Fix vsnprintf mocks for Release builds.

[hidmic]

This pull request deals with binary API nuances. Fixes regressions introduced by #272.

CI up to rcutils:

• Linux (default)
• Linux (release)

[ivanpauno]

I think this kind of mocking is really fragile.

I prefer having less coverage and not doing this, but I guess it's not my call.

[hidmic]

I think this kind of mocking is really fragile.

Indeed mocking system libraries this way is fragile, but as of today the alternative is not having coverage at all. Perhaps, eventually we can migrate towards fault injection (which has its own share of ifs and buts, like Release binary and Debug binaries not being the same).

I prefer having less coverage

Generally speaking, I don't agree with reducing test coverage simply because it's inconvenient. I'd much rather look for a replacement if this becomes a pain point.

[hidmic]

Thanks for the reviews!

[nuclearsandwich]

@hidmic the change introduced here is causing build failures in Release configuration on CentOS:

There are several related errors but here's the first one:

In file included from /home/jenkins-agent/workspace/nightly_linux-centos_extra_rmw_release/ws/src/ros2/rcutils/test/test_shared_library.cpp:20:
/home/jenkins-agent/workspace/nightly_linux-centos_extra_rmw_release/ws/src/ros2/rcutils/test/test_shared_library.cpp: In member function ‘virtual void TestSharedLibrary_basic_load_Test::TestBody()’:
/home/jenkins-agent/workspace/nightly_linux-centos_extra_rmw_release/ws/src/ros2/rcutils/test/./mocking_utils/stdio.hpp:67:12: error: ‘__vsnprintf_chk’ was not declared in this scope
     scope, __vsnprintf_chk, [&](char * __str, size_t, int, size_t __size, \
            ^~~~~~~~~~~~~~~
/home/jenkins-agent/workspace/nightly_linux-centos_extra_rmw_release/ws/src/ros2/rcutils/test/./mocking_utils/patch.hpp:334:36: note: in definition of macro ‘patch’
   make_patch<__COUNTER__, decltype(function)>( \
                                    ^~~~~~~~
/home/jenkins-agent/workspace/nightly_linux-centos_extra_rmw_release/ws/src/ros2/rcutils/test/test_shared_library.cpp:64:32: note: in expansion of macro ‘patch_vsnprintf’
     auto mock = mocking_utils::patch_vsnprintf(
                                ^~~~~~~~~~~~~~~
/home/jenkins-agent/workspace/nightly_linux-centos_extra_rmw_release/ws/src/ros2/rcutils/test/./mocking_utils/stdio.hpp:67:12: note: suggested alternative: ‘__asprintf’
     scope, __vsnprintf_chk, [&](char * __str, size_t, int, size_t __size, \
            ^~~~~~~~~~~~~~~
/home/jenkins-agent/workspace/nightly_linux-centos_extra_rmw_release/ws/src/ros2/rcutils/test/./mocking_utils/patch.hpp:334:36: note: in definition of macro ‘patch’
   make_patch<__COUNTER__, decltype(function)>( \
                                    ^~~~~~~~
/home/jenkins-agent/workspace/nightly_linux-centos_extra_rmw_release/ws/src/ros2/rcutils/test/test_shared_library.cpp:64:32: note: in expansion of macro ‘patch_vsnprintf’
     auto mock = mocking_utils::patch_vsnprintf(
                                ^~~~~~~~~~~~~~~

The failure does not occur in Debug or None configurations, only Release and removing this commit unbreaks the build. However since this is a fix for an issue on tier 1 platforms I am not proposing a revert.

[nuclearsandwich]

This change also does not appear to address the test failures in the nightly_linux_thread_sanitizer job.

[hidmic]

Hmm, this is becoming increasingly painful. I'm surprised about tsan changing binary APIs, though at this point it seems clear that portability in system libraries applies to sources only. I'll check how much coverage worsens without this and revert if we're still in the clear.