(j3.2006) REAL with NaN argument

[Malcolm Cohen]

>Should one expect
>
>  double precision :: D
>  D = IEEE_Value(1.0e0,IEEE_Signaling_NaN)
>
>to produce a double-precision signaling NaN?

No.

"any operation signaling an invalid operation exception and for which a 
floating-point result is to be delivered shall deliver a quiet NaN"

and

"Signaling NaNs shall ... signal the invalid operation exception ... for every 
general-computational and signaling-computational opertion except for the 
conversions listed in 5.12"

Converting one floating-point format to another is a general-computational 
operation described in 5.4.2, therefore converting single precision SNaN to 
double precision is required to signal (and return a quiet NaN).

Of course that's for IEEE-compliant conversions, which we don't actually 
require, so you're not *guaranteed* to get the exception signalled plus a quiet 
NaN, but it seems likely on most systems.

>7.2.1.2, 7.2.1.3 (in particular, tables 7.8 and 7.9), and 13.7.138,
>don't mention IEEE exceptional values.

Good.

>  Should they?

No.

Cheers,
-- 
................................Malcolm Cohen, Nihon NAG, Tokyo.