Network Working Group                                       T. Showalter
Internet Draft: Sieve                                    Carnegie Mellon
Document: draft-showalter-sieve-04.txt                       August 1998
Expire in six months (31 January 1999)

                   Sieve -- a Mail Filtering Language

Status of this memo

   This document is an Internet-Draft.  Internet-Drafts are working
   documents of the Internet Engineering Task Force (IETF), its areas,
   and its working groups.  Note that other groups may also distribute
   working documents as Internet-Drafts.

   Internet-Drafts are draft documents valid for a maximum of six months
   and may be updated, replaced, or obsoleted by other documents at any
   time.  It is inappropriate to use Internet-Drafts as reference
   material or to cite them other than as ``work in progress.''

   To learn the current status of any Internet-Draft, please check the
   ``1id-abstracts.txt'' listing contained in the Internet-Drafts Shadow
   Directories on (Africa), (Europe), (Pacific Rim), (US East Coast), or (US West Coast).

   The protocol discussed in this document is experimental and subject
   to change.  Persons planning on either implementing or using this
   protocol are STRONGLY URGED to get in touch with the author before
   embarking on such a project.


   Copyright (C) The Internet Society 1998.  All Rights Reserved.


   This document describes a mail filtering language for filtering
   messages at time of final delivery.  It is designed to be independent
   of protocol, and implementable on either a mail client or mail
   server.  It is meant to be extensible, simple, and independent of
   access protocol, mail architecture, and operating system.  It is
   suitable for running on a mail server where users may not be allowed
   to execute arbitrary programs, such as on black box IMAP servers, as
   it has no variables, loops, or ability to shell out to external

Showalter                 Expires 31 Jan 1999                   [Page 1]

Internet DRAFT                   Sieve                    August 7, 1998

                           Table of Contents

Status of this memo
0. Meta-information on this draft
0.1. Discussion
0.2. Known Problems
0.2.1. Probable Extensions
0.2.2. Known Bugs
0.3. Open Issues
0.3. Noted Changes
1. Introduction
1.1. Conventions used in this document
1.2. Example mail messages
2. Design
2.1. Form of the language
2.2. Whitespace
2.4. Literal data
2.4.1. Numbers
2.4.2. Strings String lists Headers Addresses
2.5. Tests
2.5.1. String Comparison
2.6.1. Match Keyword
2.6.2. Comparators
2.7. Tagged Arguments
2.8. Blocks
2.9. Commands
2.9.1. Positional Arguments
2.9.2. Optional Arguments
2.9.3. Blocks as Arguments
2.11. Evaluation
2.11.1. Implicit keep
3. Conditionals and Control Structures
4. Actions
4.1. Action reject
4.2. Action fileinto
4.3. Action forward
4.4. Action keep

Showalter                 Expires 31 Jan 1999                   [Page 2]

Internet DRAFT                   Sieve                    August 7, 1998

4.6. Action stop
4.7. Action discard
5. Tests
5.1. Test allof
5.2. Test anyof
5.3. Test exists
5.4. Test false
5.5. Test header
5.6. Test not
5.7. Test size
6. Errors in Processing a Script
7. Extensibility
7.1. Capability String
7.2. Registry
7.3. Capability Transport
8. Transmission
9. Acknowledgments
10.  Formal Grammar
11. Security Considerations
12. Author's Address
Appendix A.  References
Appendix B. Full Copyright Statement

Showalter                 Expires 31 Jan 1999                   [Page 3]

Internet DRAFT                   Sieve                    August 7, 1998

Showalter                 Expires 31 Jan 1999                   [Page 2]

Internet DRAFT                   Sieve                    August 7, 1998

0. Meta-information on this draft

   This information is intended to facilitate discussion.  It will be
   removed when this document leaves the Internet-Draft stage.

0.1. Discussion

   This draft is being discussed on the MTA Filters mailing list at
   <>.  Subscription requests can be sent to
   <> (send an email message with the
   word "subscribe" in the body).  More information on the mailing list
   along with a WWW archive of back messages is available at

0.2. Known Problems

0.2.1. Probable Extensions

   The following suggestions have been made, and will probably be
   addressed by extensions.

   An extension for regular expressions will be written.  While regular
   expressions are of questionable utility for most users, the
   programmers writing implementations desperately want regular

   Envelope-matching commands are not readily supported by all mail
   systems, and putting them in the draft will result in a system that
   cannot be implemented by a mail architecture that does not adequately
   store envelopes.

   "Detailed" addressing or "sub-addressing" (i.e., the "foo" in an
   address "") is not handled, and will be moved
   to an extension for those systems that offer it.

   A vacation command has been requested for an extension; a preliminary
   draft exists and will be submitted to the internet-drafts repository.
   Vacation functionality is isn't in the draft because having vacation
   assumes you can store the addresses of people who have already
   received vacation notifications, which isn't always the case.

