[patch] Fix LTO warning on types with variable offsets

[Richard Guenther]

On Sun, May 16, 2010 at 12:15 PM, Eric Botcazou <ebotcazou@adacore.com> wrote:
>> Note that merging too many types is as problematic as merging not enough
>> types.  I at least wanted to ask why you compare DECL_SIZE in
>> gimple_compare_field_offset when it explicitely says it will compare
>> the offset only.
>
> That's explained in the comment, isn't it?

Is it?

+              /* Once gimplification is done, self-referential offsets are
+                 instantiated as operand #2 of the COMPONENT_REF built for
+                 each access and reset.  Therefore, they are not relevant
+                 anymore and fields are interchangeable provided that they
+                 represent the same access.  */

This function is not about field interchangability but about fields
having the same offset.

>> The size check should be covered by the caller comparing the type of the
>> field, no?

Which means the represent the same access if the types are equal.

> No, think of bitfields.

I don't see how bitfields are any different.  We might want to compare
DECL_BIT_FIELD as well in the callers, and also compare
DECL_BIT_FIELD_TYPE in that case.

Richard.

> --
> Eric Botcazou
>