https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117805
--- Comment #20 from mjr19 at cam dot ac.uk --- I am not convinced that gfortran's current behaviour is wholly consistent with what a mathematician would reasonably expect. When I was taught complex arithmetic, multiplication by one and addition of zero were both identity operations. But in gfortran they can move one across a branch cut. program test print *,sqrt(-(1.0,0.0)) ! Below branch cut print *,sqrt(-(1.0,-0.0)) ! Above branch cut print *,sqrt(-(1*(1.0,-0.0))) ! Now below branch cut print *,sqrt((-1.0,-0.0)) ! Below branch cut print *,sqrt((-1.0,-0.0)+0) ! Now above branch cut end program test I don't think it is possible to produce rules for complex arithmetic based on a Cartesian representation with signed zeros which do not produce surprises similar to the above.
