Request for Comments: 713                           Jack Haverty  (JFH@MIT-DMS)
NIC #34739                                                             Apr 1976







I. ABSTRACT


A mechanism is defined for use by message servers in
transferring data between hosts.  The mechanism, called the
MSDTP, is defined in terms of a model of the process as a
translation between two sets of items, the abstract entities
such as 'strings' and 'integers', and the formats used to
represent such data as a byte stream.

A proposed organization of a general data transfer
mechanism is described, and the manner in which the MSDTP
would be used in that environment is presented.





































                                -1-

II. REFERENCES


Black, Edward H., "The DMS Message Composer", MIT Project
MAC, Programming Technology Division Document
SYS.16.02.

Burchfiel, Jerry D., Leavitt, Elsie M., Shapiro, Sonya and
Strollo, Theodore R., compilers, "Tenex Users' Guide",
Bolt Beranek and Newman, Cambridge, Mass., May 1971,
revised January 1975, Descriptive sections on the TENEX
subsystems: MAlLER, p. 116-11; MAlLSTAT, p. 118-119;
READMAIL, p. 137; and SNDMSG, p. 165-170.

Haverty, Jack, "Communications System Overview", MIT Project
MAC, Programming Technology Division Document
SYS.16.00.

Haverty, Jack, "Communications System Daemon Manual", MIT
Project MAC, Programming Technology Division Document
SYS.16.01.

ISI Information Automation Project, "Military Message
Processing System Design," Internal Project
Documentation (Out of Print), Jan. 1975

Message Services Committee, "Interim Report", Jan. 28, 1975

Mooers, Charlotte D., "Mailsys Message System: Manual For
Users", Bolt Beranek and Newman, Cambridge, Mass., June
1975 (draft).

Myer, Theodore H., "Notes On The BBN Mail System", Bolt
Beranek and Newman, November 8, 1974.

Myer, Theodore H., and Henderson, D. Austin, "Message
Transmission Protocol", Network Working Group RFC 680,
NIC 32116, April 30, 1975.

Postel, Jon, "The PCPB8 Format", NSW Proposal, June 5, 1975

Tugender, R., and D. R. Oestreicher, "Basic Functional
Capabilities for a Military Message Processing
Service," ISI?RR-74-23., May 1975

Vezza, Al, "Message Services Committee Minority Report",
Jan. 1975










                                   -2-

III. OVERVIEW


This document describes a mechanism developed for use
by message servers communicating over an eight-bit
byte-oriented network connection to move data structures and
associated data-typing information.  It is presented here in
the hope that it may be of use to other projects which need
to transfer data structures between dissimilar hosts.

A set of abstract entities called PRIMITIVE ITEMS is
enumerated.  These are intended to include traditional data
types of general utility, such as integers, strings, and
arrays.

A mechanism is defined for augmenting the set of
abstract data entities handled, to allow the introduction of
application-specific data, whose format and semantics are
understood by the application programs involved, but which
can be transmitted using common coding facilities.  An
example might be a data structure called a 'file
specification', or a 'date'.  Abstract data entities defined
using this mechanism will be termed SEMANTIC ITEMS, since
they are typically used to carry data having semantic
content in the application involved.

Semantic and primitive items are collectively referred
to simply as ITEMS.

The protocol next involves the definition of the format
of the byte stream used to convey items from machine to
machine.  These encodings are described in terms of OBJECTS,
which are the physical byte streams transmitted.

To complete the protocol, the rules for translating
between objects and items are presented as each object is
defined.

An item is transmitted by being translated into an
object which is transmitted over the connection as a stream
of bytes to the receiver, and reconstructed there as an
item.  The protocol mechanism may thus be viewed as a simple
translator.  It enumerates a set of abstract entities, the
items, which are known to programmers, a set of entities in
byte-stream format, the objects, and the translation rules
for conversion between the sets.  A site implementing the
MSDTP would typically provide a facility to convert between
objects and the local representation of the various items
handled.  Applications using the MSDTP define their
interactions using items, without regard to the actual
formats in which such items are represented at various
machines.  This permits programs to handle higher-level
concepts such as a character string, without concern for its
numerous representational formats.  Such detail is handled
by the MSDTP.

                                -3-


Finally, a discussion of a general data transfer
mechanism for communication between programs is presented,
and the manner in which the particular byte-oriented
protocol defined herein would be used in that environment is
discussed.

Terminology, as introduced, is defined and highlighted
by capitalizing.


IV. PRIMITIVE DATA ITEMS

The primitive data items include a variety of
traditional, well-understood types, such as integers and
strings.  Primitive data items will be presented using
mnemonic names preceded by the character pair "p-", to serve
as a reminder that the named object is primitive.

These items may be represented in various computer
systems in whatever fashion their programmers desire.