   A suggestion was made to set IMAP flags on delivery (e.g., \Flagged,
   \Deleted, \Answered, \Seen).

   An "include" command is not included, but has been suggested  for  an

Showalter                 Expires 31 Jan 1999                   [Page 4]

Internet DRAFT                   Sieve                    August 7, 1998

0.2.2. Known Bugs

   The formal grammar probably still has some holes in it.

   The reject command needs to be rechecked against the DSN

   The error-handling clauses of this specification may not be
   completely sensible, and may conflict.

   The discussion of the limits of actions is not there.  Only one
   forward should be allowed per message.  Keep and reject are mutually

0.3. Open Issues

   In the event that there is an error while processing a script, what
   happens?  The draft implies you file into INBOX, but what if you've
   already taken actions before you do this?  (The parts of the draft
   that require syntax checking stuff are all SHOULDs.)

   I tried to fill in some of the blanks in previous versions; among
   them, the description of what a rejected input message looks like,
   but it's still nearly incomplete.

   I moved the substring matching stuff out of the  header  command  and
   into  a  section  of  its  own  as  it  is  reusable  by  extensions.
   Suggestions on this section would be appreciated.

   I tried to fill in the blanks in the  section  on  extensibility  and
   borrowed  some stuff from the ACAP spec (specifically, the comparator
   registry), but it's probably not complete or good enough.

0.3. Noted Changes

   This draft was unfortunately rushed and probably contains numerous
   errors.  This is purely the fault of the editor.  For this reason,
   readers are asked to subscribe to the mailing list noted above for
   discussion, as well as some inevitable corrections.

   The grammar has changed substantially to allow easier modification by
   extensions at the expense of some additional definition in the rest
   of the draft.

   Support and reply have been removed, and the format of stringlists
   has changed.

Showalter                 Expires 31 Jan 1999                   [Page 5]

Internet DRAFT                   Sieve                    August 7, 1998

1. Introduction

   This memo documents a language that can be used to create filters for
   electronic mail. It is not tied to any particular operating system or
   mail architecture.  It requires the use of [IMAIL]-compliant
   messages, but otherwise should generalize to other systems that meet
   these criteria.

   The language is powerful enough to be useful, but limited in power in
   order to allow for a safe server-side filtering system.  The
   intention is to make it impossible for users to do anything more
   complex (and dangerous) than write simple mail filters, along with
   facilitating GUI-based editors. The language is not Turing-complete,
   and provides no way to write a loop or a function.  Variables are not

   Implementations of the language are expected to take place at time of
   final delivery, when the message is moved to the user-accessible
   mailbox.  In systems where the MTA does final delivery, such as and
   traditional UNIX mail, is reasonable to sort when the MTA deposits
   mail into the user's mailbox.  If the MTA does not do final delivery,
   or lacks the power to sort into separate mailboxes, as is the case
   under POP3, the MUA must do filtering into local-disk folders.

   There are a number of reasons to use a filtering system.  Mail
   traffic for most users has been increasing due both to increased
   usage of e-mail, the emergence of unsolicited email as a form of
   advertising, and increased usage of mailing lists.

   Experience at Carnegie Mellon has shown that if a filtering system is
   made available to users, many will make use of it in order to file
   messages from specific users or mailing lists.  However, many others
   did not make use of the Andrew system's FLAMES [FLAMES] filtering
   language due to difficulty in setting it up.

   Because of the expectation that users will make use of filtering if
   it is offered and easy to use, this language has been made simple
   enough to allow many users to make use of it, but rich enough that it
   can be used productively.  However, it is expected that GUI-based
   editors will be the preferred way of editing filters for a large
   number of users.

1.1. Conventions used in this document

   In examples, line breaks have been inserted for readability.

   In the sections of this document that discuss the requirements of
   various keywords and operators, the following conventions have been

Showalter                 Expires 31 Jan 1999                   [Page 6]

Internet DRAFT                   Sieve                    August 7, 1998


   The key words "MUST", "MUST NOT", "SHOULD", "SHOULD NOT", "CAN", and
   "MAY" in this document are to be interpreted as defined in

   Each section on a test, action, or control structure has a line
   labeled "Syntax:".  This line describes the syntax of the command,
   including its name and its arguments.  Required arguments are listed
   inside angle brackets ("<" and ">").  Optional arguments are listed
   inside square brackets ("[" and "]").  However, the formal grammar
   for these commands in section 10 and is the authoritative reference
   on how to construct these commands.

