The problem was in my commit. It turns out that Matlab is more permissive than Octave. In set_prior.m, I used initialization of cellarray bayestopt_.name to [], when I should have used {} In Matlab a= [[]; {'sss'}] is a 1x1 cellarray, the same as [{}; {'sss'}] In Octave a=[[]; {'sss'}] is a 2x1 cellarray with an empty matrix as the first element and a=[{}; {'sss'}] is a 1x1 cellarray I think Octave does the right thing Best Michel On 06/14/2012 10:28 AM, Marco Ratto wrote:

Hi Sebastien,

I have been stacked until today for family problems. I will hande these issues as soon as I can thanks for tracking the problems, this help a lot

best Marco

On 6/14/2012 10:24 AM, Sébastien Villemot wrote:

...

Hi Marco,

The following two tests are currently broken in master and 4.3 (only under Octave):

* identification/kim/kim2.mod * identification/as2007/as2007.mod

I tracked down the problem to commit 3ff832ddccd085426eaa79fd0f29ae3e9400eac5 (using bisection). I fail to understand how this commit can break the tests.

Also note that gsa/ls2003.mod randomly fails on both MATLAB and Octave. This is probably of lesser priority.

Best,