The rule which
requires MIME MUAs to treat unrecognized subtypes of text as "text/plain"
is documented in both RFC 2046 and RFC 2049. The fact that three MIME
MUAs failed one of the most basic MIME requirements is unacceptable.
Isn't the problem here really that (1) customer demand trumps standards,
and (2) as Laurence pointed out, some folks are registering (and using!)
invalid subtypes of "text", like "text/html", "text/sgml", and
"text/rtf"? MUA vendors don't want to put gibberish up on their
customer's screens, and there are registered subtypes of "text" which
look like gibberish to normal users when presented directly, so the only
sensible MUA rule for commercial vendors is to censor anything they
don't know to be correctly presentable.
Perhaps, but the data from this survey doesn't support this conclusion.
Specifically, we have three MUAs that have problems, but only one of them
(Outlook) is a relatively recent design. At least one of the other two
(ZMailer) was designed quite some time ago, considerably before such problems
were readily evident. I also suspect that even with Outlook the issue is less
one of "mustn't throw up unintelligible stuff" than it is one of "how we map an
open-ended set of types into our internal typing system that's based on file
extensions".
There is also considerable evidence that when shoe is on the other foot, so to
speak, and it is noncompliance with MIME that leads to bad displays full of
garbage, this fact hasn't stopped MUAs from breaking the rules. For example, if
you look at proper handling of the charset parameter, you'll find that many
agents don't bother to check it even though MIME says you're supposed to. And
similarly, if you look at compliance in the area of message/rfc822 handling,
you'll find that a fair number of agents didn't bother to do it until we raised
the bar a bit. And failure to comply here often leads to screens full of
base64.
What I do think we have is evidence of something we've known from the start,
which is "developers often implement things incorrectly and once implemented,
things tend not to get fixed, even when users complain". Scarcely a major
insight, to be sure, but about the only one I think can be had from the data.
I also believe that text/html is actually legal and registered. The others
aren't, but they are also fairly rare.
I remember raising this point during the original MIME discussions (when
some folks were proposing text/postscript :-). The fact is, by allowing
arbitrary registration, we remove the ability to say anything in general
about the so-called top-level media types, because people are always
going to misuse them. I'd suggest just facing that fact, and removing
any RFC requirements that are supposed to apply to all instances of any
top-level type.
Actually, we do not allow arbitrary registrations and never have. The number of
things that have been registered until text/ is actually quite small. And while
you may argue that text/html was a mistake, it isn't like it wasn't discussed
at considerable length. We have only ourselves to blame if it is a botch.
Ned