.

It would be nicer if you just had a class in the test case instead of depending on TypeElement#asType. Now to understand the test, one has to look at the annotated JDK and try to see what annotations are on that method.

You only updated the test in the issue. You should add test cases to this PR to ensure the issue is actually fixed.

Okay, I will add it as I find the minial test case without TypeElement#asType dependency.