https://travis-ci.org/ned14/boost-outcome
The first one is for boost-outcome, correct? It only tests g++-6 with c++14, not g++-7 (in 14 or 17) (where the ICE occurs), or or clang+libc++ (where your other errors were).
The same constexpr ICE problem afflicts GCC 6. But similarly it happens randomly across machines.
Aren't the second and third for standalone outcome? That isn't being reviewed here, correct? (i.e. Standalone outcome doens't even ICE MSVC2017 with Vinnie's test case, but boost-outcome, the one being submitted for review, does?).
I didn't think it necessary to run more CIs than at present. I know they appear to be free to us, but I felt it anti-social to be using more jobs than is strictly necessary. And the existing CIs cover all of MSVC, clang and GCC, with Dinkumware, libstdc++ and libc++ on Linux, Windows and MacOS. That's a good cross section of the most common configurations. I am surprised about the clang + libc++ missing header because the XCode 9 CI job ought to have caught that, but that happens sometimes. boost-outcome is the same library as Outcome standalone, if you do a side by side diff you'll see that. The only change is the use of Boost.Config instead of SD6 feature macros, and Boost.Exception. Hence I felt that boost-outcome didn't need much testing above making sure it works inside Boost, which is what the CI job does. Niall -- ned Productions Limited Consulting http://www.nedproductions.biz/ http://ie.linkedin.com/in/nialldouglas/