[llvm-commits] [patch] AlignAs fixes

[David Blaikie]

Ping.

On Tue, Sep 4, 2012 at 5:26 PM, David Blaikie <dblaikie at gmail.com> wrote:
> Ping.
>
> On Wed, Aug 29, 2012 at 8:56 AM, David Blaikie <dblaikie at gmail.com> wrote:
>> Updated based on feedback from IRC. Updated some comments, made Clang
>> use the standard feature rather than the __attribute__, removed some
>> unused template definitions.
>>
>> On Tue, Aug 28, 2012 at 3:56 PM, David Blaikie <dblaikie at gmail.com> wrote:
>>> Here's an attempt to fix the incorrect/non-portable code in AlignAs.h
>>> and its unit test.
>>>
>>> We were previously trying to align typedefs, which isn't valid in
>>> C++11. Even attempting to use alignas correctly won't work on any
>>> compiler other than Clang anyway, even those (such as GCC 4.7)
>>> reporting C++11 compliance through the __cplusplus macro value.
>>>
>>> So instead:
>>> for __GNUC__, use __attribute__((aligned(x))) of a char
>>> for __cplusplus == 201103L || __has_feature(cxx_alignas), use alignas
>>> on a char (this code path isn't actually tested at the moment since
>>> clang reports __GNUC__ so it hits the first case)
>>> for _MSC_VER, use __declspec(align(x)) on a char
>>>
>>> This doesn't try to align a typedef as we were doing before, so I cut
>>> a few test cases & extra types out of the test file - this doesn't
>>> change the implementation much because we were stuffing a variable of
>>> the typedef'd type in an anonymous struct anyway - so now we just use
>>> the template with the aligned member directly.
>>>
>>> Yes/no/maybe? (I haven't actually looked at the code to see if this is
>>> producing correct code - just updated the test cases fairly
>>> mechanically & checked that they pass)
>>>
>>> - David