ALGOL 68-R was the first implementation of the Algorithmic language ALGOL 68.
In December 1968 the report on the Algorithmic language ALGOL 68 was published. On 20–24 July 1970 a working conference was arranged by the IFIP to discuss the problems of implementation of the language,[1] a small team from the Royal Radar Establishment attended to present their compiler, written by I.F. Currie, Susan G. Bond and J.D. Morrison. In the face of estimates of up to 100 man-years to implement the language, using up to 7 pass compilers they described how they had already implemented a one-pass compiler which was in production use in engineering and scientific applications.
Contents |
The ALGOL 68-R compiler was initially written in a local dialect of ALGOL 60 with extensions for address manipulation and list processing. The parser was written using J.M. Foster's Syntax Improving Device (SID) parser generator.
| “ | About 20K of this is program, which we feel is too large. -- Currie[2] |
” |
The first version of the compiler occupied 34K words. It was later rewritten in ALGOL 68-R,[3] taking around 36K words to compile most programs.[4]
ALGOL 68-R was implemented under the GEORGE 3 operating system on an ICL 1907F. The compiler was distributed without charge by ICL on behalf of the RRE.
| “ | It is a question of morality. We have a Bible and you are sinning! |
” |
In order to allow one pass compilation ALGOL 68-R implemented a sub-set of the language defined in the original report:[6]
Many of these restrictions were adopted by the revised report on ALGOL 68.
To allow compiling in one pass ALGOL 68-R insisted that all identifiers were specified (declared) before use.
The standard program:
proc even = (int number) bool: ( number = 0 | true | odd (abs (number - 1))); proc odd = (int number) bool: ( number = 0 | false | even (abs (number - 1)));
would have to be re-written as:
proc (int) bool odd; proc even = (int number) bool : ( number = 0 | true | odd (abs (number - 1))); odd := (int number) bool : ( number = 0 | false | even (abs (number - 1)));
To allow recursive declarations of modes (types) a special stub mode declaration was used to inform the compiler that an up-coming symbol was a mode rather than an operator:
mode b; mode a = struct (ref b b); mode b = [1:10] ref a;
In the standard language the proceduring coercion could, in a strong context, convert an expression of some type into a procedure returning that type. This could be used to implement call by name.
Another case where proceduring was used was the declaration of procedures, in the declaration:
proc x plus 1 = int : x + 1;
the right hand side was a cast of x + 1 to integer, which was then converted to procedure returning integer.
The ALGOL 68-R team found this too difficult to handle and made two changes to the language. The proceduring coercion was simply dropped and the form mode : expression was redefined as a procedure denotation, casts being indicated by an explicit val symbol:
real : x co a cast to real in ALGOL 68 co real val x co a cast to real in ALGOL 68-R co
Code that had a valid use for call by name (For example Jensen's device) could simply pass a procedure denotation:
proc sum = (int lo, hi, proc (int) real term) real :
begin
real temp := 0;
for i from lo to hi do
temp +:= term (i);
temp
end;
print (sum (1, 100, (int i) real: 1/i))
In the version of the language defined in the revised report these changes were accepted, although the form of the cast was slightly changed to mode (expression).
real (x) co a cast to real in revised ALGOL 68 co
In the original language the void mode was represented by an empty mode:
: x := 3.14; co cast (x := 3.14) to void co proc endit = goto end; co a procedure returning void co
The ALGOL 68-R team decided to use an explicit void symbol in order to simplify parsing (and increase readability):
void val x := 3.14; co cast (x := 3.14) to void co proc endit = void : goto end; co a procedure returning void co
This modification to the language was adopted by the ALGOL 68 revised report.
Formal declarers are the modes on the left hand side of an identity declaration, or the modes specified in a procedure declaration. In the original language they could include array bounds and specified whether the matching actual declarer was fixed, flex or either:
[ 15 ] int a; co an actual declarer, bounds 1:15 co ref [ 3 : ] int b = a; co This is an error co proc x = (ref [ 1 : either] int a) : ...
| “ | I think it was a reasonable thing myself to omit the bounds from the formal-declarers but I think it was a terrible crime to omit the either or the flex |
” |
The ALGOL 68-R team redefined formal declarers to be the same as virtual declarers which include no bound information. They found that this reduced the ambiguities in parsing the language and felt that it was not a feature that would be used in real programs.
If a procedure needed certain bounds for its arguments it could check them itself with the upb (upper bound) and lwb (lower bound) operators.
In ALGOL 68-R the example above could be recoded like this: (the bounds of a in the procedure would depend on the caller).