IV.1 -- Set Of Primitive Items


The set of primitive items defined includes p-INT,
p-STRING, p-STRUC, p-BITS, p-CHAR, p-BOOL, p-EMPTY, and
p-XTRA.

Since the protocol was developed primarily for use in
message services, items such as p-FLOAT are not included
since they were unnecessary.  Additional items may be easily
added as necessary.

A p-INT performs the traditional role of representing
integer numbers.  A p-BITS (BIT Stream) item represents a
bit stream.  The two possible p-BOOL (BOOLean) items are
used to represent the logical values of *TRUE* and *FALSE*.
The single p-EMPTY item is used to, for example, indicate
that a given field of a message is empty.  It is provided to
act as a place-holder, representing 'no data', and appears
as *EMPTY*.

The p-STRUC (STRUCture) item is used to group together
a collection of items as a single value, maintaining the
ordering of the elements, such as a p-STRUC of p-INTs.

A p-CHAR is a single character.  The most common
occurrence of character data, however, will be as p-STRINGs.
A p-STRING should be considered to be a synonym for a
p-STRUC containing only p-CHARs.  This concept is important
for generality and consistency, especially when considering
definitions of permissible operations on structures, such as
extracting subsequences of elements, etc.

                                   -4-

Four p-XTRA items, which can be transmitted in a single
byte, are made available for higher level protocols to use
when a frequently used datum is handled which can be
represented just by its name.  An example would be an
acknowledgment between two servers.  Using p-XTRAs to
represent such data permits them to be handled in a single
byte.  There are four possible p-XTRA items, termed *XTRA0*,
*XTRA1*, *XTRA2*, and *XTRA3*.  These may be assigned
meanings by user protocols as desired.


IV.2 -- Printing Conventions


The following printing conventions are introduced to
facilitate discussion of the primitive items.

When a specific instance of a primitive data item is
presented, it will be shown in a traditional representation
for that kind of data.  For example, p-INTs are shown as
sequences of digits, e.g. 100, p-STRINGs, as sequences of
characters enclosed in double-quote characters, for example
"ABCDEF".

As shown above, the two possible p-BOOL items are shown
as *TRUE* or *FALSE*.  The object p-EMPTY appears as
*EMPTY*.  A bit stream, i.e. p-BITS, appears as a stream of
1s and 0s enclosed in asterisks, for example *100101001*.  A
p-CHAR will be presented as the character enclosed in single
quote characters, e.g., 'A'.

P-STRUCs are printed as the representations of their
elements, enclosed in parentheses, for example (1 2 3 4) or
("XYZ" "ABC" 1 2) or ((1 2 3) "A" "B"). Note that because
p-STRINGs are simply a class of p-STRUCs assigned a special
name and printing format for brevity and convenience, the
items "ABC" and ('A' 'B' 'C') are identical, and the latter
format should not be used.

To present a generic p-STRUC, as in specifying formats
of the contents of something, the items are presented as a
mnemonic name, optionally followed by a colon and the
permissible types of values for that datum.  When one of
several items may appear as the value for some component,
the permissible ones appear separated by vertical-bar
characters.  For example, p-INT|p-STRING represents a single
item, which may be either a p-INT or a p-STRING.

To represent a succession of items, the Kleene star
convention is used.  The specification p-INT[*] represents
any number of p-INTs.  Similarly, p-INT[3,5] represents from
3 to 5 p-INTs, while p-INT[*,5] specifies up to 5 and
p-iNT[5,*] specifies at least 5 p-INTs.



                                   -5-

For example, a p-STRUC which is used to carry names and
numbers might be specified as follows.

(name:p-STRING number:p-INT)

In discussing items in general, when a specific data
value is not intended, the name and types representation may
be used, e.g., offset:p-INT to discuss an 'offset' which has
a numeric value.


V. SEMANTIC ITEM MECHANISM


The semantic item mechanism provides a means for
program designers to use a variety of application-specific
data items.

This mechanism is implemented using a special tagged
structure to carry the data type information as well as the
actual components of the particular semantic item.  For
discussion purposes.  Such a special p-STRUC will be termed a
p-EDT (Extended Data Type).

When p-EDTs are transferred, their identity as a p-EDT
is maintained.  So that an applications program receives the
corresponding semantic item instead of a simple p-STRUC.  A
p-EDT is identical to a p-STRUC in all other respects.


V.1 -- Format of p-EDTs


A prototypical p-EDT follows.  It is printed as if it
were a normal p-STRUC.  Since p-EDTs are converted to
semantic items for presentation to the user, a p-EDT will
never be used except in this protocol definition.

(type:p-INT|p-STRING version:p-INT com1:any
com2:any ...)

