💾 Archived View for gemini.bortzmeyer.org › rfc-mirror › rfc515.txt captured on 2022-01-08 at 20:42:45.

View Raw

More Information

⬅️ Previous capture (2021-11-30)

-=-=-=-=-=-=-







Network Working Group                                          R. Winter
Request for Comments: 515                Computer Corporation of America
NIC 16446                                                    6 June 1973


              Specifications for Datalanguage, Version 0/9

Preface

   Datalanguage is the language processed by the Datacomputer, a data
   utility system being developed for the Arpanet.  The Datacomputer
   performs data storage and data management functions for the benefit
   of computers on the network.

   Version 0/9 is currently running at CCA.  This version is extremely
   primitive; however, it does offer an opportunity for experience with
   the Datacomputer and with fundamental Datalanguage concepts.

   Subsequent versions will provide greater portions of the full
   Datalanguage capability, which has been described earlier
   (Datalanguage, Working Paper No. 3, Datacomputer Project, October,
   1971, NIC 8028).  For example, one of the primary restrictions in
   0/9--elementary data items must be fixed-length ASCII strings--will
   be eliminated in Version 0/10, which is currently being implemented.

   Based on the experience gained in the implementation of these early
   versions, and based on the feedback from their use, a revised
   specification of the full language will be issued.

1. Introduction

   This document presents a precise and complete specification of
   Datalanguage, Version 0/9.  It is organized into 11 sections, of
   which this introduction is the first.  Section 2 discusses the
   capabilities of Version 0/9 in general terms.  Sections 3 and 4 are
   concerned with data description and the directory.  Sections 5
   through 8 cover the expression of data management operations.
   Section 9 discusses the recognition of names.  Section 10 covers
   miscellaneous topics and Section 11 specifies the syntax in BNF.

   This specification is to be followed with a user manual, which will
   present the language in tutorial form and treat components of the
   Datacomputer-user interface other than the language.








Winter                                                          [Page 1]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


2. Capabilities of Version 0/9

   Version 0/9 of Datalanguage has capabilities for the storage of
   files; for addition of data to existing files, and for the deletion
   of files.  Retrievals can output whole files as well as subsets of
   files.  Data can be selected from files by content, using expressions
   formed from boolean and inequality operators.

   At the option of the file creator, an inversion is constructed and
   maintained by the Datacomputer.  The inversion increases the
   efficiency of selective retrieval, at the cost of storage space and
   file maintenance effort.  Users other than the file creator need not
   be aware of the existence of the inversion, or of which fields are
   inverted file keys.  The language is designed so that they state the
   desired result of a retrieval, and the Datacomputer uses the
   inversion as much as the request permits.

   Elementary data items are fixed-length ASCII strings.  Files are a
   restricted class of hierarchical structures.

   Many of the restrictions mentioned in this memo will be short-lived.
   In particular, those statements followed with 3 asterisks (***) refer
   to restrictions that will be considerably weakened or eliminated
   entirely in the next version of the software.

3. Data Description

   A container is a variable whose value is a data object of general
   character and arbitrary size (In Version 0/9, size is restricted.
   See section 3.4).  Examples of containers which are implemented in
   other systems are files, records, fields, groups, and entries.

   The container is distinct from the data in the container.  For
   example, space allocation is an operation on a container, while
   changing the unit price field from 25 to 50 is an operation on data
   in a container.

   A container may enclose other containers.  When a container is not
   enclosed by another container, it is said to be outermost.  If
   container A encloses container B, and no other container in A also
   encloses B, then A immediately encloses B.

   A Datalanguage description is a statement of the properties of a
   container.

   All containers have the attributes ident and type.  Ident is a
   character string by which users refer to the container.  Type
   determines the form of the container's value; the value can be



Winter                                                          [Page 2]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


   elementary, or it can consist of other containers.  There are 3
   types: LIST, STRUCT, and STRING(***).  A LIST contains a group of
   containers having the same description.  A STRUCT contains a group of
   containers, each of which has its own description.  A STRING is a
   sequence of ASCII characters.  While a STRING is not really an
   elementary item, it is handled as one in Version 0/9.

   Certain containers can have other attributes.  An outermost container
   has a function.  The function attribute specifies whether the
   container is to be used for storage or for transmission.

   Size is some meaningful dimension of the container, which is type-
   dependent.  It is used for space allocation and data stream parsing.

   An aggregate container (i.e., one that contains other containers) has
   as an attribute the description or descriptions of its components.
   Thus if S is a STRUCT containing A, B, and C, then the descriptions
   of A, B, and C are attributes of S.

   A STRING defined in certain contexts can have an inversion attribute.
   This is an access property that is not really local to the STRING,
   but is associated with it for convenience.

3.1 Ident

      The ident of a container is composed of alphanumeric characters,
      the first of which is alphabetic.  It may not consist of more than
      100 characters.

      The elements of a STRUCT must have idents unique in the STRUCT.

3.2 Function

      The function of a container is either FILE, PORT, or TEMPORARY
      PORT.  When the function is FILE, then the container is used for
      storage of data at the Datacomputer.  When the function is PORT,
      then the container is used for transmission of data into or out of
      the Datacomputer.  When the function is TEMPORARY PORT (which may
      be abbreviated TEMP PORT), the container behaves like a PORT;
      however, its description is not retained in the Datacomputer
      beyond the session in which it is created.










Winter                                                          [Page 3]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


3.3 Type

      Type is one of: LIST, STRUCT, or STRING.  These are defined on the
      preceding page.

      In an occurrence of a STRUCT, the elements appear in the order in
      which their descriptions appear in the STRUCT description.  All
      elements are present in each occurrence of the STRUCT.

      An element of a STRUCT or LIST can be a container of any datatype.
      However, the outermost container must be a LIST(***).

3.4 Size

      The size of a STRING is the number of characters in it.  The size
      of a STRUCT is not defined (***).  The meaning of the size of a
      LIST depends upon other properties of the LIST (***).

      Ordinarily, the size of a LIST is the number of LIST-members.  An
      exception is the case of the outermost-LIST.  In an outermost-LIST
      with a function of FILE, the size is the number of LIST-members
      for which space should be allocated.  When no size is present in
      this case, the system computes a default.  In an outermost-LIST
      with a function of PORT, the size is ignored (***).

