encoding.)

    (6)   Trailing "white space" characters (SPACE, TAB (HT)) on
          a line may be discarded by some transport agents, while
          other transport agents may pad lines with these
          characters so that all lines in a mail file are of
          equal length.  The persistence of trailing white space,
          therefore, must not be relied on.

    (7)   Many mail domains use variations on the US-ASCII
          character set, or use character sets such as EBCDIC
          which contain most but not all of the US-ASCII
          characters.  The correct translation of characters not
          in the "invariant" set cannot be depended on across
          character converting gateways.  For example, this
          situation is a problem when sending uuencoded
          information across BITNET, an EBCDIC system.  Similar
          problems can occur without crossing a gateway, since
          many Internet hosts use character sets other than US-
          ASCII internally.  The definition of Printable Strings
          in X.400 adds further restrictions in certain special
          cases.  In particular, the only characters that are
          known to be consistent across all gateways are the 73
          characters that correspond to the upper and lower case
          letters A-Z and a-z, the 10 digits 0-9, and the
          following eleven special characters:

            "'"  (US-ASCII decimal value 39)
            "("  (US-ASCII decimal value 40)
            ")"  (US-ASCII decimal value 41)
            "+"  (US-ASCII decimal value 43)
            ","  (US-ASCII decimal value 44)
            "-"  (US-ASCII decimal value 45)
            "."  (US-ASCII decimal value 46)
            "/"  (US-ASCII decimal value 47)
            ":"  (US-ASCII decimal value 58)
            "="  (US-ASCII decimal value 61)
            "?"  (US-ASCII decimal value 63)

          A maximally portable mail representation will confine
          itself to relatively short lines of text in which the
          only meaningful characters are taken from this set of
          73 characters.  The base64 encoding follows this rule.

    (8)   Some mail transport agents will corrupt data that
          includes certain literal strings.  In particular, a




 
RFC 2049                    MIME Conformance               November 1996


          period (".") alone on a line is known to be corrupted
          by some (incorrect) SMTP implementations, and a line
          that starts with the five characters "From " (the fifth
          character is a SPACE) are commonly corrupted as well.
          A careful composition agent can prevent these
          corruptions by encoding the data (e.g., in the quoted-
          printable encoding using "=46rom " in place of "From "
          at the start of a line, and "=2E" in place of "." alone
          on a line).

   Please note that the above list is NOT a list of recommended
   practices for MTAs.  RFC 821 MTAs are prohibited from altering the
   character of white space or wrapping long lines.  These BAD and
   invalid practices are known to occur on established networks, and
   implementations should be robust in dealing with the bad effects they
   can cause.

4.  Canonical Encoding Model

   There was some confusion, in earlier versions of these documents,
   regarding the model for when email data was to be converted to
   canonical form and encoded, and in particular how this process would
   affect the treatment of CRLFs, given that the representation of
   newlines varies greatly from system to system.  For this reason, a
   canonical model for encoding is presented below.

   The process of composing a MIME entity can be modeled as being done
   in a number of steps.  Note that these steps are roughly similar to
   those steps used in PEM [RFC-1421] and are performed for each
   "innermost level" body:

    (1)   Creation of local form.

          The body to be transmitted is created in the system's
          native format.  The native character set is used and,
          where appropriate, local end of line conventions are
          used as well.  The body may be a UNIX-style text file,
          or a Sun raster image, or a VMS indexed file, or audio
          data in a system-dependent format stored only in
          memory, or anything else that corresponds to the local
          model for the representation of some form of
          information.  Fundamentally, the data is created in the
          "native" form that corresponds to the type specified by
          the media type.