A new publishing paradigm

Anyone that has been working with science has thought about the publishing process and the pitfalls that goes with it. So, I will not list those here. Instead I would like to propose an alternative to traditional publishing.
The main idea is very simple: instead of publishing individual articles, we should copy the modern software development process more or less straight-off. Someone, or several, maintains a "publication". This is simply a git (or some other vcs) repository containing something resembling a book. Any scientist that has something to contribute to the book can send in a traditional pull request to complement the publication.
There will be so many advantages with this: less writing for each published unit, more consistency, an updated modern and efficient peer review process, open access, simple referencing (by making versioned releases or referencing a commit hash).