You say that there are implementations out there that implement the
older syntax, and that other implementations use a 2822 compatible
syntax. So for this particular issue, I think you should:
Explain (and describe) the two implemented syntaxes; one that is
incompatible with 2822 and a subset that IS compatible. Then explain
that new implementations MUST use the compatible format.
The same goes for any other issue: start your description with what's
currently out there. Then, if necessary, describe what new
implementations SHOULD or MUST do.
Remember, people aren't just writing new code to generate the stuff.
People also have to implement code that accepts the older stuff.
Maybe we could use the example I gave in a previous message about the
Received-SPF header. Should I document what was allowed under the
older SPF specs, which doesn't even match RFC2822's "obsolete" syntax,
or should I document something that is largely compatible with all
implementations that I know of and compatible with RFC2822? I think
it reasonable to say that an already large percent of Received-SPF
headers conform to what is in the I-D (e.g. RFC2822 compatible), and
that in the future, this percentage will increase.