Re: [XEN PATCH 3/5] xen/sort: address violations of MISRA C:2012 Rule 8.2

[Jan Beulich <jbeulich@xxxxxxxx>]

On 22.11.2023 02:19, Stefano Stabellini wrote:
> On Tue, 21 Nov 2023, Jan Beulich wrote:
>> On 21.11.2023 01:04, Stefano Stabellini wrote:
>>> On Mon, 20 Nov 2023, Jan Beulich wrote:
>>>> On 20.11.2023 14:13, Federico Serafini wrote:
>>>>> On 20/11/23 10:02, Jan Beulich wrote:
>>>>>> On 17.11.2023 09:40, Federico Serafini wrote:
>>>>>>> --- a/xen/include/xen/sort.h
>>>>>>> +++ b/xen/include/xen/sort.h
>>>>>>> @@ -23,8 +23,8 @@
>>>>>>>   extern gnu_inline
>>>>>>>   #endif
>>>>>>>   void sort(void *base, size_t num, size_t size,
>>>>>>> -          int (*cmp)(const void *, const void *),
>>>>>>> -          void (*swap)(void *, void *, size_t))
>>>>>>> +          int (*cmp)(const void *key, const void *elem),
>>>>>>
>>>>>> Why "key" and "elem" here, but ...
>>>>>>
>>>>>>> +          void (*swap)(void *a, void *b, size_t size))
>>>>>>
>>>>>> ... "a" and "b" here? The first example of users of sort() that I'm
>>>>>> looking at right now (x86/extable.c) is consistent in its naming.
>>>>>>
>>>>>
>>>>> On the Arm side there are {cmp,swap}_memory_node() and
>>>>> {cmp,swap}_mmio_handler(): "key"/"elem" are used for the comparison
>>>>> and "_a"/"_b" for the swap.
>>>>
>>>> So - re-raising a question Stefano did raise - is Misra concerned about
>>>> such discrepancies? If yes, _all_ instances need harmonizing. If not, I
>>>> see no reason to go with misleading names here.
>>>
>>> Federico confirmed that the answer is "no".
>>>
>>> I think we can use "key" and "elem" in this patch as they are more
>>> informative than "a" and "b"
>>
>> Except that "key" and "elem" are (imo) inapplicable to sort() callbacks
>> (and inconsistent with the naming in the 2nd callback here); they _may_
>> be applicable in bsearch() ones. Note also how in the C99 spec these
>> parameters of callback functions don't have names either.
> 
> Yes, reading the example in extable.c I think you are right. Maybe it is
> better to use "a" and "b" in both cmp and swap if you agree.

Using a and b is (as it looks) in line with at least some uses we have, so
less code churn than going with some other, more descriptive names (like
left/right). So yes, I'm okay with using a/b.

Jan