1.2. Example mail messages

   The following mail messages will be used throughout this document  in

   Message A
   Date: Tue, 1 Apr 1997 09:06:31 -0800 (PST)
   Subject: I have a present for you

   Look, I'm sorry about the whole anvil thing, and I really
   didn't mean to try and drop it on you from the top of the
   cliff.  I want to try to make it up to you.  I've got some
   great birdseed over here at my place -- top of the line
   stuff -- and if you come by, I'll have it all wrapped up
   for you.  I'm really sorry for all the problems I've caused
   for you over the years, but I know we can work this out.
   Wile E. Coyote       "Super Genius"

Showalter                 Expires 31 Jan 1999                   [Page 7]

Internet DRAFT                   Sieve                    August 7, 1998

   Message B
   From: youcouldberich!@reply-by-postal-mail
   Date:  Mon, 31 Mar 1997 18:26:10 -0800 (PST)
   Subject: $$$ YOU, TOO, CAN BE A MILLIONAIRE! $$$

   $20,000 IN LESS THAN TWO MONTHS!  AND IT'S LEGAL!!!!!!!!!
   !!!!!!!!!!!!!!!!!!111111111!!!!!!!11111111111!!1  JUST

2. Design

2.1. Form of the language

   This language is made up as a set of commands.  Commands can take a
   number of arguments; arguments can be either literal data, tests, or
   blocks of commands.

2.2. Whitespace

   Whitespace is used to separate commands.  Whitespace is made up of
   tabs, newlines (CRLF, never just CR or LF), and the space character.
   The amount of whitespace used is not significant.


   Comments begin with a "#" character that is not contained within a
   string and continue until the next CRLF.

   Example:  if size over 100K { # this is a comment

   XXX this example is broken

2.4. Literal data

   Literal data means data that is not executed and is supplied as
   arguments, such as numbers and strings.

Showalter                 Expires 31 Jan 1999                   [Page 8]

Internet DRAFT                   Sieve                    August 7, 1998

2.4.1. Numbers

   Numbers are given as ordinary decimal numbers.  However, those
   numbers that have a tendency to be fairly large, such as message
   sizes, may have a "K", "M", or "G" appended to indicate a multiple of
   a base-two number.  To be comparable with the power-of-two-based
   versions of SI units that computers frequently use, K specifies kilo,
   or 1,024 (2^10) times the value of the number; M specifies mega, or
   1,048,576 (2^20) times the value of the number; and G specifies giga,
   or 1,073,741,824 (2^30) times the value of the number.

   Implementations MUST provide 31 bits of magnitude in numbers, but may
   provide more.

   Negative, fractional, and decimal numbers are not permitted  by  this

2.4.2. Strings

   Scripts involve large numbers of strings, as they are used for
   pattern matching, addresses, and textual bodies, etc.  Typically,
   short quoted strings suffice for most uses, but a more convenient
   form is provided for longer strings such as bodies of messages.

   A quoted string starts and ends with a single double quote (the <">
   character).  A backslash ("\") inside of a quoted string is followed
   by either another backslash or a double quote.  This two-character
   sequence represents a single backslash or double-quote within the
   string, respectively.

   Other escape sequences may be permitted depending on context.  An
   undefined escape sequence (such as "\a" in a context where "a" has no
   special meaning) is interpreted as if there were no backslash (in
   this case, "\a" is just "a").

   Non-printing characters such as tabs, CR and LF, and control
   characters are permitted in strings.  NUL (ASCII 0) is not allowed in

   For entering larger amounts of text, such as an email message, a
   multi-line form is allowed.  It starts with the keyword "text:",
   followed by a CRLF, and ends with the sequence of a CRLF, a single
   period, and another CRLF.  In order to allow the message to begin
   lines with a single-dot, lines are dot-stuffed.  That is, when
   composing a message body, an extra `.' is added before each line
   which begins with a `.'.  When the server interprets the script,
   these extra dots are removed.

Showalter                 Expires 31 Jan 1999                   [Page 9]

Internet DRAFT                   Sieve                    August 7, 1998

   Note that a comment may occur in between the "text:" and the CRLF,
   but not within the string itself. String lists

   When matching patterns, strings frequently come in groups.  For this
   reason, a list of strings is allowed in many tests, implying that if
   the test is true using any one of the strings, then the test is true.
   Implementations are encouraged to use short-circuit evaluation in
   these cases.

   For instance, the test `header ["To", "Cc"] contains
   ["", ""]' is true if either the
   To header or Cc header of the input message contains either of the
   e-mail addresses "" or "".

   Conversely, in any case where a list of strings would be appropriate,
   a single string is allowed without being a member of a list; it is
   equivalent to a list with a single member.  So the test `exists "To"'
   is equivalent to the test `exists ["To"]'. Headers

   Headers are a subset of strings.  In the Internet Message
   Specification [IMAIL], each header line is allowed to have whitespace
   nearly anywhere in the line, including after the field name and
   before the subsequent colon.  Extra spaces between the header name
   and the ":" in a header field are ignored by the interpreter.

   A header name never contains a colon.  The "From" header refers to a
   line beginning "From:" (or "From   :", etc.).  No header will match
   the string "From:" due to the trailing colon. Addresses

   A number of commands call for email addresses, which are also a
   subset of strings.  These addresses must be compliant with [IMAIL].
   Implementations MUST ensure the addresses are syntactically valid,
   and need not ensure that they are actually deliverable.

