Ah wait you mentioned a branch sites test, right that makes sense. You've got two or more identical sequences in the analysis. Ensure there are no identical sequences in your alignments and ... PRESTO! It will work :-)

Below are my original thoughts before I re-read your question. I've kept the notes because it helps explain the theory a bit. There is no need to supply the tree, it should be resolved.

--------

The tree likelihood looks to be decreasing nicely, so it can see both the tree and the codons in the alignment.

At a guess the error relates to two or more indentical nucleotide sequences in your alignment. Thus zero non-synonymous sites (amino acid change) and zero synonymous (silent) sites. This error would only occur if you were calculating selection for each 'node' in the tree (internal branch).  Alternatively, possibly if the number of synonymous sites was zero, regardless of the non-synonymous mutations this error could appear, but this doesn't seem to fit the error output.

I don't know "Godon", but recommend `PAML` for this analysis. Admittedly it is complicated to deploy, but it is a very robust algorithm and hugely respected.