Re: [patch v2] aligned_alloc: conform to C17

[Florian Weimer <fweimer@redhat.com>]

* DJ Delorie:

> Florian Weimer <fweimer@redhat.com> writes:
>> Why do we need to export the __libc_aligned_alloc symbol?  That doesn't
>> seem to be right.  We should simply change the behavior of the existing
>> aligned_alloc symbol.
>
> I did it only because memalign has a __libc_memalign symbol.  The old
> symbol used to be a weak alias for __libc_memalign, so to change it, we
> need a separate function - which would thus be __libc_aligned_alloc, so
> that aligned_alloc() can continue to be a weak function.

None of this should alter the exported symbol list, though.  It is
possible to use versioned_symbol, libc_hidden_ver, maybe with #ifdef
SHARED conditionals, to avoid the need for additional exported signal.

__libc_memalign was added when we started using it in the dynamic loader
by mistake.  That has now been fixed, and maybe we should turn it into a
compat symbol.

>> If you do not want to change behavior for existing binaries (why?),
>
> I'm going on the premise that aligned_alloc() is new to C11 and was
> always documented to fail in the way I'm making it now fail (although
> there are other "documented to fail" cases we're not implementing).
> Thus there is no need to support a pre-C11 functionality.

Sounds good to me.

Thanks,
Florian