Re: RFC: MI - Detecting change of string contents with variable objects

[Vladimir Prus <ghost at cs dot msu dot su>]

On Thursday 04 January 2007 23:34, Nick Roberts wrote:
>  > > + 	      char* print_value = value_get_print_value (value, var->format);
>  > 
>  > Use "char *print_value", please.
> 
> OK.
>  
>  > >   	      gdb_assert (!value_lazy (var->value));
>  > >   	      
>  > > ! 	      if (var->print_value)
>  > > ! 		{
>  > > ! 		  if (strcmp (var->print_value, print_value))
>  > > ! 		    {
>  > > ! 		      xfree (var->print_value);
>  > > ! 		      var->print_value = print_value;
>  > > ! 		      changed = 1;
>  > > ! 		    }
>  > > ! 		}
>  > > ! 	      else
>  > > ! 		var->print_value = print_value;
>  > 
>  > Should we set changed = 1 in the "else"?
> 
> No.  This is only reached in the initial call to install_new_value i.e
> -var-create.

How so? For -var-create the 'initial' parameter to install_new_value should be
1, so we have:

   if (!initial && changeable)
    {
       ................

  	      gdb_assert (!value_lazy (var->value));

              if (var->print_value)
               {
                 if (strcmp (var->print_value, print_value))
                   {
                     xfree (var->print_value);
                     var->print_value = print_value;
                     changed = 1;
                   }
               }
             else
               var->print_value = print_value;      
    }

It seems to be this code is *never* reached during call from -var-create.
In fact, I don't seem to understand how var->print_value will be initialized
by -var-create. 


>  > Otherwise the patch seems fine, if it tests OK, but I'm still a little
>  > nervous about it.  For example, you'll call val_print on a struct
>  > or array to see if it's changed.  Depending on things like "set print
>  > elements", that might not print out the whole string.  This is
>  > probably a behavior change.  Is it a harmless one?  If it is, then
>  > should we be sharing the code with c_value_of_variable that avoids
>  > printing structs, unions, and arrays, and never mark them as changed
>  > unless their types change?
> 
> The function val_print is already used for -var-evaluate-expression.  AFAICS
> "set print elements" has no effect on variable objects, perhaps because
> val_print is only called on the leaves but I can see that using it might
> expose MI to the vagaries of CLI.  Currently, however, string changes don't
> get reported at all, without the user changing configuration values.

The code in question is executed only for changeable values, and so will never
be executed for structures.

- Volodya