That said there are probably ways a LLM could improve a decompiler in a way that does not impact its correctness. Like deriving class and variables names based on context, when symbols are missing or obfuscated.

I used it for game modding and documentation (and caught/reported a few game bugs + vulnerabilities along the way). I'd pull game files from Steam depots with steamkit, decompile with CFR, and run the resulting java through doxygen.

C# also has huge amounts of syntactic sugar: `yield return` and `await` compile into huge state machines; `fixed` statements come with similar problems as "finally" in java (including the possibility of exponential code growth during decompilation).

Personally, I don't get the sentiment. Yeah, decompiling might not produce the original source code, which is fair. It's possible to generate code using invokeDynamic and what not - still being valid code if a compiler opts to do so.

When decomiling bytecode there has to be a reason for, and a good one. There has to be a goal.

If the code is somewhat humanly understandable that's ok. if it's more readable than just bytecode, that's already an improvement.

Reading bytecode alone is not hard when it comes to reverse engineering. Java already comes with methods and fields available by design. Having local variable names and line numbers preserved is very common, due to exception stack traces being an excellent debugging tool. Hence debugging info gets to be preserved.

try/finally shares the same issues, albeit less pronounced.