[Cfp-interest 3174] atan2()

[Damian McGuckin]

Looking at a range of libraries' implementations of atan() and atan2(), 
and especially those based on FDLIBM from Sun, I notice that all of the 
atan2(y, x) violate the now standard for the case of y/x being subnormal. 
But atan(x) follows the now standard for x being subnormal.

I do note that in N1124.pdf (2005), neither routine specified that a range 
error occurred.

When did they change and did both change at the same time?

Thanks - Damian