The Communications manual says "Other Octave functions work correctly with Galois Fields and so overloaded
versions are not necessary. This include such functions as size and polyval."

Clearly, this is no longer the case. The culprit could be http://hg.savannah.gnu.org/hgweb/octave/rev/23523700b6b8 ("polyval.m:  ...Use isfloat to require that inputs be floating point numeric (Matlab compatibility)."), or maybe an earlier change to polyval.

For now I commented out the failing tests in comms. In the longer term, it would be nice to have an overloaded version of polyval for Galois field inputs.

Running "test comms" in octave 10 and with hg snapshot of communications I see:

polyval: P must be a numeric floating point vector

This occur at

y0 = polyval (poly1, x0);

and perhaps later in

ck1 = polyval (poly1, roots1);

"poly1" is a Galois field array:

octave:5> poly1 = gf ([2, 4, 5, 1], 3)
poly1 =
GF(2^3) array. Primitive Polynomial = D^3+D+1 (decimal 11)

Array elements =

   2   4   5   1

octave:6> whos poly1
Variables visible from the current scope:

variables in scope: top scope

  Attr   Name        Size                     Bytes  Class
  ====   ====        ====                     =====  =====
         poly1       1x4                         16  galois

Total is 4 elements using 16 bytes

Apparently the code worked before, but not anymore.
Not sure what have changed and what is the proper fix.

Dmitri.
--