Premier IT Outsourcing and Support Services within the UK

User Tools

Site Tools


Network Working Group 4691 RFC-5 Jeff Rulifson

                                                              June 2, l969

:DEL, 02/06/69 1010:58 JFR ; .DSN=1; .LSP=0; ['=] AND NOT SP ; ['?]; dual transmission?


 The Decode-Encode Language (DEL) is a machine independent language
 tailored to two specific computer network tasks:
    accepting input codes from interactive consoles, giving immediate
    feedback, and packing the resulting information into message 
    packets for network transmissin.
    and accepting message packets from another computer, unpacking
    them, building trees of display information, and sending other
    information to the user at his interactive station.
 This is a working document for the evolution of the DEL language.
 Comments should be made through Jeff Rulifson at SRI.


 The initial ARPA network working group met at SRI on October 25-26,
    It was generally agreed beforehand that the runmning of interactive
    programs across the network was the first problem that would be
    This group, already in agreement about the underlaying notions of
    a DEL-like approach, set down some terminology, expectations for
    DEL programs, and lists of proposed semantic capability.
    At the meeting were Andrews, Baray, Carr, Crocker, Rulifson, and
 A second round of meetings was then held in a piecemeal way.
    Crocker meet with Rulifson at SRI on November 18, 1968.  This
    resulted in the incorporation of formal co-routines.
    and Stoughton meet with Rulifson at SRI on Decembeer 12, 1968.  It
    was decided to meet again, as a group, probably at UTAH, in late
    January 1969.
 The first public release of this paper was at the BBN NET meeting in
 Cambridge on February 13, 1969.


 NST   The NST library is the set of programs necessary to mesh
 efficiently with the code compiled at the user sites from the DEL
 programs it receives.  The NST-DEL approach to NET interactive system
 communication is intended to operate over a broad spectrum.
 The lowest level of NST-DEL usage is direct transmission to the
 server-host, information in the same format that user programs
 would receive at the user-host.
    In this mode, the NST defaults to inaction.  The DEL program
    does not receive universal hardware representation input but 
    input in the normal fashion for the user-host.
    And the DEL 1 program becomes merely a message builder and
 A more intermediate use of NST-DEL is to have echo tables for a
 TTY at the user-host.
    In this mode, the DEL program would run a full duplex TTY for
    the user.
    It would echo characters, translate them to the character set 
    of the server-host, pack the translated characters in messages,
    and on appropriate break characters send the messages.
    When messages come from the server-host, the DEL program would
    translate them to the user-host character set and print them on
    his TTY.
 A more ambitious task for DEL is the operation of large,
 display-oriented systems from remote consoles over the NET.
    Large interactive systems usually offer a lot of feedback to
    the user.  The unusual nature of the feedback make it
    impossible to model with echo table, and thus a user program
    must be activated in a TSS each time a button state is changed.
       This puts an unnecessarily large load on a TSS, and if the
       system is being run through the NET it could easily load two
       To avoid this double overloading of TSS, a DEL program will
       run on the user-host.  It will handle all the immediate
       feedback, much like a complicated echo table.  At appropriate
       button pushes, message will be sent to the server-host and
       display updates received in return.
    One of the more difficult, and often neglected, problems is the
    effective simulation of one nonstandard console on another non-
    standard console.
       We attempt to offer a means of solving this problem through
       the co-routine structure of DEL programs.  For the
       complicated interactive systems, part of the DEL programs
       will be constructed by the server-host programmers.
       Interfaces between this program and the input stream may
       easily be inserted by programmers at the user-host site.


 To minimize the number of translators needed to map any facility's
 user codes to any other facility, there is a universal hardware
 This is simply a way of talking, in general terms, about all the
 hardware devices at all the interactive display stations in the initial
 For example, a display is thought of as being a square, the
 mid-point has coordinates (0.0), the range is -1 to 1 on both
 axes.  A point may now be specified to any accuracy, regardless of
 the particular number of density of rastor points on a display.
 The representation is discussed in the semantic explanations
 accompanying the formal description of DEL.


