[PATCH 0/2] linux: Keep time64 stat layout independent of feature macros

[Adhemerval Zanella Netto]

On 05/08/26 15:45, Matthias Goergens wrote:
> Hi Adhemerval,
> 
> Thanks for taking this up.  The squash in ecf95727a787 looks right to
> me: the struct_stat_time64_helper.h changes match the series (the
> endian include, plus dropping the two trailing reserved words), and
> the __fieldts macro in bits/struct_stat.h covers the arc/or1k
> alignment case we had not modelled.  Moving the test to a
> glibcextract-based script is a clear improvement — it runs in
> cross-compilation where our io/ tests could not.
> 
> One clarification: you wrote "we will need to fix it along with
> BZ#34466" — I assume you meant BZ#32119 (this bug).  BZ#34466 is my
> other, unrelated patch (POSIX.1-2024 string declarations), still
> awaiting review.

Oops, indeed. I was reading you BZ#34466 patch and mixed up things.

> 
> Happy to help with ABI testing: the compile-time layout models from
> our original analysis cover i386, arm, ppc, mips, and riscv32, and I
> can add arc and or1k and run the lot against the squashed change if
> that is useful alongside your own testing.  Just say the word.

For such chage that involves potential kernel or glibc changes I make/check 
all the potential affected ABIs (essentially the one with a libc.abilist).
I haven't see any issue, but I would also run some make on native builds
along with qemu-system.