      Only outermost containers may be larger than a TENEX page (2560
      ASCII characters)(***).

3.5 Inversion

      An inversion is an auxiliary data structure used to facilitate
      retrieval by content.

      Its basic application is the fast retrieval of sets of outermost-
      LIST-members (this can be extended to other container sets, and
      will be after release 1).  Consider a list of weather
      observations, stored as a file on the Datacomputer.  If quick
      retrieval of observations by COUNTRY is desired, then this is
      indicated in the description of the COUNTRY container.  According
      to common usage in information retrieval, this makes COUNTRY a key
      in the retrieval of observations.

      Note that the inversion option only affects the efficiency of
      retrieval by COUNTRY, not the ability to retrieve by COUNTRY.







Winter                                                          [Page 4]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      There are restrictions on use of the inversion option.  First, it
      can be applied only to STRINGs.  Second a STRING having the
      inversion option must occur only once in each
      outermost-LIST-member.  Third, it is ignored when applied to
      STRINGs in PORT descriptions.

      Eventually there will be several types of inversion option; in
      Version 0/9 there is only the 'D' option (for distinct).

3.6 Syntax

      The description is simply an enumeration of properties; these
      properties are specified in the order:

            <ident> <function> <type> <size> <other>

      Properties which do not apply are omitted.  An example:

            F FILE LIST (25) A STR (10)

      Here 'F' is the <ident>, 'FILE' is the <function>, 'LIST' is the
      <type>, '(25)' is the size, and 'A STR (10)' is the <other> of one
      description.  Of course, 'A STR (10)' is itself another
      description: the description for members of the LIST named F.

      An example of a complete description for a file of weather
      observations keyed on location:

            WEATHER FILE LIST
            OBSERVATION STRUCT
                 LOCATION STRUCT
                       CITY STR (10), I=D
                       COUNTRY STR (10), I=D
                       END
                 TIME STRUCT
                       YEAR STR (2)
                       DAY STR (3)
                       HOUR STR (2)
                       END
                 DATE STRUCT
                       TEMPERATURE STR (3)
                       RAINFALL STR (3)
                       HUMIDITY STR (2)
                       END
                 END
      The ENDs are needed to delimit the list of elements of a STRUCT.
      `, I=D' indicates that the string is to be an inversion key for
      the retrieval of outermost-LIST-members.



Winter                                                          [Page 5]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


4. Directory

   The directory is a system file in which the names and descriptions of
   all user-defined containers are kept.

   The directory is structured as a tree.  Each node has an ident, which
   need not be unique.  There is a single path from the root of the tree
   to any node.  The idents of the nodes along this path are
   concatenated, separated by periods, to form a pathname, which
   unambiguously identifies the node (e.g., A.B.C could be a pathname
   for node with an ident of C).

   In a later version of the software, the directory will be generalized
   to provide for links between nodes, so that it will not properly be a
   tree.  For now, however, the tree model is convenient and adequate.

   A node may represent a container, or it may simply hold a place in
   the space of pathnames.  When it represents a container, it cannot
   (currently) have subordinate nodes.

   Eventually, it is planned to model the directory as a structure of
   containers, with its description distributed throughout the
   structure.  Most operations defined on the directory will be defined
   on user data, and vice versa.  Access privileges and privacy locks
   will be part of the data description and will likewise be applicable
   both to directory nodes and data structures below the node level.

4.1 CREATE

      A CREATE-request either; (a) adds a node to the directory,
      optionally associating the description of either a PORT or a FILE
      with the node, or (b) creates a temporary container which is not
      entered in the directory, but has a description and can be
      referenced in requests.  If the description defines a file, CREATE
      causes space to be allocated for the file.

      To create a node with a description:
           CREATE <pathname> <description> ;
      To create a node with no description:
           CREATE <pathname> ;
      Note that the description determines whether or not the container
      is temporary (see section 3.2 for details).









Winter                                                          [Page 6]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      A CREATE-request adds a single node to the directory.  Thus to add
      CCA.RAW.F to an empty directory, three requests are needed:
           CREATE CCA ;
           CREATE CCA.RAW ;
           CREATE CCA.RAW.F ;
      Notice that the last ident of the pathname doubles as the first
      ident of the description:
           CREATE CCA.RAW.G FILE LIST A STR (5) ;

      That is, G is both the ident of a node and the ident of an
      outermost container of type LIST.

4.2 DELETE

      A DELETE-request deletes a tree of nodes and any associated
      descriptions or data.  The syntax is:
           DELETE <pathname> ;
      The named node and any subordinates are deleted.

      Note that to delete data while retaining the directory entry and
      description, DELETE should not be used (see section 6.3 for the
      proper method).

4.3 LIST

      The LIST-request is used to display system data of interest to a
      user.  It causes the data specified to be transmitted through the
      Datalanguage output port.

      Several arguments of LIST apply to the directory.  LIST %ALL
      transmits all pathnames in the directory.  LIST %ALL.%SOURCE
      transmits all descriptions in the directory.  Instead of %ALL, a
      pathname can be used:
           LIST <pn>.%ALL
      Lists pathnames subordinate to <pn>.
           LIST <pn>.%SOURCE
      lists descriptions subordinate to the node represented by <pn>.

      For details about the LIST-request, see section 10.1.

5. Opening and closing containers

   Containers must be open before they can be operated on.

   A container is open when it is first created.  It remains open until
   closed explicitly by a CLOSE-request or implicitly by a DELETE-
   request or by session end.




Winter                                                          [Page 7]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


   A closed container is opened by an OPEN-request.  A temporary
   container is always open; a CLOSE-request deletes it.

5.1 Modes

      An open container has a mode, which is one of: READ, WRITE, or
      APPEND.  The mode determines the meaning and/or legitimacy of
      certain operations on the container.

      The mode is established by the operation which opens the
      container.  It can be changed at any time by a MODE-request.  A
      CREATE leaves the container in WRITE mode.  An OPEN either
      specifies the mode explicitly or implicitly sets the mode to READ.

5.2 Syntax

      To open a container:
           OPEN <pathname> <mode> ;
      or:
           OPEN <pathname> ;
      where <mode> is defaulted to READ.

