Re: [RFC PATCH bpf-next] bpftool: Mark generated skeleton headers as system headers

[Yonghong Song <yhs@fb.com>]

On 7/29/22 3:12 AM, Quentin Monnet wrote:
> On 28/07/2022 19:25, Yonghong Song wrote:
>>
>>
>> On 7/28/22 9:56 AM, Jörn-Thorben Hinz wrote:
>>> Hi,
>>>
>>> after compiling a skeleton-using program with -pedantic once and
>>> stumbling across a tiniest incorrectness in skeletons with it[1], I was
>>> debating whether it makes sense to suppress warnings from skeleton
>>> headers.
>>>
>>> Happy about comments about this. This change might be too suppressive
>>> towards warnings and maybe ignoring only -Woverlength-strings directly
>>> in OBJ_NAME__elf_bytes() be a better idea. Or keep all warnings from
>>> skeletons available as-is to have them more visible in and around
>>> bpftool’s development.
>>
>> This is my 2cents. As you mentioned, skeleton file are per program
>> and not in system header file directory. I would like not to mark
>> these header files as system files. Since different program will
>> generate different skeleton headers, suppressing warnings
>> will prevent from catching potential issues in certain cases.
>>
>> Also, since the warning is triggered by extra user flags like -pedantic
>> when building bpftool, user can also add -Wno-overlength-strings
>> in the extra user flags.
> 
> I agree with Yonghong, I don't think it's a good idea to mark the whole
> file as a system header. I would maybe consider the other solution where
> we can disable the warning locally in the skeleton, just around
> OBJ_NAME__elf_bytes() as you suggested. Although I suppose we'd need
> several pragmas if we want to silence it for GCC and clang, for example?
> It looks like your patch was only addressing GCC?

Quentin, "#pragma GCC system_header" also works for clang. Basically
clang implemented system_header pragma identical to gcc counterpart.

> 
> Thanks for the contribution,
> Quentin