Re: Bug or feature?

[Adam Beneschan <adambeneschan@gmail.com>]

On Wednesday, May 14, 2014 12:57:47 PM UTC-7, Adam Beneschan wrote:

> It does appear that GNAT may not always check the range properly when a
> function whose result is Positive tries to return a negative value due to
> overflow.  That would explain why the recursive version doesn't get an
> exception.

Here's an example:

with Ada.Integer_Text_IO;
with Ada.Text_IO;

procedure Test is
   Result : Integer;

   function Add (X1, X2 : Positive) return Positive is
   begin
      return X1 + X2;
   end Add;

   function Multiply (X1, X2 : Positive) return Positive is
   begin
      return X1 * X2;
   end Multiply;

begin
   Result := Add(2_000_000_000, 2_000_000_000);
   Ada.Integer_Text_IO.Put (Item => Result, Width => 1);
   Result := Multiply(6, 2_000_000_000);
   Ada.Integer_Text_IO.Put (Item => Result, Width => 1);
end Test;

Compiled without -gnato.  The result is that a negative number is output after the call to Add, but the call to Multiply raises an exception.  So it looks like when the function result is Positive, a "return" that returns the result of an addition doesn't check to make sure the result is in range (when overflow is possible), but a "return" that returns the result of a multiplication does check.  That would also explain why the original Buffer_Overflow test case works but the recursive Fibonacci routine doesn't.

                                 -- Adam