Unexpected NaN value

Unexpected NaN value

[Luca Donatini]

Dear all,

I noticed a strange behaviour of the gfortran compiler.

In the source file below, when the real variable b has an "integer-like"
value (e.g. "2", "2.", "2.0" ...) the code correctly writes out a numerical
result. But when a "real-like" value is assigned to b (e.g. "2.1",
"2.0000009", "2.83" ...) , the code writes a NaN value.
This does not occurr when the sample code is compiled with an Intel
compiler (I tested version 2018a).

I have the NaN value problem with both:

   - gfortran 8.2.1 20180831 (Arch Linux package)
   - gfortran 6.3.0 20170516 (Debian Stretch package).


Is this a bug? A known one?

Many thanks in advance for all the clarifications,


Luca Donatini

*** example source file ***

program test
    implicit none

    real :: a = -1.57079637
    real :: b = 2.1

    write(*,*) cos(a)
    write(*,*) cos(a) **(2*b)

end program

Re: Unexpected NaN value

[Thomas Koenig]

Hi Luca,

> program test
>      implicit none
> 
>      real :: a = -1.57079637
>      real :: b = 2.1
> 
>      write(*,*) cos(a)
>      write(*,*) cos(a) **(2*b)
> 
> end program

cos(a) happens to be negative, so cos(a)**(2*b) is undefined,
because 2*b is real.  And yes, this is expected.

Regards

	Thomas

Re: Unexpected NaN value

[Luca Donatini]

Dear Thomas,

Many thanks for the clarification!


Best regards,


Luca

Il gio 4 ott 2018, 22:30 Thomas Koenig <tkoenig@netcologne.de> ha scritto:

> Hi Luca,
>
> > program test
> >      implicit none
> >
> >      real :: a = -1.57079637
> >      real :: b = 2.1
> >
> >      write(*,*) cos(a)
> >      write(*,*) cos(a) **(2*b)
> >
> > end program
>
> cos(a) happens to be negative, so cos(a)**(2*b) is undefined,
> because 2*b is real.  And yes, this is expected.
>
> Regards
>
>         Thomas
>