Lars T. Kyllingstad: > real.init is a signaling NaN, real.nan is not. I don't know if this is > by design, but I suppose it may be: Thank you for your probably correct hypothesis :-) Bye, bearophile