[cfe-dev] libc++ miscompilation on trunk?

[İsmail Dönmez]

Btw, exact failure list looks like this:

libc++ :: std/input.output/iostream.format/ext.manip/get_money.pass.cpp
libc++ :: std/input.output/iostream.format/ext.manip/get_time.pass.cpp
libc++ :: std/input.output/iostream.format/ext.manip/put_money.pass.cpp
libc++ :: std/input.output/iostream.format/ext.manip/put_time.pass.cpp
libc++ :: std/localization/locale.categories/category.ctype/locale.ctype.byname/narrow_1.pass.cpp
libc++ :: std/localization/locale.categories/category.ctype/locale.ctype.byname/narrow_many.pass.cpp
libc++ :: std/localization/locale.categories/category.time/locale.time.get.byname/date_order_wide.pass.cpp


On Mon, Jun 29, 2015 at 2:08 PM, İsmail Dönmez <ismail at donmez.ws> wrote:
> Hi,
>
> A regression between r240624 and r240761 resulted in new regressions
> on my Linux x86-64 box and one of them is
> test/std/input.output/iostream.format/ext.manip/put_money.pass.cpp:
>
> ./test/std/input.output/iostream.format/ext.manip/put_money.pass.cpp:80:
> int main(): Assertion `sb.str() == L"-$1,234,567.89"' failed.
> Aborted
>
> Looking under gdb it looks like the data is corrupted:
>
> (gdb) frame 4
> #4  0x0000000000403bab in main () at
> ./test/std/input.output/iostream.format/ext.manip/put_money.pass.cpp:80
> 80              assert(sb.str() == L"-$1,234,567.89");
> (gdb) print sb.str()
> $1 = {<std::__1::__basic_string_common<true>> = {<No data fields>},
>   __r_ = {<std::__1::__libcpp_compressed_pair_imp<std::__1::basic_string<wchar_t,
> std::__1::char_traits<wchar_t>, std::__1::allocator<wchar_t> >::__rep,
> std::__1::allocator<wchar_t>, 2>> = {<std::__1::allocator<wchar_t>> =
> {<No data fields>},
>       __first_ = {{__l = {__cap_ = 9, __size_ = 6, __data_ = 0x609160
> L"-$\x60e9b0.\x60e9b0\x60e9b0"}, __s = {{
>               __size_ = 9 '\t', __lx = 9 L'\t'}, __data_ =
> L"\000\006\000\x609160"}, __r = {__words = {9, 6,
>               6328672}}}}}, <No data fields>}, static npos =
> 18446744073709551615}
>
> This only seems to happen with wchar_t types so the next test on line
> 89 fails too. Just for reference I build clang in two stages so libcxx
> is built with stage2 compiler. I see the failure on openSUSE 13.2
> where no system package has changed so this is unlikely to be an
> outside bug.
>
> Any ideas?
>
> Regards,
> ismail