Moreover, 2.10 says (albeit oddly) that CRLF gets normalized
to LF everywhere, and the LF would get normalized to a single
space inside of an attribute (or Public Identifier) value.
There's an inconsistency somewhere with respect to CRLF handling,
but the "always normalize CRLF to LF" approach makes things
generally consistent. One issue is that 2.10 only identifies
this in a parenthetical (non-normative?) comment, and the text
there that's normative is more restrictive.
- Dave