For 3028bis, 2.7.2, paragraph 2, how about:
Comparisons are performed in Unicode. Implementations convert
text from header fields in all charsets [HEADER-CHARSET] to
Unicode as input to the comparator (see 2.7.3). Implementations
must be capable of decoding US-ASCII, ISO-8859-1, the US-ASCII
subset of ISO-8859-* character sets, and UTF-8.
Yeah, that's good.
Hmm, I think the paragraph needs to also specify that text in unknown
charsets never matches, no?
I think you're right.
Tim