Re: RFC: [PATCH] Add warn_if_not_aligned attribute

[Martin Sebor <msebor at gmail dot com>]

On 06/05/2017 11:45 AM, H.J. Lu wrote:
> On Mon, Jun 5, 2017 at 8:11 AM, Joseph Myers <joseph@codesourcery.com> wrote:
> > The new attribute needs documentation.  Should the test be in c-c++-common
>
> This feature does support C++.  But C++ compiler issues a slightly
> different warning at a different location.
>
> > or does this feature not support C++?
>
> Here is the updated patch with documentation and a C++ test.  This
> patch caused a few testsuite failures:
>
> FAIL: gcc.dg/compat/struct-align-1 c_compat_x_tst.o compile
>
> /export/gnu/import/git/sources/gcc/gcc/testsuite/gcc.dg/compat//struct-align-1.h:169:1:
> warning: alignment 1 of 'struct B2_m_inner_p_outer' is less than 16
>
> FAIL: g++.dg/torture/pr80334.C   -O0  (test for excess errors)
>
> /export/gnu/import/git/sources/gcc/gcc/testsuite/g++.dg/torture/pr80334.C:4:8:
> warning: alignment 1 of 'B' is less than 16

Users often want the ability to control a warning, even when it
certainly indicates a bug.  I would suggest to add an option to
make it possible for this warning as well.

Btw., a bug related to some of those this warning is meant to
detect is assigning the address of an underaligned object to
a pointer of a natively aligned type.  Clang has an option
to detect this problem: -Waddress-of-packed-member.  It might
make a nice follow-on enhancement to add support for the same
thing.  I mention this because I think it would make sense to
consider this when choosing the name of the GCC option (i.e.,
rather than having two distinct but closely related warnings,
have one that detects both of these alignment type of bugs.

Martin