The first element, the 'type' is generally a p-INT, and
is used to identify the particular type of semantic item.
Types are assigned numeric codes in a controlled fashion.
The type may alternatively be specified by a p-STRING, to
permit development of new data types for possible later
assignment of codes.  Each type has an equivalent p-STRING
name.  These may be used interchangeably as 'type' elements,
primarily to maintain upward compatibility.

The second element of a p-EDT is always an p-INT, the
'version', and specifies the exact format of the particular
datum.  A semantic item may undergo several revisions of its
internal structure.  Which would be evident through assigning
different versions to each revision.

                                   -6-

Successive components.  The 'com' elements, if any.
carry the actual data of the semantic item.  As each
semantic item is defined, conventions on permissible values
and interpretation of these components are presented.  Such
definitions may use any types of items to specify the format
of the semantic item.  Use of lower level concepts, such as
objects, in these definitions is prohibited.

Semantic items will be printed as the mnemonic for the
type involved, preceded by the character pair "s-", to
signify that the data item is handled by this mechanism.


V.2 -- Printing Conventions


A semantic item is represented as if it were a p-STRUC
containing only the components, if any, but preceded by the
semantic type name and a # character.  The version number is
assumed to be 1 if unspecified.  For later versions, the
version number is attached to the type name, as in, for
example, FILE-2 to represent version 2 of the FILE data
type.

For example, a semantic item called a 'file
specification' might be defined, containing two components,
a host number and pathname.  A specific instance of such an
item might appear as #FILE(69 "DIRECTORY.NAME-OF-FILE"),
while a generic s-FILE might be presented as the following.

#FILE(host:p-INT|p-STRING pathname:p-STRING)


the item, which may be either a p-INT or p-STRING, and
'pathname' is the second component, which must be a
p-STRING.  The full definition would present interpretation
rules for these components.


VI.  ENCODING OBJECTS


This section presents the set of objects which are used
to represent items as byte streams for inter-server
transmission.  Objects will be presented using mnemonic
type-names preceded by the character pair "b-", indicating
their existence only as byte streams.

All servers are required to be capable of decoding the
entire set of objects.  Servers are not required to transmit
certain objects which are available to improve channel
efficiency.




                               -7-

The encodings are designed to facilitate programming
and efficiency of the receiving decoder.  In all cases, the
type and length in bytes of objects is supplied as the first
information sent.  This characteristic is important for ease
of implementation.  The type information permits a decoder to
be constructed in a modular fashion.  The most important
advantage of including size information is that the receiver
always knows how many bytes it must read to discover what to
do next, and knows when each object terminates.  This
requirement avoids many potential problems with timing and
synchronization of processes.

Two varieties of objects are defined.  The first will
be called ATOMIC, and includes objects used to efficiently
encode the most common data.  The second variety is termed
NON-ATOMIC, and is used to encode larger or less common
items.

In all cases, a data object begins with a single byte,
which will be termed the TYPE-BYTE, a field of which
contains the type code of the object.  The following bytes,
if any, are interpreted according to the type involved.


VI.1 -- Presentation Conventations


In discussing formats of bytes, the following
conventions will be employed.  The individual bits of a byte
will be referenced by using capital letters from A to H,
where A signifies the highest order bit, and H the lowest.
The entire eight bit value, for example, could be referred
to as ABCDEFGH.  Similarly, subfields of the byte will be
identified by such sequences.  The CDEF field specifies the
middle four bits of a byte.

In referring to values of fields, binary format will be
used, and small letters near the end of the alphabet will be
used to identify particular bits for discussion.  For
example, we might say that the BCD field of a byte contains
a specifier for some type, and define its value to be
BCD=11z.  In discussions of the specifier usage, we could
refer to the cases where z=l and where z=0, as shorthand
notation to identify BCD=111 and BCD=110, respectively.


V1.2 -- Type-Byte Bit Assignment


To assist in understanding the assignment of the
various type-byte values, the table and graph below are
included, showing representations of the eight bits.




                               -8-

OXXXXXXX -- CHAR7 (CHARacter, 7 bit)
10XXXXXX -- SINTEGER (Small INTEGER)
l10XXXXX -- NON-ATOM (NON-ATOMic objects)
11100XXX -- LINTEGER (Large INTEGER)
11101XXX -- reserved
11110XXX -- SBITSTR (Short BIT STReam)
111110XX -- XTRA (eXTRA single-byte objects)
1111110X -- BOOL (BOOLean)
11111110 -- EMPTY (EMPTY data item)
11111111 -- PADDING (unused byte)


In each case, the bits identified by X's are used to
contain information specific to the type involved.  These
are explained when each type is defined.

