Unfortunately the FSF thinks the GPL end justifies any misleading means

First a request, could you please improve the layout of your comments by highlighting quoted passages more clearly? I see some people doing it with color, while I use the blockquote tag, but the way you quote entire passages in quotes (which passages then sometimes also contain quotes) is a bit confusing. Thanks.

I agree that there are two questions - is there a patent licence and, if there is one, what is its extent.

I don't deny that those are important questions, but the extent of the license has a lot to do with whether it's an implicit or explicit one. If it's an explicit one, then the language of the explicit grant defines it. If it's an implicit one, then there's no language of a grant to refer to and the scope will be what a court would consider a reasonable scope of an implicit license.

if it is effective and broad, who cares if it is implicit ? :-)

As I just explained: the scope depends on the nature. Theoretically, an explicit license could be anything from extremely narrow to extremely broad, while an implicit one will be resaonable, which is neither of the two extremes.

And I simply don't buy the argument that any change in the whole of Java means that none of the patents are being used as they were in the Sun/Oracle version

The most conservative position on the reasonable scope of an implicit license would be that any change can put the program outside that scope. That's similar to a programmer pointing out that any change in the program can make any other part unstable. It doesn't mean that it necessarily happens, but if all you know is that something changed, then that's the risk to assume (meaning you'd have to test everything again).

If a patent holder wants to limit the scope of the implicit patent license as much as possible, he will argue that whatever changes were made turned the program into something new that wasn't intended to be allowed under the license. Looking at the Oracle-Google example, the patents Oracle asserts appear to relate to the overall functioning of the virtual machine as opposed to individual programming commands. In the event Google changed anything about the commands (supporting fewer or supporting additional ones, or a combination of both), Oracle could argue that the selection of programming commands going through the patented mechanisms wasn't meant to be allowed under the implicit license.

The phrase says "limited to the use" not "limited to the code".

That's true but not helpful in practical terms. Assuming there is an implicit patent license in place and you take the relevant code and just change variable names, the algorithm itself remains the same. So it's the same use and would also in my understanding still be licensed. But once you change the logic of the relevant code, the patent holder will claim it's a new use. Since the implicit patent license should be reasonable, the nature and stature of any changes to the logic of the code should be taken into account. Absolutely minor changes should not deprive the forker and his users of the patent license, but once the changes are significant, the patent holder may have a case that the implicit license really just relied to what he (the patent holder) published.

Sun/Oracle would then have to argue that patents (originally taken out to protect Java) are used DIFFERENTLY in their Java and in the (GPL) Dalvik.

That's right. As I mentioned further above, even if there's simply a difference in terms of supported commands or whatever else that gets processed by the patented algorithms, Oracle could argue that it's a new use. If I say "could argue", that doesn't mean they'd be certain to win, but I said earlier on in this thread that the way the FSF wanted its statement to be understood was that the GPL would have solved the problem, so any reasonable doubt is enough to say that the FSF overstated.

("even if you receive the code direct from the owner of the patents" was referring to the little discussion of the "REdistribute" wording in the GPL2, where it seemed to me that the protection was less clear if you received the code direct from an owner with patents.)

Thanks for clarifying. The term "protection" is difficult to use here. If someone takes a GPL'd work and contrary to the REdistribute part simply imposes restrictions, then you are the licensee are still not directly protected. However, the forker who did that would have distributed code under the GPL without permission to do so. So in the end both of you would have a problem. The forker would have a copyright problem but you might still have a patent problem :-)