Re: [PATCH] io: Do not implement fstat with fstatat

[Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>]

On 12/09/23 02:31, Florian Weimer wrote:
> * Adhemerval Zanella Netto:
> 
>> On 11/09/23 16:32, Linus Torvalds wrote:
>>> On Mon, 11 Sept 2023 at 06:11, Adhemerval Zanella Netto
>>> <adhemerval.zanella@linaro.org> wrote:
>>>>
>>>> I used the same fstatat logic, but using __NR_fstat should be fine.
>>>
>>> I think you should keep using the same logic as in fstatat().
>>>
>>> Using "#ifdef __NR_newfstatat" basically checks for "not stat64".
>>>
>>> So, for example, x86-64 (and x64) have __NR_newfstatat, but 32-bit
>>> i386 has __NR_fstatat64.
>>>
>>> Now, maybe the other tests effectively already capture this (ie I
>>> suspect FSTATAT_USE_STATX may already be the thing that makes 32-bit
>>> i386 different), but I do think it's actually better the way it was.
>>
>> The FSTATAT_USE_STATX already handles it, and there is a explicit comment
>> later at !FSTATAT_USE_STATX for which ABIs are affected regarding glibc
>> support.  So either way (check __NR_newfstat and __NR_fstat) should be
>> ok.
> 
> If __NR_newfstatat is clear to the subject matter experts, I won't
> object to it.  But please add a comment.

I sent a newer version that checks for __NR_stat instead, with comments
to which ABI selects what [1].

[1] https://patchwork.sourceware.org/project/glibc/patch/20230911132548.1981093-1-adhemerval.zanella@linaro.org/