he's not saying that the GPL has a poor legal basis, just that the FSF definition of a derivitive work hoes beyond where it has a good legal basis.
all the GPL enforcement court cases that I am aware of has involved clear copying of the work without including the source.
I am not aware of any GPL enforcement court actions where a library was under the GPL, proper source for that library was provided, and the enforcement action was that because the library was linked to another program, that other program had to be released under the GPL.
I know this is stated frequently as a requirement, and I fully believe that the threats around this may have forced some companies to GPL their software, but just because threats have worked doesn't mean it has a solid legal basis.
Yes, this does mean that I question if there is really an effective legal difference between the GPL and the LGPL.
the idea that software A could be considered derived from library B one day (because it only works with library B), and then the next day someone releases library C with the same interface as library B changing the status of Software A to no longer be considered a derivative work of library B (because it now works with both library B and library C) seems like nonsense to me. The status of software A as a derivative of some other piece of software needs to be based on what the developers A did while they were developing the software, not anything that anyone else does.
arguing that dynamic linking is Ok, but static linking makes something a derived work smacks of the same silliness as people claiming that because a computer copies a binary from the hard drive to ram (or from the ram to the CPU cache/registers) to execute it, copyright law now allows them to control the use of the software.
I also don't think that loosing these edge cases of derived would necessarily be a big deal for the community. I think more damage would be done by people no longer trusting the claims than by the loss of the license coverage.