An equivalent tree representation follows, for those
who prefer it.
start with high order bit
 |
 |
 |
 0-----0-----0-----0-----0-----0-----0-----0-----X
 |     |     |     |     |     |     |     |   PADDING
0|    0|    0|    0|    0|    0|    0|    0|
 |     |     |     |     |     |     |     |
 X     |     X     |     X     |     X     X
CHAR7  | NON-ATOM  |    BITS   |   BOOL   EMPTY
 (7)   |   (5)     |    (3)    |   (1)
       |        0| |           |
   SINTEGER        |          XTRA
      (6)          |           (2)
               LINTEGER
                  (3)

        Type-Byte Bit Assignment Scheme




This picture is interpreted by entering at the top, and
taking the appropriate branch at each node to correspond to
the next bit of the type-byte, as it is scanned from left to
right.  When a type is assigned, the branch terminates with
an "X' and the name of the type of the object, with the
number of remaining bits in parentheses.  The individual
object definitions specify how these bits are used for that
particular type.


V1.3 -- Atomic Objects


Atomic objects are identified by specific patterns in a
type-byte.  Receiving servers must be capable of recognizing


                               -9-

and handling all atomic types, since the size of the object
is not explicitly present in a uniform fashion.


================================
| Atomic Object: B-CHAR7       |
================================


The b-CHAR7 (CHARacter 7 bit) object is introduced to
handle transmission of characters, in 7-bit ASCII format.
Since the vast majority of message-related data involves
such objects, they are designed to be very efficient in
transmission.  Other formats, such as eight bit values, can
be introduced as non-atomic objects.  The format of a b-CHAR7
follows:

A=0 identifying the b-CHAR7 data type
BCDEFGH=tuvwxyz containing the character
code

The tuvwxyz objects contain the ASCII code of the
character.  For example, transmission of a "space' (ASCII
code 32, 40 octal) would be accomplished by the following
byte.

00100000
ABCDEFGH

A=0 to identify this byte as a b-CHAR7.  The remaining
bits contain the 7 bit code, octal 40, for space.

A b-CHAR7 standing alone is presented as a p-CHAR.
Such occurrences will probably be rare if they are used at
all.  The most common use of b-CHAR7's is as elements of
b-USTRUCs used to transmit p-STRINGS, as explained later.


=============================
| Atomic Object: B-SINTEGER |
=============================

The b-SINTEGER (Small INTEGER) object is used to
transmit very small positive integers, of values up to 64.
It always translates to an p-INT, and any p-INT between 0
and 63 may be encoded as a b-SINTEGER for transmission.  The
format of an b-SINTEGER follows.

AB=10 identifying the object as a b-SINTEGER
CDEFGH=uvwxyz containing the actual number

For example, to transmit the integer 10 (12 octal), the
following byte would be transmitted:

10001010
ABCDEFGH

                               -10-

AB=10 to specify a b-SINTEGER.  The remaining six bits
contain the number 10 expressed in binary.

=============================
| Atomic Object: B-SINTEGER |
=============================

The b-SINTEGER (Large INTEGER) object is used to
transmit p-INTs to any precision up to 64 bits.  It is
always translated as a p-INT.  Sending servers are permitted
to choose either b-SINTEGER or b-SINTEGER format for
transmission of numbers, as appropriate.  When possible,
b-SINTEGERs can be used for better channel efficiency.  The
format of a b-SINTEGER follows:

ABCDE=11100 specifying that this is a b-SINTEGER.
FGH=xyz containing a count of number of bytes to follow.

The xyz bits are interpreted as a number of bytes to
follow which contain the actual binary code of the the
integer in 2's complement format.  Since a zero-byte integer
is disallowed, the pattern xyz=000 is interpreted as 1000,
specifying that 8 bytes follow.  The number is transmitted
with high-order bits first.  This format permits
transmission of integers as large as 64 bits in magnitude.

For example, if the number 4096 (10000 octal) is to be
transmitted, the following sequence of bytes would be sent:

11100010 00010000 00000000
ABCDEFGH ---actual data---

ABCDE=11100, identifying this as a b-LINTEGER, E=0,
specifying a positive number, and FGH=010, specifying that 2
bytes follow, containing the actual binary number.

============================
| Atomic Object: B-SBITSTR |
============================

The b-SBITSTR (Short BIT STReam) object is used to
transmit a p-BITS of length 63 or less.  For longer bit
streams, the non-atomic object b-LBITSTR may be used.  The
format of a b-SBITSTR follows.

ABCDE=11110 specifying the type as b-SBITSTR
FGH=xyz specifying the number of bytes
following.







                               -11-
The xyz value specifies the number of additional bytes
to be read to obtain the bit stream values.  As in the case
of b-SINTEGER, the value xyz=000 is interpreted as 1000,
specifying that 8 bytes follow.

