11706 vs ([lno] Canonical iv creation)

[Nathan Myers]

Apologies if I am missing the essence of this discussion, I just
wanted to make sure a few points are not missed...

First, the requirement on ::pow vs. std::pow is that, whatever else
you do, for the "pow(double, double)" version, &::pow == &std::pow.

Second, for <cmath>, ::pow may be visible, but need not be.  (In 
principle it's better if it's not visible, but we haven't done that
for anything else.)

Third, if we provide a pow (and other things) for complex, it would
surprise users less if both <cmath> and <complex> at least declare 
the same set of operations -- i.e., both should include something 
like bits/complex_fwd.h.  Otherwise, very surprising things can 
happen, when part of a set of overloads is seen, and then used (e.g. 
in a user header), and then the rest of the overloads appear.  

Gaby probably understands all this in detail, but I'm not sure if
he's explained it to everybody who needs to know.

Nathan Myers
ncm-nospam@cantrip.org