[compiler-rt] r233071 - [ASan] Do not link ASan OSX runtime with -lc++abi (which was added in r233036)

Kuba Brecka kuba.brecka at gmail.com
Wed Mar 25 10:40:19 PDT 2015 

Aah, it’s because we have `-undefined dynamic_lookup` linker flag in the Makefile-based build, but not in CMake.  Compare the CMake build:

$ nm -mg cmake-build/.../libclang_rt.ubsan_osx_dynamic.dylib | grep ZTIN
  (undefined) external __ZTIN10__cxxabiv117__class_type_infoE (from libc++abi)
  (undefined) external __ZTIN10__cxxabiv120__si_class_type_infoE (from libc++abi)
  (undefined) external __ZTIN10__cxxabiv121__vmi_class_type_infoE (from libc++abi)

with the dylib produced by Make:

$ nm -mg make-build/.../libclang_rt.ubsan_osx_dynamic.dylib | grep ZTIN
  (undefined) external __ZTIN10__cxxabiv117__class_type_infoE (dynamically looked up)
  (undefined) external __ZTIN10__cxxabiv120__si_class_type_infoE (dynamically looked up)
  (undefined) external __ZTIN10__cxxabiv121__vmi_class_type_infoE (dynamically looked up)

We should probably choose only one method (dynamic_lookup or linking libc++abi) and use the same in both build systems.

