Vint Cerf - UCLA
Eric Harslem - Rand
RFC 194 John Heafner - Rand
NIC 7139
Category: D.4 Bob Metcalfe - MIT
Updates: None
Obsoletes: None Jim White - UCSB
THE DATA RECONFIGURATION SERVICE --
COMPILER/INTERPRETER IMPLEMENTATION NOTES
I. NEW FEATURES OF THE LANGUAGE
1. The meaning of S(#,E,,l) is only find an arbitrary
number (<=256) of EBCDIC characters and store them in
identifier S. This descriptor is terminated only by
an invalid EBCDIC or by exceeding maximum permissible
character count (256).
2. The assignment (S .<=. T) causes all attributes of
identifier T to be given to S, i.e., length, type,
and contents.
3. (S .<=. T || X) concatenates X onto the right-hand
side of T and stores the result in S. If T and X
are binary the resulting value has a length equal
to the sum L(T) + L(X).
4. T(X) joins L(X) and V(X) as a built-in identifier
function.
T(X) = type of identifier X.
L(X) = length of contents of X.
V(X) = contents of X converted to binary
(decimal - binary is presently the only
transformation).
5. New types ED and AD are EBCDIC and ASCII encoded
decimal, respectively. These have been added to
complement the V(X) function.
6. New type SB has been added as signed binary. Type B
is a logical binary string.
7. The syntactic notation for return-from-a-form has
been changed. See new syntax.
II. NEW SYNTAX
form :: = rule | form
rule :: = label inputstream outputstream;
label :: = INTEGER | NULL
inputstream :: = terms | NULL
terms :: = term | terms, term
outputstream :: = :terms | NULL
term :: = identifier | identifier descriptor |
descriptor | comparator
identifier :: = <alpha followed by 0-3 alphanumerics>
descriptor :: = (replicationexpr, datatype, valueexpr,
lengthexpr control)
comparator :: = (concatexpr connective concatexpr control) |
(identifier .<=. concatexpr control)
replicationexpr :: = # | arithmetricexpr | NULL
datatype :: = B | O | X | E | A | ED | AD | SB | T (identifier)
valueexpr :: = concatexpr | NULL
lengthexpr :: = arithmeticexpr | NULL
connective :: = .LE. | .LT. | .GT. | .GE. | .EQ. | .NE.
concatexpr :: = value | concatexpr value
value :: = literal | arithmeticexpr
arithmeticexpr :: = primary | arithmeticexpr operator primary
primary :: = identifier | L(identifier) | V(identifier) |
INTEGER
operator :: = + | - | * | /
literal :: = literaltype "string"
literaltype :: = B | 0 | X | E | A | ED | AD | SB
string :: = <from 0 to 256 chars>
control :: = :options | NULL
options :: = SFUR (arithmeticexpr) | SFUR (arithmeticexpr),
SFUR (arithmeticexpr)
SFUR :: = S | F | U | SR | FR | UR
III. THE FORM INTERPRETER
Interpreter Overview
The interpreter is a simple minded machine having the virtue of
helping the compiler writer by providing a rather powerful instruction
set for hard-to-compile operations. Figure 1 shows the machine
configuration:
+-------------+ +--------------+
| inputstream | | outputstream |
+-------------+ +--------------+
/\ /
\ /
\ /
\ \/
+-----------------------+
| CPU |
+-----------------------+
| /\
| |
| |
\/ |
+-----------------------+
Storage: | Instruction |
| Sequence |
+-----------------------+
| Label Table |
+-----------------------+
| Literal/Identifier |
| Pool |
+-----------------------+
| Variable length |
| string area |
+-----------------------+
Fig. 1. Form Interpreter
The CPU is a box full of miscellaneous parts, the most important
being the Arithmetic Logic Unit and the instruction decoding unit. The
CPU also maintains a collection of state registers to keep track of what
it is doing. Figure 2 shows the rough layout.
