Re: Behavior of feval when it fails?

[John W. Eaton]

On 11/02/2016 05:40 PM, Rik wrote:
> All,
>
> I was looking into a bug report and found that the feval function fails
> silently.  See test code below.
>
> +verbatim+
> octave:1> feval ('sin', 1)
> ans =  0.84147
> octave:2> feval ('sinABC', 1)
> octave:3>
> -verbatim-
>
> It seems like feval should most likely throw an error if the function
> doesn't exist.  But maybe we get in to weird loops where feval is called
> repeatedly.  The code for feval is in oct-parse.cc.
>
> +verbatim+
> octave_value_list
> feval (const std::string& name, const octave_value_list& args, int nargout)
> {
>   octave_value_list retval;
>
>   octave_value fcn = symbol_table::find_function (name, args);
>
>   if (fcn.is_defined ())
>     retval = fcn.do_multi_index_op (nargout, args);
>   else
>     {
>       try
>         {
>           maybe_missing_function_hook (name);
>         }
>       catch (octave::execution_exception& e)
>         {
>           error (e, "feval: function '%s' not found", name.c_str ());
>         }
>     }
>
>   return retval;
> }
> -verbatim-
>
> and the maybe_missing_function_hook function calls feval.
>
> +verbatim+
> void maybe_missing_function_hook (const std::string& name)
> {
>   // Don't do this if we're handling errors.
>   if (buffer_error_messages == 0 && ! Vmissing_function_hook.empty ())
>     {
>       octave_value val = symbol_table::find_function (Vmissing_function_hook);
>
>       if (val.is_defined ())
>         {
>           // Ensure auto-restoration.
>           octave::unwind_protect frame;
>           frame.protect_var (Vmissing_function_hook);
>
>           // Clear the variable prior to calling the function.
>           const std::string func_name = Vmissing_function_hook;
>           Vmissing_function_hook.clear ();
>
>           // Call.
>           feval (func_name, octave_value (name));
>         }
>     }
> }
> -verbatim-
>
> Maybe the Vmissing_function_hook function should return a value that
> indicates whether the hook handled the exception or not.  If it didn't then
> it could report that back to the caller?  Anyone have a good idea on this?

As John D noted on the tracker, it used to generate an error if feval 
was called for a nonexistent function.  I assume this is an 
unintentional change in behavior that happened when error_state was 
eliminated.

I can look at it if you want me to.  My first step might be to try using 
gdb (set up to catch exceptions when they are thrown) to see exactly 
what path is taken when feval is called for a missing function.

jwe