> If the original program is licensed under certain versions of the GNU GPL, the translated program must be covered by the same versions of the GNU GPL.
That sounds extremely hard to enforce, especially between languages that aren’t similar enough for a clear one-to-one correlation to appear. I could see this being an issue between CL and Scheme or two different kinds of shell scripting, but less so with Rust and elisp.
Most likely but I think the FSF has sufficient legal and technical chops to figure out if, in this instance, they have a good case and should contact the developer.
As I recall, Stallman used to do something similar early on pre-GNU and stopped when he realized it wasn’t technically necessary, wasn’t even particularly helpful to himself in particular, and might even have legal implications to reimplement software with that method.
In fact, I remember clearly that the FSF was extremely persistent that any tool that was going into user land in GNU had to be implemented in an obviously different way. This is one of the reasons why Bison produces faster code than Yacc. I think grep was another one of those, "Well we can't do it the normal way, so let's find a crazy way that will be super fast".
If you’re in the mental space of thinking about mathematical problems, then you’re probably very far from the legal meaning of ‘translation’. IANAL, but I have taken some IP Law classes, and these two domains don’t use the same kind of formalisms. (In practice, the meaning of ‘transformative work’ is not specified with anything in the same continent as mathematical precision; the canonical definition of obscenity is probably closer.)
> If the original program is licensed under certain versions of the GNU GPL, the translated program must be covered by the same versions of the GNU GPL.