+-----------------+ +---------------+
| Instruction | | Instruction |
| Counter | | Register |
+-----------------+ +---------------+
|
|
V
+----------------+
| Operation Code |
| Decoding |
Run Time Stack +----------------+
+------------------+ / | \
| Operands | / | \
+------------------+ \/ V \/
| | +-----------------+
+------------------+ / Instruction \
| | | Interpreter |
+------------------+ | Routines |
| | \ /
+------------------+ +---------------+
| | | /\
+------------------+ | |
| | | |
+------------------+ V |
| | +---------------+
+------------------+ <------------- | Arithmetic |
| | -------------> | Logic Unit |
+------------------+ +---------------+
| |
+------------------+
| |
+------------------+
+------------------+ +------------------+
|Initial Input Ptr.| | Output pointer |
+------------------+ +------------------+
+------------------+ +------------------+
|Current Input Ptr.| | True/False Flag |
+------------------+ +------------------+
Fig. 2. The Central Processor
The CPU is a stack machine driven by a Polish postfix instruction
sequence. Operands placed on the Run Time Stack are used for arithmetic
expression evaluation and for parameter passing between the interpreter
and the built-in functions.
The Current Input Pointer and the Output Pointer keep track of the
two data streams. Two input pointers are needed because of the backup
requirement in the event of rule failure. All of these pointers are bit
pointers into the two streams.
Various implementations of the Run Time Stack are independent of
the interpretation of the DRS machine's instruction set. It is
suggested that the stack will contain instruction operands from the
instruction stream.
The format of a compiled instruction sequence for a form is shown
in Fig. 3.
16 bits
+--------/\---------+
/ \
+---------------------+
| length n in bytes |
+-- +---------------------+
| | |
| | compiled |
| | 16-bit |
n < | instructions |
| | |
| | |
| | |
+-- +---------------------+
Fig. 3. Compiled Instruction Sequence Format
The format of the compiled Label Table is shown in Fig. 4.
16 bits
+-----/\-------+
/ \
+-----------------+
| length n |
| in bytes |
+-- +------------------+-----------------+
| | numeric value of | byte offset |
| | statement number | in inst. seq. |
| +------------------+-----------------+
| | : : |
n < | : : |
| | : : |
| | |
| | |
| | |
+-- +------------------------------------+
\_________________ _________________/
V
32 bits
Fig. 4. Compiled Label Table
Literals and Identifiers are compiled as shown in fig. 5.
2 2
+----/\----+ +----/\----+
/ \ / \
+-------------+--------------+
1 1 | length n | length n |
___/\____ ___/\____ | in bytes | in bytes |
+---------+----------+-------------+--------------+
/ | |//////////| | |
| | Type |//////////| bit length | byte offset |
| | |//////////| | |
| +---------+----------+-------------+--------------+
5*n < | : |
| | : |
| | : | Identifiers
| | |
\ | |
+-------------------------------------------------+
/ | |
| | literals are |
| | byte-aligned | Literals
m < | |
| | |
| | |
\ +-------------------------------------------------+
Legend:
Type 0 = undefined
1 = B (binary)
2 = 0 (octal)
3 = X (hexadecimal)
4 = E (EBCDIC)
5 = A (ASCII)
6 = ED (EBCDIC encoded decimal)
7 = AD (ASCII encoded decimal)
8 = SB (signed binary, two's complement)
Fig. 5. Compiled Literals and Identifiers
Types B, 0, X, AD, ED, and SB point to 32-bit word- aligned data
shown below.
+---+---+-----+-------+ +-------------------+ word-aligned,
| T |///| L | ---+-----> | | 32-bit right-
+---+---+-----+-------+ +-------------------+ justified
Types E and A point to byte-aligned symbol streams
as shown below.
byte-aligned, L <= 256
+---+---+-----+-------+ +------------------------+
| T |///| L | ---+-----> | |
+---+---+-----+-------+ +------------------------+
Instruction Format
Since literals and identifiers will be stored in the same data
area, more than 256 literals plus identifiers might be encountered so
more than 8 bits are needed to reference literal/id pool. Furthermore,
such references must be distinguished from operators in the instruction
stream, so a 16-bit instruction will be used, as shown below.
+--------+------------------------+
| 4 | 12 |
+--------+------------------------+
|
/
/
/
|
V
LD = 0 literal or identifier reference (12-bit positive integer)
IC = 1 12-bit two's complement integer constant
OP = 2 operator
AD = 3 address (12-bit positive integer)
ARB = 4 indefinite replication factor
NULL = 5 missing attribute of term
The operation code decoder picks up types 0, 1, 3, 4,
and 5 and deposits them on top of the stack (TOS). LD is an
index into the literal/identifier table, and AD is an index
into the instruction sequence.
