Dear Rob,
mpc_abs(NaN+I*Inf) = Inf is consistent with hypot(qNaN,Inf) = Inf in IEEE 754,
Hi Paul,
Then, I suppose my point is that the mpc_abs() implementation is inconsistent with the basic definition of doing sqrt(NaN^2 + Inf^2).
Given that hypot(qNaN, Inf) = Inf, I wonder if hypot() is really the correct function for mpc_abs() to be using for these specific values.
Anyway - it's a fairly inconsequential quibble, and you guys are better judges of this than I am.
BTW, I didn't know that IEEE 754 had decreed that Inf trumps NaN in this one case.
Cheers,
Rob