[ 15 ] int a; co an actual declarer, bounds 1:15 co ref [] int b = a [ at 3]; co use slice so b has bounds 3:17 co proc x = (ref [] int a) void: ... co bounds given by caller co
In the revised report on ALGOL 68 formal bounds were also removed, but the flex indication was moved in position so it could be include in formal declarers:
[ 1: flex ] int a; co original ALGOL 68, or ALGOL 68-R co flex [ 1: ] int a; co revised ALGOL 68, co
proc x = (ref [ 1: flex ] int a) : ... co Original ALGOL 68 co proc x = (ref [ ] int a) void: ... co ALGOL 68-R co proc x = (ref flex [ ] int a) void: ... co Revised ALGOL 68 co
In ALGOL 68 code can be run in parallel by writing par followed by a collateral clause, for example in:
par begin producer, consumer end
the procedures producer and consumer will be run in parallel. A semaphore type (sema) with the traditional P (down) and V (up) operators is provided for synchronisation between the parts of the parallel clause,
This feature was not implemented in ALGOL 68-R.
An extension known as ALGOL 68-RT was written which used the subprogramming feature of the ICL 1900 to provide multithreading facilities to ALGOL 68-R programs with semantics similar to modern thread libraries.[8] No changes were made to the compiler, only the run-time library and the linker.
In ALGOL 68 the goto symbol could be omitted from a jump:
proc stop = : ...; ... begin if x > 3 then stop fi; co a jump, not a call co ... stop: skip end
As ALGOL 68-R was a one pass compiler this was too difficult, so the goto symbol was made obligatory.
The same restriction was made in the official sublanguage, ALGOL 68S.[9]
In ALGOL 68 uniting is the coercion that produces a union from a constituent mode, for example:
mode ibool = union (int, bool); co an ibool is an int or a bool co ibool a = true; co the bool value true is united to an ibool co
In standard ALGOL 68 uniting was possible in firm or strong contexts, so for example could be applied to the operands of formulas:
op istrue = (ibool a) bool: ...; if istrue 1 co legal because 1 (int) can be united to ibool co then ...
The ALGOL 68-R implementers found this gave too many ambiguous situations so restricted the uniting coercion to strong contexts.
The effects of this restriction were rarely important and, if necessary, could be worked around by using a cast to provide a strong context at the required point in the program.
The ALGOL 68-R compiler initialised unused memory to the value -6815700.[10][11]
This value was chosen because:
F00L.The same value was used to represent nil.
| “ | I notice, in some of your sample programs, that you are not underlining or stropping anything. |
” |
In ALGOL family languages it is necessary to distinguish between identifiers and basic symbols of the language. In printed texts this was usually accomplished by printing basic symbols in boldface or underlined (begin or begin for example).
In source programs some stropping technique had to be used. In many ALGOL like languages before ALGOL 68-R this was accomplished by enclosing basic symbols in single quote characters ('begin' for example). In ALGOL 68-R basic symbols could be distinguished by writing the in upper case, lower case being used for identifiers.
As ALGOL 68-R was implemented on a machine with 6 bit bytes (and hence a 64 character set) this was quite complicated and, at least initially, programs had to be composed on paper tape using a Flexowriter.
Partly based on the experience of ALGOL 68-R the revised report on ALGOL 68 specified hardware representations for the language, including UPPER stropping.
ALGOL 68-R included extensions for separate compilation and low-level access to the machine.
Since ALGOL 68 is a strongly typed language the simple library facilities used by other languages on the ICL 1900 system were insufficient. ALGOL 68-R was delivered with its own library format and utilities which allowed sharing of modes, functions, variables and operators between separately compiled segments of code which could be stored in albums.[13]
A segment to be made available to other segments would end with a list of declarations to be made available:
graphlib co the segment name co begin mode graphdata = struct ( ... ); mode graph = ref graphdata; proc new graph = ( ... ) graph : ...; proc draw graph = (graph g) void : ...; ... end keep graph, new graph, draw graph finish
And then the graph functions could be used by another segment:
myprog with graphlib from graphalbum
begin
graph g = new graph (...);
...
draw graph (g);
...
end
finish
As a strongly typed high level language ALGOL 68 prevented the user from directly accessing the low level hardware, there are no operators for address arithmetic for example.
Since ALGOL 68-R didn't compile to standard ICL semicompiled (link-ready) format it was necessary to extend the language to provide features in ALGOL 68-R to write code that would normally be written in assembler. Machine instructions could be written inside code ... edoc sections and the address manipulation operators inc, dec, dif, as were added.[14]
An example, using a GEORGE peri operation to issue a command:
[1 : 120] CHAR buff;
INT unitnumber;
STRUCT (BITS typemode, reply, INT count, REF CHAR address)
control area := (8r47400014,0,120,buff[1]);
...;
CODE 0,6/unitnumber; 157,6/typemode OF control area EDOC
A copy of the ALGOL 68-R compiler, runnable under David Holdsworth's George 3 emulator, is available at http://sw.ccs.bcs.org/CCs/g3/index.html