Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!orion.oac.uci.edu!ucivax!milne
From: milne@ics.uci.edu (Alastair Milne)
Newsgroups: comp.lang.pascal
Subject: Re: need a little flexibility
Message-ID: <2818F692.27076@ics.uci.edu>
Date: 27 Apr 91 03:57:05 GMT
References: <1991Apr24.183315.7997@ux1.cso.uiuc.edu>  <1991Apr24.235739.25115@watmath.waterloo.edu> <91115.111155IO92203@MAINE.BITNET>
Distribution: na
Organization: UC Irvine Department of ICS
Lines: 79

In <91115.111155IO92203@MAINE.BITNET> IO92203@MAINE.BITNET (Scott Maxell) writes:

>In article <1991Apr24.235739.25115@watmath.waterloo.edu>,
>>   x : byte;
>>   y : byte;
>>   case boolean of
>>    True : (data : array[1..252] of byte);
>>    False: (dataw: array[1..126] of word);
>>end;
>>
....
>>  b.data[i] := hi(w);
>>  b.dataw[i] := w;
>>end.
>>

>   This last approach won't work as stated because a variant record can
>only contain one of the possible fields declared. The field in the variant
>should also have a declared variable and not just a type.

   Not true!!

   Pascal does *not* guard against using whichever of the variants you
   want at any time.  Ada has constructions which do, but Pascal does not.

   Nor does the tagfield actually have to be an allocated field,
   rather than just a type.  It *can* be, of course, if you want your
   program to be able to ask as which variant a give instance of the 
   record was created, but there's no obligation.

>Type
>  Block = RECORD
>            X,
>            Y : Byte;
>            Case WordData : Boolean OF
>              True  : ( DataW : ARRAY [1..126] OF Word );
>              False : ( Data  : ARRAY [1..252] OF Byte );
>          END;  (* record *)

>VAR
>  WordData : BOOLEAN;
>  OneWord  : Word;
>  OneByte  : Byte;

>BEGIN
>  B.WordData := True;   (* or false depending on what your data is. *)
>  OneByte    := ????;   (* Your byte data.                          *)
>  OneWord    := ????;   (* Your word data.                          *)
>  n          := ????;   (* Array index.                             *)
>  Case B.WordData OF
>    True  : B.Data  [n] := OneByte;
>    False : B.DataW [n] := OneWord;

      { All this is internally consistent, but you could as easily write
	FALSE: B.Data[N] := OneByte; 
	and Pascal wouldn't care at all. }
>  END;  (* case *)
>END;

>  When adding data to a variant record, there needs to be a value placed in
>the tag field, or you will generally get an error.

    You will get errors if the program is counting on telling by the 
    tag field which variant is being used, since not being able to 
    tell may cause it to assign to the wrong variant, possibly with 
    the wrong size, overwrite memory, etc. etc. .  Then an uninitialised
    tag field will wreak havoc -- but not because of any internal bindings,
    simply because it's supplying the assignment algorithm with false
    information.  But this is typically in the case of dynamically allocated 
    variant records, not the static ones given here.

    But if your routines are not themselves making use of the tag field,
    there is no need to have one.

    Note I am speaking of UCSD Pascal and Turbo Pascal.  I don't have the 
    ANSI standards available -- they may be more rigid.


    Alastair Milne