likPoisson fails on Octave

The following code, slightly modified from the Getting Started code for a Poisson likelihood fails on Octave:

x = gpml_randn(0.8, 20, 1);                 % 20 training inputs
z = sin(3*x) + 0.1*gpml_randn(0.9, 20, 1);  % 20 noisy training targets
y = randp(z)';
xs = linspace(-3, 3, 61)';                  % 61 test inputs

meanfunc = @meanConst;                    % Constant mean function
covfunc = @covSEiso;              % Squared Exponental covariance function
likfunc = @likPoisson;              % Poisson likelihood

hyp = struct('mean', 0, 'cov', [0 0], 'lik', []);

hyp2 = minimize(hyp, @gp, -100, @infLaplace, meanfunc, covfunc, likfunc, x, y);

The error message:

warning: Inference method failed [xnorm: wrong type argument 'sq_string'] .. attempting to continue

I've tried dev, master, and octave branches. Any help would be much appreciated.