RFC INDEX PAGE
Google
 

 
 
Network Working Group                        Marshall T. Rose (Delaware)
Request for Comments: 934                       Einar A. Stefferud (NMA)
                                                            January 1985
 
              Proposed Standard for Message Encapsulation
 
 
STATUS OF THIS MEMO
 
   This RFC suggests a proposed protocol for the ARPA-Internet
   community, and requests discussion and suggestions for improvements.
   Distribution of this memo is unlimited.
 
Introduction, Scope, and Motivation
 
   The services that a user agent (UA) can offer are varied.  Although
   all outgoing mail may be thought of as going through a single posting
   slot to connect to the message transport system (MTS), it is possible
   to consider a message draft being posted as described by one of the
   following four types of postings:
 
      Originate - a new message is composed from scratch, which, to the
      knowledge of the UA, is unrelated to any message previously
      handled by the user.
 
      Reply - a message is composed as a reply to a message previously
      received by the user.  In most circumstances, the UA aids the user
      in composing the reply by constructing the header portion of the
      message draft, using components extracted from the received
      message headers.
 
      Forward - one more more messages previously received by the user
      are formatted by the UA as a part of the body portion of the
      draft.  In this sense, a "digest" for an interest group may be
      considered as forwarding.  Similarly, an argument may be made that
      "blind-carbon-copies" should also be handled in this fashion.
 
      Distribute - a message previously received by the user is
      re-posted to the MTS.  The draft being re-posted is identical to
      the original message with the exception that certain "ReSent-XXX"
      headers are appended to the headers portion of the draft, and the
      "Return-Path" header is reset to reference the re-sender's
      address.  (See [RFC-821] for a discussion of the Return-Path
      header.)
 
   Most user agents support the first two of these activities, many
   support the first three, and a few support all four.
 
   This memo concerns itself only with the third type, which is message
   forwarding.  (For a brief treatment of the semantics of message
   components with respect to replies, see [RFC-822].) In many ways,
 
 
Rose & Stefferud                                                [Page 1]