Kuba

> On Mar 25, 2015, at 3:10 PM, Kuba Brecka <kuba.brecka at gmail.com> wrote:
> 
> Looks like it builds and passes tests now.  But I’m very curious about why the Makefile-based build doesn't require “-lc++abi” and it still builds fine.  That seems very suspicious to me…
> 
> Kuba
> 
>> On Mar 24, 2015, at 11:00 PM, Alexey Samsonov <vonosmas at gmail.com <mailto:vonosmas at gmail.com>> wrote:
>> 
>> Thanks! Submitted as r233122. Let me know (and feel free to revert) if it still breaks Jenkins or another build.
>> 
>> On Tue, Mar 24, 2015 at 2:44 PM, Anna Zaks <ganna at apple.com <mailto:ganna at apple.com>> wrote:
>> Alexey,
>> 
>> I've applied the following patch on top of yours and tested with cmake ninja build plus "ninja check-ubsan" and "ninja check-asan".
>> 
>> Cheers, 
>> Anna. 
>> 
>> zaks$ git diff
>> diff --git a/CMakeLists.txt b/CMakeLists.txt
>> index 3e67883..c6cd87d 100644
>> --- a/CMakeLists.txt
>> +++ b/CMakeLists.txt
>> @@ -304,8 +304,7 @@ if(APPLE)
>>    set(DARWIN_iossim_CFLAGS
>>      -mios-simulator-version-min=7.0 -isysroot ${IOSSIM_SDK_DIR})
>>    set(DARWIN_osx_LINKFLAGS -mmacosx-version-min=${SANITIZER_MIN_OSX_VERSION}
>> -     -stdlib=libc++
>> -     -lc++abi)
>> +     -stdlib=libc++)
>>    set(DARWIN_iossim_LINKFLAGS
>>      -Wl,-ios_simulator_version_min,7.0.0
>>      -mios-simulator-version-min=7.0
>> diff --git a/lib/ubsan/CMakeLists.txt b/lib/ubsan/CMakeLists.txt
>> index 8c2d30b..a8b3f61 100644
>> --- a/lib/ubsan/CMakeLists.txt
>> +++ b/lib/ubsan/CMakeLists.txt
>> @@ -31,7 +31,8 @@ if(APPLE)
>>      add_compiler_rt_darwin_dynamic_runtime(clang_rt.ubsan_${os}_dynamic ${os}
>>        ARCH ${UBSAN_SUPPORTED_ARCH}
>>        SOURCES $<TARGET_OBJECTS:RTUbsan.${os}>
>> -              $<TARGET_OBJECTS:RTSanitizerCommon.${os}>)
>> +              $<TARGET_OBJECTS:RTSanitizerCommon.${os}>
>> +      LINKFLAGS -lc++abi)
>>  
>>      add_dependencies(ubsan clang_rt.ubsan_${os}_dynamic)
>>    endforeach()
>> 
>>> On Mar 24, 2015, at 11:36 AM, Alexey Samsonov <vonosmas at gmail.com <mailto:vonosmas at gmail.com>> wrote:
>>> 
>>> On Tue, Mar 24, 2015 at 11:24 AM, Anna Zaks <ganna at apple.com <mailto:ganna at apple.com>> wrote:
>>> 
>>>> On Mar 24, 2015, at 10:46 AM, Justin Bogner <mail at justinbogner.com <mailto:mail at justinbogner.com>> wrote:
>>>> 
>>>> Alexander Potapenko <glider at google.com <mailto:glider at google.com>> writes:
>>>>> Author: glider
>>>>> Date: Tue Mar 24 08:10:55 2015
>>>>> New Revision: 233071
>>>>> 
>>>>> URL: http://llvm.org/viewvc/llvm-project?rev=233071&view=rev <http://llvm.org/viewvc/llvm-project?rev=233071&view=rev>
>>>>> Log:
>>>>> [ASan] Do not link ASan OSX runtime with -lc++abi (which was added in r233036)
>>>>> This should fix the OSX Jenkins build at
>>>>> http://lab.llvm.org:8080/green/job/clang-stage1-cmake-RA_check/ <http://lab.llvm.org:8080/green/job/clang-stage1-cmake-RA_check/>
>>>> 
>>>> This didn't quite work - compiler-rt doesn't even link after this
>>>> change:
>>>> 
>>>>  Undefined symbols for architecture x86_64:
>>>>    "typeinfo for __cxxabiv1::__class_type_info", referenced from:
>>>>        __ubsan::checkDynamicType(void*, void*, unsigned long) in ubsan_type_hash.cc.o
>>>>        isDerivedFromAtOffset(__cxxabiv1::__class_type_info const*, __cxxabiv1::__class_type_info const*, long) in ubsan_type_hash.cc.o
>>>>        findBaseAtOffset(__cxxabiv1::__class_type_info const*, long) in ubsan_type_hash.cc.o
>>>> 
>>>> See here:
>>>> 
>>>>  http://lab.llvm.org:8080/green/job/clang-stage1-cmake-RA_build/2492/consoleFull#-158682280549ba4694-19c4-4d7e-bec5-911270d8a58c <http://lab.llvm.org:8080/green/job/clang-stage1-cmake-RA_build/2492/consoleFull#-158682280549ba4694-19c4-4d7e-bec5-911270d8a58c>
>>>> 
>>>> I've reverted both r233071 and r233036 in r233097 to get things working
>>>> again.
>>> 
>>> If we are reverting all this, 233035 needs to be reverted as well.
>>> 
>>> Yes.
>>> 
>>> 
>>> Maybe we should XFAIL the two tests instead?
>>> 
>>> No, I just think we should provide -lc++abi only for UBSan. I.e. remove it from DARWIN_osx_LINKFLAGS and instead provide:
>>> 
>>>     add_compiler_rt_darwin_dynamic_runtime(clang_rt.ubsan_${os}_dynamic ${os}
>>>       ARCH ${UBSAN_SUPPORTED_ARCH}
>>>       SOURCES $<TARGET_OBJECTS:RTUbsan.${os}>
>>>               $<TARGET_OBJECTS:RTSanitizerCommon.${os}>
>>>       LINKFLAGS -lc++abi)
>>> 
>>> Could you check if it works?
>>>  
>>> 
>>> CC-ing Alexey.
>>> 
>>>> 
>>>>> Modified:
>>>>>    compiler-rt/trunk/CMakeLists.txt
>>>>> 
>>>>> Modified: compiler-rt/trunk/CMakeLists.txt
>>>>> URL:
>>>>> http://llvm.org/viewvc/llvm-project/compiler-rt/trunk/CMakeLists.txt?rev=233071&r1=233070&r2=233071&view=diff <http://llvm.org/viewvc/llvm-project/compiler-rt/trunk/CMakeLists.txt?rev=233071&r1=233070&r2=233071&view=diff>
>>>>> ==============================================================================
>>>>> --- compiler-rt/trunk/CMakeLists.txt (original)
>>>>> +++ compiler-rt/trunk/CMakeLists.txt Tue Mar 24 08:10:55 2015
>>>>> @@ -304,8 +304,7 @@ if(APPLE)
>>>>>   set(DARWIN_iossim_CFLAGS
>>>>>     -mios-simulator-version-min=7.0 -isysroot ${IOSSIM_SDK_DIR})
>>>>>   set(DARWIN_osx_LINKFLAGS -mmacosx-version-min=${SANITIZER_MIN_OSX_VERSION}
>>>>> -     -stdlib=libc++
>>>>> -     -lc++abi)
>>>>> +     -stdlib=libc++)
>>>>>   set(DARWIN_iossim_LINKFLAGS
>>>>>     -Wl,-ios_simulator_version_min,7.0.0
>>>>>     -mios-simulator-version-min=7.0
>>>>> 
>>>>> 
>>>>> _______________________________________________
>>>>> llvm-commits mailing list
>>>>> llvm-commits at cs.uiuc.edu <mailto:llvm-commits at cs.uiuc.edu>
>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits <http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits>
>>>> _______________________________________________
>>>> llvm-commits mailing list
>>>> llvm-commits at cs.uiuc.edu <mailto:llvm-commits at cs.uiuc.edu>
>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits <http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits>
>>> 
>>> 
>>> 
>>> -- 
>>> Alexey Samsonov
>>> vonosmas at gmail.com <mailto:vonosmas at gmail.com>
>> 
>> 
>> 
>> -- 
>> Alexey Samsonov
>> vonosmas at gmail.com <mailto:vonosmas at gmail.com>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150325/a2b7fa93/attachment.html>

More information about the llvm-commits mailing list