Ok, how about the following:
The backward compatibility problem probably mainly exists for
unpack("C*") which has traditionally been used to get to the
underlying bytes of an encoded string. In my patch C was a bit
special anyways (I made >= 256 a croak to make it easy to catch
that style of usage), and for pack it has the property of wrapping
for >= 256. With the companion patch I plan for pack that would no
longer be needed, but changing it would be another break with backward
compatibility. So what if I leave "C" as old style "look through
the encoding", make all other formats (except "c" and "C") encoding
neutral and introduce a new letter for encoding neutral "character",
let's say "E" (suggestions for a better letter welcome, a pity "u" is
already taken).
I'd still like to make U0 and C0 mode into my "reversed" interpretation,
but now that unpack("C*") ignores the mode I think the impact will be
very low and worth doing to make pack and unpack consistent.
