Use simple name constructor in inner classes

[Rawi01]

This PR fixes #2571.

Using @With in inner classes generates a constructor using the name of the outer and the inner class. ecj does not like that if there also is a generic. To solve this lombok now generates constructor class using raw types: new Outer<T>.Inner<U>() -> new Outer.Inner().

[Gaibhne]

I manually built the jar to test it, and both the test case as well as the project the issue originally occurred in compile fine with my Eclipse now. Thank you!

[rspilker]

I'm not sure it solves the right problem. In general, having a non-static inner type where the outer type contains generics is asking for compiler problems.

Maybe @rzwitserloot can share his experiences with this problem?

[Rawi01]

Is there a case that is not covered by test cases?

Information about inner classes + inner constructors + generics are quite rare and it seems like that even compilers disagree about how that should be handled 🤷‍♂️

[Gaibhne]

For what it's worth: the bug this is addressing is a regression - the code that triggers the compiler error without this PR used to compile just fine a while ago.

[rzwitserloot]

The problem is, I recently changed lombok to go from generating new Inner() to new Outer.Inner() to work around another bug. So, apparently, we're stuck between a rock and a hard place. Will investigate

[rzwitserloot]

I added all this to fix issue #2268; this revert would mean 2268 is then broken again. We can debate two things:

• Is it even possible to fix both this issue as well as 2268? If not, is ecj just broken here and should we be filing bugs over there instead?
• Failing that, is undoing the fix for 2268 and thereby fixing 2571 the lesser of the two evils?

But.. not for the next release, it'll have to wait.

[Rawi01]

@rzwitserloot Reworked the solution, it now uses raw types for the constructor call in eclipse/ecj. I also reverted the javac changes because I think that we should not worsen the delombok output to fix this weird ecj behaviour.