Pushed https://sourceforge.net/p/octave/signal/ci/c6e0c8b249ef581af233c37c839b479057f8baca/ with change and additional test

The FIR part of residuez() are conjugate w.r.t. to their true values.

The only test that shows this (with a non-real result) asserts f wrongly to -2j: the true result is (+)2j:

%! B=[1,6,6,2]; A=[1,-(2+j),(1+2j),-j];
%! [r,p,f,m] = residuez(B,A);
%.....
%! assert(f,-2j,1E-6);
...

The test case was recalculated manually and with Python's scipy.signal.residuez: the FIR part comes out as (+)2j.

So the last line of the function

f = conj(fliplr(f));

should read

f = fliplr(f);

This behavior is carried over to residuez() (residuez.R) in R's gsignal package, BTW.