The decoder examines OP elements further:
4 4 8
+--------+--------+----------------+
| 0010 | |////////////////|
+--------+--------+----------------+
OP |
+----------> 0 = binary operator
1 = unary operator
2 = special operator
Binary Operators (*)
Let the TOS contain y and the next level, x. The binary operators
compute x <bop> y, popping both x, y from stack, and put the result
back on top of the stack.
+---+ <-- TOS +-----+ <-- TOS
| y | | x-y |
e.g. x-y => +---+ ===> +-----+
| x | |/////|
+---+ +-----+
Binary Operator Encoding
4 4 4 4
+--------+--------+--------+--------+
| 0010 | 0000 | |////////|
+--------+--------+--------+--------+
|
+--------------------------+
|
V
0 = integer +
1 = integer -
2 = integer x
3 = integer : (or /), no remainder
4 = concatenate ||
All binary operations except concatenate expect the top
two elements on the stack to describe type B, 0, X, or SB. The
result is always a 32-bit type B element. The concatenate
operator fails unless both types are identical. For example:
-------
(*) As suggested above, the stack really contains instruction
operands that describe data; for convenience in illustrations
the data rather than their descriptors are shown on the stack.
type L value T L V
+------+------+------+ +------+------+------+
TOS -> | B | 32 | 4 | | B | 32 | 12 | <- TOS
+------+------+------+ ==> +------+------+------+
| B | 8 | 16 | |//////|//////|//////|
+------+------+------+ +------+------+------+
Before-operation after-operation
+------+------+------+ +------+------+------+
TOS -> | A | 2 | DE | | A | 5 |ABCDE | <- TOS
+------+------+------+ ==> +------+------+------+
| A | 3 | ABC | |//////|//////|//////|
+------+------+------+ +------+------+------+
Before || operation after || operation
No binary operator has any effect on the TRUE/FALSE flag.
Unary Operators
4 4 4 4
+--------+--------+--------+--------+
| 0010 | 0001 | | |
+--------+--------+--------+--------+
| |
+--------------+ |
| |
V |
0 = integer minus V
1 = load identifier 0 = evaluated contents
(after dec - binary
conversion)
1 = length field
2 = type field
2 = Label Table Reference
For the unary minus operator the data described by the top of the
stack is replaced with its 2's complement. The form fails if the TOS
type is not SB, B, 0, or X.
The Load identifier expects the TOS to describe an index into the
literal/identifier pool (that is, an LD instruction) . The TOS
described data is replaced by 32-bit type B values. The operation fails
if the contents cannot be converted from encoded decimal to binary. B,
0, and X types are treated as unsigned integers, SB is treated as 2's
complement.
The Label Table Reference operator expects a 32-bit type B value
described by TOS and searches for this label in the label Table. If
found, the TOS described data is replaced by the relative address in the
instruction sequence of the label (in the form of an AD instruction).
If not found, the form fails. No Unary operator has any effect on the
TRUE/FALSE flag.
Special Operators
4 4 4 4
+--------+--------+--------+--------+
| 0010 | 0010 | | |
+--------+--------+--------+--------+
| |
+-----------------------+ /
| /
V /
0 = store TOS |
1 = return V
2 = branch 0 = true, 1 = false, 2 = unconditional
3 = compare 0 = .EQ. 2 = .LE. 4 = .GE.
1 = .NE. 3 = .LT. 5 = .GT.
4 = move input ptr 0 = store current into initial
1 = store initial into current
5 = input call 0 = no compare
1 = compare
6 = output call
Store TOS
The TOS describes an index into the ID table and the next lower
element in the stack describes a value to be stored. After execution,
both elements are popped off the stack.
Return
The TOS describes a value to be returned to the routine which
initiated the FORM MACHINE. The actual mechanism will be implementation
dependent, but the FORM MACHINE will relin- quish control after this
instruction completes execution.
Branch
The TOS describes an index into the instruction sequence to be used
as the new instruction counter (IC) if the branch conditions are
satisfied. The branch instruction checks the state of the TRUE/FALSE
flag register and either increments the IC by 1 or replaces it with the
TOS described element. In any case, the TOS is popped.
Compare
The compare operator takes the two elements described by the two
top stack entries and compares them (.EQ.,.LT.,etc.). If n is at the
top of the stack, and m is just below, then m.xx.n is performed, and the
TRUE/False flag is set accordingly. For .xx. = .EQ. or .NE. we must
have identical type, length, and content for equality to hold.