 Suppose that a user at a remote site, say Utah, is entered in the
 AHI system and wants to run NLS.
 The first step is to enter NLS in the normal way.  At that time
 the Utah system will request a symbolic program from NLS.
    REP   This program is written in DEL.  It is called the NLS
    Remote Encode Program (REP).
    The program accepts input in the Universal Hardware
    Representation and translates it to a form usable by NLS.
    It may pack characters in a buffer, also do some local
 When the program is first received at Utah it is compiled and
 loaded to be run in conjunction with a standard library.
 All input from the Utah console first goes to the NLS NEP.  It is
 processed, parsed, blocked, translated, etc.  When NEP receives a
 character appropriate to its state it may finally initiate
 transfers to the 940.  The bits transferred are in a form
 acceptable to the 940, and maybe in a standard form so that the
 NLSW need not differentiate between Utah and other NET users.


 After each node has implemented the library part of the NST, it
 need only write one program for each subsystem, namely the
 symbolic file it sends to each user that maps the NET hardware
 representation into its own special bit formats.
    This is the minimum programming that can be expected if 
    console is used to its fullest extent.
    Since the NST which runs the encode translation is coded at the
    user site, it can take advantage of hardware at its consoles to
    the fullest extent.  It can also add or remove hardware 
    features without requiring new or different translation tables
    from the host.
    Local users are also kept up to date on any changes in the system
    offered at the host site.  As new features are added,
    the host programmers change the symbolic encode program.  When
    this new program is compiled and used at the user site, the new
    features are automatically included.
 The advantages of having the encode translation programs
 transferred symbolically should be obvious.
    Each site can translate any way it sees fit.  Thus machine code
    for each site can be produced to fit that site; faster run
    times and greater code density will be the result.
    Moreover, extra symbolic programs, coded at the user site, may
    be easily interfaced between the user's monitor system and the
    DEL program from the host machine.  This should ease the
    problem of console extension (e.g. accommodating unusual keys and
    buttons) without loss of the flexibility needed for man-machine
 It is expected that when there is matching hardware, the symbolic
 programs will take this into account and avoid any unnecessary
 computing.  This is immediately possible through the code
 translation constructs of DEL.  It may someday be possible through
 program composition (when Crocker tells us how??)


    The right side of the picture represents functions done at the
    user's main computer; the left side represents those done at the
    host computer.
       Each label in the picture corresponds to a statement with the
       same name.
       There are four trails associated with this picture.  The first
       links (in a forward direction) the labels which are concerned
       only with network information.  The second links the total
       information flow (again in a forward direction).  The last two
       are equivalent to the first two but in a backward direction.
       They may be set with pointers t1 through t4 respectively.
       [">tif:] OR I" >nif"]; ["<tif:] OR ["<nif"];


 Keyboard is the set of input devices at the user's console.
 Input bits from stations, after drifting through levels of monitor
 and interrupt handlers, eventually come to the encode translator.
 Encode maps the semi-raw input bits into an input stream in a
 form suited to the serving-host subsystem which will process the
 input.  [>nif(hrt)<nif(keyboard)]
    The Encode program was supplied by the server-host subsystem
    when the subsystem was first requested.  It is sent to the user
    machine in symbolic form and is compiled at the user machine
    into code particularly suited to that machine.
    It may pack to break characters, map multiple characters to
    single characters and vice versa, do character translation, and
    give immediate feedback to the user.
 1 dm    Immediate feedback from the encode translator first goes to
 local display management, where it is mapped from the NET standard
 to the local display hardware.
    A wide range of echo output may come from the encode
    translator.  Simple character echoes would be a minimum, while
    command and machine-state feedback will be common.
    It is reasonable to expect control and feedback functions not
    even done at the server-host user stations to be done in local
    display control.  For example, people with high-speed displays
    may want to selectively clear curves on a Culler display, a
    function which is impossible on a storage tube.
 Output from the encode translator for the server-host goes to the
 invisible IMP, is broken into appropriate sizes and labeled by the
 encode translator, and then goes to the NET-to-host translator.
    Output from the user may be more than on-line input.  It may be
    larger items such as computer-generated data, or files
    generated and used exclusively at the server-host site but
    stored at the user-host site.
    Information of this kind may avoid translation, if it is already in
    server-host format, or it may undergo yet another kind of translation
    if it is a block of data.
 hrp  It finally gets to the host, and must then go through the
 host reception program.  This maps and reorders the standard
 transmission-style packets of bits sent by the encode programs
 into messages acceptable to the host.  This program may well be
 part of the monitor of the host machine. [>tif(net mode)<nif(code)]


