Re: [Gm2] Fwd: CASE: gm2-4.1.2 of 2011-11-08 bug (or possibly philosophical difference)

[Fischlin Andreas]

Dear Gaius,

> Hi Andreas,
> 
> overlapping variant tag values, for example:
> 
> TYPE
>   mytag = [0..4] ;
> 
>   foo = RECORD
>            CASE :mytag OF
> 
>            0..2:  x: INTEGER |
>            1..4:  y: CARDINAL
> 
>            END
>         END ;
> 
> should gm2 complain that the two tag ranges overlap?  [*]

for all what I know above is illegal, a plain syntax error (multiple defined 
cases). Perhaps what is intended should be defined similar to this

MODULE TstCase;

  TYPE
    Mystruct = RECORD
                CASE c:CARDINAL OF
                 0,2: x: INTEGER;  |
                 1,3..4: y: CARDINAL; |
                END;
                END(*RECORD*);
                
  VAR
    x: Mystruct;
    
  TYPE
    Myenum = (a, b, c, d);
    Mystruct2 = RECORD
                CASE e:Myenum OF
                 a..b: x: INTEGER;  |
                 c..d: y: CARDINAL; |
                END;
                END(*RECORD*);
                
  VAR                
    y: Mystruct2;
                    
BEGIN
  x.c := 4; x.x := 77; 
  y.e := b; y.y := 88;
END TstCase.


Regards,
Andreas


ETH Zurich
Prof. Dr. Andreas Fischlin
Systems Ecology - Institute of Integrative Biology
CHN E 21.1
Universitaetstrasse 16
8092 Zurich
SWITZERLAND

address@hidden
www.sysecol.ethz.ch

+41 44 633-6090 phone
+41 44 633-1136 fax
+41 79 221-4657 mobile

             Make it as simple as possible, but distrust it!
________________________________________________________________________



On 13/11/2011, at 15:47 , Gaius Mulley wrote:

> "Fischlin  Andreas" <address@hidden> writes:
> 
>> Dear Gaius,
>> 
>> What exactly do you mean by overlap? You mean memory allocation
> 
> 
> Hi Andreas,
> 
> overlapping variant tag values, for example:
> 
> TYPE
>   mytag = [0..4] ;
> 
>   foo = RECORD
>            CASE :mytag OF
> 
>            0..2:  x: INTEGER |
>            1..4:  y: CARDINAL
> 
>            END
>         END ;
> 
> should gm2 complain that the two tag ranges overlap?  [*]
> 
> 
>> including alignment in memory or ambiguous case selection? The latter
> 
> yes I was referring to ambiguous case selection - ok yes it will be
> flagged as such!  [*]
> 
>> is clearly an error, the first should be possible or you defy lots of
>> legacy code, especially PIM code (and I know thousands and thousands
>> lines of such code ;-) ). For PIM code the assumption is that the
> 
> indeed alignment overlap is very necessary for many applications.
> 
>> compiler allocates memory for the largest variant (unless requested
>> otherwise with a construct such as TSIZE(T(t1)) used as parameter with
>> ALLOCATE (not NEW)). Alignment in memory is done without gaps,
>> assuming a byte addressing machine, meaning that the shortest variant
>> of the record wastes the most memory. Today such memory waste is not
>> very costly and if a user is concerned about this, he/she could
>> program different records from the very beginning, instead of a case
>> variant (or use ALLOCATE(p, TSIZE(T(tmin))) and make sure she never
>> changes the case during use of a given memory block). So the
>> programmer has full control over this anyway.
>> 
>> BTW, to illustrate this, some code excerpt of a little routine that
>> depends heavily on coercion made possible through case variant records
>> and that support greatly portable programing among platforms. We use
>> this to determine big vs. little endianess and consequently all REAL
>> to string conversions in an endianess independent manner.
>> 
>>  PROCEDURE MachineIsBigEndian(): BOOLEAN; 
>>    TYPE
>>      LongIntVar = RECORD
>>        CASE :BOOLEAN OF
>>            TRUE : c : LONGCARD;
>>          | FALSE: b : ARRAY [1..4] OF CHAR; (* assumes TSIZE(CHAR) = 
>> TSIZE(BYTE) *)
>>        END;
>>      END;
>>    VAR int : LongIntVar; one: CHAR;
>>  BEGIN (* MachineIsBigEndian *)
>>    int.c := 1; one := 01C;
>>    IF (int.b[4] = one) THEN
>>      (* test whether really big endian *)
>>      IF (int.b[1] <> 0C) OR (int.b[2] <> 0C) OR (int.b[3] <> 0C) THEN 
>>        (* middle endian, should not occur *) 
>>        HALTOnMachineDependency("Portab","MachineIsBigEndian","Unsupported 
>> architecture encountered (middle endian?)");
>>      END(*IF*);
>>    ELSE
>>      (* test whether really little endian *)
>>      IF (int.b[2] <> 0C) OR (int.b[3] <> 0C) OR (int.b[4] <> 0C) THEN 
>>        (* middle endian, should not occur *)
>>        HALTOnMachineDependency("Portab","MachineIsBigEndian","Unsupported 
>> architecture encountered (middle endian?)");
>>      END(*IF*);
>>    END(*IF*);
>>    RETURN (int.b[4] = one)
>>  END MachineIsBigEndian;
>> 
>> If in memory int.c would not overlap with int.b the entire algorithm
>> would completely fail. BTW, I believe a Modula-2 compiler not
>> supporting such an overlap would be pretty much useless and AFAIK to
>> loads of legacy Modula-2 code (and ISO too).
> 
> yes I totally agree - network protocol implementation is also made
> slightly cleaner with this feature.
> 
> regards,
> Gaius
> 
>> 
>> Regards,
>> Andreas