2.5. Tests

   Tests are given as arguments to commands in order to control how the
   run.  Generally, a test is used to decide if a block of code should
   be evaluated.

Showalter                 Expires 31 Jan 1999                  [Page 10]

Internet DRAFT                   Sieve                    August 7, 1998

2.5.1. String Comparison

   When matching one string against another, there are a number of ways
   of performing the match.  These are accomplished with three matches
   -- an exact match, a substring match, and a wildcard glob-style
   match.  In order to provide for matches between character sets and
   case insensitivity, Sieve borrows ACAP's comparator registry.

2.6.1. Match Keyword

   There are two allowed match keywords describing the allowed match in
   this draft; they are ":is" and ":contains".  Match keywords are
   supplied to those commands which allow them to specify whether the
   match is to be a complete match or not.

   These are used as tagged arguments to tests that perform string
   comparison.  Exactly one of them is necessary for a command.

   The ":contains" version describes a substring match.  If the value
   argument contains the key argument as a substring, the match is true.
   For instance, the string "frobnitzm" contains "frob" and "nit", but
   not "fbm".  The null key ("") is contained in all values.

   The ":is" version describes an absolute match; if the contents of the
   first string are absolutely the same as the contents of the second
   string, they match.  Only the string "frobnitzm" is the string
   "frobnitzm".  The null key only "is" the null value.

   In order to specify what type of match is supposed to happen,
   commands that support matching take optional tagged arguments ":is"
   and ":contains".  Commands default to using ":is" matching.  Note
   that these modifiers may interact with comparators; in particular,
   some comparators are not suitable for matching with ":contains".  It
   is an error to use a comparator with ":contains" that is not
   compatible with it.

2.6.2. Comparators

   In order to allow for character set-independent matches, the match
   keyword may be coupled with a comparator name.  Comparators are
   described for [ACAP]; a registry is defined for ACAP, and this
   specification uses that registry.

   ACAP defines multiple comparator types.  Only equality types are used
   in this specification.

Showalter                 Expires 31 Jan 1999                  [Page 11]

Internet DRAFT                   Sieve                    August 7, 1998

   All implementations MUST  support  the  "i;octet"  comparator,  which
   simply  compares  one  octet with the next.  If left unspecified, the
   default is octet.

   Some comparators may not be usable with substring matches; that is,
   they may only work with ":is".  It is an error to try and use a
   comparator with "matches" or "contains" that is not compatible with

   A comparator is specified with commands that support matching by the
   ":comparator" option.  This option is followed by a string providing
   the name of the comparator to be used.  So in this example,

   Example:  if header "Subject" :contains :comparator "i;ascii-casemap"
                "make money fast" {

   would discard any message with subjects such as "Make Money Fast" and
   "MaKe MONEY fAST".

   OPEN:   Are there any other comparators that SHOULD or MUST be

2.7. Tagged Arguments

   This document provides for tagged arguments in the style of

   A tagged argument is an an argument for a command that begins with
   ":", and consists of a tag naming the argument, such as ":contains".
   This argument means that zero or more of the next tokens have some
   particular meaning, depending on the argument.  These next tokens may
   be numbers or strings, but are never blocks.

   To keep the language simple, tagged arguments should not take tagged
   arguments as arguments.

   One case where this is useful is the ":comparator" argument, which
   allows the user to specify which ACAP comparator will be used to
   compare two strings, since different languages may impose different
   orderings on UTF-8 [UTF-8] characters.

   Tagged arguments may appear in any order, and may be interspersed
   with positional arguments.

   OPEN:   Perhaps tagged arguments should always be before positional

Showalter                 Expires 31 Jan 1999                  [Page 12]

Internet DRAFT                   Sieve                    August 7, 1998

2.8. Blocks

   Blocks are sets of commands enclosed within curly braces.  Blocks are
   supplied to commands so that the commands can implement control

   So a control structure is just a command that happens to take a test
   and a block as its arguments; depending on the result of the control
   structure, it runs the code in the block zero or more times.  (Note
   that by the commands supplied in the specification, there are no
   loops, so the control structures supplied--if, elsif, and else--run a
   block either once or not at all.)

2.9. Commands

   Sieve scripts are made up of commands.  Commands can take any of the
   tokens above as arguments, and arguments may be either tagged or
   positional arguments.

   A command begins with a name, which is a simple token.  It ends with
   either a semicolon or a block.  (Commands ending with blocks are used
   to implement control structures.)  Commands never take both a
   semicolon and a block, nor do they ever take more than one block as
   an argument.

2.9.1. Positional Arguments

   Positional arguments are familiar from any programming language.  A
   command takes zero or more untagged positional arguments in order to
   specify its behavior.  Positional arguments are given their value
   based on their order in the command.

2.9.2. Optional Arguments

   Optional arguments are tagged arguments that may be omitted; when
   omitted, they are given default values.