 decode   Output from the server-host initially goes through decode,
 a translation map similar to, and perhaps more complicated than,
 the encode map.  [>nif(urt)>tif(imp ctrl)<tif(net mode)]
    This map at least formats display output into a simplified
    logical-entity output stream, of which meaningful pieces may be
    dealt with in various ways at the user site.
       The Decode program was sent to the host machine at the same
       time that the Encode program was sent to the user machine.
       The program is initially in symbolic form and is compiled
       for efficient running at the host machine.
       Lines of charaters should be logically identified so that
       different line widths can be handled at the user site.
       Some form of logical line identification must also be made.
       For example, if a straight line is to be drawn across the
       display this fact should be transmitted, rather than a
       series of 500 short vectors.
       As things firm up, more and more complicated structural
       display information (in the manner of LEAP) should be sent
       and accommodated at user sites so that the responsibility for
       real-time display manipulation may shift closer to the user.
    imp ctrl   The server-host may also want to send control
    information to IMPs.  Formatting of this information is done by
    the host decoder.  [>tif(urt) <tif(decode)]
    The other control information supplied by the host decoder is
    message break up and identification so that proper assembly and
    sorting can be done at the user site.
 From the host decoder, information does to the invisible IMP, and
 directly to the NET-to-user translator.  The only operation done
 on the messages is that they may be shuffled.
 urt   The user reception translator accepts messages from the
 user-site IMP 1 and fixes them up for user-site display.  
 [>nif(d ctrl)>tif(prgm ctrl)<tif(imp ctrl)<nif(decode)]
    The minimal action is a reordering of the message pieces.
    dctrl   For display output, however, more needs to be done.  The
    NET logical display information must be put in the format of
    the user site.  Display control does this job.  Since it
    coordinates between (encode) and (decode) it is able to offer
    features of display management local to the user site.
    prgmctrl   Another action may be the selective translation and
    routing of information to particular user-site subsystems.
       For example, blocks of floating-point information may be
       converted to user-style words and sent, in block form, to a
       subsystem for processing or storage.
       The styles and translation of this information may well be a 
       compact binary format suitable for quick translation, rather
       than a print-image-oriented format.
    (display)   is the output to the user.  [<nif(d ctrl)]
    (net mode)   This is the mode where a remote user can link to a node
    indirectly through another node.   [<nif(decode)<tif(hrt)]



    All statements in this branch which are not part of the compiler
    must end with a period.
    To compile the DEL compiler:
       Set this pattern for the content analyzer ( (symbol for up arrow)P1
       SE(P1) <-"-;). The pointer "del" is on the first character of pattern.
       Jump to the first statement of the compiler.  The pointer "c"
       is on this statement.
       And output the compiler to file  ( '/A-DEL' ).  The pointer "f"
       is on the name of the file for the compiler output -
  1. meta file (k=100.m=300,n=20,s=900)
       file = mesdecl $declaration $procedure "FINISH";
       procedure =
         procname (
               type "FUNCTION" /
               "PROCEDURE" ) .id (type .id / -empty)) /
            "CO-ROUTINE") ' /
         $declaration labeledst $(labeledst ';) "endp.";
       labeledst = ((left arrow symbol).id ': / .empty) statement;
       type = "INTEGER" / "REAL" ;
       procname = .id;
    Functions are differentiated from procedures to aid compilers in
    better code production and run time checks.
       Functions return values.
       Procedures do not return values.
    Co-routines do not have names or arguments.  Their initial
    envocation points are given the pipe declaration.
    It is not clear just how global declarations are to be??


    declaration = numbertype / structuredtype / label / lcl2uhr /
    uhr2rmt / pipetype;
    numbertype = : ("REAL" / "INTEGER") ("CONSTANT" conlist /
    conlist =
       .id '(left arrow symbol)constant
       $('. .id '(left arrow symbol)constant);
    varlist =
       .id ('(left arrow symbol)constant / .empty)
       $('. .id('(left arrow symbol)constant / .empty));
    idlist = .id $('. .id);
    structuredtype = (tree" / "pointer" / "buffer" ) idlist;
    label = "LABEL1" idlist;
    pipetype = PIPE" pairedids $(', pairedids);
    pairedids = .id .id;
    procname = .id;
    integerv = .id;
    pipename = .id;
    labelv = .id;
 Variables which are declared to be constant, may be put in
 read-only memory at run time.
 The label declaration is to declare cells which may contain the
 machine addresses of labels in the program as their values.  This 
 is not the B5500 label declaration.
 In the pipe declaration the first .ID of each pair is the name of
 the pipe, the second is thke initial starting point for the pipe.


