Ok... I need to rework constExprHash so no C++14 features are used

I haven't tried to see if it give the same result, but maybe something like this ?

constexpr size_t hash(const char* str, size_t val = 14695981039346656037ULL)
{
    return *str == '\0'
        ? val
        : hash(++str, val ^ static_cast<size_t>(*str) * 1099511628211ULL);
}

https://godbolt.org/z/7vqYqPTPh

EDIT: It probably won't give the same result because the order of evaluation of the parameters is unspecified, maybe ++str will evaluate first, maybe *str. To fix that we can simply write str + 1 instead of ++str, so the variable won't change its value.

EDIT 2: And it seems that the multiply operator take precedence over the bitwise XOR operator, yet again something that makes it non conforment (but easily fixable with parenthesis).

Coverage decreased (-4.0e-05%) to 99.925% when pulling 2230a37 on otrempe:Fix-CollidingMockingContextIds into 78ca536 on eranpeer:master.

You need to add the new test files to https://github.com/eranpeer/FakeIt/blob/master/build/sources.mk as well, otherwise they won't run for GCC / Clang.

Coverage decreased by 4.0e-05% 😮 😆

The Visual Studio build yields lots of warning C4307: '*': integral constant overflow
However, Microsoft stopped issuing this warning starting with MSVC v19.23.

I guess my .h/.cpp helper files need to be added to the sources.mk as well

You're right, but only the .cpp.

Thanks for the fix, I'll try to look more into it next week, I'll do some benchmark because I'm curious if it affect compilation time or not.

I didn't see any measurable difference in compilation time with the fix for the FakeIt unit tests with GCC. I guess it's because our test files are small and the includes take a lot of time to compile, maybe with a test file that is a lot bigger (and pre-compiled header ?) we could notice a difference, or maybe the overhead is so insignificant that we wouldn't notice anything anyway. Either way it looks good to me.

If using precompiled headers (e.g. with VS) it might make the difference more noticeable (if fakeit.h is added to that).

Superseded by #336. I went with a different approach that I explained in the PR.