People,

looking at all the threads abouth this topic leaves me confused - I
could not find an answer to the following question:

if( var/var /= 1) then
write(*,*) 'This is a NAN or INF'
endif

i) can I test wether a real variable is a number with this test?
ii) or is it highly dependent on what platform/compiler you are
using ?
iii) does it always work?

Cheers
-Ralf

Ralf Schaa wrote:

With the right code generation, it will work on any hardware which
supports IEEE754 and the like. (I assume var is a real data type).
A more reasonable way to write an equivalent test would be
if(var /= var)then
write(*,*)'Nan value' ...

Ralf Schaa wrote:

It is highly dependent on the platform and compiler. No, it doesn't
always work. I wouldn't even count on it usually working. It is likely
to cause an abort on some platforms. That's not to speak of the fact
that 0.0 is a perfectly valid number that is neither NaN nor Inf, but is
quite unlikely to give you 1 as a result. In fact, based on that, I
would venture to say that there are probably no compilers where this
test is accurate.

Ralf Schaa wrote:

On Jul 13, 12:02 pm, nos...@see.signature (Richard Maine) wrote:

[snip]

Thanks Richard and Tim.

It's never easy ...

Ralf Schaa wrote:

(snip)

I thought the test for NaN was supposed to be var.ne.var, but compilers
might optimize that one away.

Are you only considered IEEE standard floating point? Without that
I don't believe you can safely test 0./0..

-- glen

On Jul 13, 2:00 pm, glen herrmannsfeldt ugcs.caltech.edu> wrote:
[snip]

Glen,

the problem would concern F95 (as of 2006) and the Lahey v5.7
compilers on Windows PC's - if this is the right answer to that
question at all (?)

-Ralf

Ralf: You might see the references listed at:

http://ftp.aset.psu.edu/pub/ger/fortran/hdk/nan.f90

Skip Knoble

On Thu, 12 Jul 2007 18:37:50 -0700, Ralf Schaa wrote:

Ralf Schaa wrote:

Ralf, ...