Suppose there is only one parameter estimated, with bound constraints (no other constraints). Then line 606 of cmlmt.src will create np=1. By default, qp_a=nan. Then line s 1146-1147 (if scalmiss(qp_a); qp_a = zeros(1,np-1)~1;) creates trouble "Argument out of range " because of zeros(1,np-1).
CMLMT2.0 has been released for several years. I paid for this, yet it is so disappointing! (I pinned down another serious bug a couple of years ago regarding MLE/QMLE robust covariance matrix; it appears that it is still there!)
What version of CMLMT to you have. The current version, 3.0.1, has the following for handling the single parameter model.
if np > 1;
qp_a = zeros(1,np-1)~1;
qp_a = 1;
qp_b = -1e256;
numeq = 0;
With regard to the QML bug you mentioned, could you pass on a more specific description to email@example.com? Thank you.