Revision histories in user docs

[Petr Bokoc]

Hi docs,

For the longest time we used to maintain a revision history for each 
book we published on docs.fp.o. However, they weren't particularly 
useful, each revision was noted by the date, author, and an unhelpful 
note like "updating for Fedora 21" or "async update", without a list
of 
specific changes made. The revisions were basically only added because 
the old toolchain used for Red Hat docs required it, and contributors 
from Red Hat built a habit of adding them and did it in Fedora as well.

The current books (Release Notes, Installation Guide, and System 
Administrator's Guide) still have revision histories, but the Install 
Guide and Sysadmin Guide have last entries from 2016 and 2017 and the 
Release Notes have an empty one (which makes sense since we completely 
rewrite the book for each release, but I don't think people add new 
entries if they republish). It's more confusing than useful at this 
point, and I think we should do something about it.

The way I see it, we have two choices:

* Start maintaining proper revision histories with much more detail. 
This would require every PR or commit to also add something to the 
revision history, which means another thing to keep in mind and a little 
bit of extra work. There's also a potential problem about the date: when 
you add a revhistory entry, you have no way of knowing when it's 
actually going to be published, so you have to use the current date - 
but the reader doesn't care about when you commited it, the only use for 
a date in a revhistory is for the reader to see if anything changed 
since they last checked.

* Just get rid of revision histories altogether. We'd lose a potentially 
useful feature but in its current state it's useless anyway.

A third way would be to automatically insert a "last modified" timestamp 
in every page's top bar; the timestamp would ideally contain a link to a 
set of commit diffs relevant to that page with dates. That seems 
difficult to implement though, so I don't think that's an option - 
unless we could insert that timestamp during the CI/CD process when we 
have one, I suppose.

What do you all think?

Petr