Keith Moore writes:
I don't share your confidence that this happens in every environment,
for every program that generates header fields.
That isn't your problem. If you want C-normalized text on the wire,
specify C-normalized text on the wire. You don't need to know whether
software accomplishes this by having text C-normalized in the keyboard
interface (as everyone does now) or in the MUA (which would work too).
Anyway, your hypothetical Unicode normalization problems would also
arise in normalizing RFC 2047, so you can't use normalization as an
argument against moving to UTF-8.
Which is precisely why 2047 was never intended for anything that needs
to be interpreted by machine.
Even with that (shortsighted) restriction, RFC 2047 was badly designed.
For example, you should have declared encoding in a separate header
field, not by inserting bytes into existing header fields. You should
have required 8-bit-clean software in 1991; pandering to 7-bit software
was a huge mistake. You should have required UTF-8 support in 1994. You
should have required UTF-8 as a default in 1998.
---D. J. Bernstein, Associate Professor, Department of Mathematics,
Statistics, and Computer Science, University of Illinois at Chicago