      To close a container:
           CLOSE <ident> ;
      where <ident> is the name of an outermost container.

      Two containers with the same outermost <ident> can not be
      opened at the same time (***).

      To change the mode of an open container:
           MODE <ident> <newmode> ;

5.3 LIST

      LIST %OPEN transmits name, mode and connection status of each open
      outermost container through the Datalanguage output port. (The
      Datalanguage output port is the destination to which all
      Datacomputer diagnostics and replies are sent.  It is established
      when the user initially connects to the Datacomputer.)  For
      details of the LIST-request, see section 10.1.

6. Assignment

   Assignment transfers data from one container to another.

   The equal sign ('=') is the symbol for assignment.  The value of the
   operand on the right of the equal sign is transferred to the operand
   on the left.  (Eventually, both operands will be weakly-restricted



Winter                                                          [Page 8]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


   Datalanguage expressions, which may evaluate to sets as well as to
   single containers.  Now, the left must be a container name, the right
   may be a container name or a constant.)

   Assignment is defined for all types of containers.  When the
   containers are aggregates, their elements are paired and data is
   transferred between paired elements.  Elements of the target
   container that do not pair with some source element are handled with
   a default operation (currently they are filled with blanks).

   The operands of an assignment must have descriptions that match.  The
   idea of matching is that the descriptions must be similar enough so
   that it is obvious how to map one into the other.

6.1 Conditions for legitimate assignment

      Assignment must reference objects, not sets.  An object is:
         (a)   an outermost container, or
         (b)   a constant, or
         (c)   in the body of a FOR-loop, either
               (c1) a member of a set defined by a FOR-OPERAND, or
               (c2) a container which occurs once in such a member

      In the case of a reference of type (c1), the object referenced is
      taken to be the current member.  In the case of (c2), the object
      referenced is that which occurs in the current member.  This is
      explained further in section 7.

      The left operand of an assignment is subject to further
      restriction.  If it is an outermost container, it must be open in
      either WRITE- or APPEND-mode.  If it is not an outermost
      container, then the reference is of type (c), which means that
      some FOR-operand has established a context in which the assign-
      operand is an object.  The FOR-operand which establishes this
      context must be the output-operand of the FOR.

      When the assign-operand is an outermost container, it must be
      open.  Such an operand must be referenced by its simple container
      ident(***), not its directory pathname.

      In the body of a loop nested in one or more other loops,
      assignments are further restricted, due to a 0/9 implementation
      problem.  See section 7.2 for details.








Winter                                                          [Page 9]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      Finally, the descriptions of the operands must match.  If one is a
      constant, then the other must be a STRING(***).  If both are
      containers, then in the expression:
         A = B;
      the descriptions of containers A and B match if:
      1. A and B have the same type
      2. If A and B are LISTs, then they have equal numbers of
         LIST-members, or else A is an outermost-LIST.
      3. If A and B are aggregates, then at least one container
      immediately enclosed in A matches, and has the same ident as, one
      container immediately enclosed in B.

6.2 Result of assignment

      If the operands are STRINGs, then the value of B, left-justified,
      replaces the value of A.  If B is longer than A, the value is
      truncated.  If B is shorter than A, then A is filled on the right
      with blanks as necessary.

      If the operands are STRUCTs, then assignment is defined in terms
      of the STRUCT members.  If a member of A, mA, matches and has the
      same name as a member of B, mB, then mB is assigned to mA.  If no
      such mB exists, then mA is filled with blanks.

      If the operands are LISTs, the result depends on several factors.
      First, notice that the descriptions of the LIST-members must
      match;  otherwise the assignment would not be legitimate by the
      matching rules of 6.1.

      If A is an outermost-LIST, then it can be in either of two modes:
      WRITE or APPEND.  If A is in WRITE-mode, its previous contents are
      first discarded; it is then handled as though it were in APPEND-
      mode.

      If A is not an outermost-LIST, then it is always effectively in
      WRITE-mode.

      After taking the mode of A into account, as described above, the
      procedure is:
         for each member of LIST B
         (a) add a new member to the end of A
         (b) assign the current number of B to the new member of A









Winter                                                         [Page 10]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


6.3 Deletion of Data Through Assignment

      If A is an outermost container in WRITE-mode, and B is a container
      with description that matches A, and if B contains no data, then
      A=B has the effect of deleting all data from A.  Note that if A is
      in APPEND-mode in these circumstances, then A=B is a no-operation
      (i.e., has no effect).

7. FOR

      FOR <output set spec>, <input set spec> <body> END ;

   The output set is optional: that is, FOR need not produce output.
   When the output set is omitted, the syntax is:

      FOR <input set spec> <body> END ;

   The operations specified in the body are performed once for each
   member of the input set.  References in the body to the input set
   member are treated as references to the current input set member.
   When an output set is present, a new member is created and added to
   the output set for each iteration (i.e., for each input set member).
   References to the output set member, similarly, are treated as
   references to the current output set member.

   The output set spec must be the name of a LIST member.  Each
   iteration of the FOR will create one such member, and add it to its
   LIST (hereafter called the output LIST).  The body terminates the
   value that the new member receives.  Any STRING in the new member
   which is not given a value by the body receives he default value of
   all blanks.

   The input set spec must be an expression evaluating to a set of
   LIST-members (see section 7.1 for details of input set
   specification).  Each iteration for the FOR will input one such
   member; the FOR will terminate when each member of the set has been
   processed.  The LIST from which the input set members are drawn is
   called the input LIST.

   FOR is effectively a means of accomplishing variants of assignment
   between a pair of LISTS.  FOR is less concise than assignment, but
   offers more flexibility.  Its advantages are:
      (a) not all the input LIST-members need be transferred to the
          output LIST.  A subset can be selected by content.
      (b) the user has explicit control over the assignment of values to
          output LIST-members.





