Courts have interpreted derivative works to be so when at least some parts of it contain the original copyrighted work. If that test fails, the whole derivative work argument is null and void.
I wouldn't want to be the one with my neck on the line in court over this.
Consider that .torrent files also don't have the content, yet torrent sites which serve them and provide tracking for torrent clients are having to play cat and mouse with copyright holders.
There have been court cases on this. In Galoob v. Nintendo, the Ninth Circuit ruled that Galoob did not infringe copyright by distributing a device (the Game Genie) that allowed users to modify copyrighted video games. The case established, at the time, that modifying a computer program or other copyrighted work for personal use was fair use.
That said, so much of what we knew about copyright and software in the 90s has been overturned by subsequent rulings or even statutes (like the DMCA) that if you wish to avoid being burned by copyright suits, you must retain an attorney knowledgeable in this area of law. (Which is true in any case.) As an example, look-and-feel copyrights are back on the table thanks to The Tetris Company LLC.
I agree that a bunch of people who are all licencees of a software work should be able to exchange instructions with each other for how to modify that to work better for them; I'd like courts to see it that way.
These instructions do not increase the number of people who have copies of the work, and so no copying is taking place; copyright should not be applicable here. The licensor may argue that license clauses is being violated, but not copyright; i.e. the users are using the software in ways that they don't like (use is not copying or redistribution).
I wouldn't want to be the one with my neck on the line in court over this.
Consider that .torrent files also don't have the content, yet torrent sites which serve them and provide tracking for torrent clients are having to play cat and mouse with copyright holders.