Re: [PATCH] Fix building tst-socket-timestamp-compat on older Linux

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

On 14/02/2022 10:08, Adhemerval Zanella wrote:
> 
> 
> On 11/02/2022 18:40, Florian Weimer via Libc-alpha wrote:
>> * Tulio Magno Quites Machado Filho:
>>
>>> Florian Weimer via Libc-alpha <libc-alpha@sourceware.org> writes:
>>>
>>>> * Tulio Magno Quites Machado Filho via Libc-alpha:
>>>>
>>>>> diff --git a/sysdeps/unix/sysv/linux/tst-socket-timestamp-compat.c b/sysdeps/unix/sysv/linux/tst-socket-timestamp-compat.c
>>>>> index 0ff1a214e6..68b0a0e412 100644
>>>>> --- a/sysdeps/unix/sysv/linux/tst-socket-timestamp-compat.c
>>>>> +++ b/sysdeps/unix/sysv/linux/tst-socket-timestamp-compat.c
>>>>> @@ -23,6 +23,23 @@
>>>>>  #include <support/xunistd.h>
>>>>>  #include <stdbool.h>
>>>>>  
>>>>> +/* The following macros are only available on ABIs that support 32 bit time_t.
>>>>> +   This test has a runtime check to guarantee it runs only on these ABIs.
>>>>> +   However, we need to create fake values to guarantee this test is built
>>>>> +   for all the ABIs.  */
>>>>
>>>> Isn't this related to the kernel header version, and not just time32
>>>> support?
>>>
>>> The runtime test does depend on the kernel header version because it depends
>>> on the time_t size.
>>> However, the definition of these macros depends only on the value of
>>> __TIMESIZE, which doesn't depend on the kernel header [1].
>>>
>>> [1] https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/powerpc/bits/socket-constants.h;hb=HEAD#l50
>>
>> What I meant is this: I see the same build failure on aarch64 with our
>> 4.18-derived headers, but not with our 5.14-derived headers.  This is on
>> the same architecture, so it can't be glibc's time32 support state.
> 
> There are two issue in fact, one is that socket-constants.h is only included
> for !__USE_MISC, which is then disable for _DEFAULT_SOURCE (enabled with
> _GNU_SOURCE).  The another issue is socket-constants.h only defines
> the *_OLD and *_NEW variant in some cases, different than kernel header.
> 
> I think we should always include socket-constants.h, only define the values
> if there are not already defined by the kernel headers, and always define
> the *_OLD and *_NEW constants.  Something like the below (it requires to
> adapt to other arch-specific socket-constants.h as well).
> 

Another way is to move the test to a internal one and use the values
on socket-constants-time64.h instead.