Winter                                                         [Page 11]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


   This is most easily understood by an example:
       P PORT LIST               F FILE LIST
         R STRUCT                  R STRUCT
           B STR                     A STRUCT
           C STR                       A1 STR
           END                         A2 STR
                                     B STR
                                     C STR
                                     END

   (1)   P = F ;
   (2)   FOR P.R, F.R
               P.R = F.R ;
               END ;
   (3)   FOR P.R, F.R WITH A1 EQ 'XY' OR A2 GE 'AB'
               B = C ;
               C = A2 ;
               END

   Here, (1) and (2) are entirely equivalent requests.  However, (3) is
   quite different and is not expressible as assignment.  It selects a
   subset of F.Rs.  The values it gives to the P.Rs could not result
   from application of the matching rules to F and P.

   Because FOR is effectively assignment between a pair of LISTs, the
   LISTs referenced by a legitimate FOR-operation are largely subject to
   the same restrictions as LISTs referenced in an assignment.  One
   exception is that the descriptions of the LIST-members need not
   match.

   These restrictions are:
   (a)   both LISTs must be objects in the context in which the FOR
         appears.
   (b)   both LISTs must be open or contained in open outermost
         containers.
   (c)   if the output LIST is an outermost container, it must be in
         WRITE- or APPEND-mode.
   (d)   If the output LIST is not outermost, the LIST which most
         immediately encloses it must be the output LIST of an enclosing
         FOR.

   The mode of the output LIST of the FOR affects the result much as it
   would in an assignment: that is, a FOR outputting to a LIST in
   WRITE-mode overwrites previous contents, while a FOR outputting to a
   LIST in APPEND-mode adds to previous contents.






Winter                                                         [Page 12]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


   CAUTION TO THE READER: For convenience, these specifications use
   phrases such as 'LISTs referenced by a FOR'.  Recall that such a
   phrase is not literally correct, in the sense that the operands of a
   FOR are always LIST members, not LISTs.

7.1 Details of input set specification

      The input set is specified by a Datalanguage expression that
      evaluates to a set of LIST-members.  Such an expression can be
      simply the set of all members of a LIST, or it can be a subset of
      the members of a LIST, specified by content.  For example, with
      the description:

          F FILE LIST
               R STRUCT
                     A STR (1)
                     B STR (2)
                     END

      the expression:
         F.R
      references all R's on the LIST F.  However:
         F.R WITH A EQ '5'
      references only those R's containing an A having the value '5'.
      The expressions permitted as input set specifications are of the
      form:
         <list-member-name> WITH <boolexp>
      The <boolexp> is constructed of comparison expressions joined by
      the Boolean operators AND and OR.  Any expression can be negated
      with NOT.

      Comparison operators have the highest precedence.  Next highest is
      AND, then OR, then NOT.

      The comparison expressions are restricted to the form:
         <container name> <comop> <constant>
      where:
      (a)   <constant> is a string constant enclosed in single quotes
            (see section 10.2 for a discussion of constants)
      (b)   <comop> is one of six operators:
                      EQ          equal
                      NE          not equal
                      LT          less than
                      GT          greater than
                      LE          less than or equal to
                      GE          greater than or equal to
      (c)   <container name> is the name of a STRING that appears once
            in each LIST-member.



Winter                                                         [Page 13]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      The constant is truncated or padded with blanks on the right to
      make it equal in size to the container to which it is being
      compared.  Notice that padding on the right is not always
      desirable (users will have control over the padding in a future
      release).  In particular, care must be exercised when using
      numbers in Version 0/9.  (A number represented as a STRING should
      actually be described as a number; eventually it will be possible
      to do this).

7.2 FOR-body

      Two operations are legitimate in a FOR-body:  FOR and assignment.

      These are subject to the restrictions discussed in Section 6.1 and
      in the introduction to Section 7.  The restrictions are related to
      three requirements:  (1) that the names be recognizable (see
      Section 9 for details), (2) that a request be consistent regarding
      direction of data transfer between containers, both within itself
      and with the MODE of outermost containers, and (3) that transfers
      occur between objects, not sets of objects.  The first two
      requirements are permanent, but will become weaker in later
      versions of the language.  The last requirement is temporary and
      will be present only in early versions.

      Due to an implementation problem associated with Version 0/9,
      there is a somewhat bizarre restriction applied to references made
      in the body of a loop nested in another loop.  This restriction is
      not expected to pose any practical problems for users, and is not
      part of the language design, but is discussed here for
      completeness.

      The restriction is most easily understood by example:

      given the description
          F LIST
                R STRUCT
                      A STR (3)
                      BL LIST (3)
                            B STR (3)
                      C STR (3)
                      END
      and the request fragment:
          FOR ...,R
                FOR ...,B
                      ... = A ;
                      ... = C ;
                      END
                END



Winter                                                         [Page 14]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      observe:
      (a)   The outer loop processes the set of R's in F.
      (b)   For each R in F, the inner loop processes the set of B's in
            the BL contained in that R.
      (c)   In the body of the inner loop, there are references to A and
            C, which do not occur in B, but do occur in R.  That is, the
            objects referenced in the inner loop body are defined by the
            outer loop, not the inner loop.  In general, this is fine;
            in the case of C, however, we have a problem.
      (d)   C occurs beyond the end of BL.

      The 0/9 compiler is capable of neither (1) looking ahead enough to
      locate C before it compiles code for the loop, nor (2) while
      generating code to loop on the B's in BL, generating a separate
      body of code that skips to the end of BL to locate C.  Thus it can
      handle A, which has been located before it begins loop generation,
      but it cannot handle C, which requires a little foresight.

      The request fragment shown would not cause problems if the
      description were changed to:

          F LIST
                R STRUCT
                      A STR (3)
                      C STR (3)
                      BL LIST (3)
                            B STR (3)
                      END

      Then both A and C would have been found before code for the inner
      loop was generated.

8.  Data Transmission

   Data is transferred from container to container by execution of
   assignment and FOR operations.  The outermost containers involved in
   transfers can be files or ports.  If both are files, then the
   transfer is internal to the Datacomputer.  If either is a port, then
   an address in the external world is needed to accomplish the data
   transmission.











Winter                                                         [Page 15]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


   Such an address is supplied through a CONNECT-request, which
   associates a container (having a function of PORT) with an external
   address:
      CONNECT <ident> TO <address> ;
   Here <address> is either a specifications of host and socket number,
   or a TENEX file designator (for CCA's TENEX) enclosed in single
   quotes.  The host and socket form is:
      <socket> AT <host>
   where <socket> is a decimal number, and <host> is either a decimal
   number or a standard host name (since standard host names don't exist
   right now, it has to be the TENEX 'standard' name for the host.
   Contact the author for the latest list.)  If <host> is omitted, it is
   taken to be the host from which the Datalanguage is being
   transmitted.

