Sorry for my tardy response ...
Rick Jelliffe wrote:
Perhaps it would be better to redefine text/* to clarify that readability
by naive readers is not the defining characteristic, but that the freedom
of intermediate systems to transcode it (perhaps even at the cost of
losing characters, how awful) is.
It sounds like you're suggesting that only text/* types can be
transcoded. I disagree. text/* types are currently (AFAIK) the only
types to be transcoded because the user agent has additional information
about the encodings on these types that allow it to always access the
content. While it's true that processors of application/* types don't
have this additional information available to them, the use of "+xml"
changes that, and allows transcoding.
Note that this isn't written down anywhere. However, in considering
application/xhtml+xml vs. text/xhtml+xml, I did thoroughly review the
MIME RFCs and draft-murata-xml to ensure that nothing precluded
application/xhtml+xml (or any application/*+xml type) described-content
from being transcoded.
I hope this is consistent with others' views.
MB