The other boolean comparators will not be attempted if types are
different (i.e., form fails), but for same types, B, 0, X cause binary-
justified compares, and A, E, AD, ED cause left-justified string
compares with the shorter string padded with blanks.
Move Input Pointer
This operator (no operands) replaces the Current Input Pointer with
the Initial Input Pointer (back-up), or the Initial Input Pointer with
the current one (entry to rule).
Input Call
This is the most complex operator thus far encountered. It requires
four operands from the run-time stack:
TOS +----------------------------+
| binary or null | length to find
+----------------------------+
| LD to literal or null | value (literal)
+----------------------------+
| binary code | input data type
+----------------------------+
| binary, arbitrary, or null | replication count
+----------------------------+
The input call operator can be invoked with the "no compare" flag
set, in which case the value expression parameter is ignored and only
the input type and length expressions are used. In this case, the input
routine tries to find in the input stream as many characters of the
required type (bits, digits, etc.) as needed to fill the length
expression requirement. If successful, the TRUE/FALSE flag is set TRUE,
the stack is popped to remove the input parameters, and the string
obtained is described by the TOS. If the input stream cannot be matched
then the parameters are popped off the stack, and the TRUE/FALSE flag is
set FALSE.
If the "compare" flag is set, the input stream must be searched for
the value expression. However, we must take some care here to be sure
we know what to look for. There are several cases:
a) The length expression parameter is greater than the
length of the value expression but the type of input de-
sired is the same as the value expression type. For B, 0
and X types, right-justify value expression in length-
expression field, sign bit is extended left if type BS.
If type A, E, AD, or ED pad on the right with blanks. b) Same as
a) but length is too small. B, 0, and X type strings
are truncated on the left. A, E, AD and ED are truncated
on the right. c) The type of the value expression and the type
parameter
differ. This case is deferred for discussion and pre-
sently is considered an error causing form failure.
If the input string matches, then the TRUE/FALSE flag is set true,
the parameters are popped from the stack, and the resulting string is
described by the TOS. Otherwise, the FALSE flag is set and the
parameters are popped.
When a successful match is found the input subroutine always
advances the Current Input Pointer by the appropriate amount. Since we
are dealing at the bit level this pointer must be maintained as a bit
pointer!
Output Call
This routine utilizes the same parameters as the input call, but
operates on the output stream. The TRUE/FALSE flag is not distributed
by this operator. As for input, there are four parameters on top of the
stack, the length expression value, the value expression value, the
desired output type, and the replication expression value. When there
is a mis- match between the output type and the value expression type, a
conversion must take place. The value expression is trans- formed into
the desired output type and fitted into the field length specified by
the length expression.
Truncation and Padding Rules
a) Character -> character (A,E,AD,ED -> A,E,AD,ED) conversion
is left-justified and truncated or padded with blanks
on the right. b) Character -> numeric and numeric -> character
conversion is
right-justified and truncated or padded on the left with
zeros. Beware! Two's complement numbers may be bollixed
by this. c) Numeric -> character conversion is right-justified and
left padded with blanks or left-truncated. As for the unary
operators, a numeric bit-string is treated as unsigned, except SB which
is treated as two's complement. Thus we have:
(1,ED,X"FF",3) = E'255'
(1,ED,X"100",3) = E'256'
but (1,ED,SB"10000000",4) = E'-256'
If the output routine is able to perform the desired action, it
advances the Output Stream Pointer, and pops all parameters from the
run-time stack.
V. INSTRUCTION SET
it/id ref LD <num> Literal or identifier
reference -> TOS
int const IC <num> small 2's comp. integer
constant -> TOS
address AD <num> Address -> TOS
null parameter NULL missing term attribute
add ADD TOS = x,y x + y -> TOS
subtract SUB TOS = x,y x - y -> TOS
multiply MUL TOS = x,y x * y -> TOS
divide DIV TOS = x,y x/y -> TOS
concatenate CON TOS = x,y x||y -> TOS
unary minus UNIN TOS = x -x -> TOS
load id value LIV TOS = LD x V(LD x) -> TOS
load id length LIL TOS = LD x V(LD x) -> TOS
load id type LIT TOS = LD x V(LD x) -> TOS
look up label LVL TOS = x AD x -> TOS
sto STO TOS = x,y y -> x
return RET TOS = x return to
caller with x
branch true BT TOS = AD x AD x -> Instr.