   The address associated with a port can be changed by issuing another
   CONNECT-request.

   A DISCONNECT-request simply breaks an existing port/address
   association without establishing a new one.  (A CLOSE-request that
   references an open port executes a DISCONNECT.)  The syntax of
   DISCONNECT is:
      DISCONNECT <ident> ;

   A port is disconnected when: (a) no successful CONNECT-request has
   ever been issued for it, or (b) a DISCONNECT for the port has been
   executed since the last successful CONNECT.

   When a disconnected port is referenced in an assignment, it is
   connected by default either to:
   (a) the connection used for the transmission of Datalanguage to the
         Datacomputer, or
   (b) the connection used for the transmission of Datacomputer
         diagnostics to the user
   The choice between (a) and (b), of course, depends on whether the
   reference is for input or output.  These connections are established
   by the network user's ICP to the Datacomputer at the beginning of the
   session.

   Note that CONNECT and DISCONNECT do not open files or network
   connections.  They simply make address associations within the
   Datacomputer.  The files and connections are opened before each
   request and closed after each request.








Winter                                                         [Page 16]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


9.  Names in Datalanguage

   A name is recognized when it has been associated with a particular
   data container or set of containers.

   Datalanguage has mechanisms for the recognition of names in contexts.
   That is, the meaning of the name is related to where it appears.

   This makes it possible to attach natural meanings to partially
   qualified names.

   For example:

       WEATHER FILE LIST
             STATION STRUCT
                   CITY STR (15)
                   STATE STR (15)
                   DATA LIST (24)
                         OBSERVATION STRUCT
                               HOUR STR (2)
                               TEMPERATURE STR (3)
                               HUMIDITY STR (2)
                               PRESSURE STR (4)
                               END
                   END

       RESULTS PORT LIST
             RESULT STRUCT
                   CITY STR (15)
                   HOUR STR (2)
                   TEMPERATURE STR (3)
                   END

       FOR STATION WITH STATE EQ 'CALIFORNIA'
             FOR RESULT, OBSERVATION WITH HOUR GT '12'
                         AND HUMIDITY LT '75'
                   CITY = CITY ;
                   HOUR = HOUR ;
                   TEMPERATURE = TEMPERATURE ;
                   END ;
             END ;

   in the assignment 'CITY = CITY', the first CITY is understood to be
   RESULT.CITY and the second is understood to be STATION.CITY.







Winter                                                         [Page 17]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


9.1 Informal Presentation of Recognition Rules

      'Ident' is used in the sense of section 3.  For example, in the
      description:
         F FILE LIST R STRUCT A STR (1) B STR (1) END
         F, R, A and B are idents.

      A context is a tree whose nodes are idents.  In such a tree, the
      terminal nodes are idents of STRINGs.  The ident of a LIST is
      superior to the ident of the LIST-member.  The ident of a STRUCT
      is superior to the idents of the STRUCT elements.  The context
      whose top node is F is said to be the context of F.

                  +-----+
                  !  F  !
                  +-----+
                     !
                     !
                  +-----+
                  !  R  !
                  +-----+
                     !
                     !
          +----------+----------+
          !                     !
          !                     !
       +-----+               +-----+
       !  A  !               !  B  !
       +-----+               +-----+



       Figure 9.1-1  The context of F

      A pathname is a sequence of idents, naming nodes along a path from
      one node to another.  A full pathname in the context starts at the
      topmost node.  Thus F.R.B is a full pathname in the context of F.
      A partial pathname starts at a node other than the topmost node
      (e.g.  R.B, B).

      In Datalanguage, pathnames omitting intermediate nodes, such as
      F.B (which omits 'R'), are not permitted.  Thus partial pathnames
      are partial only in that additional names are implied on the left.








Winter                                                         [Page 18]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      Three attempts at recognition of a pathname, PN, in a context, CX,
      are made:

      (a)   recognition of PN as a full pathname in CX
      (b)   recognition of PN as a partial pathname in which only the
            topmost node of CX is omitted
      (c)   recognition of PN as an arbitrary partial pathname occurring
            only once in CX.

      The attempts are made in the above order, and the recognition
      process halts with the first successful attempt.

      As an example, consider the description:

          F FILE LIST
                R STRUCT
                      A STR
                      B STR
                      S STRUCT
                            R STR

      which defines the context in Figure 9.1-2.


                                  +-----+
                                  !  F  !
                                  +-----+
                                     !
                                     !
                                  +-----+
                                  !  R  !
                                  +-----+
                                     !
                                     !
                    +----------------+----------------+
                    !                !                !
                    !                !                !
                 +-----+          +-----+          +-----+
                 !  A  !          !  B  !          !  S  !
                 +-----+          +-----+          +-----+
                                                      !
                                                      !
                                                   +-----+
                                                   !  R  !
                                                   +-----+

                  Figure 9.1-2      Example Context




Winter                                                         [Page 19]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      In this context, F.R.A is a full pathname.  Thus, F.R.A is
      recognized in attempt (a).  R is a partial pathname in which only
      the topmost node is omitted.  Thus R is recognized in attempt (b).
      Note carefully that R is recognized as a reference to F.R, not to
      F.R.S.R.  Finally, B is an arbitrary partial pathname occurring
      only once in the context.  Thus B is recognized in attempt (c).

      Two stacks of contexts are maintained:  one for names used in an
      input sense, and one for names used in an output sense.  When a
      name is to be recognized, it is first decided whether the
      reference is an input reference or an output reference.  An input
      reference is (a) the right hand operand of an assign, or (b) a
      name in the input set spec of a FOR.  An output reference is (a)
      the left operand of an assign, or (b) the output operand of a FOR.
      The first context on the appropriate context stack is then
      searched, according to the procedure outlined on the previous
      page.  If the name is neither recognized nor ambiguous in that
      context, search continues in the next context on the stack.  If
      the name can be recognized in none of the contexts on the
      appropriate stack, it is unrecognizable.