To avoid requiring specification of exactly the number
of bits contained, the following convention is used.  The
first data byte is scanned from left to right until the
first 1 bit is encountered.  The bit stream is defined to
begin with the immediately following bit, and run through
the last bit of the last byte read.  In other words, the bit
stream is 'right-adjusted' in the collected bytes, with its
left end delimited by the first "on' bit.

For example, to send the bit stream *001010011* (9
bits), the following bytes are transmitted.

11110010 00000010 01010011
ABCDEhij klmnopqr stuvwxyz

The hij=010 value specifies that two bytes follow.  The
q bit, which is the first 1 bit encountered, identifies the
start of the bit stream as being the r bit.  The rstuvwxyz
bits are the bit stream being handled.

=========================
| Atomic Object: b-BOOL |
=========================

The b-BOOL (BOOLean) object is used to transmit
p-BOOLs.  The format of b-BOOL objects follows.

ABCDEFG=1111110 specifying the type as
b-BOOL
H=z specifying the value

The two possible translations of a b-BOOL are *FALSE*
and *TRUE*.

11111100 represents *FALSE*
11111101 represents *TRUE*
ABCDEFGz

if z=0, the value is FALSE, otherwise TRUE.



========================================
| Atomic Object: B-EMPTY |
========================================

The b-EMPTY object type is used to transmit a 'null'
object, i.e. an *EMPTY*.  The format of an b-EMPTY follows.

ABCDEFGH=11111110 specifying *EMPTY*

                                -12-
=========================
| Atomic Object: B-XTRA |
=========================

The b-XTRA objects are used to carry the four possible
p-XTRA items, i.e., *XTRA0*, *XTRA1*, *XTRA2*, and *XTRA3*.
These four items correspond to the binary coding of the
remaining two bits after the b-XTRA type code bits.  The
format of a b-XTRA follows.

ABCDEF=111110 to specify the type b-XTRA
GH=yz to identify the particular p-XTRA item
carried

The GH bits of the byte are decoded to produce a
particular p-XTRA item, as follows.

GH=00 -- *XTRA0*
GH=01 -- *XTRA1*
GH=10 -- *XTRA2*
GH=11 -- *XTRA3*

The b-XTRA object is included to provide the use of
several single-byte data items to higher levels.  These
items may be assigned by individual applications to improve
the efficiency of transmission of several very frequent data
items.  For example, the message services protocols will use
these items to convey positive and negative acknowledgments,
two very common items in every interaction.

========================================
| Atomic Object: B-PADDING
========================================

This object is anomalous, since it represents really no
data at all.  Whenever it is encountered in a byte stream in
a position where a type-byte is expected, it is completely
ignored, and the succeeding byte examined instead.  Its
purpose is to serve as a filler in byte streams, providing
servers with an aid in handling internal problems related to
their specific word lengths, etc.  The encoders may freely
use this object to serve as padding when necessary.

All b-PADDING data objects exist only within an encoded
byte stream.  They never cause any data item whatsoever to
be presented externally to the coder module.  The format of a
b-PADDING follows.

ABCDEFGH=11111111

Note that this does not imply that all such 'null'
bytes in a stream are to be ignored, since they could be
encountered as a byte within some other type, such as
b-LINTEGER.  Only bytes of this format which, by their
position in the stream, appear as a 'type' byte are to be
ignored.

                                -13-
VI.4 -- Non-Atomic Objects


Non-atomic objects are are always transmitted preceded
by both a single type byte and some small number of size
byte(s).  The type byte identifies that the data object
concerned is of a non-atomic type, as well as uniquely
specifying the particular type involved.  All non-atomic
objects have type byte values of the following form.

ABC=110 specifying that the object is
non-atomic
DEFGH=vwxyz specifying the particular type
of object

The vwxyz value is used to specify one of 31 possible
non-atomic types.  The value vwxyz=00000 is reserved for use
in future expansion.

In all non-atomic data objects, the byte(s) following
the type-byte specify the number of bytes to follow which
contain the data object.  In all cases, if the number of
bytes specified are processed, the next byte to be seen
should be another type-byte, the beginning of the next
object in the stream.

The number of bytes containing the object size
information is variable.  These bytes will be termed the
SIZE-BYTES.  The first byte encountered has the following
format.

A=s specifying the manner in which the size
information is encoded
BCDEFGH=tuvwxyz specifying the size, or
number of bytes containing the size

The tuvwxyz values supply a positive binary number.  If
the s value is a one, the tuvwxyz value specifies the number
of bytes to follow which should be read and concatenated as
a binary number, which will then specify the size of the
object.  These bytes will appear with high order bits first.
Thus, if s=1, up to 128 bytes may follow, containing the
count of the succeeding data bytes, which should certainly
be sufficient.