counter
branch false BF TOS = AD x AD x -> Instr.
counter
branch BU TOS = AD x AD x -> Instr.
counter
compare equal CEQ TOS = x,y (y.EQ.x) ->
TRUE/FALSE
flag
compare not equal CNE TOS = x,y (y.NE.x) -> T/FF
compare <= CLE TOS = x,y (y.LE.x) -> T/FF
call output OUT TOS = r,t,v,l (r,t,v,l) -> output
call input IN ( INC = compare TOS = r,t,v,l (r,t,v,l) -> TOS
INN = no compare )
current -> initial SCIP CIP -> IIP (store current input
ptr - initial IP)
initial -> current SICP IIP -> CIP (store initial input
ptr - CIP)
VI. EXAMPLE COMPILATION
FORM SOURCE GENERATED POLISH INSTRUCTION SEQUENCE
ADDR. INSTR. COMMENTS
(NUMB.<=.1); 0 SICP RULE PRELUDE
1 IC 1
2 LD 0 REFERENCE TO NUMB
3 STO STORE IN NUMB
4 SCIP RULE POSTLUDE
1 CC(,E,,1:FR(99)), 5 SICP RULE PRELUDE
6 NULL NO REPLICATION EXPRESSION
7 IC 4 TYPE EBCDIC
8 NULL NO VALUE EXPRESSION
9 IC 1 LENGTH
10 INN INPUT CALL WITH NO COMPARE
11 AD 15
12 BT SKIP RETURN IF INN SUCCEEDS
13 IC 99 RETURN CODE
14 RET RETURN TO CALLER IF FAILED
15 LD 1 REFERENCE TO CC
16 STO STORE INPUT DATA IN CC
LINE(,E,,121: 17 NULL NO REPLICATION EXPRESSION
FR(99)), 18 IC 4 TYPE IS EBCDIC
19 NULL NO VALUE EXPRESSION
20 IC 121 LENGTH
21 INN INPUT WITH NO COMPARE
22 AD 26
23 BT SKIP RETURN IF OK
24 IC 98 RETURN CODE
25 RET RETURN TO CALLER IF FAILED
26 LD 2 REFERENCE TO LINE
27 STO STORE INPUT IN LINE
:CC, 28 SCIP SUCCESSFUL INPUT
29 NULL NO REPLICATION FACTOR
30 LD 1 REFERENCE TO CC
31 LIT TYPE OF CC
32 LD 1 REFERENCE TO VALUE OF CC
33 LD 1 CC AGAIN
34 LIL LENGTH OF CC
35 OUT OUTPUT CC
(,ED,NUMB,2), 36 NULL NO REPLICATION
37 IC 6 TYPE IS ED
38 LD 0 REFERENCE TO VALUE OF NUMB
39 IC 2 LENGTH OF OUTPUT FIELD
40 OUT OUTPUT NUMB AS EBCDIC DEC.
(,E,E".",1), 41 NULL NO REPLICATION
42 IC 4 TYPE IS EBCDIC
43 LD 3 REFERENCE TO E"."
44 IC 1 LENGTH TO OUTPUT
45 OUT OUTPUT THE PERIOD
(,E,LINE,117), 46 NULL NO REPLICATION
47 IC 4 TYPE IS EBCDIC
48 LD 2 REFERENCE TO LINE
49 IC 117 LENGTH TO OUTPUT
50 OUT PUT OUT CONTENTS OF LINE
(NUMB.<=.NUMB+1: 51 LD 0 REFERENCE TO NUMB
U(1)); 52 IC 1 AMOUNT TO ADD
53 ADD ADD TO NUMB
54 LD 0 REFERENCE TO NUMB
55 STO STORE BACK INTO NUMB
56 AD 5 PLACE TO GO
57 B UNCONDITIONAL BRANCH BACK
LITERAL/IDENTIFIER TABLE
0 NUMB
1 CC
2 LINE
3 E"."
LABEL TABLE
LABEL OFFSET
1 5
[ This RFC was put into machine readable form for entry ]
[ into the online RFC archives by Simone Demmel 6/97 ]
|
Comment about this RFC, ask questions, or add new information about this topic: