S'Is_nan or S'Is_inf?

S'Is_nan or S'Is_inf?

[Warren]

Is there any is not-a-number or is-infinity test support in
Ada05+ (for floats)?  Is there any being planned?  

Just shooting from the hip, it would seem to be a standard 
thing that should be included these days. 

Obviously you'd also need Booleans to tell you if they are 
supported or not, on your given platform.

Warren

Re: S'Is_nan or S'Is_inf?

[Rod Chapman]

On Jul 15, 10:00 pm, Warren <ve3...@gmail.com> wrote:
> Is there any is not-a-number or is-infinity test support in
> Ada05+ (for floats)?  Is there any being planned?  

What is S'Valid supposed to return for an S which is
a floating-point NaN???
 - Rod

Re: S'Is_nan or S'Is_inf?

[J-P. Rosen]

Warren a �crit :
> Is there any is not-a-number or is-infinity test support in
> Ada05+ (for floats)?  Is there any being planned?  
> 
If X /= X, then X is a NaN (see http://en.wikipedia.org/wiki/NaN)

Not sure if there is such a simple test for infinities...
-- 
---------------------------------------------------------
           J-P. Rosen (rosen@adalog.fr)
Visit Adalog's web site at http://www.adalog.fr

Re: S'Is_nan or S'Is_inf?

[Simon Wright]

Rod Chapman <roderick.chapman@googlemail.com> writes:

> On Jul 15, 10:00 pm, Warren <ve3...@gmail.com> wrote:
>> Is there any is not-a-number or is-infinity test support in
>> Ada05+ (for floats)?  Is there any being planned?  
>
> What is S'Valid supposed to return for an S which is
> a floating-point NaN???

I thought it was supposed to produce False, but all the AARM seems to
say (13.9.2) is that the value needs to be normal and have a valid
representation for 'Valid to return True. It also lists the
circumstances under which invalid data can be created: they don't look
to me as though they include

   NaN : constant Float := 0.0 / 0.0;

GNAT certainly reports 'Valid as False for +/-Inf & NaN.

Re: S'Is_nan or S'Is_inf?

[Simon Wright]

Warren <ve3wwg@gmail.com> writes:

> Is there any is not-a-number or is-infinity test support in Ada05+
> (for floats)?  Is there any being planned?

With GNAT, using a subtype

   subtype Checked_Float is Float range Float'Range;

will give you a Constraint_Error for NaN or +/-Inf.

Re: S'Is_nan or S'Is_inf?

[Dmitry A. Kazakov]

On Thu, 15 Jul 2010 23:07:51 +0100, Simon Wright wrote:

> Warren <ve3wwg@gmail.com> writes:
> 
>> Is there any is not-a-number or is-infinity test support in Ada05+
>> (for floats)?  Is there any being planned?

with Ada.Text_IO;  use Ada.Text_IO;
procedure IEEE is -- Only if Float is IEEE!
   Zero : Float := 0.0;
   Inf  : Float := 1.0 / Zero;
   NaN  : Float := 0.0 / Zero;
begin
   Put_Line ("Valid    " & Boolean'Image (Inf'Valid));
   Put_Line ("In range " & Boolean'Image (Inf <= Float'Last));
   Put_Line ("Self NaN " & Boolean'Image (NaN = NaN));
end IEEE;

On an IEEE machine it could print 3x FALSE.

> With GNAT, using a subtype
> 
>    subtype Checked_Float is Float range Float'Range;
> 
> will give you a Constraint_Error for NaN or +/-Inf.

Yes, one of the reasons not to use built-in types is that the floating
point ones most likely are IEEE with all nasty consequences.

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

Re: S'Is_nan or S'Is_inf?

[Warren]

Dmitry A. Kazakov expounded in news:1tbp3geoa5yna$.171cmlfdrbm98$.dlg@
40tude.net:

> with Ada.Text_IO;  use Ada.Text_IO;
> procedure IEEE is -- Only if Float is IEEE!
>    Zero : Float := 0.0;
>    Inf  : Float := 1.0 / Zero;
>    NaN  : Float := 0.0 / Zero;
> begin
>    Put_Line ("Valid    " & Boolean'Image (Inf'Valid));
>    Put_Line ("In range " & Boolean'Image (Inf <= Float'Last));
>    Put_Line ("Self NaN " & Boolean'Image (NaN = NaN));
> end IEEE;
> 
> On an IEEE machine it could print 3x FALSE.

On a non IEEE machine, is there going to be an exception 
raised when dividing by zero?

Warren

Re: S'Is_nan or S'Is_inf?

[Dmitry A. Kazakov]

On Fri, 16 Jul 2010 17:06:38 +0000 (UTC), Warren wrote:

> Dmitry A. Kazakov expounded in news:1tbp3geoa5yna$.171cmlfdrbm98$.dlg@
> 40tude.net:
> 
>> with Ada.Text_IO;  use Ada.Text_IO;
>> procedure IEEE is -- Only if Float is IEEE!
>>    Zero : Float := 0.0;
>>    Inf  : Float := 1.0 / Zero;
>>    NaN  : Float := 0.0 / Zero;
>> begin
>>    Put_Line ("Valid    " & Boolean'Image (Inf'Valid));
>>    Put_Line ("In range " & Boolean'Image (Inf <= Float'Last));
>>    Put_Line ("Self NaN " & Boolean'Image (NaN = NaN));
>> end IEEE;
>> 
>> On an IEEE machine it could print 3x FALSE.
> 
> On a non IEEE machine, is there going to be an exception 
> raised when dividing by zero?

I would say yes. But theoretically if a non-IEEE machine has special
representations of x/0, as IEEE does, then the compiler vendor is permitted
not to raise Constraint_Error.

Language lawyers?

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

Re: S'Is_nan or S'Is_inf?

[Warren]

Dmitry A. Kazakov expounded in
news:1tbp3geoa5yna$.171cmlfdrbm98$.dlg@40tude.net: 

> with Ada.Text_IO;  use Ada.Text_IO;
> procedure IEEE is -- Only if Float is IEEE!
>    Zero : Float := 0.0;
>    Inf  : Float := 1.0 / Zero;
>    NaN  : Float := 0.0 / Zero;
> begin
>    Put_Line ("Valid    " & Boolean'Image (Inf'Valid));
>    Put_Line ("In range " & Boolean'Image (Inf <= Float'Last));
>    Put_Line ("Self NaN " & Boolean'Image (NaN = NaN));
> end IEEE;
> 
> On an IEEE machine it could print 3x FALSE.

To detect NaN then, this seems to work:

    function Is_Nan(F : Float) return Boolean is
    begin
        if not F'Valid then
            return not ( Is_Infinity(F) or Is_Neg_Infinity(F) );   
        else
            return False;
        end if;
    end Is_Nan;

where Is_Infinity(F) etc. is implemented as suggested. Essentially
if not valid, but not one of the infinities(+ or -), then it 
must be NaN. This works with Gnat on Cygwin. I'll need
to test it on other platforms, but hopefully most if not all
IEEE platforms will support this.

Warren

Re: S'Is_nan or S'Is_inf?

[Simon Wright]

Warren <ve3wwg@gmail.com> writes:

> To detect NaN then, this seems to work:

See above, where Jean-Pierre said

> If X /= X, then X is a NaN (see http://en.wikipedia.org/wiki/NaN)

Re: S'Is_nan or S'Is_inf?

[Dmitry A. Kazakov]

On Fri, 16 Jul 2010 22:35:14 +0000 (UTC), Warren wrote:

> I'll need
> to test it on other platforms, but hopefully most if not all
> IEEE platforms will support this.

IEEE representations have more than +Inf, -Inf and NaN. There are  other
objects: +0, -0, denormalized numbers (see attribute 'Denorm). Some have
dubious semantics:

with Ada.Text_IO;  use Ada.Text_IO;
procedure IEEE_Zeros is
   Zero : Float := 0.0;
   N0   : Float := 0.0 / (-1.0 / Zero);
begin
   Put_Line ("-0 =" & Float'Image (N0) & " 0 = " & Float'Image (Zero));
   Put_Line ("-0 = 0 " & Boolean'Image (N0 = Zero));
   Put_Line ("-0 < 0 " & Boolean'Image (N0 < Zero));
end IEEE_Zeros;

Negative zero is zero but not equal to, etc.

What are you trying to achieve? Because in automation we have the rule
never ever let IEEE cripples slip through. Don't read them, don't write
them, don't compute them.

The solution Simon suggested is basically everything Ada programmer should
know about IEEE 754! (:-))

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

Re: S'Is_nan or S'Is_inf?

[Warren]

Simon Wright expounded in news:m2tynytykr.fsf@pushface.org:

> Warren <ve3wwg@gmail.com> writes:
> 
>> To detect NaN then, this seems to work:
> 
> See above, where Jean-Pierre said
> 
>> If X /= X, then X is a NaN (see http://en.wikipedia.org/wiki/NaN)

Thanks. I later had a head slapping moment on the
weekend where I had figured that out. ;-)

Warren

Re: S'Is_nan or S'Is_inf?

[Warren]

Dmitry A. Kazakov expounded in
news:5xj0ja9xywto$.3ncqua6onpoa.dlg@40tude.net: 

> On Fri, 16 Jul 2010 22:35:14 +0000 (UTC), Warren wrote:
> IEEE representations have more than +Inf, -Inf and NaN. There are 
> other objects: +0, -0, denormalized numbers (see attribute 'Denorm).
> Some have dubious semantics:

Ya, I know about those.  As indicated in my other reply,
I did eventually figure out that if F /= F, then it is NaN  
(for some reason, I didn't see this at first).

> What are you trying to achieve? Because in automation we have the rule
> never ever let IEEE cripples slip through. Don't read them, don't
> write them, don't compute them.

In my Basic interpreter, I'll raise an error if you try to
use these IEEE values in an argument to a function or try to 
compute something with an operator (here NOT 'Valid is sufficient).  
However, if the expression result is one of these, I do allow 
it to be assigned to a [basic] variable.

So then the user needs a (basic) function to test if the
variable V is +/- infinity or NaN. So in basic, ISNAN(X)
returns 1 if NaN, else zero, for example.

In the C version of the interpreter, I relied upon C functions
for these tests. Ada clearly does not need these C functions, 
which is bliss for the Ada rewrite.

Thanks for everyone's help.

Warren

Re: S'Is_nan or S'Is_inf?

[Randy Brukardt]

"Warren" <ve3wwg@gmail.com> wrote in message 
news:Xns9DB6AD1651560WarrensBlatherings@81.169.183.62...
> Is there any is not-a-number or is-infinity test support in
> Ada05+ (for floats)?  Is there any being planned?

We considered it for Ada 2005 (see AI95-0315-1). It was abandoned because of 
the large impact on implementations and the weak demand (hardly any user 
requests).

John Barnes mentions this in the Ada 2005 Rational: 
http://www.adaic.com/standards/05rat/html/Rat-9-3-3.html

It wasn't resurrected for Ada 2012, so there won't be any change here.

BTW, GNAT seems to pass through a lot of IEEE stuff, thus you can write 
examples using that. Other Ada compilers (at least as of 2003-4 timeframe) 
vary widely in what is exposed (Janus/Ada only supports IEEE denormal 
numbers; everything else raises Constraint_Error before it is stored).

                                  Randy.

Re: S'Is_nan or S'Is_inf?

[Warren]

Randy Brukardt expounded in news:i22m9q$cd4$1@munin.nbi.dk:

> "Warren" <ve3wwg@gmail.com> wrote in message 
> news:Xns9DB6AD1651560WarrensBlatherings@81.169.183.62...
>> Is there any is not-a-number or is-infinity test support in
>> Ada05+ (for floats)?  Is there any being planned?
> 
> We considered it for Ada 2005 (see AI95-0315-1). It was abandoned
> because of the large impact on implementations and the weak demand
> (hardly any user requests).
> 
> John Barnes mentions this in the Ada 2005 Rational: 
> http://www.adaic.com/standards/05rat/html/Rat-9-3-3.html
> 
> It wasn't resurrected for Ada 2012, so there won't be any change here.
> 
> BTW, GNAT seems to pass through a lot of IEEE stuff, thus you can
> write examples using that. Other Ada compilers (at least as of 2003-4
> timeframe) vary widely in what is exposed (Janus/Ada only supports
> IEEE denormal numbers; everything else raises Constraint_Error before
> it is stored). 
> 
>                                   Randy.

Very interesting. My project is fairly dependant on gnat (at the 
moment at least), so things should work ok. I'm not personally
a fan of these special IEEE values, but they crop up in 3rd party
libraries like the GSL etc. or in the reading of binary values from
a file.

Warren

Re: S'Is_nan or S'Is_inf?

[Dmitry A. Kazakov]

On Tue, 20 Jul 2010 14:12:01 +0000 (UTC), Warren wrote:

> I'm not personally
> a fan of these special IEEE values, but they crop up in 3rd party
> libraries like the GSL etc. or in the reading of binary values from
> a file.

This was a motivation behind

   http://www.dmitry-kazakov.de/ada/components.htm#IEEE_754
 
Inputs and outputs in IEEE 754 representations can be converted to and from
Ada floating-point types in a portable way. Within the program you forget
about IEEE.

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