Since many non-atomic objects will be fairly short, the
s=0 condition is used to indicate that the 7 bits contained
in tuvwxyz specify the actual data byte count.  This permits

objects of sizes up to 128 bytes to be specified using one
size-information byte.  The case tuvwxyz=0000000 is
interpreted as specifying 128 bytes.

For example, a data object of some non-atomic type
which requires 100 (144 octal) bytes to be transmitted would
be sent as follows.

                                -14-

110XXXXX -- identifying a specific
non-atomic object
01100100 -- specifying that 100 bytes follow
.
.
data -- the 100 data bytes
.
.

Note that the size count does not include the
size-specifier byte(s) themselves, but does include all
succeeding bytes in the stream used to encode the object.

A data object requiring 20000 (47040 octal) bytes would
appear in the stream as follows.

110XXXXX -- identifying a specific
non-atomic object
10000010 -- specifying that the next 2 bytes
contain the stream length
01001110 -- first byte of number 20000
00100000 -- second byte
.
.
data -- 20,000 bytes
.
.

Interpretation of the contents of the 20000 bytes in
the stream can be performed by a module which knows the
specific format of the non-atomic type specified by DEFGH in
the type-byte.

The remainder of this section defines an initial set of
non-atomic types, the format of their encoding, and the
semantics of their interpretation.


================================
| Non-atomic Object: B-LBITSTR |
================================

The b-LBITSTR (Long BIT Stream) data type is introduced
to transmit p-BITS which cannot be handled by a b-SBITSTR.
A b-LBITSTR may be used to transmit short p-BITS as well.
Its format follows.










                                -15-

11000001 size-bytes data-bytes
ABCDEFGH

ABC=110 identifies this as a non-atomic object.
DEFGH=00001 specifies that it is a b-LBITSTR.  The standard
sizing information specifies the number of succeeding bytes.
Within the data-bytes, the first object encountered must
decode to a p-INT.  This number conveys the length of the
bit stream to follow.  The actual bit stream begins with the
next byte, and is left-adjusted in the byte stream.  For
example to encode *101010101010*, the following b-LBITSTR
could be used, although a b-SBITSTR would be more compact.

11000001 -- identifies a b-LBITSTR
00000010 -- b-SINTEGER, to specify length
10001100 -- size = 2
10101010 -- first 8 data bits
10100000 -- last 4 data bits



==============================
| Non-atomic Object: B-STRUC |
==============================

The b-STRUC (STRUCture) data type is used to transmit
any p-STRUC.  The translation rules for converting a b-STRUC
into a primitive item are presented following the discussion
of b-REPEATs.  The b-STRUC format appears as follows.

11000010 size-bytes data-bytes
ABCDEFGH

ABC=110 identifies this as a non-atomic type.
DEFGH=00010 specifies that the object is a b-STRUC.  Within
the data-bytes stream, objects simply follow in order.  This
implies that the b-STRUC encoder and decoder modules can
simply make use of recursive calls to a standard
encoder/decoder for processing each element of the b-STRUC.

Note that any type of object is permitted as an element of a
b-STRUC, but the size information of the b-STRUC must
include all bytes used to represent the elements.

Containment of b-STRUCs within other b-STRUCs is
permitted to any reasonable level.  That is, a b-STRUC may
contain as an element another b-STRUC, which contains
another b-STRUC, and so on.  All servers are requires to
handle such containment to at least a minimum depth of
three.

Examples of encoded structures appear in a later
section.


                                -16-
============================
| Non-atomic Object: B-EDT |
============================

A b-EDT is the object used as the carrier for p-EDTs in
transmission of semantic items.  It is functionally
identical to a b-STRUC, but has a different type code to
permit it to be identified and converted to a semantic item
instead of a p-STRUC.  The format of a b-EDT follows.

11000011 size-bytes data-bytes
ABCDEFGH

As with all non-atomic types, ABC=110 to identify this
as such, and DEFGH=00011 to specify a b-EDT.  The objects in
the data-bytes are decoded as for b-STRUCs.  However, the
first object must decode to a p-iNT or p-STRING and the
second to a p-INT, to conform to the format of p-EDTs.



===============================
| Non-atomic Object: b-REPEAT |
===============================


The b-REPEAT object is never translated directly into
an item.  It is legal only as an component of an enclosing
b-STRUC, b-USTRUC, b-EDT, or b-REPEAT.  A b-REPEAT is used to
concisely specify a set of elements to be treated as if they
appeared in the enclosing structure in place of the
b-REPEAT.  This provides a mechanism for encoding a sequence
of identical data items or patterns efficiently for
transmission.

A common example of this would be in transmission of
text, where line images containing long sequences of spaces,
or pages containing multiple carriage-return, line-feed
pairs, are often encountered.  Such sequences could be
encoded as an appropriate b-REPEAT to compact the data for
transmission.  The format of a b-REPEAT is as follows.