2.9.3. Blocks as Arguments

   Commands may take blocks as arguments.  A block is always the last
   argument to a command, and when it exists, it replaces the semicolon
   that would otherwise end the command.

2.11. Evaluation

   Precedence is not important in any of the commands in this base
   specification.  However, as an extension might make order of
   operation important, all arguments to rules MUST be evaluated in

Showalter                 Expires 31 Jan 1999                  [Page 13]

Internet DRAFT                   Sieve                    August 7, 1998

   left-to-right order.  Those operations that can implement short-
   circuit evaluation (such as "allof" and "anyof") MUST do so.

   Sieve imposes specific limits on actions; for instance, a rejected
   message may not also be filed into a mailbox.  These restrictions are
   noted on a per-command basis.

   OPEN:   Or rather, they should be.

2.11.1. Implicit keep

   If evaluation of a script fails to result in one "fileinto", "keep",
   or "reject", a "keep" action is implicitly taken.  So the message is
   filed into the user's primary mailbox; that is,

   For instance, with any of the short messages offered above, the
   following script produces no actions.

   Example:  if size over 500K discard;

3. Conditionals and Control Structures

   In order for a script to do more than one set of actions, control
   structures are needed.  In Sieve, a control structure is a command
   that takes a block as an argument.

   In this document, only the "if" control structure is provided.  There
   are three pieces to if: "if", "elsif", and "else".

   Syntax:   if <test> <command>

   Syntax:   elsif <test> <command>

   Syntax:   else <block>

   The semantics are similar to any other programming language this
   appears in.  When the interpreter sees an "if", it evaluates the test
   associated with it.  If the test is true, it executes the block
   associated with it.

   If the test of the "if" is false, it evaluates the test of the first

Showalter                 Expires 31 Jan 1999                  [Page 14]

Internet DRAFT                   Sieve                    August 7, 1998

   "elsif" (if any).  If the test of "elsif" is true, it runs the
   elsif's block.  An elsif may be followed by an elsif, in which case,
   the interpreter repeats this process until it runs out of elsifs.

   When the interpreter runs out of elsifs, there may be an "else" case.
   If there is, and none of the if or elsif tests were true, the
   interpreter runs the else case.

   This provides a way of performing exactly one of the blocks in the

   In the following example, both Message A and B are dropped.

   Example:  if header "from" contains "coyote" {
             } elsif  header ["subject"] :contains ["$$$"] {
             } else fileinto "INBOX";

   In the script below, when run over message A, forwards the message to;  message B, to; any other
   message is forwarded to

   Example:  if header ["From"] contains ["coyote"] {
                forward "";
             } else if header "Subject" contains "$$$" {
                forward "";
             } else
                forward "";

4. Actions

   This document supplies six actions that may be taken on a message:
   keep, fileinto, forward, reject, discard, and stop.

4.1. Action reject

   Syntax:   reject <reason-string>

   The "reject" action resends the message to the sender, wrapping it in
   a "reject" form, noting that it was rejected by the recipient.  In
   the following script, message A is rejected and returned to the

   Example:  if header "from" contains "" {

Showalter                 Expires 31 Jan 1999                  [Page 15]

Internet DRAFT                   Sieve                    August 7, 1998

                reject "I am not taking mail from you, and I don't want
                your birdseed, either!";

   A reject message MUST takes the form of a failed DSN as specified  by
   [DSN].    The  human-readable  portion  of  the  message,  the  first
   component of the DSN, contains the human readable message  describing
   the  error,  although  it SHOULD contain additional text alerting the
   original sender that mail was refused by a filter.  This part of  the
   DSN might appear as follows:

   Message was refused by recipient's mail filtering program.
   Reason given was as follows:

   I am not taking mail from you, and I don't want your
   birdseed, either!

   The action-value field as defined in the DSN specification MUST be

   A rejected message may not be filed, forwarded, or kept.  A message
   that triggers a "reject" action is never allowed to be kept by the
   user, and the "reject" overrides all other actions.

   A message may only be rejected once.

4.2. Action fileinto

   Syntax:   fileinto <folder>

   The "fileinto" action drops the message into a named folder.
   Implementations SHOULD support fileinto, but may not be able to in
   cases where the filtering agent is not able to write to the users'
   folders (such as a [POP3] implementation running inside the mail
   server where the folders are stored on the users' local disks).

   In  the  following  script,  message   A   is   filed   into   folder

   Example:  if header ["to"] contains "coyote" {
                fileinto "INBOX.harassment";

Showalter                 Expires 31 Jan 1999                  [Page 16]

Internet DRAFT                   Sieve                    August 7, 1998

4.3. Action forward

   Syntax:   forward <address>

   The "forward" action is used to forward the message to another user
   at the supplied address, as a mail forwarding feature does.  The
   "forward" action makes no changes to the message body or headers, and
   only modifies the envelope recipient.

