For example, I might use a manual by tearing it into pieces and using the individual pages as confetti for a parade. But I cannot copy GFDL'd manuals and then do this.
I congratulate you on your imagination--it never occurred to me to think about this as a use of a manual. As it happens, you are free to do that, because copyright does not cover tearing up a manual. You don't need a license to give you permission to do that. And then, in between the silly ones and the reasonable ones, there are a whole lot more, with some pretty darn ambiguous cases. Yes, there are gray areas where it is hard to decide. I had to think for months about whether the TeX license qualified as free, since it makes the whole of the original TeX source code invariant. And I had to think for weeks about a LaTeX license, that required changing the name of any file that you modify. I eventually concluded that LaTeX was free despite this requirement, but only because it has a remapping feature that lets you say "Use file myfoo.sty when the document asks for foo.sty". My previous question is still the right one, I think. How would you go about explaining why "send $1 to the author for permission to make changes to this program" is not a mere "requirement", but actually kills freedom? That question is straightforward: if you have to pay for permission to do something, you do not have permission to do it. Debian has a way of answering that question: but our way, which involves the DFSG, would say that "send $1 to the author for permission to make changes" is wrong for the same reasons that "send $1 to the author for permission to make copies", and is wrong for the same reason that we think that invariant sections are wrong. The DFSG doesn't say anything about invariant sections; you're assuming a very strict interpretation. You're also assuming that the DFSG should be applied to manuals as well as software, and that the interpretation should be the same. I'm arguing for a less strict interpretation of the DFSG.