This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [ob] unbreak MI


On Tuesday 27 November 2007 09:27:37 Nick Roberts wrote:
>  > > Generally, with a NULL pointer, or and address that can't be dereferenced,
>  > > MI prints out the value field as value="".
>  > > 
>  > > What is the problem in this case?  Why isn't the right fix to add a
>  > > check_typedef somewhere?
>  > 
>  > check_typedef? The original problem was that check_typedef was getting
>  > called on NULL pointer, so adding more check_typedef calls won't help.
>  > Probably:
>  > 
>  >         if (!gdb_type)
>  >                 ui_out_field_string (uiout, "value", "");
>  >         else if (mi_print_value_p (gdb_type, print_values))
>  >                 ui_out_field_string (uiout, "value", varobj_get_value (var));
>  > 
>  > is the right logic?
> 
> It's probably the right logic, but it seems to cure the symptom rather than the
> cause.  What I mean't, I guess, was where/how does check_typedef is get passed
> a NULL pointer?  And can't that call be conditioned (i.e. "add a *check* to
> check_typedef") , e.g., something like:
> 
> if (!gdb_type)
>    check_typedef (gdb_type)

Just look at mi_print_value_p, and you'll see a call to check_typedef. Actually,
the code previously looked like:

	if (type != NULL)
		type = check_typedef (type);

It was changed in revision 1.38, with the following comment:

	2007-08-28  Michael Snyder  <msnyder@access-company.com>

	* mi/mi-cmd-var.c (mi_print_value_p): No longer necessary to
	check for null before calling check_typedef.

However, apparently check_typedef still crashes when passed NULL,
and it can be passed NULL.

The original code, in fact, was in error too, because of this:

  return (TYPE_CODE (type) != TYPE_CODE_ARRAY
	  && TYPE_CODE (type) != TYPE_CODE_STRUCT
	  && TYPE_CODE (type) != TYPE_CODE_UNION);	

This will crash if 'type' is NULL. Testsuite fails to detect this because presently
type is NULL only for C++ pseudo-fields  ('public'/'private') and the code
above is only executed for --simple-values.

Does this clarify things?

- Volodya


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]