<chrono>: building with -J emits C4365

[tiagomacarios]

Building with -J will trigger C4365.

https://godbolt.org/z/dxbj64q36

#include <chrono>

output:

example.cpp
C:/data/msvc/14.34.31931-Pre/include\xcharconv_ryu.h(907): error C2220: the following warning is treated as an error
C:/data/msvc/14.34.31931-Pre/include\xcharconv_ryu.h(907): warning C4365: '=': conversion from 'int' to 'char', signed/unsigned mismatch
Compiler returned: 2

[strega-nil-ms]

We do not support building with /Wall, and I believe that this warning is off by default. I do not believe that we should change code due to this warning, since c = c + 1; is completely reasonable code and it's perfectly well defined in this case no matter what c is.

[tiagomacarios]

Office compiles with Wall. We can fix our fork of the STL, but we would prefer to have fixes upstreamed and reduce our maintenance surface.

[strega-nil-ms]

Alright, we'll discuss it; I'm personally against changing anything here, since this warning is (imo) quite silly. We could add the warning to our "disabled warnings" set fairly easily, but I'm concerned about setting a precedent for supporting /Wall.

[CaseyCarter]

C4365 is one of very few warnings in our supported off-by-default warning set; we enable it in our "std" test suite, but in neither tr1 nor libcxx:

STL/tests/std/tests/prefix.lst

Line 6 in 8ddf4da

PM_CL="/FIforce_include.hpp /w14365 /D_ENFORCE_FACET_SPECIALIZATIONS=1 /D_STL_CALL_ABORT_INSTEAD_OF_INVALID_PARAMETER"

I'm perfectly happy to disable it in <xcharconv_ryu.h> just as we do in several other places. (IMO the real issue here is /J, which IIRC we support but probably shouldn't since we obviously have no test coverage.)