Re: Last Call: 'The APPLICATION/MBOX Media-Type' to Proposed Standard

On Mon, 16 Aug 2004 22:47:52 EDT, Tony Hansen said:

> The claim in Appendix A is that there were no authoritative sources of 
> documentation for the mbox formats and otherwise it's "only documented 
> in anecdotal form". I'm sorry, but the the definitions ARE there, and 
> ARE almost always authoritative for those systems.

Somehow, I can't get thrilled by the concept of saying a format is documented
because we have (for example) 3 systems, and each has an authoritative
definition of the version it uses, and the definitions are incompatible (and
yes, the Solaris 'content-length:' scheme and '>from ' escaping are basically
incompatible - there exist messages that can't be converted from one to the
other without information loss).

> Because Solaris 8 is System Vr4-derived, you should look at 'man mail' 
> for the definitive definition. You'll find Content-Length: documented there.

It says:

     A letter is composed of some  header  lines  followed  by  a
     blank line followed by the message content. The header lines
     section of the letter consists of one  or  more  UNIX  post-
     marks:

           From     sender     date_and_time     [remote     from
          remote_system_name]

     followed by one or more standardized message header lines of
     the form:

           keyword-name: [printable text]

     where keyword-name  is  comprised  of  any  printable,  non-
     whitespace  characters  other  than  colon (`:'). A Content-
     Length: header line, indicating the number of bytes  in  the
     message  content  will  always  be present unless the letter
     consists of only header lines with  no  message  content.

For bonus points - is the 'crlf-crlf' between the header and the body included
in the Content-Length:?  There's other issues as well - what if the
Content-Length: is computed across a non-canonified message - how do
you send it across the wire?

'man mail' doesn't mention escaping a 'From ' inside a message,
except for this:

     The default mode for printing messages is  to  display  only
     those header lines of immediate interest. These include, but
     are not limited to,  the  UNIX  From  and  >From  postmarks,
     From:,  Date:,  Subject:,  and Content-Length: header lines,
     and any recipient header lines such as To:, Cc:,  Bcc:,  and
     so  forth.  After the header lines have been displayed, mail

Of course, that's because Solaris doesn't use '>From ' escaping
because it has Content-Length instead.

Should other systems trust the value of a Content-Length:?

Should other systems be required to include a Content-Length?

Should other systems escape a 'From ' iff there's no Content-Length?

What if an mbox file has a Content-Length on some items but not others?

How do you recover from a corrupted Content-Length?

So - where is the *one true canonical* definition of an mbox that actually
answers all these basic questions that an implementer *needs* to know the
answer to?