[PATCH v2] Replace copysign (0,-1) with -0.0 in initializer

[Maciej W. Rozycki]

On Tue, 17 Dec 2024, H.J. Lu wrote:

> > OK but let's see what Maciej says.
> 
> There are other places where -0.0 is used:
> 
> stdlib/strtod_l.c:      RETURN (negative ? -0.0 : 0.0,
> stdlib/strtod_l.c:       RETURN (negative ? -0.0 : 0.0, tp);
> stdlib/strtod_l.c:          result = negative ? -0.0 : 0.0;
> stdlib/strtod_l.c:    return negative ? -0.0 : 0.0;

 Thank you for the reference, however existing use may not be sufficient 
an argument.

 The sign of 0.0 doesn't matter in many computational scenarios (of course 
it does in some or we wouldn't have it), and the purpose of these test 
cases is to exercise standards compliance of output conversion where the 
sign does matter, rather than the numeric value where it may or may not.

 Existing use indicates GCC gets it right, but it would be a shame if we 
lost coverage because a compiler implementation ignores the sign of zero 
in initialisation.

> If there are no objections, I will check it tomorrow.

 I was not able to track down a definitive standards reference requiring 
IEEE 754 compiler implementations to respect the sign of zero in literals, 
would you or anyone else be able to point me at one?

 Otherwise I wonder whether the test cases just ought to set the sign bit 
of the expected negative part of the array at run time, via a constructor.

  Maciej