11000100   -- identifyIng the object as a
                b-REPEAT
size-bytes -- the standard non-atomic object
                size information
countspec  -- an object which translates to a p-INT
.
.
data -- the objects which define the pattern
.
.

The 'countspec' object must translate to an p-INT to
specify the number of times that the following data pattern
should be repeated in the object enclosing the b-REPEAT.

                                -17-

The remaining objects in the b-REPEAT constitute the
data pattern which is to be repeated.  The decoding of the
enclosing structure will be continued as if the data pattern
objects appeared 'countspec' times in place of the b-REPEAT.
Zero repeat counts are permitted, for generality.  They
cause no objects to be simulated in the enclosing structure.

An encoder does not have to use b-REPEATs at all, if
simplicity of coding outweighs the benefits of data
compression.  In message services, for example, an encoder
might limIt itself to only compressing long text strings.  It
is important for compatibility, however, to have the ability
in the decoders to handle b-REPEATs.

===============================
| Non-atomic Object: B-USTRUC |
===============================

The b-USTRUC (Uniform Structure) object type is
provided to enable servers to convey the fact that a p-STRUC
being transferred contains items of only a single type.  The
most common example would involve a b-USTRUC which
translates to a p-STRUC of only p-CHARs, and hence may be
considered to be a p-STRING.  Servers may use this
information to assist them in decoding objects efficiently.
No server is required to generate b-USTRUCs.

The internal construction of a b-USTRUC is identical to
that of a b-STRUC, except for the type-byte.  The format of a
b-USTRUC follows.

11000101 size-bytes data-bytes
ABCDEFGH

ABC=110 to identify a non-atomic object.  DEFGH=00101
specifies the object as a b-USTRUC.

===============================
| Non-atomic Object: B-STRING |
===============================

The b-STRING object is included to permit explicit
specification of a structure as a p-STRING.  This
information will permit receiving servers to process the
incoming structure more efficiently.  A b-STRING is
formatted similarity to a b-USTRUC, except that its type-byte
identifies the object as a b-STRI/NG.  The normal sizing
information is followed by a stream of bytes which are
interpreted as b-CHAR7s, Ignoring the high-order bit.  The
format of a b-STRING follows.

11000110 size-bytes data-bytes
ABCDEFGH

ABC=110 to identify a non-atomic object.  DEFGH=00110
specifies the object as a b-STRING.

                                -18-

VI.5 -- Structure Translation Rules


A b-STRUC is translated into a p-STRUC.  This is
performed by translating each object of the b-STRUC Into its
corresponding item, and saving it for inclusion In the
p-STRUC being generated.  A b-USTRUC is handled similarly,
but the coding programs may utilize the information that the
resultant p-STRUC will contain items of uniform type.  The
preferred method of coding p-STRINGS is to use b-USTRUCs.

If all of the elements of the resultant p-STRUC are
p-CHARs, it is presented to the user of the decoder as a
p-STRING.  A p-STRING should be considered to be a synonym
for a p-STRUC containing only characters.  It need not
necessarily exist at particular sites which would present
p-STRUCs of p-CHARs to their application programs

The object b-REPEAT is handled in a special fashion
when encountered as an element.  When this occurs, the data
pattern of the b-REPEAT is translated into a sequence of
items, and that sequence is repeated in the next higher
level as many times as specified in the b-REPEAT.
Therefore, b-REPEATS are legal only as elements of a
surrounding b-STRUC, b-USTRUC, b-EDT, or b-REPEAT.

In encoding a p-STRUC or p-STRING for transmission, a
translator may use b-REPEATs as desired to effect data
compression, but their use is not mandatory.  Similarly,
b-STRINGS may be used, but are not mandatory.

A b-EDT is translated into a p-EDT to identify it as a
carrier for a semantic item.  Otherwise, it is treated
identically to a b-STRUC.


VI.6 -- Translation Summary


The following table summarizes the possible
translations between primitive items and objects.

p-INT    <--> b-LINTEGER, b-SINTEGER
p-STRING <--> b-STRING, b-STRUC, b-USTRUC
p-STRUC  <--> b-STRING, b-STRUC, b-USTRUC
p-BITS   <--> b=SBITSTR, b-LBITSTR
p-CHAR   <--> b-CHAR7
p-BOOL   <--> b-BOOL
p-EMPTY  <--> b=EMPTY
p-XTRA   <--> b-XTRA
p-EDT    <--> b-EDT (all semantic items)
-none-   <--> b-PADDING
-none-   <--> b-REPEAT (only within structure)

Note that all semantic items are represented as p-EDTs
which always exist as b-EDTs in byte-stream format.

