> I really like that, but what about citing books?
Public libraries! Good enough to cite, good enough to ask your library system to carry.
The exception is textbooks, but these are 1. frequently also considered secondary sources, and 2. have their own weird economics that are just as bad as that of paid journals, and so need a separate fix.
> Sometimes, there are significant differences between the ArxiV version and the journal version. I've seen stuff where the journal had better figures, etc. In that case, a reader would be better of reading the journal article than the ArxiV version.
Yes, in fact, there are always at least minor differences; this is because, right now, the IP of a journal paper belongs to the publishing journal, so you can't publish exactly the same version of the paper anywhere else. (Just like a freelance writer who sells a short story to a magazine can't also publish that short story on their own website.) Paper authors get around this by publishing pre-prints—literally, the final-minus-one draft of the paper, rather than the final draft. Sometimes the changes between versions N-1 and N are trivial, but sometimes they're substantive and change the citability of the work.
In a world where journals didn't own, but rather licensed, the IP of the papers they "carry", this wouldn't be a problem, and paper authors would be publishing final prints of their papers in arXiv, because they'd have no reason not to. (Specifically, they'd likely publish the pre-print, and then—if there are substantive changes—they'd publish an updated version as well, and modify the metadata for the preprint so the paper's web-indexed DOI-metadata page would mention, and link to, the updated version, as well as the previous version.)