      When a stack is empty, the recognition procedure is different.
      The search is carried on in a special context: The context of
      %OPEN.  Its top node, %OPEN, is a built in system ident.
      Subordinate to %OPEN is a context for each open directory node.
      Each such context represents all the idents defined in the
      directory nodes having data descriptions:
         F FILE LIST R STRUCT A STR (1) B STR (1)
      and:
         P PORT LIST R STRUCT A STR (1) B STR (1)
      then the context of %OPEN would be as in Figure 9.1-3.




















Winter                                                         [Page 20]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


                                 +-------+
                                 ! %OPEN !
                                 +-------+
                                     !
                                     !
                         +-----------+-----------+
                         !                       !
                         !                       !
                 +-----+                      +-----+
                 !  A  !                      !  S  !
                 +-----+                      +-----+
                    !                            !
                    !                            !
                 +-----+                      +-----+
                 !  R  !                      !  R  !
                 +-----+                      +-----+
                    !                            !
                    !                            !
              +-----+-----+                +-----+-----+
              !           !                !           !
              !           !                !           !
           +-----+     +-----+          +-----+     +-----+
           !  A  !     !  B  !          !  A  !     !  B  !
           +-----+     +-----+          +-----+     +-----+


                  Figure 9.1-3      The Context of %OPEN

      When a directory node is closed, the corresponding context is
      removed from the context of %OPEN.  When a node is opened, the
      associated context is added as the rightmost context subordinate
      to %OPEN.

9.2 Context Stack Maintenance

      The context stacks are always empty between requests.  The FOR-
      operator adds entries to the stacks.  FOR A adds the context of A
      to the input context stack.  FOR A, B ... adds the context of A to
      the output stack and the context of B to the input stack.

      When adding to an empty stack, FOR adds two contexts instead of
      one.  The second addition to the stack is the context of the
      looparg; the first addition is the context of the outermost
      container which encloses the looparg.







Winter                                                         [Page 21]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      For example, given a context of %OPEN as in Figure 9.1-3, and
      empty context stacks, the fragment 'FOR F.R ...' adds two
      contexts: (1) the context of F, and (2) the context of F.R.

      Contexts are removed from the stacks by the END matching the FOR
      which added them.

10.  Miscellaneous Topics

10.1 The LIST-request

      The LIST-request provides a means for the user to inspect system
      data of interest to him.  The user can determine the contents of
      the directory, the source or parsed form of any data description
      in the system, and the mode and connection status of any open file
      or port.

      The LIST operator has a single operand, which can have any of
      several forms.  The action of the operator is to output a list of
      values on the Datalanguage output port.

      To output a list of pathnames, the operand used is '%ALL'.  When
      '%ALL' appears alone, all pathnames in the directory are listed.
      When '%ALL' appears after the last ident in a directory pathname,
      the full pathnames of all nodes subordinate to the named node are
      listed.

      To output one or more source descriptions, the operand '.%SOURCE'
      is used.  '.%SOURCE' is preceded with one of (a) '%ALL', (b)
      '%OPEN', or (c) the ident of an open outermost container.  The
      output is either (a) all descriptions, (b) all open descriptions,
      or (c) a particular open description.

      To output a parsed description, the operand '.%DESC' is used
      ('%DESCRIPTION' is also accepted).  This operand is preceded
      either with (a) '%OPEN', or (b) the ident of an open outermost
      container.














Winter                                                         [Page 22]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      Examples:

         Let P be the ident of an open PORT.  Let A.B.C be a
         directory pathname.
             LIST %ALL
             LIST A.B.C.%ALL
             LIST %OPEN
             LIST %ALL.%SOURCE
             LIST %OPEN.%SOURCE
             LIST P.%SOURCE
             LIST %OPEN.%DESC
             LIST P.%DESC

         Note that 'LIST A.B.C.%SOURCE' is not legal - '.%SOURCE'
         must be preceded with the ident of an open container, not a
         pathname.  A similar restriction applies to '.%DESC'.

10.2 Constants in Datalanguage

      A constant of type STRING can be included in a Datalanguage
      request by enclosing it in single quotes:
         'ABC'
      A single quote is included in a constant by preceding it with a
      double quote:
         'FATHER"'S'
      Likewise, a double quote is included by preceding it with a double
      quote:
         'JOHN SAID ""HELLO""'

      Such constants can be used on the right of comparison operators
      and of assignment operators which reference strings.

      Eventually, Datalanguage will contain facilities for the inclusion
      of constants of all datatypes; such constants are simply a special
      case of the Datalanguage expression and will be permitted wherever
      such expressions are permitted.

10.3 Character Set

      Internally, Version 0/9 of the Datacomputer software operates in
      7-bit ASCII characters.  Its output to the ARPANET is converted to
      8-bit ASCII.  On input from the ARPANET, it expects 8-bit
      characters, which it converts to 7-bit characters.








Winter                                                         [Page 23]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      To convert from 7- to 8-bit characters, a '0' bit is prefixed.  To
      convert from 8- to 7-bit characters, the high-order bit is
      checked.  If the high-order bit is a '0', the bit is discarded and
      the character is accepted as a 7-bit character.  If the high-order
      bit is a '1', then the character is discarded.

      (In the following discussion, as elsewhere in this memo, all
      references to numerical character codes are in decimal).

      The remainder of this section discusses treatment of codes 0
      through 127, when they appear in Datalanguage requests.

      In general, printing characters are acceptable in requests, while
      control characters are not.  There are some exceptions, which are
      detailed below.  The printing characters are codes 32-126.  The
      control characters are codes 0-31 and 127.

      Certain control characters are accepted:

         Tab(9) is accepted freely in requests.  It functions as a
         separator (explained below).

         EOL(31), meaning end-of-line, is accepted in requests,
         functioning both as a separator and an activator (a).  EOL has
         a special meaning in data, and should not be introduced into
         STRING constants(***).

         Control-L(12) is an activator and a high-level request
         delimiter.  It terminates the test of any request being
         processed when it is encounter in the input stream.  It is
         useful in Datacomputer-user program synchronization.