   A simple script can be used for forwarding:

   Example:  forward "";

   The forward command performs an MTA-style forward--that is, what  you
   get  from  a .forward file using sendmail under UNIX.  The address on
   the SMTP envelope is replaced with the one on the forward command and
   the  message  is  sent  back out.  (This is not an MUA-style forward,
   which creates a new message with a different sender and  message  ID,
   wrapping the old message in a new one.)

   OPEN:   At least one person rejects to this definition, claiming that
           a sendmail-style forward is inherently broken.

4.4. Action keep

   Syntax:   keep

   The "keep" action is whatever action is taken in lieu of all other
   actions, if no filtering happens at all; generally, this simply means
   to file the message into the user's main mailbox.  This command
   provides a way to execute this action without needing to know the
   name of the user's main mailbox, providing a way to call it without
   needing to understand the user's setup, or the underlying mail

   Example:  if size under 1M keep; else discard;

4.6. Action stop

   Syntax:   stop

   The "stop" action ends all processing.  If no actions have been
   executed, then the keep action is taken.

Showalter                 Expires 31 Jan 1999                  [Page 17]

Internet DRAFT                   Sieve                    August 7, 1998

   Example:  [There needs to be an example here.]

4.7. Action discard

   Syntax:   discard

   Discard drops the message.  In the following script, any mail from
   "" is thrown out.

   Example:  if header ["from"] contains [""]

   Discard takes no arguments.

   While an important part of this language, "discard" has the potential
   to create serious problems for users: A student leaving themselves
   logged in to a machine in a computer lab may find their script
   changed to just "discard".  In order to protect users in this
   situation (along with similar situations), implementations MAY keep
   messages destroyed by a script for an indefinite period, and MAY
   disallow scripts that throw out all mail.

5. Tests

   Tests are used in conditionals to decide which part(s) of the
   conditional to execute.
5.1. Test allof

   Syntax:   allof ( <test> , <test> , ... <test> )

   The allof test preforms a logical AND on the tests supplied to it.

   Example:  allof (false, false)  =>   false
             allof (false, true)   =>   false
             allof (true, true)    =>   true

   The allof test takes as its argument a test-list.

Showalter                 Expires 31 Jan 1999                  [Page 18]

Internet DRAFT                   Sieve                    August 7, 1998

5.2. Test anyof

   Syntax:   anyof ( <test> , <test> , ... <test> )

   The anyof test preforms a logical OR on the tests supplied to it.

   Example:  anyof (false, false)  =>   false
             anyof (false, true)   =>   true
             anyof (true, true)    =>   true

5.3. Test exists

   Syntax:   exists <header-name-list>

   The "exists" test is true if the headers listed in the
   <header-name-list> argument exist within the message.  All of the
   headers must exist or the test is false.

   The following example throws out mail that doesn't have a From header
   and a Date header.

   Example:  if not exists ["From","Date"] {

5.4. Test false

   Syntax:   false

   The "false" test always evaluates to false.

5.5. Test header

   Syntax:   header <header-name-list> <match-keyword> <key-list>

   The "header" test evaluates to true if the any header name matches
   any key.  How the match is done is described by the second argument,
   which is one of the string comparison arguments discussed in section
   2.6.  The first argument to header, the header-name-list, is a list
   of headers to get values from to be searched.  The key-list is a list
   of keys.

Showalter                 Expires 31 Jan 1999                  [Page 19]

Internet DRAFT                   Sieve                    August 7, 1998

   If a header  listed  in  the  header-name-list  argument  exists,  it
   contains  the  null  key  ("").   However, if the named header is not
   present, it does not contain the null key. So if a message  contained
   the header

           X-Caffeine: C8H10N4O2

   these tests on that header evaluate as follows:

           header ["X-Y-Z"] is [""]         => false
           header ["X-Y-Z"] contains [""]   => true

5.6. Test not

   Syntax:   not <test>

   The "not" test takes some other test as an argument, and yields the
   opposite result.

5.7. Test size

   Syntax:   size <":over" / ":under"> <limit [quantifier]>

   The "size" test deals with the size of a message.  It takes either a
   tagged argument of ":over" or ":under", followed by a number
   representing the size of the message.

   If the argument is ":over", and the size of the message is greater
   than the number provided, the test is true; otherwise, it is false.

   If the argument is ":under", and the size of the message is less than
   the number provided, the test is true; otherwise, it is false.

   The size of a message is defined to be the number of octets from the
   initial header until the last character in the message body.

6. Errors in Processing a Script

   In any programming language, errors are inevitable.  Users are
   expected to make errors, and even if a script works correctly today,
   it may fail tomorrow due to quotas, mailboxes being removed or
   renamed, or some piece of hardware being down.  It is imperative that
   mail get through.

   Implementations SHOULD check a script before it is run in order to
   ensure that it is valid.  Implementations SHOULD NOT try and recover

Showalter                 Expires 31 Jan 1999                  [Page 20]

Internet DRAFT                   Sieve                    August 7, 1998

   from a script with errors, and should instead file mail into the
   user's primary mailbox.

