[Cfp-interest 2607] Re: [SC22WG14.22883] N3071 exactly representable

[Jim Thomas]

Right. the statement and example are not consistent. 

5.2.4.2.2 #21 says "If an optional unary + or − operator followed by a signaling NaN macro is used as the initializer for initializing an object of the same type that has static or thread storage duration, the object is initialized with a signaling NaN value.” I believe this applies to initialization of parts of a complex object. If so, the example should be:

constexpr double _Complex dc1 = DBL_SNAN; // ok

One problem here is that 5.2.4.2.2 #21 doesn’t cover constexpr but that should be fixed, e.g. by inserting “or that is declared with storage-class specifier constexpr” after "that has static or thread storage duration” (as has been proposed to accommodate constexpr in other place).

- Jim Thomas

> On Jan 13, 2023, at 2:35 PM, Joseph Myers <joseph at codesourcery.com> wrote:
> 
> I'd like to make one comment on the changes CFP proposed for this issue in 
> N3081:
> 
> The proposed normative wording says "If the initializer has real type and 
> a signaling NaN value, the unqualified versions of the type of the 
> initializer and the corresponding real type of the object declared shall 
> be compatible.".  But then the examples include
> 
> constexpr double _Complex dc1 = DBL_SNAN; // constraint violation
> 
> and I think showing that example as a constraint violation is not 
> consistent with the "corresponding real type" part of the normative 
> wording - that is, this example might be shown as "ok", or conceivably 
> argued to be unspecified whether it's OK, but shouldn't be shown as 
> required to be a constraint violation given the normative text proposed.
> 
> -- 
> Joseph S. Myers
> joseph at codesourcery.com