problem with consistency of root finding and sorting

[Ben Abbott]

I have Octave 2.9.14 on Mac OS X (intel & PPC) installed via Fink and  
am concerned that my installations are giving different results than  
others.

During the process of modifying residue.m so that it would function  
in a manner consistent with Matlab's I've become suspicious that my  
Octave installation  does not work the same as others.

The difference is small, but is troublesome.

Given the following short example,

        r = sort(roots([1 0 18 0 81]))
        r_err = r - round(r)

I obtain the results below

r = sort(roots([1 0 18 0 81]))
r =

   0.0000 - 3.0000i
   0.0000 + 3.0000i
  -0.0000 - 3.0000i
  -0.0000 + 3.0000i

r_err = r - round(r)
r_err =

  2.5814e-08 + 1.6920e-08i
  2.5814e-08 - 1.6920e-08i
  -2.5814e-08 - 1.6920e-08i
  -2.5814e-08 + 1.6920e-08i

I suspect that others are getting slightly different results.  
Specifically, some installations appear to be returning conjugate of  
"r" instead. I've posted the "r_err" because it appears larger than  
I'd expect.

I'd appreciate if others would try to short script and post the  
result ... in the hope of isolating the problem.

TiA