I think this is a bug in the compiler; similar code without all of the stuff around it does not fire this warning. I would prefer we fix the compiler rather than work around it in the library, (or at least mark it with a transition).

Okay, now that I look at it, it is an off-by-default warning; I personally think it's very silly to warn for:

unsigned char c = ...;
c = c + 1;

and I'd prefer we just turned off that warning or something along those lines.

We already have the same static_cast for fixing the warning:

I disagree with that - __digits is not the same type as the type we're assigning into, which means it makes sense (to me) to cast it down - it's a narrowing cast.

The unsigned char thing is just a silly result of C's arithmetic promotion rules. It should not be treated as something to warn on.

We talked about this at the weekly maintainer meeting; warning C4365 is indeed one of the few off-by-default warnings we attempt to be clean for:

Thanks for the fix and updated test coverage! This is a minimally invasive change for the Ryu-derived code.

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

I had to push an additional commit to fix the internal test harness, as /clr:pure is incompatible with /J. I've chosen to change the usual_matrix to the impure_matrix, omitting a bit of unnecessary test coverage. (There are an extremely small number of /clr:pure users, virtually all of which are in C++14 mode, so the combination of /clr:pure and /Zc:alignedNew- isn't worth testing.)

Thanks for fixing this warning! 🛠️ ⚠️ ✨