6

I've decompiled yet another Pascal compiler for the BESM-6.

While it does not implement the full language (for example, packed records and arrays are not implemented, helper routines SEL and INS are offered instead to manipulate parts of words), there is a feature which would have been a good addition to the standard, if Wirth had thought of it in time, namely using its range check feature not only as a "behind the scenes" mechanism, but as a user-accessible predicate.

It is well known that although Pascal allowed to declare "set" types, e.g.

type bits = 0..31; bitset = set of bits;

the maximum capacity of such types was limited. The early implementations restricted a set type to a single word with whatever word size the host computer had. For example, CDC 6600 had 60, BESM-6 had 48.

With set types of limited use, and the language having range subtypes as first-order constructs, one might ask, what would be a better way to succinctly write a check for a value being within an arbitrary range of a discrete type. Suppose, I have

type range = 1000000..2000000;

and I want to check if an integer variable is within that range. The best solution the standard can offer is to write

const rangeLow = 1000000; rangeHigh = 2000000;
type range = rangeLow..rangeHigh;
function isInRange(i:integer):boolean;
begin
  isInRange := (rangeLow <= i) and (i <= rangeHigh)
end;

However, when one has

var r:range;i:integer;

and writes r := i;, the compiler inserts a range check which makes the program terminate if unsuccessful. There is no way to use that internal range-checking mechanism to a user's advantage.

The solution implemented and employed in the BESM-6 Pascal compiler was:

(*=P-,T-,M-*)
_PROGRAM MAIN;
_TYPE RANGE = 1000000..2000000;
_VAR I,R,S:INTEGER;
_BEGIN
  S := 0;
  R := 1;
  _FOR I := 1 _TO 1000000 _DO _BEGIN
    _IF R _IN RANGE _THEN S := S + 1;
    R := (R * 1001 + 1234567) _MOD 4294967296
  _END;
  WRITELN(S/1000000);
_END.

It prints

     2.3100Е-04

When the compiler encounters EXPR in TYPE in an if operator, it generates code as if EXPR was about to be assigned to a variable of type TYPE, pointing the "failure" target to the else clause (or beyond the then clause), as opposed to the PMD routine for the actual range checks.

The range-checking code is then associated with TYPE and is reused if other instances of EXPR in TYPE are encountered.

This looks more succinct and user-friendly than requiring the user to write the check explicitly, as above.

Has there been another Pascal compiler with such a feature? Or is it only in the USSR where they dared to extend the language in such a way?

4
  • 2
    "This looks more succinct and user-friendly than requiring the user to write the check explicitly, " We already went over this a few times in your earlier questions, but Wirth was a minimalist: He'd rather take things away from the language (and he frequently did during development of his languages), if the benefits didn't justify the added complexity. So he would have disagreed very much with your above statement: two comparisons with an AND are simple enough. (And Pascal dialects all over the world "dared" to extend it in lots of different directions, much to Wirth's chagrin).
    – dirkt
    Commented May 29, 2021 at 16:47
  • 2
    @dirkt That's why I'm not asking why didn't Wirth add that to the language. I'm asking if there was another dialect among those all over the world with that little addition, given that there is no added complexity to the syntax or code generation. The feature already exists in the language, the small step taken is to allow access to it. Supporting it in the BESM-6 compiler takes 86 instructions.
    – Leo B.
    Commented May 29, 2021 at 17:14
  • 1
    Does the compiler use stropping to designate reserved words (as implied by your example code)? If so, why? To simplify lexical analysis? Does it allow identifiers with the same spelling as reserved words?
    – texdr.aft
    Commented May 29, 2021 at 22:18
  • 1
    @texdr.aft It is a restricted variant of stropping. An underscore before an identifier, otherwise delimited in a usual way, assigns a different token class to it, but extra underscores are an error (or, rather, underscores serve as separators). Yes, it allows identifiers matching reserved words; I use that in my decompilation. Regarding stropping, it appears that its author was somewhat influenced by Algol, allowing both _GOTO and _GO _TO; as it turns out, supporting that adds a trivial amount of code.
    – Leo B.
    Commented May 30, 2021 at 1:19

0

You must log in to answer this question.

Browse other questions tagged .