    exp = "IF" conjunct "THEN" exp "ELSE" exp;
    sum = term (
       '+ sum /
       '- sum /
  1. empty);
    term = factor (
       '* term /
       '/ term /
       '(up arrow symbol) term /
    factor = '- factor / bitop;
    bitop = compliment (
       '/' bitop /
       '/'\ bitop /
       '& bitop / (
    compliment = "--" primary / primary;
 (symbol for up arrow) means mod. and /\ means exclusive or.
 Notice that the uniary minus is allowable, and parsed so you can
 write x*-y.
 Since there is no standard convention with bitwise operators, they
 all have the same precedence, and parentheses must be used for
 Compliment is the l's compliment.
 It is assumed that all arithmetic and bit operations take place in
 the mode and style of the machine running the code.  Anyone who
 takes advantage of word lengths, two's compliment arithmetic, etc.
 will eventually have problems.


    primary =
       constant /
       builtin /
       variable / (
       block /
       '( exp ');
    variable = .id (
       '(symbol for left arrow) exp /
       '( block ') /
    constant =  integer / real / string;
    builtin =
       mesinfo /
       cortnin /
       ("MIN" / "MAX") exp $('. exp) '/ ;
 parenthesized expressions may be a series of expressions.  The
 value of a series is the value of the last one executed at run time.
 Subroutines may have one call by name argument.
 Expressions may be mixed.  Strings are a big problem?  Rulifson
 also wants to get rid of real numbers!!


    conjunct = disjunct ("AND" conjunct / .empty);
    disjunct = negation ("OR" negation / .empty);
    negation = "NOT" relation / relation;
    relation =
       '( conjunct ') /
       sum (
         "<=" sum /
         ">=" sum /
         '< sum /
         '> sum /
         '= sum /
         '" sum /
 The conjunct construct is rigged in such a way that a conjunct
 which is not a sum need not have a value, and may be evaluated
 using jumps in the code.  Reference to the conjunct is made only
 in places where a logical decision is called for (e.g. if and
 while statements).
 We hope that most compilers will be smart enough to skip
 unnecessary evaluations at run time.  I.e a conjunct in which the
 left part is false or a disjunct with the left part true need not
 have the corresponding right part evaluated.


    statement = conditional / unconditional;
    unconditional = loopst / cases / cibtrikst / uist / treest /
    block / null / exp;
    conditional = "IF" conjunct "THEN" unconditional (
       "ELSE" conditional /
    block = "begin" exp $('; exp) "end";
 An expressions may be a statement.  In conditional statements the
 else part is optional while in expressions it is mandatory.  This
 is a side effect of the way the left part of the syntax rules are


    treest = setpntr / insertpntr / deletepntr;
    setpntr = "set" "pointer" pntrname "to" pntrexp;
    pntrexp = direction pntrexp / pntrname;
    insertpntr = "insert" pntrexp "as"
       (("left" / "right") "brother") /
       (("first" / "last: ) "daughter") "of" pntrexp;
    direction =
       "up" /
       "down" /
       "forward" /
       "backward: /
       "head" /
    plantree = "replace" pntrname "with" pntrexp;
    deletepntr = "delete: pntrname;
    tree = '( tree1 ') ;
    tree1 = nodename $nodename ;
    nodename = terminal / '( tree1 ');
    terminal = treename / buffername / point ername;
    treename = id;
    treedecl = "pointer" .id / "tree" .id;
 Extra parentheses in tree building results in linear subcategorization,
 just as in LISP.


