Daniel Jacobowitz wrote:

> On Sun, Dec 31, 2006 at 03:01:42AM +0300, Vladimir Prus wrote:
>> @@ -1061,6 +1061,14 @@ varobj_update (struct varobj **varp, str
>>    if (new == NULL)
>>      {
>>        (*varp)->error = 1;
>> +      /* Also set the value to NULL, so that
>> +     when the value becomes valid in future,
>> +     -var-update notice the change.  */
>> +      if ((*varp)->value != NULL)
>> +    {
>> +      value_free ((*varp)->value);
>> +      (*varp)->value = NULL;
>> +    }
>>        return -1;
>>      }
>>  
> 
> "in the future", "notices the change".  Should we be doing this
> everywhere that we set var->error (probably add a new helper to do
> that)?  There's about a half-dozen places.

I believe this was the only place where we don't set value to NULL on
errors. This reminded me that I wanted to remove var->error for quite some
time -- it's used in a single place, and using 'value == NULL' to indicate
error is quite as reasonable.

What about the attached?

- Volodya

2006-12-31  Daniel Jacobowitz  <dan@codesourcery.com>
            Vladimir Prus  <vladimir@codesourcery.com>

        Fix 'selected frame' varobjs.
        * varobj.c (struct varobj): Remove the error field.
        (varobj_set_value): Don't check var->error.
        (install_new_value): Don't set var->error.
        (varobj_update): Always pass the new value
        of the root via install_new_value.
        (create_child): Don't set error field.
        (new_variable): Likewise.
        (c_value_of_root): Always reevaluate the value
        of selected frame varobjs in the selected frame.
        Don't call reinit_frame_cache.