a few comments on the first part of the draft
[this bounced for some reason, so I'm resending]
In clause 2.3, the doctype decl is given as:
'<!DOCTYPE' (Name | S)+ ('[' [^]]* ']')? '>'
Unless I'm confused by the notation, that appears to allow Name
to come after DOCTYPE with no intervening space. It also appears
to allow any number of space-separated Names before the internal subset.
(I think Chris Maden mentioned this too.) Finally, it appears not
to allow any ] characters in the internal subset, but not only
could they appear in attribute values, but the conditionalSect
construct uses them. In fact, it uses ']]>' so any trivial
grammar that stops at the first occurrence of ]> is going to
screw up the internal subset.
Also in the trivial grammar, the processing instruction is given as:
'<?' [^?]* ('?' [^>]+)* '?>'
That seems to say that, once you've seen the first ? in the content
of the PI, you can no longer have any >'s in the PI. Is that correct?
Would the latter be better:
'<?' [^?]* ('?' [^>] [^?]*)* '?>'
(Likewise in clause 2.5. Processing Instructions.)
Also in clause 2.3, the spec suggests &sqot; and " for single
and double quotes respectively "to allow attribute values to contain
both single and double quotes." Whereas " is defined in ISO
Numeric and Special Graphic characters, &sqot; is not in any of
the defns with which I'm familiar. Nor can I find an entity among
the ISO set for single, neither left nor right, not rising or whatever
quote. Instead of inventing one, I'd suggest just not defining
anything for a single quote. You only need to define one for
double quote to allow attribute values (delimited by double quotes)
to contain both single quotes and double quotes. Let's toss this
&sqot; (or whatever, regardless of the name) and just mention ".
I decided not to work on  the MsData definition at this time of night.
In 2.9, is the value of the <?XML RMD=... PI supposed to be quoted
or not? Though it may not have to be quoted grammar-wise, I think
so many people will be used to quoting such things that we should
at least allow it optionally to be quoted.