Figure out something for CI so we don't have to build w/ -D_GLIBCXX_USE_CXX11_ABI=0

[arlolra]

See #79

[eighthave]

FYI, the Android builds are using -D_GLIBCXX_USE_CXX11_ABI=1 #90

[arlolra]

FYI, the Android builds are using -D_GLIBCXX_USE_CXX11_ABI=1

But that's using gitlab-ci, not travis.

[eighthave]

The Android builds in general need to use `-D_GLIBCXX_USE_CXX11_ABI=1` whether run by gitlab-ci, manually, or otherwise. As far as I can tell, the Android build of Chromium/libwebrtc seems to force it somehow.

[uumaro]

The current situation is awkward... @arlolra, do I understand right that the only reason for setting -D_GLIBCXX_USE_CXX11_ABI=0 in .pc files is compatibility with the Travis CI environment?

It seems that setting =0 is working against defaults. The Tor Browser build already uses =1, just by using the defaults in projects/webrtc, which requires externally overriding the .pc file value on the command line.

How does one test changes to the .travis.yml file? It looks like Travis now supports dist: xenial (16.04); I'd like to try that, using prebuilt libraries built with =1.

@eighthave:

The Android builds in general need to use -D_GLIBCXX_USE_CXX11_ABI=1 whether run by gitlab-ci, manually, or otherwise. As far as I can tell, the Android build of Chromium/libwebrtc seems to force it somehow.

Yes indeed: in order to get the linux-amd64 and linux-arm libraries built with with =0, I had to hack the webrtc source :( I didn't find a way to control it externally.

[eighthave]

The Travis-CI enviroment is ancient: Ubuntu 14.04 running Go 1.6. I see no reason to support something that old, especially for a project based in Go, which as a platform seems all about leaving the past behind.

[arlolra]

The current situation is awkward... @arlolra, do I understand right that the only reason for setting -D_GLIBCXX_USE_CXX11_ABI=0 in .pc files is compatibility with the Travis CI environment?

Yes, looking back at when this was filed, I had said,

I briefly tried setting -D_GLIBCXX_USE_CXX11_ABI=1 in CI but couldn't get it to work. I think the toolchain is built with --disable-libstdcxx-dual-abi so that definition is moot.

#79 (comment)

How does one test changes to the .travis.yml file? It looks like Travis now supports dist: xenial (16.04); I'd like to try that, using prebuilt libraries built with =1.

Seems reasonable.

Since you don't have permissions on this repo, you'll need to fork and enable Travis on a personal repo. It can be set to run on every push to a branch. Alternatively, you can just submit a pull request (and continue force pushing to a branch created for that) and it'll run, since it's enable on pulls here.

[uumaro]

I did a test in #92, which changes dist: trusty to dist: xenial and removes -D_GLIBCXX_USE_CXX11_ABI=0 from the .pc files (and rebuilds the magic). It failed, but because of an asan error, which is not what I was expecting.

It seems like the error may be related to the ABI change; it happens under FakeDataChannel::label which only constructs and returns a std::string. Unless there is some weird automatic variable lifetime issue I'm missing. If I patch CGO_Channel_Label to return strdup("fake"), then a similar error occurs in CGO_Channel_Protocol.

I'm dropping the log here until I can look at it more closely.

