[PPL-devel] Bug in ppl_new_LinExpression_with_dimension

[Roberto Bagnara]

Pedro Vasconcelos wrote:
> While experimenting with the Haskell PPL binding, I noticed that the C
> interface function "ppl_new_LinExpression_with_dimension" always creates
> an expression with one plus the specified dimension. The following code
> illustrates the problem: 
> 
> #include <ppl_c.h>
> 
> main () {
>   ppl_LinExpression_t e;
>   ppl_dimension_type dim_in=3, dim_out;;
> 
>   ppl_initialize();
> 
>   ppl_new_LinExpression_with_dimension(&e, dim_in);
>   dim_out = ppl_LinExpression_space_dimension(e);
>   printf("%d, %d\n", dim_in, dim_out);    /* 3, 4 rather than 3,3 as expected */
> 
>   ppl_finalize();
> }
> 
> Looking at the source code interface/C/ppl_c.cc, it becomes clear that
> the expression created has a single Variable(d), therefore dimension d+1:
> 
> int
> ppl_new_LinExpression_with_dimension(ppl_LinExpression_t* ple,
> 				     ppl_dimension_type d) try {
>   *ple = to_nonconst(new LinExpression(0*Variable(d)));  /* <----- should be Variable(d-1) */
>   return 0;
> }
> CATCH_ALL
> 
> This is of course trivial to correct, but I thought it best to report so
> that it gets corrected in the newer releases.

Dear Pedro,

thanks a lot for your report.  You spotted a genuine bug in the
C interface.  The fix is almost as you indicate, with the exception
that the case where d == 0 must also be handled properly.  The
patch I applied to the CVS head version is the following:

   http://www.cs.unipr.it/cgi-bin/cvsweb.cgi/ppl/interfaces/C/ppl_c.cc.diff?cvsroot=ppl&r1=1.110&r2=1.111

(notice that the interface of ppl_new_LinExpression_with_dimension
has changed since the release of PPL 0.6.1).
All the best,

     Roberto

-- 
Prof. Roberto Bagnara
Computer Science Group
Department of Mathematics, University of Parma, Italy
http://www.cs.unipr.it/~bagnara/
mailto:bagnara at cs.unipr.it