Problem with discrete colors in images using Gnuplot backend

[E. Joshua Rigler]

I'm trying to determine if an image plotting bug is caused by Octave,
Gnuplot, or some MS Windows strangeness.

On a linux box, I run Octave 3.4.2 with Gnuplot 4.4.2. I plot an image
with only 9 discrete integers (0-8) using a 9-row color table with rgb
values specified to be really obvious (e.g., red=[1 0 0], cyan = [0 1
1], black=[0 0 0], etc.). I set caxis([-.5 8.5]), so that each
discrete color is centered. I get the results I expect.

Now, on my fresh Octave for Windows installation (3.4.3, with Gnuplot
4.4.3), I do the same thing and get strange results. Some colors come
out as expected, but most are way off. I read in the Gnuplot 4.4.3
changelog that they 'fixed' a bug with discrete color palette. FWIW,
when I use the fltk graphics toolkit everything looks correct, but
there are other issues with fltk that prevent me from using it in
production.

I've tried looking into the Octave plotting back-end source code
previously and become totally confused. So, I'll just ask if there is
some sort of inline fix, perhaps to adjust image data or color tables
in Octave prior to sending the data to Gnuplot, that no longer works
correctly with the 'fixed' Gnuplot. My apologies for beginning this
inquiry by assuming the issue is with Octave, but that is simply what
I am most familiar with. I kind of suspect that the problem lies with
the 'fixed' Gnuplot or Windows, but the prospect of tracking that down
is daunting.

-EJR