                                -19-
V1.7 -- Structure Coding Examples


The following stream transmits a b-STRUC containing 3
b-SINTEGERs, with values 1, 2, and 3, representing a p-STRUC
containing three p-INTs, i.e. (1 2 3).

11000010 -- b-STRUC
00000011 -- size=3
10000001 -- b-SINTEGER=1
10000010 -- b-SINTEGER=2
10000011 -- b-SINTEGER=3

The next example represents a b-STRUC containing the
characters X and Y, followed by the b-LINTEGER 10,
representing a p-STRUC of 2 p-CHARs and a p-INT, i.e., ('X'
'Y' 10).  Note that the p-INT prevents considering this a
p-STRING.

11000010 -- b-STRUC
00000100 -- size=4
01011000 -- b-CHAR7 'X'
01011001 -- b-CHAR7 'Y'
11100001 -- b-LINTEGER
00001010 -- 10

Note that a better way to send this p-STRUC would be to
represent the integer as a b-SINTEGER, as shown below.

11000010 -- b-STRUC
00000011 -- size=3
01011000 -- b-CHAR7 'X'
01011001 -- b-CHAR7 'Y'
10001010 -- b-SINTEGER=10

The next example shows a b-STRUC of b-CHAR7s.  It is
the translation of the b-STRING "HELLO".

11000010 -- b-STRUC
00000101 -- size=5
01001000 -- b-CHAR7 'H'
01000101 -- b-CHAR7 'E'
01001100 -- b-CHAR7 'L'
01001100 -- b-CHAR7 'L'
01001111 -- b-CHAR7 'O'

This datum could also be transmitted as a b-STRING.
Note that the character bytes are not necessarily b-CHAR7s,
since the high-order bit is ignored.

11000110 -- b-STRING
00000101 -- size=5
01001000 -- 'H'
01000101 -- 'E'
01001100 -- 'L'
01001100 -- 'L'
01001111 -- 'O'

                                -20-
To encode a p-STRING containing 20 carriage-return
line-feed pairs, the following b-STRUC containing a b-REPEAT
could be used.

11000010 -- b-STRUC
00000101 -- size=5
11000100 -- b-REPEAT
00000011 -- size=3
10010100 -- count, b-SINTEGER=20
00001101 -- b-CHAR7, "CR'
00001010 -- b-CHAR7, 'IF'

To encode a p-STRUC of p-INTs, where the sequence
contains a sequence of thirty 0's preceded by a single 1,
the following b-STRUC could be used.

11000010 -- b-STRUC
00000110 -- size=6
10000001 -- b-SINTEGER=1
11000100 -- b-REPEAT
00000010 -- size=2
10011110 -- count, b-SINTEGER=30
10000000 -- b-SINTEGER=0


VII. A GENERAL DATA TRANSFER SCHEME


This section considers a possible scheme for extending
the concept of a data translator into an multi-purpose data
transfer mechanism.

The proposed environment would provide a set of
primitive items, including those enumerated herein but
extended as necessary to accommodate a variety of
applications.  Communication between processes would be
defined solely in terms of these items, and would
specifically avoid any consideration of the actual formats
in which the data is transferred.

A repertoire of translators would be provided, one of
which is the MSDTP machinery, for use in converting items to
any of a number of transmission formats.  Borrowing a
concept from radio terminology, each translator would be
analogous to a different type of modulation scheme, to be
used to transfer data through some communications medium.
Such media could be an eight-bit byte-oriented connection,
36-bit connection, etc.  and conceivably have other
distinguishing features, such as bandwidth, cost, and delay.
For each media which a site supports, it would provide its
programmers with a module for performing the translations
required.




                                -21-

Certain media or translators might not handle various
items.  For example, the MSDTP does not handle items which
might be termed p-FLOATs, p-COMPLEXs, p-ARRAY, and so on.  In
addition, the efficiency of various media for transfer of
specific items may differ drastically.  MSDTP, for example,
transfers data frequently used in message handling very
efficiently, but is relatively poor at transfer of very
large or deep tree structures.

Available at each site as a process or subroutine
package wouLd be a module responsible for interfacing with
its counterpart at the other end of the media.  These
modules would use a protocol, not yet defined, to match
their capabilities, and choose a particular media and
translator, when more than one exists, for transfer of data
items.

Such a facility could totally insulate applications
from need to consider encoding formats, machine differences,
and so on, as well as eliminate duplication of effort in
producing such facilities for every new project which
requires them.  In addition, as new translators or media are
introduced, they would become immediately available to
existing users without reprogramming.

Implementation of such a protocol should not be very
difficult or time-consuming, since it need not be very
sophisticated in choosing the most appropriate transfer
mechanism in initial implementations.  The system is
inherently upward-compatible and easily expandable.























                                -22-