Add StringMaker for std::variant, std::monostate

[melak47]

Description

Adds StringMaker specializations for std::variant and std::monostate, which can be enabled with CATCH_CONFIG_ENABLE_VARIANT_STRINGMAKER or CATCH_CONFIG_ENABLE_ALL_STRINGMAKERS.

Also includes configuration macros for detecting C++17 and header presence, so it doesn't break the build for anyone on C++ < 17 who's using CATCH_CONFIG_ENABLE_ALL_STRINGMAKERS.

Tests are modelled after the tuple tests.

[melak47]

Hm, seems the tests fail to compile on clang 6 + libstdc++ because of a bug: https://bugs.llvm.org/show_bug.cgi?id=31852
Newer clang should have a fix, newer libstdc++ should have a workaround, but I'm not sure what to do now :/

Edit: clang trunk has the fix, not sure about the upcoming 7.0 release, and gcc 8.2.0's libstdc++has the workaround for clang 6.

[horenmar]

You can try modifying .travis.yml to use a newer libstdc++ for the Clang+C++17 builds.

Alternatively, I'll see what I can do on Monday.

[codecov]

Codecov Report

Merging #1380 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master    #1380   +/-   ##
=======================================
  Coverage   80.08%   80.08%           
=======================================
  Files         119      119           
  Lines        3389     3389           
=======================================
  Hits         2714     2714           
  Misses        675      675

[codecov]

Codecov Report

Merging #1380 into master will decrease coverage by 0.03%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #1380      +/-   ##
==========================================
- Coverage   80.11%   80.08%   -0.03%     
==========================================
  Files         119      119              
  Lines        3384     3389       +5     
==========================================
+ Hits         2711     2714       +3     
- Misses        673      675       +2

[melak47]

I tried changing it to libstdc++-8-dev, but that doesn't seem to make a difference.
Also, I just noticed that the output says clang 5.0.0 despite the travis configuration asking for 6.0.

[horenmar]

The output is the image's clang, later on CXX is set to ${COMPILER} which is clang++-6.0 for that image.

---

Anyway, that sounds like it is time for more compile time configuration, specifically _GLIBCXX_RELEASE(https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_macros.html).

The basic scheme is to have a check like this

#include <ciso646> // Get libstdc++ and libc++ version macros
#if defined(__CLANG__) && defined(_GLIBCXX_RELEASE) && (CLANG_MAJOR < 7) && (_GLIBCXX_RELEASE < 8)
// This combination of Clang and libstdc++ can't compile std::variant
#define CATCH_INTERNAL_CONFIG_NO_VARIANT
#endif

I don't actually remember Clang's macros, but you get the idea 😃
The "fun" part is that including <ciso646> has observable side effect under MSVC, so it has to be guarded against that as well.

[melak47]

I did some more testing, libstdc++ 8.2 has the workaround, 8.1 doesn't, clang trunk appears to have the fix, clang 7 nightly doesn't.
_GLIBCXX_RELEASE only gives the major version, so it ended up begin a little uglier than your sample :)

[melak47]

Apparently __GLIBCXX__ is useless to detect the actual libstdc++ version, since it can be the datestamp of whenever the snapshot was taken: https://gcc.gnu.org/onlinedocs/libstdc++/manual/abi.html#abi.versioning.__GLIBCXX__

So that leaves disabling the feature for all versions of libstdc++ <= 8, and letting the user override with CATCH_CONFIG_CPP17_VARIANT :/

[horenmar]

Yeah, __GLIBCXX__ does not work.

I am going to do a quick fixup and then squash + merge the PR into master.