jpmag

The call stack indicates the reason: to be able to use the proper version tags in the installed bin. Maybe this could be set up differently, but this is legitimate...

Well, ATM git is required to ~build~ (edit: install). That's it. This is a reasonable expectation when cmake is being run - means it's a developer (edit: or a development...

Also, I'm curious: why is it that you can't just install git?

Thanks for the PR, and sorry for the delay -- github's notifications are really broken for me. I am not familiar with Bazel, but AFAICT the structure is OK, and...

@zaucy do you intend to proceed with this PR?

The test failure is curious. It occurs on the last line of [this test](https://github.com/biojppm/c4core/blob/3729e3fdfa5df6072e1b2d0dcea755d7e977ad58/test/test_charconv.cpp#L2141), so what's happening here is that `atof("nan", &v)` does not result in `std::numeric_limits::quiet_NaN()` (same for double)....

@zaucy thanks for all the hard work, your PR is very thorough. The remarks are all minor, and the main sticking point is to find out whether the problem with...

ok, so given that all CI runs other than bazel are ok, this leads us to conclude that bazel, or the way it runs the tests, has something to do...

It also occurred to me that it may have something to do with how bazel compiles the code. The fact that it happens with `std::numeric_limits::quiet_NaN()` and nowhere else also suggests...

I've looked through the compile flags and nothing there seems to warrant this behavior. One last idea/try on this front: in the failing test, add `static_assert(std::numeric_limits::has_quiet_NaN, "quiet indeed")` and trigger...