Re: PATCH: Start Fortran support for variable objects.

[Daniel Jacobowitz <drow@false.org>]

On Mon, Jul 04, 2005 at 01:41:13PM +1200, Nick Roberts wrote:
>  > >  > First of all, never reference ->main_type - see above for the right way
>  > >  > to get the low bound.  An even better way (it seems) is to call
>  > >  > get_discrete_bounds.  Take a look at value_subscript for an example.
>  > > 
>  > > Better than TYPE_LOW_BOUND?
>  > 
>  > I suppose.  I don't know which one is preferred; some day, someone
>  > should go through and clean them all up to be consistent.  I'm fine
>  > with either choice.
> 
> To keep things simple I've used TYPE_LOW_BOUND.  I've tested with the examples
> I've posted before and it works.  Presumably there should also be a test case,
> so I'll create one for mi-var-child.exp and mi2-var-child.exp (you still
> haven't approved my patch for mi2-cmd-stack.exp (28 Jun 2005 01:53:52 +1200).

You posted nothing to gdb-patches on June 27th, 28th, or 29th (except
for the first version of this patch).  I vaguely remember seeing a
patch on gdb@ when Mark complained about your introducing regressions.
But if you'd like it approved, please post it to the patches list; I am
methodical about processing gdb-patches mail because it has a clearly
defined request-reply format, and gdb@ discussions tend to wander off
on tangents (like that one did).

BTW, found the patch in the archives - the changelog entry is for the
wrong file.  Also, can we just remove the failing test, instead of
adding new tests to mi2?  We really need to get a coherent story
together on what "is" mi2, but I don't think we need to add tests for
new commands to it.

> 2005-06-30  Nick Roberts  <nickrob@snap.net.nz>
> 
> 	* varobj.c (varobj_list_children): Allow non-zero offsets for
> 	languages like Fortran.

Retcode is unused.

Can't we get here with struct types?  In which case this will blow up:

> +      j = i + TYPE_LOW_BOUND (TYPE_INDEX_TYPE (var->type));
> +
>        /* check if child exists, if not create */
> -      name = name_of_child (var, i);
> +      name = name_of_child (var, j);
>        child = child_exists (var, name);
>        if (child == NULL)
> -	child = create_child (var, i, name);
> +	child = create_child (var, j, name);
>  
>        *((*childlist) + i) = child;
>      }

Also, I'm beginning to wonder if you're doing this in the right place. 
Not that it matters a whole lot, but index is 0-based in every other
case, including for structs.  Maybe the children of arr(4) should be
arr.0 == arr(1), arr.1 == arr(2), arr.2 == arr(3), arr.3 == arr(4). 
Then you'd add the lower bound in c_value_of_child.  Does that work?
Do you have an opinion on which is "more right"?

-- 
Daniel Jacobowitz
CodeSourcery, LLC