Re: EXC_ARITHMETIC when writing a floating point variable

[Jerry DeLisle <jvdelisle at frontier dot com>]

On 10/31/2010 08:55 AM, Richard Daal wrote:
> Dear all,
>
> I am struggling with a problem for some time now, where I do not know
> how to proceed. I would greatly appreciate, if I could get some
> advise.
>
> I try to write a double precision variable "Test = 3.456" with
> "write(*, *) Test" to the terminal.
> That produces a floating point exception. With gdb, I get the
> following error message:
> "Program received signal EXC_ARITHMETIC, Arithmetic exception.
> 0x000000001001c8132 in __dtoa ()"
>
> However, at some place in the code double precision variables of an
> allocatable array that belongs to a derived type are written to the
> terminal and this works fine. When removing the write statement that
> produces the error, the program works fine.
> I do not understand, what could go wrong with this "write" statement?
> How could I check what the problem is? Its a relatively complex code.
>
> I use the following flags with gfortran 4.5.1 (on Mac OS 10.6.4) for
> compilation:
> -ffree-form -fimplicit-none -frange-check -std=f2003
> -fdefault-double-8 -fdefault-integer-8 -fdefault-real-8 -fmax-errors=0
> -pedantic -Wall -fconvert=native -fbounds-check
> -fmax-stack-var-size=32768 -g -ffpe-trap=invalid,zero,
> overflow,underflow,precision,denormal
> ... and for linking:
> -arch_errors_fatal -arch x86_64 -g
>
> I thank you very much in advance for your help..
>
> With kind regards,
>
> Richard

My first suggestion is don't use these:

 -fdefault-double-8 -fdefault-integer-8 -fdefault-real-8

It is better to use explicit kind specifications and set a parameter in the 
program for the kind type to use.

I suppose your code is legacy and does not have any easy way to do this. The 
only way that I can see that the write statement would fail like that is if the 
variable has been stomped on because something is out of bounds. Of course there 
coud be a real exception in __dtoa () which would be a bug.

Try eliminating chunks of code until the problem goes away.  This might give a hint.

Jerry