Problems with propagating Gaussian covariance matrices

[iron76]

Dear dimple developers,
I am experiencing a problem when using the Gaussian solver with Multivariate variables. In particular the following code:

fg = FactorGraph();
fg.Solver = 'Gaussian';
N = 3;
x1 = RealJoint(N); x2 = RealJoint(N); q12 = RealJoint(N);
x1_meas = 1;
x1.FixedValue = ones(N,1)._x1_meas;
q12_meas = 3;
q12.FixedValue = ones(N,1)._q12_meas;
fg.addFactor(@add, q12, x1, x2);
fg.solve();

does not return a zero covariance matrix for the variable x2.Belief which is instead expected since q12 = x1 + x2 with q12 and x1 given. I get instead a covariance matrix eye(3)*1e-7 which is then propagated causing huge numerical errors. Anyone know how this can be solved?
Francesco

[analog-cbarber]

Internally this is https://bugzilla.cdcdesign.analog.com/show_bug.cgi?id=53