Re: Incorect result in Gnumeric solver

[Jean Bréfort <jean brefort normalesup org>]

Le samedi 19 mars 2005 à 13:45 +0100, Michal Kaukic a écrit :
>   According to my last message (Incorrect result in solver),
> when LP-solver was given an objective function of the form:
>
>   sum   (c_i * x_i) + K
> (i=1..n)
>
> the original problem was transformed into incorrect one,
> the first variable x_1 having decreased the coefficients in
> matrix of constraints by -K.
> The constant K is also stripped from obj. function, which is O.K.
>
>   We are talking about Gnumeric 1.4.3 official version sources.
>   I replaced in solver.c the line 419 which reads:
>
>                 x0 = base;
>
> with lines:
>
>                 cell_eval (target);
>                 x0 = base = value_get_as_float (target->value);
>
>   I have no full (maybe, nor partial :-) understanding of source code,
> but x0=base referred here to "base" - variable assigned on line 363,  
> when we dealt with objective function. Now, the constraints settings  
> for solver are to be made  - why we nead refer to something 
> about obj. function?
>
>   The above-mentioned change fixed for me the problem. Maybe, it breaks
> something else, etc. But, I think, this additional information can help
> the developers to identify the problem (it must be either 
> in function "get_lp_coeff", or in something like above    
> - incorrect initialization of variables.)
>
>                                      Mike

I confirm the problem, please, fill a bug report. Bugzilla
(http://bugzilla.gnome.org) is a much better place than the mailing list
to discuss bugs, and this IS a bug.

Thanks,
Jean