   Users MUST be notified of errors in processing a script.  The method
   by which users are notified is implementation defined, but a mail
   message clearly describing the error is suggested if a preferable
   alternative cannot be found.

   In an implementation that allows for a script to be checked when it
   is turned over to the server, the script can be checked for errors
   before it is submitted.  Implementations SHOULD notify the user of
   the error and refuse to accept a syntactically invalid script or one
   that makes use of extensions that the server does not report.

   Implementations MUST allow mail to be filed without filtering in case
   of a syntax error in the script.  Implementations MUST avoid sending
   multiple messages describing the same error.

   Implementations are REQUIRED to notify users of errors in filtering
   scripts.  If there are errors in the script being used, mail SHOULD
   be filed into the user's main mailbox.  Implementations MUST NOT
   discard mail unless a command explicitly demands it.

7. Extensibility

   New control structures, actions, and tests can be added to the
   language.  Sites must make these features known to their users; this
   document does not define a way to discover the list of extensions
   supported by the server.

   Any extensions to this language MUST define a string that uniquely
   identifies that extension.  If a new version of an extension changes
   the functionality of a previously defined extension, it MUST use a
   different name.  The purpose of such a string is for the "support"
   test, which mandates that script requires the use of that extension.

   Additionally, in a situation where there is a submission protocol and
   an extension advertisement mechanism aware of the details of this
   language, scripts submitted can be checked against the mail server to
   prevent use of an extension that that the server does not support.

7.1. Capability String

   Capability strings are typically short strings describing what
   capabilities are supported by the server.  The following capability
   strings are defined by this document:

   fileinto    The string "fileinto" indicates the implementation

Showalter                 Expires 31 Jan 1999                  [Page 21]

Internet DRAFT                   Sieve                    August 7, 1998

               supports filing into mailboxes.

   comparator- The string "comparator-elbonia" is provided if the
               implementation supports the "elbonia" comparator.
               Therefore, all implementations have at least the
               "comparator-i;octet" capability.

7.2. Registry

   In order to provide a standard set of extensions, a registry is
   provided by IANA.  Capability names may be registered on a first-
   come, first-served basis.  Extensions designed for interoperable use
   should be defined as standards track or IESG approved experimental

   Subject: Registration of new Sieve extension

   Capability name:
   Capability keyword:
   Capability arguments:
   Standards Track/IESG-approved experimental RFC number:
   Person and email address to contact for further information:

7.3. Capability Transport

   As the range of mail systems that this draft is intended to apply to
   is quite large, a method of advertising which capabilities an
   implementation supports is difficult due to the wide range of
   possible implementations.  Such a mechanism, however, should have the
   following properties.

   (1)  The implementation can advertise the complete set of extensions
        that it supports.

   OPEN:   There needs to be a more complete description here.

8. Transmission

   The MIME type for a SIEVE script is "application/sieve".  Scripts are
   encoded in UTF-8 during transmission.

Showalter                 Expires 31 Jan 1999                  [Page 22]

Internet DRAFT                   Sieve                    August 7, 1998

9. Acknowledgments

   I am very thankful to Chris Newman for his support and his ABNF
   syntax checker, to John Myers and Steve Hole for outlining the
   requirements for the original drafts, and to Rob Earhart for an early
   implementation and a great deal of help.  I am also indebted to all
   of the readers of the mailing list.

10.  Formal Grammar

   The grammar used in this section is the same as the ABNF described in

   In the case of alternative or optional rules in which a later rule
   overlaps an earlier rule, the rule which is listed earlier MUST take
   priority.  (This shouldn't happen.  Please let me know if it does.)

   argument = string / string-list / number / tag / test

   block = "{" [WSP] commands [WSP] "}"
           ;; C-style block

   CHAR-NOT-DOT = (%x01-2d / %x2f-%xff)
           ;; all the characters that aren't "."

   command = identifier WSP *(argument WSP) [WSP] ";"

   commands = *([WSP] command [WSP])

   comment = "#" *VCHAR CRLF

   identifier = (ALPHA / "_") *(ALPHA DIGIT "_")

   multi-line = "text:" [WSP] CRLF
           *((1*CHAR-NOT-DOT *CHAR CRLF) / ("." 1*CHAR-NOT-DOT *CHAR CRLF) /
             (".." *CHAR CRLF) / CRLF)
           "." CRLF
           ;; Note when used,
           ;; a leading ".." on a line is mapped to ".".

   number = 1*DIGIT [QUANTIFIER]
           ;; quantifier is a multiplier (or bit shift)

   QUANTIFIER = "K" / "M" / "G"
           ;; K == 2^10; M == 2^20; G = 2^30

   quoted-string = DQUOTE *CHAR DQUOTE
           ;; \" inside a string maps to "

Showalter                 Expires 31 Jan 1999                  [Page 23]

