libcxx on Linux: symbols used but not defined

Howard Hinnant hhinnant at apple.com
Thu Feb 7 07:47:44 PST 2013


On Feb 7, 2013, at 6:17 AM, Alexey Samsonov <samsonov at google.com> wrote:

> Hi!
> 
> I'm trying to build libcxx on Linux (Ubuntu Precise) using the following command:
> 
> cmake -DLIBCXX_CXX_ABI=libsupc++ '-DLIBCXX_LIBSUPCXX_INCLUDE_PATHS=/usr/include/c++/4.6;/usr/include/c++/4.6/x86_64-linux-gnu/.;' -DLIBCXX_ENABLE_EXCEPTIONS=OFF -DLIBCXX_ENABLE_SHARED=OFF -DLIBCXX_ENABLE_RTTI=OFF -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/tmp/tmp.boBpS5zFqt/libcxx_inst /tmp/tmp.boBpS5zFqt/libcxx
> 
> and I want the result to be "hermetic" in the sense that resulting static library libc++.a doesn't reference undefined symbols from std::
> However, I see the following problems: 
> 
> (libc++.a.new.cpp.o):/tmp/tmp.boBpS5zFqt/libcxx/src/new.cpp:function std::bad_alloc::~bad_alloc(): error: undefined reference to 'std::exception::~exception()'
> (libc++.a.new.cpp.o):/tmp/tmp.boBpS5zFqt/libcxx/src/new.cpp:function std::bad_array_new_length::~bad_array_new_length(): error: undefined reference to 'std::exception::~exception()'
> (libc++.a.new.cpp.o):/tmp/tmp.boBpS5zFqt/libcxx/src/new.cpp:function std::bad_array_new_length::~bad_array_new_length(): error: undefined reference to 'std::exception::~exception()'
> (libc++.a.ios.cpp.o):/tmp/tmp.boBpS5zFqt/libcxx/src/ios.cpp:function std::__1::basic_ostream<char, std::__1::char_traits<char> >::flush(): error: undefined reference to 'std::uncaught_exception()'
> (libc++.a.ios.cpp.o):/tmp/tmp.boBpS5zFqt/libcxx/src/ios.cpp:function std::__1::basic_ostream<wchar_t, std::__1::char_traits<wchar_t> >::flush(): error: undefined reference to 'std::uncaught_exception()'
> (libc++.a.ios.cpp.o):/tmp/tmp.boBpS5zFqt/libcxx/src/ios.cpp:function std::__1::basic_ostream<char, std::__1::char_traits<char> >::sentry::~sentry(): error: undefined reference to 'std::uncaught_exception()'
> (libc++.a.ios.cpp.o):/tmp/tmp.boBpS5zFqt/libcxx/src/ios.cpp:function std::__1::basic_ostream<char, std::__1::char_traits<char> >::sentry::~sentry(): error: undefined reference to 'std::uncaught_exception()'
> (libc++.a.memory.cpp.o):/tmp/tmp.boBpS5zFqt/libcxx/src/memory.cpp:function std::__1::bad_weak_ptr::~bad_weak_ptr(): error: undefined reference to 'std::exception::~exception()'
> (libc++.a.memory.cpp.o):/tmp/tmp.boBpS5zFqt/libcxx/src/memory.cpp:function vtable for std::__1::__shared_weak_count: error: undefined reference to 'std::__1::__shared_weak_count::__get_deleter(std::type_info const&) const'
> (libc++.a.thread.cpp.o):/tmp/tmp.boBpS5zFqt/libcxx/src/thread.cpp:function std::__1::thread::~thread(): error: undefined reference to 'std::terminate()'
> 
> What is the correct way to for that? I have the patch (attached) that works for me, but I'm not really sure it's correct
> (especially the first part).

Yeah, we tried this once:

> --- include/memory	(revision 173248)
> +++ include/memory	(working copy)
> @@ -3629,10 +3629,9 @@
>      long use_count() const _NOEXCEPT {return __shared_count::use_count();}
>      __shared_weak_count* lock() _NOEXCEPT;
>  
> -    // purposefully not protected with #ifndef _LIBCPP_NO_RTTI because doing so
> -    //  breaks ABI for those clients who need to compile their projects with
> -    //    -fno-rtti and yet link against a libc++.dylib compiled without -fno-rtti.
> +#ifndef _LIBCPP_NO_RTTI
>      virtual const void* __get_deleter(const type_info&) const _NOEXCEPT;
> +#endif _LIBCPP_NO_RTTI

and as the comments indicate, got bitten by it, at least on Apple platforms.

Way forward:  Perhaps you could #ifdef it a little more in such a way that it isn't going to trigger except for someone trying to do exactly what you are doing:  Build a static lib on linux with rtti and exceptions off.  Or perhaps slightly more general:  Build a static lib with rtti and exceptions off.

When building a dynamic lib I've got to assume that the result is something that essentially becomes part of an OS and serves everyone instead of just a single client.

Howard




More information about the cfe-commits mailing list