yiranwang added a comment.
Thank you, Eric. Also, could you please help to commit the change? I personally
do not have the permissions to change libc++ code, thanks a lot.
http://reviews.llvm.org/D12247
___
cfe-commits mailing list
cfe-commits@list
yiranwang added a comment.
Thanks for testing.
There is proof as the following, given that alignment of _Aligner is _Align
(the whole purpose of _Aligner here, which we do not want to double check), and
for any type, the "sizeof" is always multiple of "alignof", this change should
be ABI compat
yiranwang added a comment.
In http://reviews.llvm.org/D12247#234533, @EricWF wrote:
> So this patch LGTM. Sorry for being slow to understand it. However I want to
> see two things before it lands.
>
> 1. I would like to see a test of some sort that the resulting type has the
> same size and ali
yiranwang added a comment.
Hi Eric,
Could you please explain a bit more what is broken specifically? As we can see,
sizeof(), _Len, and _Align, and alignof() of "aligned_storage" are all not
changed.
http://reviews.llvm.org/D12247
___
cfe-commits
yiranwang added a comment.
some more explanation of last comment.
For most architecture sizeof(function) is 4*sizeof(void *), but it is
6*sizeof(void*) on AARCH64/LINUX, which does not sound so great as there are
padding of 2*sizeof(void*) bytes.
http://reviews.llvm.org/D12247
_
yiranwang added a comment.
In http://reviews.llvm.org/D12247#233349, @EricWF wrote:
> In http://reviews.llvm.org/D12247#233323, @davidxl wrote:
>
> > We certainly need a fix without breaking ABI. Is there a ABI conformance
> > test for libcxx?
>
>
> Well this patch definitely breaks the ABI. Thi
yiranwang added a comment.
A test case is as following. It has to be build by GCC 4.9 -O3 (maybe or
later), with latest libc++, and for AARCH64+ANDROID target.
AARCH64 requires 128 bit alignment for aligned_storage and 64 bit pointers,
while gcc 4.9 alias analysis will do field-sensitive points
yiranwang created this revision.
yiranwang added a subscriber: cfe-commits.
In libc++, there are some usage of aligned_storage which uses "sizeof" bytes of
raw data. This is problematic a bit, as the trailing padding area will be
counted by "sizeof", and it leads to out of bound access. For exam
