When Rip van Winkle awoke from sleep he found the model of society
changed. Similarly I couldn't understand what people were talking about
on this issue. When Dave Crocker wrote
It makes no sense to
have quoted-printable =od=oa be interpreted as newline.
I assumed that he meant that =0d=0a would be data in the line. I
gather from Alan Fontaine that in the context of the full discussion
he clearly means that =0d=0a should be illegal.
Of course I was well aware that my model no longer worked well with
respect to BASE64 so I should have been suspicious. Anyway the new
model is that the UNDERLYING form for text/anything is a sequence
of bytes in which end of line is represented by 0d0a(=CRLF), just as
it is for the ENCODED message. This has some advantages:
1. It makes BASE64 simpler (removes portable end of line).
2. Make TEXT more like other types in that its underlying representation
can be a sequence of octets instead of a sequence of lines.
The disadvantages are
1. It will be a real mistake if we later get character sets which
contain 0d0a within a multi-octet character. However allowing
0d0a in a line using q-p and not when using BASE64 would be a really
strange way to solve the problem.
2. CRLF is really a transport artifact. Having it mean EOL in the
underlying message is artificial, and it shows.
Well I liked the old system (and my model of it) better. But now that
it is changed I hope it will be explicit in the document. Meanwhile
I'll go back to sleep. Sorry if I disturbed anyone. Thanks to Alan
for explaining things to me.
Bob Smart
P.S. better make sure that <CR>=0a and =0d<LF> are also illegal :-).