> On Jun 11, 2025, at 17:09, Joseph Myers <josmy...@redhat.com> wrote:
>
> On Wed, 11 Jun 2025, Qing Zhao wrote:
>
>> Then how about the following case:
>>
>> typedef struct item3 Item3;
>> struct pointer_array_9 {
>>
>> int count3;
>> Item3 *array_3 __attribute__ ((counted_by (count3)));
>> }
>>
>> struct item3 {
>> int a;
>> float b[];
>> }
>>
>> In the above, the “Item3” is an incomplete type inside “struct
>> pointer_array_9”, whose definition is given
>> afterward.
>>
>> Theoretically, C FE should issue error for the above counted_by attribute.
>> However, when and how should we
>> Implement this in C FE:
>>
>> when we examine the counted_by attribute for “array_3” inside “struct
>> pointer_array_9”, there is no any information
>> on the TYPE of “Item3” yet, only when we see the definition of “struct
>> item3” afterward, we know that this structure
>> includes a FAM, the previous “counted_by” attribute should not be added for
>> “array_3”.
>>
>> Is there a way to trace back to “struct pointer_array_9” when the definition
>> of “struct item3” is parsed?
>> When should we issue error for such case?
>
> In general I think we'd only expect an error if the information required
> to give it is visible at the point where the counted_by attribute is used.
> There might be a possibility of giving an error for this case when the
> pointer gets dereferenced (and thus the counted_by information gets
> implicitly used).
Okay, I see.
I think that the right place to check and issue error for this is:
When generating the call to .ACCESS_WITH_SIZE for a component_ref with
counted_by attribute, if the pointee’s TYPE
is still incomplete at this moment, C FE issues error at this time. And the
call to .ACCESS_WITH_SIZE will not be generated.
WDYT?
thanks.
Qing
>
> --
> Joseph S. Myers
> josmy...@redhat.com
