> arbitrary character sets [aren't] even remotely worth the hassles
Amen. But in rfc2440bis-02, this is the complete definition:
Concatenating the following data creates ASCII Armor
and:
- "Comment", a user-defined comment.
There is nothing in the spec which requires an implementation to
drop non-ascii characters. Currently, an implementation which
emits 0x80-0xFF octets in an Armor Header is rfc2440-compliant.
I don't like this, but that's the way it reads.
Certain implementations are already emitting native 8-bit text
in comments. (...presumably by accident...)
If you want to ensure that ASCII Armor blocks contain only
the basic 95 printable characters, you'll need to put something
in section 6.2 which says that the 'data' which is concatenated
to form ASCII Armor must itself be pure 0x20-0x7E ASCII.
Currently this is not part of the spec.
Even something as simple as:
"Comment", a user-defined 7-bit ASCII comment.
might be sufficient.