[cfe-commits] r170160 - in /cfe/trunk: lib/CodeGen/CGClass.cpp test/CodeGenCXX/lambda-expressions.cpp
NAKAMURA Takumi
geek4civic at gmail.com
Thu Dec 13 19:03:48 PST 2012
Eli,
I am sorry. My local build tree might be wrong then. I don't reproduce
any odd emissions now.
I reconfirmed r170160 emits as below with -Asserts.
define internal void @"_ZZ1hvEN3$_78__invokeEv"(%struct.A* noalias sret %agg.res
ult) inlinehint align 2 {
call void @"_ZZ1hvENK3$_7clEv"(%struct.A* sret %agg.result, %class.anon.8* und
ef)
ret void
}
Yes, it's a trivial issue in -Asserts. We don't expect "entry:" in the
entry block.
Tweaked the test in r170175.
...Takumi
2012/12/14 NAKAMURA Takumi <geek4civic at gmail.com>:
> 2012/12/14 Eli Friedman <eli.friedman at gmail.com>:
>> On Thu, Dec 13, 2012 at 4:29 PM, NAKAMURA Takumi <geek4civic at gmail.com> wrote:
>>> Eli, it fails in -Asserts build. Not sure why.
>>>
>>> define internal void @"_ZZ1hvEN3$_78__invokeEv"(%struct.A* noalias
>>> sret %agg.result) inlinehint align 2 {
>>> call void @"_ZZ1hvENK3$_7clEv"(%struct.A* sret %agg.result,
>>> %class.anon.8* undef)
>>> call void @llvm.trap()
>>> unreachable
>>> ; No predecessors!
>>> ret void
>>> }
>>
>> My only theory for how you could end up in this situation is that you
>> didn't properly rebuild clang. I'll take a look, though.
>
> Eli, FYI, I have rebuilt r170160.
>
> As compared build tree between #965 and #969
>
> http://bb.pgr.jp/builders/cmake-clang-x86_64-linux/builds/965
> http://bb.pgr.jp/builders/cmake-clang-x86_64-linux/builds/969
>
> build/lib/Support/CMakeFiles/LLVMSupport.dir/CommandLine.cpp.o |
> Bin 183336 -> 183336 bytes
> build/tools/clang/include/clang/Config/config.h | 2 +-
> build/tools/clang/lib/Driver/CMakeFiles/clangDriver.dir/Driver.cpp.o |
> Bin 109032 -> 109048 bytes
>
> I am sure my builder must be fine. It is configure as optimized
> -assertions, with g++-4.4.6.
>
> Also msvc builder (msc16, Release (optimized -assertions) reports
> similar failure.
> http://bb.pgr.jp/builders/cmake-clang-i686-msvc10/builds/494
>
> ...Takumi
More information about the cfe-commits
mailing list