         Control-Z(26) means end-of-session when encountered in
         Datalanguage.  It has the properties of control-L, and in
         addition, causes the Datacomputer to execute an end-of-session
         procedure, which results in closing the Datalanguage
         connections, closing any open files or ports, etc.  The effect
         is equivalent to a [LOGOUT(which] does not exist yet).

         The two-character sequence <carriage return(13), line feed(10)>
         is equivalent to EOL (and is translated to EOL on input from
         the network).  The reverse sequence, as well as either
         character alone, is treated simply as other control characters
         (ignored).

      All other control characters are ignored.





Winter                                                         [Page 24]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      The printing characters are further divided into four groups:
      special characters, uppercase letters, lower case letters, and
      digits (the membership of these groups is defined in section 11).

      Corresponding upper and lower case letters are equivalent in
      requests, except with quoted strings.

      Certain special characters have a lexical function, which is
      either break or separator.  A break character terminates the
      current lexical item and returned itself as the next item.  A
      separator character terminates the current item but does not begin
      a new item (i.e., its only function is to separate items).
      Multiple separators are equivalent to a single separator.  A
      separator can always be inserted before or after a break
      character, without altering the meaning of the request.

      The separators are tab(9), space(32), and end-of-line(31).

      The break characters are left parenthesis(40), right
      parenthesis(41), equals(61), semicolon(59), period(46), comma(44),
      quote(39), and slash(47).

      (a)   An activator character causes the Datacomputer to process
            whatever has been received since the previous activator or
            the beginning of the request.  The meaning of a request is
            independent of the presence/absence of activators.  However,
            a request will not be executed until an activator beyond the
            end of the request is received.

      While Version 0/9 defines (carriage return, linefeed) in terms of
      EOL, network users should not think in terms of sending EOL's to
      the Datacomputer over the network.  EOL is not part of the network
      ASCII character set, and has no definite permanent place in
      Datacomputer implementation plans.

10.4 Comments

      Comments can be included in Datalanguage requests.  A comment is
      begun with the two-character sequence '/*', and ended with the
      two-character sequence '*/'.  Since slash is a break character, a
      comment does cause a lexical break; its overall effect is that of
      a separator.









Winter                                                         [Page 25]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


10.5 Reserved Identifiers

      Certain identifiers are reserved in Datalanguage, and should not
      be used to name containers or directory nodes.  These are:

      AND
      APPEND
      AT
      CLOSE
      CONNECT
      CREATE
      DELETE
      DISCONNECT
      END
      EQ
      FILE
      FOR
      GE
      GT
      LE
      LIST
      LT
      NODE
      NE
      NOT
      OPEN
      OR
      PORT
      READ
      STR
      STRUCT
      TO
      WITH
      WRITE

      More reserved identifiers will be added in Version 0/10.















Winter                                                         [Page 26]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


11.  Datalanguage Syntax Expressed in BNF

11.1 Requests

      11.1.01  <request>  ::=  ;
      11.1.02  <request>  ::=  <create>
      11.1.03  <request>  ::=  OPEN <pn>  ;
      11.1.04  <request>  ::=  OPEN <pn> <mode>  ;
      11.1.05  <request>  ::=  CLOSE <ident>  ;
      11.1.06  <request>  ::=  CONNECT <ident> TO <address>  ;
      11.1.07  <request>  ::=  DISCONNECT <ident>  ;
      11.1.08  <request>  ::=  MODE <ident> <mode>  ;
      11.1.09  <request>  ::=  DELETE <pn>  ;
      11.1.10  <request>  ::=  LIST <listarg>  ;
      11.1.11  <request>  ::=  <sr-request>  ;

11.2 Data Description and Creation

      11.2.01  <create>  ::=  CREATE <pn>
      11.2.02  <create>  ::=  CREATE <pn> <ftn> LIST <desc>
      11.2.03  <create>  ::=  CREATE <pn> <ftn> LIST <size> <desc>

      11.2.04  <desc>  ::=  <ident> <attributes>

      11.2.05  <attributes>  ::=  LIST <size> <desc>
      11.2.06  <attributes>  ::=  STRUCT <descs> END
      11.2.07  <attributes>  ::=  STR <size>
      11.2.08  <attributes>  ::=  STR <size> ,I=D

      11.2.09  <descs>  ::=  <desc>
      11.2.10  <descs>  ::=  <descs> <desc>

      11.2.11  <ftn>  ::=  PORT
      11.2.12  <ftn>  ::=  FILE
      11.2.13  <ftn>  ::=  TEMP PORT
      11.2.14  <ftn>  ::=  TEMPORARY PORT

      11.2.15  <size>  ::=  (  <integer constant>  )













Winter                                                         [Page 27]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


11.3 Data Storage and Retrieval

      11.3.01  <sr-request>  ::=  <assign>
      11.3.01  <sr-request>  ::=  <loop>

      11.3.03  <assign>  ::=  <pn> = <object>

      11.3.04  <loop>  ::=  FOR <looparg> <loopbody> END

      11.3.05  <looparg>  ::=  <exp>
      11.3.06  <looparg>  ::=  <pn> , <exp>

      11.3.07  <loopbody>  ::=  <sr-request>
      11.3.08  <loopbody>  ::=  <loopbody1> <sr-request>
      11.3.09  <loopbody>  ::=  <loopbody1>

      11.3.10  <loopbody1>  ::=  <sr-request>  ;
      11.3.11  <loopbody1>  ::=  <loopbody1> <sr-request>  ;

11.4 Expressions

      11.4.01  <exp>  ::=  <pn>
      11.4.02  <exp>  ::=  <pn> WITH <boolexp>

      11.4.03  <boolexp>  ::=  <pn> <comop> <string constant>
      11.4.04  <boolexp>  ::=  (  <boolexp>  )
      11.4.05  <boolexp>  ::=  NOT <boolexp>
      11.4.06  <boolexp>  ::=  <boolexp> AND <boolexp>
      11.4.07  <boolexp>  ::=  <boolexp> OR <boolexp>

      11.4.08  <comop>  ::=  EQ
      11.4.09  <comop>  ::=  NE
      11.4.10  <comop>  ::=  GT
      11.4.11  <comop>  ::=  LT
      11.4.12  <comop>  ::=  GE
      11.4.13  <comop>  ::=  LE















Winter                                                         [Page 28]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


