[arghwhat]

To be clear, a derivative work is copyright of the one creating the derivative, not the original author.

The question is whether creating the derivative work of that specific transformative nature is allowed. Unlike assets taken verbatim, this requires evaluating the exact instance. A binary decompilation is importantly not a simple translation, as that would be entirely unusable - rather, it is like creating blueprints for a finished building.

This is in part why licenses aim to manage and in part restrict you through a contract with the author, using a formally granted usage right to the entity as leverage for complying with a bunch of conditions, orthogonal to the copyright.

Your point stands though, my statement was not as accurate as it could have been.

[AshamedCaptain]

I really do not know what you're trying to nitpick here, but note that if you create a not-original-enough derivative work (which is what is happening here), you will not have any copyright whatsoever. Neither you nor the original author.

Binary decompilation is also the textbook definition of a "simple translation". It is almost always done with a mathematical formula aka algorithm and the resulting program is not only identical in observable behavior to the original one, it is also identical in the non-observable behavior (I.e. bit by bit identical on-memory data structures). The chance that you can end up with such identical program without looking at the original one is zero for anything but the most trivial programs.

Licenses are also almost totally irrelevant (even in this context) and for most software they are only enforceable by copyright anyway (e.g. the temporary on memory copies to load the program).

I do not understand what to see here that even has a hint of originality. This is why clean room is super important, even if not strictly necessary: it goes a long way to convince that the implementations do not come from the same source even if they are related.