 controlst = gost / subst / loopstr / casest;
    gost = "GO" "TO" (labelv / .id);
       assignlabel = "ASSIGN" .id "TO" labelv;
    subst = callst / returnst / cortnout;
       callst = "CALL" procname (exp / .emptyu);
       returnst = "RETURN" (exp / .empty);
       cortnout = "STUFF" exp "IN" pipename;
    cortnin = "FETCH" pipename;
    FETCH is a builtin function whose value is computed by envoking
    the named co-routine.
       loopst = whilest / untilst / forst;
       whilest = "WHILE" conjunct "DO" statement;
       untilst = "UNTIL" conjunct "DO" statement;
       forst = "FOR" integerv '- exp ("BY" exp / .empty) "TO" exp
       "DO" statements;
    The value of while and until statements is defined to be false
    and true (or 0 and non-zero) respectively.
    For statements evaluate their initial exp, by part, and to part
    once, at initialization time.  The running index of for
    statements is not available for change within the loop, it may
    only be read.  If, some compilers can take advantage of this
    (say put it in a register) all the better.  The increment and
    the to bound will both be rounded to integers during the


    casest = ithcasest / condcasest;
    ithcasest = "ITHCASE" exp "OF" "BEGIN" statement $(';
    statement) "END";
    condcasest = "CASE" exp "OF" "BEGIN" condcs $('; condcs)
    "OTHERWISE" statement "END";
    condcs = conjunct ': statement;
 The value of a case statement is the value of the last case executed.


 null = "NULL";


 iost = messagest / dspyst ;
       messagest = buildmes / demand;
          buildmest = startmes / appendmes / sendmes;
            startmes = "start" "message";
            appendmes = "append" "message" "byute" exp;
            sendmes = "send" "message";
         demandmes = "demand" "Message";
    mesinfo =
       "get" "message" "byte"
       "message1" "length" /
       "message" empty: '?;
    mesdecl = "message" "bytes" "are" ,byn "bits" long" '..


    dspyst = startbuffer / bufappend / estab;
    startbuffer - "start" "buffer";
    bufappend = "append" bufstuff $('& bufstuff);
    bufstuff = :
       "parameters" dspyparm $('. dspyparm) /
       "character" exp /
       "string"1 strilng /
       "vector" ("from" exp ':exp / .empty) "to" exp '. exp /
       "position" (onoff / .empty) "beam" "to" exp '= exp/
       curve" ;
    dspyparm F :
       "intensity" "to" exp /
       "character" "width" "to" exp /
       "blink" onoff /
      "italics" onff;
    onoff = "on" / "off";
    estab = "establish" buffername;
    The screen is taken to be a square.  The coordinates are
    normalized from -1 to +1 on both axes.
    Associated with the screen is a position register, called
    PREG.  The register is a triple <x.y.r> where x and y 
    specify a point on the screen and r is a rotation in
    radians, counter clockwise, from the x-axis.
    The intensity, called INTENSITY, is a real number in the
    range from 0 to 1.  0 is black, 1 is as light as your
    display can go, and numbers in between specify the relative
    log of the intensity difference.
    Character frame size.
    Blink bit.
    The terminal nodes of semi-trees are either semi-tree names
    or display buffers.  A display buffer is a series of logical
    entities, called bufstuff.
    When the buffer is initilized, it is empty.  If no
    parameters are initially appended, those in effect at the
    end of the display of the last node in the semi-tree will be in
    effect for the display of this node.
    As the buffer is built, the logical entities are added to it.
    When it is established as a buffername, the buffer is
    closed, and further appends are prohibited.  It is only a
    buffername has been established that it may be used in a tree
    building statement.
    Joy Stick
    Light Pens


 Program to run display and keyboard as tty.
 to run NLS
    input part
    display part
       While LENGTH " O DO
          ITHCASE GETBYTE OF Begin
          ITHCASE GETBYTE OF %file area uipdate% BEGIN
             %literal area%
             %message area%
             %name area%
             %sequence specs%
             %filter specs%
             %format specs%
             %command feedback line%
             %filer area%
             %date time%
             %echo register%
         BEGIN %DEL control%


 Steve Carr
    Department of Computer Science
    University of Utah
    Salt Lake City, Utah  84112
    Phone 801-322-7211 X8224
 Steve Crocker
    Boelter Hall
    University of California
    Los Angeles, California
    Phone 213-825-4864
 Jeff Rulifson
    Stanford Research Institute
    333 Ravenswood
    Menlo Park, California  94035
    Phone 415-326-6200 X4116
 Ron Stoughton
    Computer Research Laboratory
    University of California
    Santa Barbara, California  93106
    Phone 805-961-3221
 Mehmet Baray
    Corey Hall
    University of California
    Berkeley, California  94720
    Phone 415-843-2621
/data/webs/external/dokuwiki/data/pages/rfc/rfc5.txt · Last modified: 2001/05/11 03:48 by

Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki