Forcing exceptions on NANs with GNAT?

Forcing exceptions on NANs with GNAT?

[Markus Schoepflin]

Platform: Linux
Compiler: GCC/GNAT 3.4.3

Hello,

is it possible to influence the behaviour of GNAT regarding the handling of 
NANs? (Most importantly in the special case of division by zero.)

We need to get exceptions whenever a NAN is generated, is this possible 
somehow? (For example by setting Machine_Overflow to True and recompiling 
the compiler itself.)

Regards,
Markus

Re: Forcing exceptions on NANs with GNAT?

[Jean-Pierre Rosen]

Markus Schoepflin a �crit :

> is it possible to influence the behaviour of GNAT regarding the handling 
> of NANs? (Most importantly in the special case of division by zero.)
> 
Note that regular division by 0 does not generate NaN, only 0/0 does.

(For your original question, I'm afraid the answer is "No").

-- 
---------------------------------------------------------
            J-P. Rosen (rosen@adalog.fr)
Visit Adalog's web site at http://www.adalog.fr

Re: Forcing exceptions on NANs with GNAT?

[Dmitry A. Kazakov]

On Thu, 20 Nov 2008 12:09:41 +0100, Markus Schoepflin wrote:

> is it possible to influence the behaviour of GNAT regarding the handling of 
> NANs? (Most importantly in the special case of division by zero.)
> 
> We need to get exceptions whenever a NAN is generated, is this possible 
> somehow? (For example by setting Machine_Overflow to True and recompiling 
> the compiler itself.)

You can scrap IEEE stuff in favor of Ada semantics by declaring your own
floating-point [sub]type with a range specified. The compiler will be
forced to check values:

   type Safe_Float is digits 6 range -10.0E10..+10.0E10;
   
or

   subtype Safe_Float is Float range Float'Range;

then

   X : Safe_Float := 1.0;
   Y : Safe_Float := 0.0;
begin
   Y := X / Y;
exception
   when Error : others =>  -- Should print "range check failed"
      Put_Line (Exception_Message (Error));
end;

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de

Re: Forcing exceptions on NANs with GNAT?

[sjw]

On Nov 20, 1:46 pm, "Dmitry A. Kazakov" <mail...@dmitry-kazakov.de>
wrote:
>    subtype Safe_Float is Float range Float'Range;

Thanks for that, Dmitry; when I had to do the same thing at the office
I said 'range Float'First .. First'Last'. Yours is clearly the Right
Way to do this.

Re: Forcing exceptions on NANs with GNAT?

[Adam Beneschan]

On Nov 20, 3:09 am, Markus Schoepflin <nos...@no.spam> wrote:
> Platform: Linux
> Compiler: GCC/GNAT 3.4.3
>
> Hello,
>
> is it possible to influence the behaviour of GNAT regarding the handling of
> NANs? (Most importantly in the special case of division by zero.)
>
> We need to get exceptions whenever a NAN is generated, is this possible
> somehow? (For example by setting Machine_Overflow to True and recompiling
> the compiler itself.)

I really don't know that much about GNAT, but do those operations
cause SIGFPE signals to be generated, and if so, can they be caught by
an interrupt handler (see C.3)?  And can that handler raise an
exception that is propagated to the program (task) that performed the
operation?

                               -- Adam

Re: Forcing exceptions on NANs with GNAT?

[Jerry]

On Nov 20, 4:09 am, Markus Schoepflin <nos...@no.spam> wrote:
> Platform: Linux
> Compiler: GCC/GNAT 3.4.3
>
> Hello,
>
> is it possible to influence the behaviour of GNAT regarding the handling of
> NANs? (Most importantly in the special case of division by zero.)
>
> We need to get exceptions whenever a NAN is generated, is this possible
> somehow? (For example by setting Machine_Overflow to True and recompiling
> the compiler itself.)
>
> Regards,
> Markus

There was a discussion about NaNs a while back. Not sure how much of
it is useful to you, but here's the link:

http://groups.google.com/group/comp.lang.ada/browse_thread/thread/772ddcb41cd06d5b/746f9f7e98a8d2f9?hl=en#746f9f7e98a8d2f9

Jerry

Re: Forcing exceptions on NANs with GNAT?

[Markus Schoepflin]

Markus Schoepflin wrote:

> is it possible to influence the behaviour of GNAT regarding the handling 
> of NANs? (Most importantly in the special case of division by zero.)

[...]

Thanks all for the answers.

The suggestion from Dmitry (create a subtype with the same range as float) 
works fine for my purposes, GNAT then raises range check errors for both 
division by zero and overflows.

Markus