  DataChannel ✔✔✔✔✔✔✔✔✔✔
    Callbacks should fire correctly 
      OnMessage ✔✔✔✔✔✔✔✔✔✔✔✔
      StateChangeCallbacks ✔✔✔✔✔✔✔=================================================================
==5438==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7fff826448b0 at pc 0x7f0448ada964 bp 0x7fff82644860 sp 0x7fff82644008
WRITE of size 4 at 0x7fff826448b0 thread T0
    #0 0x7f0448ada963 in __asan_memcpy (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x8c963)
    #1 0x57ed3f in std::char_traits<char>::copy(char*, char const*, unsigned long) /usr/include/c++/5/bits/char_traits.h:290
    #2 0x57ed3f in std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_S_copy(char*, char const*, unsigned long) /usr/include/c++/5/bits/basic_string.h:299
    #3 0x57ed3f in std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_S_copy_chars(char*, char const*, char const*) /usr/include/c++/5/bits/basic_string.h:346
    #4 0x57ed3f in void std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_M_construct<char const*>(char const*, char const*, std::forward_iterator_tag) /usr/include/c++/5/bits/basic_string.tcc:225
    #5 0x57ee25 in void std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_M_construct_aux<char const*>(char const*, char const*, std::__false_type) /usr/include/c++/5/bits/basic_string.h:195
    #6 0x57ee25 in void std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_M_construct<char const*>(char const*, char const*) /usr/include/c++/5/bits/basic_string.h:214
    #7 0x57ee25 in std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::basic_string(char const*, std::allocator<char> const&) /usr/include/c++/5/bits/basic_string.h:457
    #8 0x57ee25 in FakeDataChannel::label[abi:cxx11]() const /home/travis/gopath/src/github.com/keroserene/go-webrtc/datachannel.cc:99
    #9 0x57ee25 in CGO_Channel_Label /home/travis/gopath/src/github.com/keroserene/go-webrtc/datachannel.cc:31
    #10 0x57c47f in _cgo_aeba63b75ef5_Cfunc_CGO_Channel_Label (/tmp/go-build842285155/github.com/keroserene/go-webrtc/_test/go-webrtc.test+0x57c47f)
    #11 0x47039f in runtime.asmcgocall (/tmp/go-build842285155/github.com/keroserene/go-webrtc/_test/go-webrtc.test+0x47039f)
Address 0x7fff826448b0 is located in stack of thread T0==5438==AddressSanitizer CHECK failed: ../../../../src/libsanitizer/asan/asan_thread.cc:231 "((ptr[0] == kCurrentStackFrameMagic)) != (0)" (0x0, 0x0)
    #0 0x7f0448aee661  (/usr/lib/x86_64-linux-gnu/libasan.so.2+0xa0661)
    #1 0x7f0448af3613 in __sanitizer::CheckFailed(char const*, int, char const*, unsigned long long, unsigned long long) (/usr/lib/x86_64-linux-gnu/libasan.so.2+0xa5613)
    #2 0x7f0448af1359  (/usr/lib/x86_64-linux-gnu/libasan.so.2+0xa3359)
    #3 0x7f0448aeb769  (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x9d769)
    #4 0x7f0448aebd4a  (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x9dd4a)
    #5 0x7f0448aed626 in __asan_report_error (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x9f626)
    #6 0x7f0448ada981 in __asan_memcpy (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x8c981)
    #7 0x57ed3f in std::char_traits<char>::copy(char*, char const*, unsigned long) /usr/include/c++/5/bits/char_traits.h:290
    #8 0x57ed3f in std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_S_copy(char*, char const*, unsigned long) /usr/include/c++/5/bits/basic_string.h:299
    #9 0x57ed3f in std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_S_copy_chars(char*, char const*, char const*) /usr/include/c++/5/bits/basic_string.h:346
    #10 0x57ed3f in void std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_M_construct<char const*>(char const*, char const*, std::forward_iterator_tag) /usr/include/c++/5/bits/basic_string.tcc:225
    #11 0x57ee25 in void std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_M_construct_aux<char const*>(char const*, char const*, std::__false_type) /usr/include/c++/5/bits/basic_string.h:195
    #12 0x57ee25 in void std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::_M_construct<char const*>(char const*, char const*) /usr/include/c++/5/bits/basic_string.h:214
    #13 0x57ee25 in std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::basic_string(char const*, std::allocator<char> const&) /usr/include/c++/5/bits/basic_string.h:457
    #14 0x57ee25 in FakeDataChannel::label[abi:cxx11]() const /home/travis/gopath/src/github.com/keroserene/go-webrtc/datachannel.cc:99
    #15 0x57ee25 in CGO_Channel_Label /home/travis/gopath/src/github.com/keroserene/go-webrtc/datachannel.cc:31
    #16 0x57c47f in _cgo_aeba63b75ef5_Cfunc_CGO_Channel_Label (/tmp/go-build842285155/github.com/keroserene/go-webrtc/_test/go-webrtc.test+0x57c47f)
    #17 0x47039f in runtime.asmcgocall (/tmp/go-build842285155/github.com/keroserene/go-webrtc/_test/go-webrtc.test+0x47039f)
exit status 1
FAIL	github.com/keroserene/go-webrtc	0.034s
The command "if [[ "$OS" == "linux" ]]; then CGO_CFLAGS="-fsanitize=address" CGO_LDFLAGS="-fsanitize=address -fuse-ld=gold" go test -v .; fi" exited with 1.

[uumaro]

It seems like the error may be related to the ABI change; it happens under FakeDataChannel::label which only constructs and returns a std::string. Unless there is some weird automatic variable lifetime issue I'm missing. If I patch CGO_Channel_Label to return strdup("fake"), then a similar error occurs in CGO_Channel_Protocol.

arlolra points out that this error is reminiscent of #43 (golang/go#16150). If I read the latter ticket correctly, it says that ASan and cgo are not generally compatible. #43 was about converting a Go string to a C++ string; and the error in this ticket is the other direction. So maybe this is not a real error? Maybe we should get rid of the ASan tests?

[arlolra]

Maybe we should get rid of the ASan tests?

But then you wouldn't have caught #95

[uumaro]

So maybe this is not a real error? Maybe we should get rid of the ASan tests?

Indeed, I removed the -fsanitize=address test in a commit in #92, and the CI build completed successfully. So the ABI situation is sorted out; the error was being caused instead by some interation with AddressSanitizer.

[uumaro]

Maybe we should get rid of the ASan tests?

But then you wouldn't have caught #95

Correct, the FakeDataChannel::label in this ticket seems like a false positive; but I'm less sure that #95 is benign.

[arlolra]

but I'm less sure that #95 is benign.

Yeah, that's why I retracted (deleted) my comment on the other ticket; it looks like something worth investigating.

[arlolra]

Resolved by #92