RFC 934 January 1985 Message Encapsulation forwarding can be thought of as encapsulating one or more messages inside another. Although this is useful for transfer of past correspondence to new recipients, without a decapsulation process (which this memo terms "bursting"), the forwarded messages are of little use to the recipients because they can not be distributed, forwarded, replied-to, or otherwise processed as separate individual messages. NOTE: RFC-822 mistakenly refers to distribution as forwarding (section 4.2). This memo suggests below, that these two activities can and should be the same. In the case of an interest group digest, a bursting capability is especially useful. Not only does the ability to burst a digest permit a recipient of the digest to reply to an individual digested message, but it also allows the recipient to selectively process the other messages encapsulated in the digest. For example, a single digest issue usually contains more than one topic. A subscriber may only be interested in a subset of the topics discussed in a particular issue. With a bursting capability, the subscriber can burst the digest, scan the headers, and process those messages which are of interest. The others can be ignored, if the user so desires. This memo is motivated by three concerns: In order to burst a message it is necessary to know how the component messages were encapsulated in the draft. At present there is no unambiguous standard for interest group digests. This memo proposes such a standard for the ARPA-Internet. Although interest group digests may appear to conform to a pseudo-standard, there is a serious ambiguity in the implementations which produce digests. By proposing this standard, the authors hope to solve this problem by specifically addressing the implementation ambiguity. Next, there is much confusion as to how "blind-carbon-copies" should be handled by UAs. It appears that each agent in the ARPA-Internet which supports a "bcc:" facility does so differently. Although this memo does not propose a standard for the generation of blind-carbon-copies, it introduces a formalism which views the "bcc:" facility as a special case of the forwarding activity. Finally, both forwarding and distribution can be accomplished with the same forwarding procedure, if a distributed message can be extracted as a separate individually processable message. With a proper bursting agent, it will be difficult to distinguish between Rose & Stefferud [Page 2]
RFC 934 January 1985 Message Encapsulation a message which has been distributed and a message which has been extracted from a forwarded message. This memo argues that there is no valuable distinction to be made, between forwarding and distribution, and that in the interests of simplicity, distribution facilities should not be generally available to the ordinary users of a message system. However, this memo also argues that such facilities should be available to certain trusted entities within the MTS. NOTE: this memo does not propose that the distribution facility be abolished. Rather it argues the case forcefully in the hope that other interested parties in the ARPA-Internet will join this discussion. Message Encapsulation This memo proposes the following encapsulation protocol: two agents act on behalf of the user, a forwarding agent, which composes the message draft prior to posting, and a bursting agent which decomposes the message after delivery. Definitions: a draft forwarding message consists of a header portion and a text portion. If the text portion is present, it is separated from the header portion by a blank line. Inside the text portion a certain character string sequence, known as an "encapsulation boundary", has special meaning. Currently (in existing digestification agents), an encapsulation boundary (EB) is defined as a line in the message which starts with a dash (decimal code 45, "-"). Initially, no restriction is placed on the length of the encapsulation boundary, or on the characters that follow the dash. 1. The Header Portion This memo makes no restriction on the header portion of the draft, although it should conform to the RFC-822 standard. 2. The Text Portion The text of the draft forwarding message consists of three parts: an initial text section, the encapsulated messages, and the final text section. 2.1. The Initial Text Section All text (if any) up to the first EB comprises the initial text section of the draft. This memo makes no restrictions on the Rose & Stefferud [Page 3]
RFC 934 January 1985 Message Encapsulation format of the initial text section of the draft. In the case of a digest, this initial text is usually the "table of contents" of the digest. 2.2. The Final Text Section All text (if any) after the last EB composes the final text section of the draft. This memo makes no restrictions on the format of the final text section of the draft. In the case of a digest, this final text usually contains the sign-off banner for the digest (e.g., "End of FOO Digest"). 2.3. Encapsulated Messages Each encapsulated message is bounded by two EBs: a pre-EB, which occurs before the message; and, a post-EB, which occurs after the message. For two adjacent encapsulated messages, the post-EB of the first message is also the pre-EB of the second message. Consistent with this, two adjacent EBs with nothing between them should be treated as enclosing a null message, and thus two or more adjacent EBs are equivalent to one EB. Each encapsulated message consists of two parts: a headers portion and a text portion. If the text portion is present, it is separated from the header portion by a blank line. 2.3.1. The Header Portion Minimally, there must be two header items in each message being forwarded, a "Date:" field and a "From:" field. This differs from RFC-822, which requires at least one destination address (in a "To:" or "cc:" field) or a possibly empty "Bcc:" field. Any addresses occuring in the header items for a message being forwarded must be fully qualified. 2.3.2. The Text Portion This memo makes no restrictions on the format of the text portion of each encapsulated message. (Actually, this memo does restrict the format of the text portion of each encapsulated message, but these restrictions are discussed later.) Before summarizing the generation/parsing rules for message encapsulation, two issues are addressed. Rose & Stefferud [Page 4]