Internet DRAFT                   Sieve                    August 7, 1998

           ;; \\ inside a string maps to \
           ;; All other characters map to themselves.
           ;; Note that newlines and other weird characters
           ;; are all allowed strings.

   string = quoted-string / multi-line

   string-list = "[" [WSP] *(string [WSP] "," [WSP]) string [WSP] "]" / string
           ;; if there is only a single string, the brackets are optional

   tag = ":" identifier

   test = identifier *(WSP argument) [WSP test-list]

   test-list = [WSP] "(" [WSP] *(test [WSP] "," [WSP])
       test [WSP] ")" [WSP]

   WSP = 1*(SP / CRLF / HTAB) / comment
           ;; just whitespace.  anyplace this is allowed, a comment is
           ;; as well

11. Security Considerations

   Users must get their mail.  It is imperative that whatever method
   implementations use to store the user-defined filtering scripts be

   It is equally important that implementations sanity-check the user's
   scripts, and not allow users to create on-demand mailbombs.  For
   instance, an implementation that allows a user to reject or forward
   multiple times to a single message might also allow a user to create
   a mailbomb triggered by mail from a specific user.

   Therefore, an implementation SHOULD only allow one "reject" per
   message processed, and MAY limit the number of forward actions taken.
   An implementation MUST refuse to forward a message to itself.  [OPEN:
   What do you do when a site limit prevents you from this?  Say I do
   three replies; which ones take effect when the limit is 1? 2? 0?]

   Several commands, such as "discard", "forward", and "fileinto" allow
   for actions to be taken that are potentially very dangerous.

12. Author's Address

   Tim Showalter
   Carnegie Mellon University
   5000 Forbes Avenue
   Pittsburgh, PA 15213

Showalter                 Expires 31 Jan 1999                  [Page 24]

Internet DRAFT                   Sieve                    August 7, 1998


Showalter                 Expires 31 Jan 1999                  [Page 25]

Internet DRAFT                   Sieve                    August 7, 1998


Appendix A.  References

   [ABNF] Crocker, D.,  "Augmented BNF for Syntax Specifications: ABNF",
   Internet Mail Consortium, RFC 2234, November 1997.

   [DSN] Moore, K., and G. Vaudreuil, "An Extensible Message Format for
   Delivery Status Notifications", RFC 1894, January 1996.

   [FLAMES] Borenstein, Nathaniel S., and Chris A. Thyberg, "Power, Ease
   of Use, and Cooperative Work in a Practical Multimedia Message
   System", Int. J. of Man-Machine Studies, April, 1991.  Reprinted in
   Computer-Supported Cooperative Work and Groupware, Saul Greenberg,
   editor, Harcourt Brace Jovanovich, 1991.  Reprinted in Readings in
   Groupware and Computer-Supported Cooperative Work, Ronald Baecker,
   editor, Morgan Kaufmann, 1993.

   [KEYWORDS] Bradner, S., "Key words for use in RFCs to Indicate
   Requirement Levels", RFC 2119, Harvard University, March 1997.

   [IMAP] Crispin, M., "Internet Mail Access Protocol - version 4rev1",
   RFC 2060, University of Washington, December 1996.

   [IMAIL] Crocker, D., "Standard for the Format of ARPA Internet Text
   Messages", STD 11, RFC 822, University of Delaware, August 1982.

   [MIME] Freed, N., and N. Borenstein, "Multipurpose Internet Mail
   Extensions (MIME) Part One: Format of Internet Message Bodies", RFC
   2045, Innosoft and First Virtual, November 1996.

   [SMTP] Postel, J., "Simple Mail Transfer Protocol", STD 10, RFC 821,
   USC/Information Sciences Institute, August 1982.

   [UTF-8] Yergeau, F. "UTF-8, a transformation format of Unicode and
   ISO 10646", RFC 2044, Alis Technologies, October 1996.

Appendix B. Full Copyright Statement

   Copyright (C) The Internet Society 1998. All Rights Reserved.

   This document and translations of it may be copied and furnished to
   others, and derivative works that comment on or otherwise explain it
   or assist in its implementation may be prepared, copied, published
   and distributed, in whole or in part, without restriction of any
   kind, provided that the above copyright notice and this paragraph are

Showalter                 Expires 31 Jan 1999                  [Page 26]

Internet DRAFT                   Sieve                    August 7, 1998

   included on all such copies and derivative works.  However, this
   document itself may not be modified in any way, such as by removing
   the copyright notice or references to the Internet Society or other
   Internet organizations, except as needed for the purpose of
   developing Internet standards in which case the procedures for
   copyrights defined in the Internet Standards process must be
   followed, or as required to translate it into languages other than

   The limited permissions granted above are perpetual and will not be
   revoked by the Internet Society or its successors or assigns.

This document will expire before January 31, 1999.

Showalter                 Expires 31 Jan 1999                  [Page 27]