Re: [Help-glpk] Got a wrong order of column by lpx_get_mat_row(lp, j, ndx, val)?

[Brady Hunsaker]

Michael Hennebry wrote:
> On Wed, 19 Jan 2005, [iso-2022-jp] $B%m!!%3%/%f%&(B wrote:
>
>
> > Dose anyone meet this problem:
> >
> > for example  in jth row :  a*x-y+b*z=0;
> >
> > by len=lpx_get_mat_row(lp, j, ndx, val), I found val=[-1,a,b], that mean the
> > order of column is y,x,z.
> > I added two new columns to this row like:
> >
> > ndx[len+1]=w, ndx[len+2]=v, val[len+1]=d, val[len+2]=e, len+=2
> > lpx_set_mat_row(lp,j,len,ndx,val)
> >
> > then I thought the jth row should be a*x-y+b*z+d*w+e*v=0, but when I tried
> > len=lpx_get_mat_row(lp, j, ndx, val) again,
> > I found it's d*w+e*v+a*x-y+b*z=0.
> >
> > I wonder whether the order of columns in row is fixed and how the order of
> > columns in a row is arranged there.
>
>
> My recollection is that there is nothing in the documentation
> that says lpx_get_mat_row will always return columns in the same order.
> Not even if you don't edit the problem.

I agree with Michael.  This issue came up when testing the GLPK 
interface to COIN-OR/OSI.  One of the tests for OSI fed in an instance 
and read some of the data back out.  It initially failed for GLPK 
because it expected the data in a different order, even though what GLPK 
is doing is perfectly correct.  In that case we had to modify the test 
itself.

--
Brady Hunsaker
Assistant Professor
Industrial Engineering
University of Pittsburgh
http://www.engr.pitt.edu/hunsaker/