RFC 934 January 1985 Message Encapsulation Compatibility with Existing User Agents The above encapsulation protocol is presently used by many user agents in the ARPA-Internet, and was specifically designed to minimize the amount of changes to existing implementations of forwarding agents in the ARPA-Internet. However, the protocol is not exactly like the pseudo-standard used by those forwarding agents that compose digests. In particular, the post-EB of all messages encapsulated in a digest is preceeded and followed by by a blank line. In addition, the first message encapsulated in a digest has a pre-EB that is followed by a blank line, but usually isn't preceeded by a blank line (wonderful). This memo recommends that implementors of forwarding agents wishing to remain compatible with existing bursting agents consider surrounding each EB with a blank line. It should be noted that blank lines following a pre-EB for an encapsulated message must be ignored by bursting agents. Further, this memo suggests that blank lines preceeding a post-EB also be ignored by bursting agents. NOTE: This recommendation is made in the interest of backwards-compatibility. A forwarding agent wishing to strictly adhere to this memo, should not generate blank lines surrounding EBs. Character-Stuffing the Encapsulation Boundary It should be noted that the protocol is general enough to support both general forwarding of messages and the specific case of digests. Unfortunately, there is one issue of message encapsulation which apparently is not addressed by any forwarding agent (to the authors' knowledge) in the ARPA-Internet: what action does the forwarding agent take when the encapsulation boundary occurs within a the text portion of a message being forwarded? Without exception, this circumstance is ignored by existing forwarding agents. To address this issue, this memo proposes the following character-stuffing scheme: the encapsulation boundary is defined as a line which starts with a dash. A special case is made for those boundaries which start with a dash and are followed by a space (decimal code 32, " "). During forwarding, if the forwarding agent detects a line in the text portion of a message being forwarded which starts with the encapsulation boundary, the forwarding agent outputs a dash followed by a space prior to outputting the line. Rose & Stefferud [Page 5]
RFC 934 January 1985 Message Encapsulation During bursting, if the bursting agent detects an encapsulation boundary which starts with a dash followed by a space, then the bursting agent does not treat the line as an encapsulation boundary, and outputs the remainder of the line instead. This simple character-stuffing scheme permits recursive forwardings. Generation/Parsing Rules for Message Encapsulation The rules for forwarding/bursting are described in terms of regular expressions. The first author originally derived simple finite-state automata for the rules, but was unable to legibly represent them in this memo. It is suggested that the implementors sketch the automata to understand the grammar. The conventions used for the grammar are simple. Each state is followed by one or more alternatives, which are separated by the "|" character. Each alternative starts with a character that is received as input. (CRLF, although two characters is treated as one character herein.) The last alternative for a state is the character "c", which represents any character not specified in the preceeding alternatives. Optionally following the input character is an output string enclosed by curly-braces. Following this is the state that the automata enters. The reader should note that these grammars are extremely simple to implement (and, in most cases, can be implemented quite efficiently). When the forwarding agent encapsulates a message, it should apply the following finite-state automaton. The initial state is S1. S1 :: CRLF {CRLF} S1 | "-" {"- -"} S2 | c {c} S2 S2 :: CRLF {CRLF} S1 | c {c} S2 This simply says that anytime a "-" is found at the beginning of a line, a "- " is output prior to outputting the line. Rose & Stefferud [Page 6]
