fix writesdpa.m sedumi form: A vs At

[plusk01]

Using YALMIP , the following model is parsed correctly:

y = sdpvar(4,1,'full');
A0 = [0 0; 0 0];
A1 = [1 0; 0 0]; A2 = [0 1; 1 0]; A3 = [0 0; 0 1]; A4 = -eye(2);
G = y(1)*A1 + y(2)*A2 + y(3)*A3 + y(4)*A4 - A0;
A = [y(1) y(2); y(2) y(3)];
F = [];
F = [F, G <= 0];
% F = [F, norm(A,2) <= 10];
optimize(F,y(4),sdpsettings('solver','csdp'))

But when the sedumi form is passed to csdp.m to write in spda form, it is parsed incorrectly, with writespda.m complaining about A being nonsymmetric:

Expanding c to the appropriate size
Number of constraints: 4 
Number of SDP blocks: 1 
Number of LP vars: 0 
Non symmetric A.s matrix! 
Non symmetric A.s matrix! 
Non symmetric A.s matrix! 

YALMIP is correctly passing At,b,c,K to csdp.m, which has the following function signature:

%
% [x,y,z,info]=csdp(At,b,c,K,pars,x0,y0,z0)
%
% Uses CSDP to solve a problem in SeDuMi format.

However, it appears that writespda.m is not actually taking in sedumi (i.e., A vs At):

%  This function takes a problem in SeDuMi MATLAB format and writes it out 
%  in SDPA sparse format.  
%
%  Usage:
%
%  ret=writesdpa(fname,A,b,c,K,pars)
%
%      fname           Name of SDPA file, in quotes
%      A,b,c,K         Problem in SeDuMi form

It seems that most of the time everything works correctly because of the A vs At check in writespda.m; except for when A is square, like in the following sdpa file (corresponding to the model listed above):

4 
1 
2 
0 0 0 1
1 1 1 1 -1
2 1 1 2 -1
3 1 2 2 -1
4 1 1 1  1
4 1 2 2  1

[CLAassistant]

All committers have signed the CLA.