[improbable22]

The norm at least in parts of physics is to update the arxiv version once the paper is accepted, with a note saying "v3: matches version to appear in XXX", with a DOI link. Where "matches" means the content, but not the journal's formatting, spelling & comma preferences, etc.

(And ideally this note would also mention any substantial changes, like "new section 4 explaining..." or "derivation in 3.2 re-written, but no change to conclusions", for the benefit of those who already read v2.)

[neffy]

I do the same (over in comp sci.) - usually the journals will let you put pre-print articles up on arxiv, researchgate, and web pages, and then the final article - sometimes with added spelling mistakes - many of the journals have off shored the editorial process - is on their site.

[frenchyatwork]

I'm curious what would happen if a new paper tried to reference previous papers published in ArXiv instead of the traditional publishers. From a strictly academic perspective, I can't think of any reason why this would be a problem, apart from the existential problem it causes for traditional publishers.

[improbable22]

Citing papers which aren't yet published by their arxiv numbers is routine. (And, before 1991, people did the same using the author's institute's preprint numbers.) If they have appeared in a journal by the time yours is accepted by a journal, then you should add the journal info too. (Although many journals now allow you to leave the arxiv numbers as an additional part of each reference.)