Re: Using exceptions in the core

[John W. Eaton]

On 10/02/2015 06:14 PM, John W. Eaton wrote:
> On 10/02/2015 03:15 PM, John W. Eaton wrote:
> > On 10/02/2015 02:39 PM, Rik wrote:
> >
> > > I also like the overloaded _value() method idea.  We could start out
> > > small by just overloading the string_value () method since this seems to
> > > be one of the most frequent occurrences where the function expected a
> > > string, for an option maybe, but numeric input was given.
> >
> > Yes, I think this is the most frequent case of something like
> >
> >    if (! args(0).is_string ())
> >      {
> >        error ("...");
> >        return retval;
> >      }
> >
> > so I think it makes sense to start with this one and see what we think.
> >   I'll work on this with my other changes.
>
> My current idea is attached.  With it, you can write something like
>
>    std::string url = args(0).string_value ("urlwrite: URL must be a
> string");
>
> Comments?

Any comments on this proposal?

I've now taken an initial pass through all of the Octave sources and 
have removed many simple uses of error_state.  I primarily went after 
things like

  if (error_state)
    return retval;

  if (! error_state)
    {
      ...
    }
  // No else or side effect.

We are now left with about 490 more uses (down from over 1600, I think). 
 Many of the remaining are of the form

  std::string = args(0).string_value ();
  // Or some other value extraction.
  if (error_state)
    {
      error ("more specific error message");
      return retval;
    }

These will mostly be easy to remove but we should probably agree on 
exactly how we would prefer to handle the error.  I suppose either 
overloading the various value extractors or writing some simple wrapper 
functions.  Either way, the goal is really the same, to encapsulate the 
use of try/catch in a few places instead of sprinkling them all over the 
code.

jwe