bugreport - Gpc

4 Nov 1999


      DAIMI, 4.11.1999
Dear sirs
For a little over a year I have been using the gpc compiler
with a reasonable degree of success.
Yesterday I had the opportunity to test the latest (according
to our systems programmer, Benedict Lofstedt) release of gpc.
This gave rise to four comments:
1. The attached program `arraytest' checks subrange types and
   subscripted variables. There seems to be no check of
   exceeding the limits. See the attached output which is 
   identical for the old and new version of gpc.
2. The attached program `arraytest2' checks what happens to the 
   surroundings. When one assigns values outside the subscript
   range. I have declared   var X, Y, Z : vektor
   and have exceeded the subscript range in Y.
   The latest release of gpc destroys the contents of Z
   whereas the old one destroys X
3. Program `virker' which includes `alfaplot' and `syntaks'
   is sent by separate mail. The program works well with the
   old gpc, but with the new release I get:
10> gpc virker.p
In file included from virker.p:17:
alfaplot: In function `Mark':
alfaplot:299: Internal compiler error in `emit_library_call_value', at calls.c:3198
Please submit a full bug report.
See URL:http://www.gnu.org/software/gcc/faq.html#bugreport for instructions.
4. If one omits the semicolon in line 63 (an error, I know,
   but one that did occur) of `syntaks', the second line of
   procedure AR
       var xa, xb, ya, yb : real   <--
   the old compiler produces so many lines of diagnostics,
   that I had a hard time finding the first (and only
   important) one.
   This is just one example, that gpc has problems recovering
   after an error.
program arraytest(output);
     (* test of overflow in subrange and array-index.
        gpc allows the following program,  pc does not  *)
type
   small = 0..100;
   vektor = array [small] of integer;
var
   X, Y : vektor;
   A : array [0..5, 0..5] of integer;
   i, n : small;
begin
   n := 98;
   for i := 1 to 7 do
   begin n := n + 1; X[n] := n+6;
         A[i,n] := i + n; writeln(n:4, X[n]:4, A[i,n]:5) end
end.
4> gpc arraytest.p
5> a.out
  99 105  100
 100 106  102
 101 107  104
 102 108  106
 103 109  108
 104 110  110
 105 111  112
program arraytest2(output);
     (* test of consequenses of overflow in array-index  *)
type
   small = 0..5;
   vektor = array [small] of integer;
var
   X, Y, Z : vektor;
   i, n : small;
begin
   for i := 0 to 5 do
   begin X[i] := i+10; Y[i] := i+30; Z[i] := i+50 end; 
   for i := 0 to 5 do writeln(i:3, X[i]:5, Y[i]:5, Z[i]:5);
   for i := 1 to 10 do Y[i] := i+100;
   for i := 0 to 5 do writeln(i:3, X[i]:5, Y[i]:5, Z[i]:5)
end.
7> gpc arraytest2.p
8> a.out
  0   10   30   50
  1   11   31   51
  2   12   32   52
  3   13   33   53
  4   14   34   54
  5   15   35   55
  0   10   30  106     output using the new release of gpc
  1   11  101  107
  2   12  102  108
  3   13  103  109
  4   14  104  110
  5   15  105   55
10> a.out
  0   10   30   50
  1   11   31   51
  2   12   32   52
  3   13   33   53
  4   14   34   54
  5   15   35   55
  0  106   30   50     output using the old one
  1  107  101   51
  2  108  102   52
  3  109  103   53
  4  110  104   54
  5   15  105   55
Sincerely
Ole Osterby
Department of Computer Science
    Aarhus University