MacOS-X Linker warning when building with -DEXIV2_TEAM_USE_SANITIZERS=On

[clanmills]

In dealing with ASAN issues on MacOS-X, I see there are warning from the linker. I don't understand them. They were present in RC2. I have not searched further back. They might be benign.

[ 94%] Linking CXX shared library ../lib/libexiv2.dylib
ld: warning: direct access in function 'Exiv2::XmpProperties::printProperty(std::__1::basic_ostream<char, std::__1::char_traits<char> >&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, Exiv2::Value const&)' from file 'CMakeFiles/exiv2lib.dir/properties.cpp.o' to global weak symbol 'typeinfo for std::__1::basic_ostream<char, std::__1::char_traits<char> >& (std::__1::basic_ostream<char, std::__1::char_traits<char> >&, Exiv2::Value const&, Exiv2::ExifData const*)' from file 'CMakeFiles/exiv2lib.dir/exif.cpp.o' means the weak symbol cannot be overridden at runtime. This was likely caused by different translation units being compiled with different visibility settings.
ld: warning: direct access in function 'Exiv2::Internal::CanonMakerNote::printCsLensType(std::__1::basic_ostream<char, std::__1::char_traits<char> >&, Exiv2::Value const&, Exiv2::ExifData const*)' from file 'CMakeFiles/exiv2lib_int.dir/canonmn_int.cpp.o' to global weak symbol 'typeinfo for std::__1::basic_ostream<char, std::__1::char_traits<char> >& (std::__1::basic_ostream<char, std::__1::char_traits<char> >&, Exiv2::Value const&, Exiv2::ExifData const*)' from file 'CMakeFiles/exiv2lib.dir/exif.cpp.o' means the weak symbol cannot be overridden at runtime. This was likely caused by different translation units being compiled with different visibility settings.
ld: warning: direct access in function 'Exiv2::Internal::printMinoltaSonyLensID(std::__1::basic_ostream<char, std::__1::char_traits<char> >&, Exiv2::Value const&, Exiv2::ExifData const*)' from file 'CMakeFiles/exiv2lib_int.dir/minoltamn_int.cpp.o' to global weak symbol 'typeinfo for std::__1::basic_ostream<char, std::__1::char_traits<char> >& (std::__1::basic_ostream<char, std::__1::char_traits<char> >&, Exiv2::Value const&, Exiv2::ExifData const*)' from file 'CMakeFiles/exiv2lib.dir/exif.cpp.o' means the weak symbol cannot be overridden at runtime. This was likely caused by different translation units being compiled with different visibility settings.
ld: warning: direct access in function 'Exiv2::Internal::printLensType(std::__1::basic_ostream<char, std::__1::char_traits<char> >&, Exiv2::Value const&, Exiv2::ExifData const*)' from file 'CMakeFiles/exiv2lib_int.dir/pentaxmn_int.cpp.o' to global weak symbol 'typeinfo for std::__1::basic_ostream<char, std::__1::char_traits<char> >& (std::__1::basic_ostream<char, std::__1::char_traits<char> >&, Exiv2::Value const&, Exiv2::ExifData const*)' from file 'CMakeFiles/exiv2lib.dir/exif.cpp.o' means the weak symbol cannot be overridden at runtime. This was likely caused by different translation units being compiled with different visibility settings.
[ 94%] Built target exiv2lib
Scanning dependencies of target exiv2
[ 95%] Building CXX object src/CMakeFiles/exiv2.dir/exiv2.cpp.o
[ 96%] Building CXX object src/CMakeFiles/exiv2.dir/actions.cpp.o
[ 97%] Building CXX object src/CMakeFiles/exiv2.dir/getopt.cpp.o
[ 98%] Building CXX object src/CMakeFiles/exiv2.dir/utils.cpp.o
[100%] Linking CXX executable ../bin/exiv2
[100%] Built target exiv2
686 rmills@rmillsmm:~/Downloads/foo/build $ cmake .. -DEXIV2_TEAM_USE_SANITIZERS=1 -DEXIV2_BUILD_SAMPLES=0 ; make 

I've tried to make sense of the first of the 4 errors and it becomes:

ld: warning: direct access in function 'Exiv2::XmpProperties::printProperty(basic_ostream, const&, Exiv2::Value const&)'
from file 'CMakeFiles/exiv2lib.dir/properties.cpp.o'
to global weak symbol
'typeinfo for std::__1::basic_ostream&, Exiv2::Value const&, Exiv2::ExifData const*)'
from file 'CMakeFiles/exiv2lib.dir/exif.cpp.o' means the weak means the weak symbol
cannot be overridden at runtime.

This was likely caused by different translation units being compiled with different visibility settings.

I don't know that this means. What I can see is that in properties.cpp, we have:

    std::ostream& XmpProperties::printProperty(std::ostream& os,
                                               const std::string& key,
                                               const Value& value)
    {
        PrintFct fct = printValue;
        if (value.count() != 0) {
            const XmpPrintInfo* info = find(xmpPrintInfo, key);
            if (info) fct = info->printFct_;
        }
        return fct(os, value, 0);
    }

I think he's complaining about fct (no pun intended).

[D4N]

I think the issue is related to visibility settings, as the linker notes itself:

This was likely caused by different translation units being compiled with different visibility settings.

You might want to check whether -fvisibility=hidden(not sure if that's the exact flag) is added to all translation units.

[clanmills]

Thanks, Dan. I'll look into that. I know @piponazo changed something concerning visibility a few months ago and said it was very interesting. I didn't believe him!

I expect you saw that our friend @cool-tomato reported 6 ASAN issues this morning. Crap. I'll deal with them and publish RC4 later in the week. For sure, I want to keep changes to the minimum possible, so a global change to visibility makes me nervous. I'll investigate visibility after I've deal with #590.

[clanmills]

I went searching for visibility in CMake. Guess what? @piponazo is the expert and answered a question on Stackoverflow about this several years ago. I don't see anything in CMakeLists.txt files about visibility.

https://stackoverflow.com/questions/17080869/what-is-the-cmake-equivalent-to-gcc-fvisibility-hidden-when-controlling-the-e

[piponazo]

The CMake configuration related to these visibility settings can be found in the file cmake/mainSetup.cmake:

set(CMAKE_CXX_VISIBILITY_PRESET hidden)
set(CMAKE_VISIBILITY_INLINES_HIDDEN 1)

I would not care too much about this at this moment. I have experienced these kind of warnings in different projects and none of them caused issues.

I could try to investigate it when I have some spare time, but I do not think we should worry about this on 0.27.

[clanmills]

I forgot to look in the cmake/*.cmake files for Visibility settings.

I think we do nothing about this. It's ugly, however there's no evidence that it's harmful. It's good to have this logged so we know where to restart the conversation. I saw it this morning and took a while to understand that it involves MacOS-X /Sanitizers and behaved like this in RC2 and RC3.

I'm going to close this.

Thanks to you and Dan for saying something about this.

You'll see that @cool-tomato dropped 6 poc's on RC3 this morning. I hope to close them by Wednesday and publish RC4 before the end of the week. If he (or other security folks) produce more POCs so late in the release, I will ask them about delaying the fixing for v0.27.1. No schedule for that yet - probably about 2019-03-31.