[testsfail] Process.* (memory leak in libhybris generated when probing android platforms)

[Alan Griffiths]

https://mir-jenkins.ubuntu.com/job/build-2-binpkg-mir/arch=amd64,compiler=clang,platform=mesa,release=vivid+overlay/1493/consoleFull

09:10:14 15: [ FAILED ] Process.a_main_fn_is_executed
09:10:14 15: [ FAILED ] Process.a_successful_exit_function_succeeds

https://mir-jenkins.ubuntu.com/job/build-2-binpkg-mir/arch=i386,compiler=gcc,platform=mesa,release=xenial+overlay/1493/consoleFull

09:06:54 17: [ FAILED ] Process.a_main_fn_is_executed
09:06:54 17: [ FAILED ] Process.a_successful_exit_function_succeeds

I can only imagine the build agent got no CPU for so long that wait_for_termination() timed out.

[Alan Griffiths (alan-griffiths)]

Here too: https://mir-jenkins.ubuntu.com/job/build-2-binpkg-mir/arch=amd64,compiler=clang,platform=mesa,release=vivid+overlay/1488/consoleFull

[Alan Griffiths (alan-griffiths)]

And again: https://mir-jenkins.ubuntu.com/job/build-2-binpkg-mir/arch=amd64,compiler=clang,platform=mesa,release=vivid+overlay/1494/consoleFull

[Alan Griffiths (alan-griffiths)]

And: https://mir-jenkins.ubuntu.com/job/build-2-binpkg-mir/arch=amd64,compiler=clang,platform=mesa,release=vivid+overlay/1495/consoleFull

[Alan Griffiths (alan-griffiths)]

https://mir-jenkins.ubuntu.com/job/build-2-binpkg-mir/arch=amd64,compiler=clang,platform=mesa,release=vivid+overlay/1496/consoleFull

[Kevin DuBois (kdub)]

iirc, Process* depends on the success of previous forks() (a bug in and of itself), but is not the root cause.

The cause seems to be a new libhybris release, which introduced a leak in the hw_get_module function. This function is used in platform probing, so it generates an error in valgrind.

unfortunately, valgrind seems to get confused about where this leak happens:
==8558==
==8558== 192 bytes in 1 blocks are definitely lost in loss record 246 of 271
==8558== at 0x4C2DB8F: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==8558== by 0xB56CE27: doc src
==8558== by 0xB56CFFB: doc src
==8558== by 0xB56C677: doc src
==8558== by 0xB77F59A: doc src
==8558== by 0xB77F97C: doc src
==8558== by 0xB77FC5E: doc src
==8558== by 0xB3697A8: doc src
==8558== by 0xAEB9030: std::_Tuple_impl<1ul, mir::ModuleDeleter<mir::test::doubles::NullPlatformIpcOperations> >::~_Tuple_impl() (tuple:339)
==8558== by 0x70469F: mir::graphics::module_for_device(std::vector<std::shared_ptr<mir::SharedLibrary>, std::allocator<std::shared_ptr<mir::SharedLibrary> > > const&, mir::options::ProgramOption const&) (platform_probe.cpp:38)

so I havent been able to come up with a suppression about it yet (which is the real fix)

[Kevin DuBois (kdub)]

oh, also, I suspect the reason that this was not reproducible was that only X+O systems would have the new libhybris, and would be the only ones with the leak.

[Daniel van Vugt (vanvugt)]

Fix committed to lp:mir/0.23 at revision 3530, scheduled for release in Mir 0.23.5

[Daniel van Vugt (vanvugt)]

This bug was fixed in the package mir - 0.23.5+16.10.20160729-0ubuntu1