11.5 Miscellaneous

      11.5.01  <address>  ::=  <quote> <TENEX file designator>
               <quote>
      11.5.02  <address>  ::=  <socket> AT <host>
      11.5.03  <address>  ::=  <socket>

      11.5.04  <socket>  ::=  <integer constant>  //INTERPRETED AS
               DECIMAL

      11.5.05  <host>  ::=  <integer constant>  //INTERPRETED AS
               DECIMAL
      11.5.06  <host>  ::=  *****  TENEX host names *****

      11.5.07  <object>  ::=  <pn>
      11.5.08  <object>  ::=  <string constant>

      11.5.09  <mode>  ::=  READ
      11.5.10  <mode>  ::=  APPEND
      11.5.11  <mode>  ::=  WRITE

      11.5.12  <listarg>  ::=  %ALL
      11.5.13  <listarg>  ::=  <pn>.%ALL
      11.5.14  <listarg>  ::=  %OPEN
      11.5.15  <listarg>  ::=  %ALL.%SOURCE
      11.5.16  [<LISTARG>]  ::=  <IDENT>.%SOURCE
      11.5.17  <listarg>  ::=  %OPEN.%SOURCE
      11.5.18  <listarg>  ::=  %OPEN.%DESC
      11.5.19  <listarg>  ::=  <ident>.%DESC

      11.5.20  <pn>  ::=  <ident>
      11.5.21  <pn>  ::=  <pn>.<ident>

      11.5.22  <ident>  ::=  <letter>
      11.5.23  <ident>  ::=  <ident> <letter>
      11.5.24  <ident>  ::=  <ident> <digit>

      11.5.25  <integer constant>  ::=  <digit>
      11.5.26  <integer constant>  ::=  <integer constant> <digit>

      11.5.27  <string constant>  ::=  <quote> <string conbody>
               <quote>

      11.5.28  <string conbody>  ::=  <nonquote>
      11.5.28  <string conbody>  ::=  <string conbody> <nonquote>






Winter                                                         [Page 29]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


11.6 Character Set

      11.6.01  <separator>  ::=   //SPACE(32)
      11.6.02  <separator>  ::=   //TAB(9)
      11.6.03  <separator>  ::=  <eol>

      11.6.04  <special>  ::=  <quote>
      11.6.05  <special>  ::=  <superquote>
      11.6.06  <special>  ::=  <special1>

      11.6.07  <letter>  ::=  A
      11.6.08  <letter>  ::=  B
       ...............
      11.6.09  <letter>  ::=  Z

      11.6.10  <letter>  ::=  a
      11.6.11  <letter>  ::=  b
       ...............
      11.6.12  <letter>  ::=  z

      11.6.13  <digit>  ::=  0
      11.6.14  <digit>  ::=  1
       ...............
      11.6.15  <digit>  ::=  9

      11.6.16  <nonquote>  ::=  <letter>
      11.6.17  <nonquote>  ::=  <digit>
      11.6.18  <nonquote>  ::=  <superquote> <quote>
      11.6.19  <nonquote>  ::=  <superquote> <superquote>
      11.6.20  <nonquote>  ::=  <special1>
      11.6.21  <nonquote>  ::=  <separator>

      11.6.22  <eol>  ::=  //EOL (31)
      11.6.23  <eol>  ::=  <carriage return> <line feed>

      11.6.24  <carriage return>  ::=  //CARRIAGE RETURN (13)
      11.6.25  <line feed>  ::=  //LINE FEED (10)

      11.6.26  <quote>  ::=  ' //SINGLE QUOTE(44)

      11.6.27  <superquote>  ::=  " //DOUBLE QUOTE(34)

      11.6.28  <special1>  ::=  ! //EXCLAMATION POINT(33)
      11.6.29  <special1>  ::=  # //NUMBER SIGN(35)
      11.6.30  <special1>  ::=  $ //DOLLAR SIGN(36)
      11.6.31  <special1>  ::=  % //PERCENT SIGN(37)
      11.6.32  <special1>  ::=  & //AMPERSAND(38)
      11.6.33  <special1>  ::=  ( //LEFT PARENTHESIS(40)



Winter                                                         [Page 30]

RFC 515       Specifications for Datalanguage, Version 0/9   6 June 1973


      11.6.34  <special1>  ::=  ) //RIGHT PARENTHESIS(41)
      11.6.35  <special1>  ::=  * //ASTERISK(42)
      11.6.36  <special1>  ::=  + //PLUS SIGN(43)
      11.6.37  <special1>  ::=  , //COMMA(44)
      11.6.38  <special1>  ::=  - //MINUS SIGN(45)
      11.6.39  <special1>  ::=  . //PERIOD(46)
      11.6.40  <special1>  ::=  / //SLASH(47)
      11.6.41  <special1>  ::=  : //COLON(58)
      11.6.42  <special1>  ::=  ; //SEMICOLON(59)
      11.6.43  <special1>  ::=  < //LEFT ANGLE BRACKET(60)
      11.6.44  <special1>  ::=  = //EQUAL SIGN(61)
      11.6.45  <special1>  ::=  > //RIGHT ANGLE BRACKET(62)
      11.6.46  <special1>  ::=  ? //QUESTION MARK(63)
      11.6.47  <special1>  ::=  @ //AT-SIGN(64)
      11.6.48  <special1>  ::=  [ //LEFT SQUARE BRACKET(91)
      11.6.49  <special1>  ::=   //BACK SLASH(92)
      11.6.50  <special1>  ::=  ] //RIGHT SQUARE BRACKET(93)
      11.6.51  <special1>  ::=  ^ //CIRCUMFLEX(94)
      11.6.52  <special1>  ::=  _ //UNDERBAR(95)
      11.6.53  <special1>  ::=  ` //ACCENT GRAVE(96)
      11.6.54  <special1>  ::=  { //LEFT BRACE(123)
      11.6.55  <special1>  ::=  | //VERTICAL BAR(124)
      11.6.56  <special1>  ::=  } //RIGHT BRACE(125)
      11.6.57  <special1>  ::=  ~ //TILDE(126)

      <EOF>


          [ This RFC was put into machine readable form for entry ]
           [ into the online RFC archives by Walter Benton 12/98 ]





















Winter                                                         [Page 31]