He's talking about distributing code that is not functional unless linked with a GPL library. Not about distributing code that replaces such GPL library.
Oracle is basically saying that you can not write a compatible implementation of Java.
It's partially relevant to the plan Mr. Haible had to get around the GPL by putting the onus on the user to combine readline and CLISP:
> I built a libnoreadline.a that can be linked together with lisp.a,
replacing libreadline.a .
> I will reorganize the distribution into 2 independent parts:
> * clisp.lzh containing lisp.a and libnoreadline.a,
> * readline.tar.Z containing libreadline.a and its source.
> The first one is enough to build a CLISP executable. It contains
no GNU parts.
If readline's API were copyrighted and unreimplementable, this would still be infringement. However, this is not actually the argument rms used; rather, he said:
> The FSF position would be that this is still one program, which has only been disguised as two. The reason it is still one program is that the one part clearly shows the intention for incorporation of the other part.
Who knows whether that would hold up in court or not.
Incidentally, these days (well, since 1997 or so?) we have editline, which, rather than being a sort of low-effort dummy implementation like the proposed "libnoreadline", is a full-fledged BSD-licensed readline replacement which provides a readline compatibility API. One possible outcome of this case, in which Google loses on both copyrightability and fair use, would make it arguable (not certain - Java is much larger than readline) that editline is infringing the GPL, which would be amusing.
I assume RMS would say that Java is non-free, and that the best move here would be to create a free compiler and runtime -- i.e. a new language.
This stands to reason, as one of the first things RMS did in his efforts to advance Free Software was create a free compiler (gcc) and a free editor (emacs).
He might note that gcc is not meant to replace all other languages. It is just an example of a well-made and libre compiler.
I'm not sure. His main argument in this case was that distribution of a non-GPL program with intention of letting user link with a GPL library was a "subterfuge".
Google's case is different. They didn't write Android so users can choose between a dummy libDoesNotRunAnything or wink, wink unlicensed Oracle JVM.
Or that he would have in 1992. The landscape has changed quite a bit since then.
And the terminology has improved greatly since then; for instance, nowadays the requirement would be described as making CLISP "GPL-compatible", rather than putting it under the GPL.
He's talking about distributing code that is not functional unless linked with a GPL library. Not about distributing code that replaces such GPL library.
Oracle is basically saying that you can not write a compatible implementation of Java.