Re: 回复: 回复: [PATCH] MIPS: fix y2106 problem on N64 with statx

[Adhemerval Zanella]

On 18/03/2021 03:05, yunqiang.su@cipunited.com wrote:
> 
> 
>> -----邮件原件-----
>> 发件人: Adhemerval Zanella <adhemerval.zanella@linaro.org>
>> 发送时间: 2021年3月17日 19:27
>> 收件人: yunqiang.su@cipunited.com; libc-alpha@sourceware.org
>> 抄送: aurelien@aurel32.net; jiaxun.yang@flygoat.com; macro@orcam.me.uk
>> 主题: Re: 回复: [PATCH] MIPS: fix y2106 problem on N64 with statx
>>
>> Even though it is not the default, non LFS interface are legacy ones.
>> y2038 support, for instance, is only supported with LFS interfaces.
>>
>> So it is up to you if you want to fix on fstatat.c, it should also fix the s390 issue
>> with missing nanosecond support.
>>
> 
> fstatat is not recommend used by any new write code?
> All new write code should use fstatat64?

It is an implementation detail in fact, glibc fstatat64 is the
interface with LFS support.  On ABIs with default LFS support it does
matter whether you use fstatat or fstatat64 (fstatat is fact an alias
to fstatat64).

On ABIs with LFS support (most of legacy 32-bits and some of 64-bit)
fstatat is used as default and fstatat64 is used as an asm alias if
_FILE_OFFSET_BITS=64 is defined or provided in the namespace if
_LARGEFILE64_SOURCE is defined.

I think we should now make LFS as default [1], at least for legacy
32-bit ABIs even though they are still namespace issues [2] [3].

[1] https://sourceware.org/bugzilla/show_bug.cgi?id=13047
[2] https://sourceware.org/bugzilla/show_bug.cgi?id=14106
[3] https://sourceware.org/bugzilla/show_bug.cgi?id=15766