RFC 934 January 1985 Message Encapsulation When the bursting agent decapsulates the text portion of a draft, it should apply the following finite-state automaton. The initial state is S1. S1 :: "-" S3 | CRLF {CRLF} S1 | c {c} S2 S2 :: CRLF {CRLF} S1 | c {c} S2 S3 :: " " S2 | c S4 S4 :: CRLF S5 | c S4 S5 :: CRLF S5 | c {c} S2 Although more complicated than the grammar used by the forwarding agent to encapsulate a single message, this grammer is still quite simple. Let us make the simplifying assumption that both the initial and final text sections of the draft are messages in addition to the encapsulated messages. To begin, the current message being burst is scanned at state S1. All characters are output until the EB is found (state S3). If "- " is found, the automaton enters state S2 and characters from the current message are continued to be output. Finally, a true EB is found (state S4). As the automaton traverses from state S3 to S4, the bursting agent should consider the current message ended. The remainder of the EB is discarded (states S4 and S5). As the automaton traverses from state S5 to S2, the bursting agent should consider a new message started and output the first character. In state S2, all characters are output until the EB is found. Blind Carbon Copies Many user agents support a blind-carbon-copy facility. With this facility a draft has two types of addressees: visible and blind recipients. The visible recipients are listed as addresses in the "To:" and "cc:" fields of the draft, and the blind recipients are listed as addresses in the "Bcc:" fields of the draft. The basis of this facility is that copies of the draft which are delivered to the recipients list the visible recipients only. Rose & Stefferud [Page 7]
RFC 934 January 1985 Message Encapsulation One method of achieving this is to post a single draft, which lacks any "Bcc:" fields, and, during posting, to interact with the MTS in such a way that copies are sent to both the visible and blind recipients. Unfortunately, a key problem with this arrangement is that the blind recipients can accidently reply to the draft in such a way that the visible recipients are included as addressees in the reply. This is socially unacceptable! To avoid this problem, the message which the visible recipients receive must be different than the message which the blind recipients receive. A second method is to post two drafts. The first, which goes to the visible recipients, is simply the draft without any "Bcc:" fields. The second, which goes to the blind recipients, is simply the draft with some string prepended to any "To:" and "cc:" field. For example, the user agent might prepend "BCC-" to these fields, so that the blind recipients get a draft with "BCC-To:" and "Bcc-cc:" fields and no "To:" or "cc:" fields. Unfortunately, this is often very confusing to the blind recipients. Although accidental replies are not possible, it is often difficult to tell that the draft received is the result of a blind-carbon-copy. The method which this memo suggests is to post two drafts, a visible draft for the visible recipients, and a blind draft for the blind recipients. The visible draft consists of the original draft without any "Bcc:" fields. The blind draft contains the visible message as a forwarded message. The headers for the blind draft contain the minimal RFC-822 headers and, if the original draft had a "Subject:" field, then this header field is also included. In addition, the user agent might explicitly show that the blind draft is the result of a blind-carbon-copy, with a "Bcc" header or prior to the first encapsulating boundary in the body. Message Distribution The main purpose of message distribution (often called redistribution or resending) is to provide to a secondary recipient, perhaps not included among the original addressees, with a "true original" copy that can be treated like an original in every respect. Such distribution is most often done by discussion group moderators who use automated agents to simply repost received messages to a distribution list. The better automatic distribution agents insert a new "Return-Path" header field to direct address failure notices to the discussion group address list maintainer, rather than to the original author. This form of distribution is encouraged because it Rose & Stefferud [Page 8]
RFC 934 January 1985 Message Encapsulation most simply serves to deliver messages to discussion group recipients as processable originals. It is performed by trusted pseudo-MTS agents. A second kind of distribution is that done by individuals who wish to transfer a processable copy of a received message to another recipient. This second form is discouraged in various new standards for message transfer. These include the NBS Standard for Mail Interchange [FIPS-98], and the recent CCITT draft MHS (Mail Handling Systems) X.400 standards [X.400]. In place of direct reposting of received messages as though they are new drafts, the recommendation is to forward the received message in the body of a new draft from which is can be extracted by its secondary recipient for further processing. It is in support of this recommendation that this standard for encapsulation/decapsulation is proposed. Rose & Stefferud [Page 9]
RFC 934 January 1985 Message Encapsulation References [RFC-822] D.H. Crocker. "Standard for the Format of ARPA-Internet Text Messages", University of Delaware. (August, 1982) [RFC-821] J.B. Postel. "Simple Mail Transfer Protocol", USC/Information Sciences Institute. (August, 1982). [FIPS-98] National Bureau of Standards. "Specification for Message Format for Computer Based Message Systems." (January, 1983). [X.400] Consultative Committee on International Telephone and Telegraph. "DRAFT Recommendation X.400. Message Handling Systems: System Model-Service Elements." Authors' Addresses Marshall T. Rose Department of Computer and Information Sciences University of Delaware Newark, DE 19716 MRose@UDel.ARPA Einar A. Stefferud Network Management Associates, Inc. 17301 Drey Lane Huntington Beach, CA 92647 Stef@UCI.ARPA Rose & Stefferud [Page 10]

The HTML presentation of this document is copyrighted by Grafman Productions.
Copyright © 1997 - Grafman Productions - ALL RIGHTS RESERVED
Grafman Productions