In Octave,

x = complex (0, 0);
x ^ x  => NaN + NaNi
0 ^ 0 => 1

but I believe Matlab the complex exponentiation also returns 1 (or complex (1,0) if narrowing doesn't apply).

Do we need to hunt down and eliminate all uses of std::pow for complex arguments?

Would it be better to eliminate direct use of std::complex and use our own class derived from std::complex and that overloads pow with a function that does what we need for compatibility?

Have we discussed this issue previously?  If so, I didn't find a bug report about it.