Re: [PATCH] libstdc++: atomic: Add missing clear_padding in __atomic_float constructor

[Jakub Jelinek <jakub@redhat.com>]

On Tue, Jan 30, 2024 at 04:34:30PM +0000, Jonathan Wakely wrote:
> > --- /dev/null
> > +++ b/libstdc++-v3/testsuite/29_atomics/atomic_float/compare_exchange_padding.cc
> > @@ -0,0 +1,50 @@
> > +// { dg-do run { target c++20 } }
> > +// { dg-options "-O0" }
> > +// { dg-timeout 10 }

Why dg-timeout?
I don't see it used in any of the libstdc++ tests and I don't see anything
expensive on this test.  Just leave it out.

> > +// { dg-additional-options "-mlong-double-80" { target x86_64-*-* } }
> > +// { dg-do run { target { ia32 || x86_64-*-* } } }

Also, the target selectors above look wrong.
Generally, one should use { i?86-*-* x86_64-*-* } because e.g. on some
OSes like Solaris, one builds i?86 compiler which supports -m32 or -m64
options.
I wouldn't add the -mlong-double-80 at all, it is an ABI changing option,
if you rely in the test on some 80-bit long double properties (but on which
ones?  Sure, if -mlong-double-64 or -mlong-double-128 is in effect, the
type won't have any padding, but __builtin_clear_padding should reflect
that), it is better to guard the part of the test with some checks, whether
std::numeric_limits<long double>::digits == 64, or
preprocessor __LDBL_MANT_DIG__ == 64.

> > +// { dg-add-options libatomic_16b }
> 
> Why not just use -latomic unconditionally here?
> 
> You're adding a new libatomic_16b option that expands to -latomic for
> x86 and nothing otherwise, and then using it in a test that only runs
> for x86. So it's just adding -latomic to all targets that run the
> test, right?
> 
> Also, this patch needs either a copyright assignment or a DCO sign-off:
> https://gcc.gnu.org/contribute.html#legal

And if somebody would need to commit it for you, also patch in non-garbled
state (e.g. sent as an attachment, or even compressed attachment if using
really bad MUAs).

	Jakub