Should one or two reserved sigils be available for implementation use?

[eemeli]

As discussed in #374 (comment):

@eemeli: Should we include some explicit language about whether implementations are allowed to define their own meanings for parts of the reserved space?

@aphillips: We should not allow that unless we specifically reserve one or more sigils for private use. It was on my mind writing this to propose private use, but I deferred those thoughts in favor of getting this in.

What we don't want is for vendor foo to use (let's say) ~ for private use and then MF2.1 to adopt ~ for something else. Private agreement is another thing and leaving room for that is probably a Good Idea.

[aphillips]

This will require making private separate from (and equal to) reserved, e.g.:

annotation = (function *(s option)) / private / reserved
; reserve sigils for private agreement use
; the contents of both 'private' and 'reserved' are undefined so that
;   implementations are free to define whatever they need
private        = private-start reserved-body
private-start  = "&" / "@" / "~" ; semi-random choices :-)
; reserve additional sigils for future use
reserved       = reserved-start reserved-body
reserved-start = "!" / "#" / "%" / "^" / "*" / "<" / ">" / "?"

[macchiati]

At least some people have misinterpreted the term 'reserved' in other contexts as meaning 'reserved for anyone to use'. Best to explicitly say:

; reserve additional sigils for future versions of this specification