Re: Unnecessary initialization in octave_matrix::double_value ?

[Julien Bect]

Le 28/08/2016 à 16:17, John W. Eaton a écrit :
> On 08/28/2016 04:12 AM, Julien Bect wrote:
> > Hi all,
> >
> > I have a (probably naive) C++ question concerning the implementation of
> > octave_matrix::double_value ().
> >
> > It is currently written like this:
> >
> > double
> > octave_matrix::double_value (bool) const
> > {
> >   double retval = lo_ieee_nan_value ();
> >
> >   if (is_empty ())
> >     err_invalid_conversion ("real matrix", "real scalar");
> >
> >   warn_implicit_conversion ("Octave:array-to-scalar",
> >                             "real matrix", "real scalar");
> >
> >   retval = matrix(0, 0);
> >
> >   return retval;
> > }
> >
> > I don't understand the point of initializing retval with the result of
> > lo_ieee_nan_value ().  Would it be correct to simply write:
> >
> > double
> > octave_matrix::double_value (bool) const
> > {
> >   if (is_empty ())
> >     err_invalid_conversion ("real matrix", "real scalar");
> >
> >   warn_implicit_conversion ("Octave:array-to-scalar",
> >                             "real matrix", "real scalar");
> >
> >   return matrix (0, 0);
> > }
>
> You are right, that initialization is no longer needed.  It is left 
> over from when the error function did not throw an exception but just 
> printed a message and set a global warning state, then returned.  So 
> in the old days the function would always return something.  A NaN was 
> better than undefined.  Although much of the work has been done to 
> adapt to the new way that error works, there are still many small 
> things like this to clean up.

Ok, I will prepare a patch for ::float_value() and ::double_value().