Bug: snopt interface for problems with "few" constraints

[hel-fried]

Dear Johan,

I am currently toying with snopt and yalmip and stumbled upon a dimension issue for problems with "few" constraints.

Example:

x = sdpvar(2,1)

obj = 0.5*(100*(x(2)-x(1)^2)^2 + (1-x(1))^2)
c = [-0 <= x <= 2]

optimize(c, obj, sdpsettings('solver', 'snopt', 'debug', 1))`

which yields on my installation (linux, Yalmip 20230622):

Error using snoptmex
iGfun has incorrect column dimension     2.  Should be length     1

Error in solve_snopt (line 17)
 xstate,Fstate,itn,mjritn] = snoptmex(mexopt, ...
                             ^^^^^^^^^^^^^^^^^^^^
Error in snopt (line 465)
 xstate,Fstate,output] = solve_snopt(istart, stopFun, name, ...
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Error in callsnopt (line 89)
    [xout,F,inform,xmul,Fmul] = snopt(x0, xlow, xupp, xmul, xstate,Flow, Fupp, Fmul, Fstate, usrf,ObjAdd, ObjRow,A,G,model.options.snopt);
                                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Error in solvesdp (line 420)
    eval(['output = ' solver.call '(interfacedata);']);
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Error in optimize (line 31)
[varargout{1:nargout}] = solvesdp(varargin{:});
                         ^^^^^^^^^^^^^^^^^^^^^

I haven't fully looked into the root cause but I think iGfun should be transposed in certain cases. Without really knowing what I'm doing I hacked the following two lines into snopt.m

if size(iGfun,1) == 1
    iGfun = iGfun'
    jGvar = jGvar'
end

which did the trick for me...

Best
Helfried

[johanlofberg]

Confirmed