I'm not sure if this interpretation of his words is what he meant to say.

I'd argue that code that "compiles to nothing" can be protected by copyright, too, even if it is not distributed to customers. E.g., extensive documentation inside comments in the source files is definitely copyrighted. Complex build rules might be copyrightable, too, although they definitely end up in the distributed application.

Therefore, it would probably be counterproductive to assert such a thing in court.

Google is trying to make the case that interfaces aren't 'code' here. That's the entire case here, right?

Yes. I just wanted to clarify that code that "compiles to nothing", e.g., build instructions, may still be copyrightable. Therefore, I do not think that Page wanted to extend his definition of "non-code" to anything besides interfaces/declarations.

No. Previous case already found APIs can be copyrighted. This one is about fair use.

Unless that claim about interfaces becomes crucial in some unforeseen way.

So fancy type-level assertions that prove the correctness of your program are also not code? That's a weak argument.

But declarations can have an effect on the instructions generated.