Re: What is the warning about builtin-function on gcc-4.6.0 ?

[Simon Wright <simon@pushface.org>]

Florian Weimer <fw@deneb.enyo.de> writes:

> * Simon Wright:
>
>> Florian Weimer <fw@deneb.enyo.de> writes:
>>
>>> * Simon Wright:
>>>
>>>> ytomino <aghia05@gmail.com> writes:
>>>>
>>>>> nn...then, there are no function to take infinity and NaN...
>>>>
>>>> 'Valid should do the trick.
>>>>
>>>>    FP_Value := 1.0 / 0.0;
>>>>    if FP_Value'Valid then
>>>>       --  it won't be
>>>
>>> This is a compiler bug.
>>
>> I don't understand what you mean.
>
> The standard requires that FP_Value'Valid is true.
>
>> If the compiler doesn't raise an exception on division by zero (it's
>> allowed not to, and GNAT doesn't) it will (in this case) set the result
>> to +Inf, and 'Valid will return False.
>
> My understanding is that an implementation must either raise
> Constraint_Error, or the evaluation of an expression must result in a
> valid value.

ARM 13.9.2 doesn't include division by zero as a means of producing an
invalid float.

However, A.5.3(12) ('Machine_Overflows) implies that Constraint_Error
doesn't _have_ to be raised. It doesn't say what happens - that's
implementation-defined, M.2(123) and G.2.1(13).

The GNAT RM
(http://gcc.gnu.org/onlinedocs/gcc-4.5.0/gnat_rm/Strict-Conformance-to-the-Ada-Reference-Manual.html#Strict-Conformance-to-the-Ada-Reference-Manual)
says

   "Note that the result of a floating point arithmetic operation in
   overflow and invalid situations, when the Machine_Overflows attribute
   of the result type is False, is to generate IEEE NaN and infinite
   values. This is the case for machines compliant with the IEEE
   floating-point standard, but on machines that are not fully compliant
   with this standard, such as Alpha, the -mieee compiler flag must be
   used for achieving IEEE confirming behavior (although at the cost of
   a significant performance penalty), so infinite and and NaN values
   are properly generated."


I don't see anything in the above that says that 'Valid should produce
False for Inf/NaN, but GNAT - on Intel & PowerPC hardware - certainly
does.