Re: [Help-gsl] possible bug for vectors and matrices

[Jerome BENOIT]

Hello List:

On 02/06/12 15:34, Rhys Ulerich wrote:
> Hi Jerome (and bug-gsl added on CC),
>
> > I have just gone through some gsl_(vector|matrix)_TYPE.h headers (with
> > TYPE=long, long_double, ...). Functions that scale by or add a
> > constant has a *double* as constant argument: should this argument
> > constant not be in type TYPE rather than in type double?
>
> Just to be concrete, you're saying all of
>
> int gsl_vector_long_double_scale (gsl_vector_long_double * a, const double x);
> int gsl_vector_long_double_add_constant (gsl_vector_long_double * a,
> const double x);
> int gsl_matrix_long_double_scale (gsl_matrix_long_double * a, const double x);
> int gsl_matrix_long_double_add_constant (gsl_matrix_long_double * a,
> const double x);
> int gsl_matrix_long_double_add_diagonal (gsl_matrix_long_double * a,
> const double x);

Indeed I meant these one, but also the other alike ones:

int gsl_vector_long_scale (gsl_vector_long * a, const double x);
int gsl_vector_long_add_constant (gsl_vector_long * a, const double x);

and so forth.

> should be taking long doubles as their second argument?  Likewise for
> other types?  I agree.
>
> For anything "smaller" than double I doubt that the bug will cause a
> loss of precision.

For consistency, it makes sense;
second, as regard the source/implementation part,
as noticed above, it would be simpler to just play
with the adequate type.

  But for long doubles and above it is indeed a
> problem.  Either way, the API isn't right.

That was my point.

> Anyone have an objection if I throw this in a ticket on savannah and
> take a stab at fixing it by correcting the templating?  (I think the
> right template macro is BASE rather than TYPE, but that's an
> implementation detail).

> - Rhys

Jerome