Fix Run-Time Check Failure

[Zenju]

Hey STL,

just watched your cppcon talk. I couldn't help myself and had to implement std::to_chars/std::from_chars immediately. Had a few issues:

1. #include <charconv> in my precompiled header file:
   => compiler error: "fatal error C1076: compiler limit: internal heap limit reached"

2. #include <charconv> in another header, widely-used in my project:
   => compiler error: "An internal error has occurred in the compiler. (compiler file 'd:\agent_work\1\s\src\vctools\Compiler\Utc\src\p2\p2symtab.c', line 2618)"

=> created a wrapper and #include <charconv> locally in a .cpp file => fine.

3. Runtime check error in std::to_chars (for double(12.34))
   => patch included

Best, Zenju

[sylveon]

Pretty sure this is not a supported scenario:

T:\Projects>type test.cpp
#include <iostream>

T:\Projects>cl /RTCc test.cpp
Microsoft (R) C/C++ Optimizing Compiler Version 19.22.27905 for x86
Copyright (C) Microsoft Corporation.  All rights reserved.

test.cpp
C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.22.27905\include\yvals_core.h(840): fatal error C1189: #error:  /RTCc rejects conformant code, so it is not supported by the C++ Standard Library. Either remove this compiler option, or define _ALLOW_RTCc_IN_STL to acknowledge that you have received this warning.

[BillyONeal]

(1) suggests that we might need to move the ryu tables to be in the import lib. Can you file that as an issue?
(2) is a compiler bug, not an STL bug. Please report through https://developercommunity.visualstudio.com/spaces/62/index.html with a self contained repro.
(3) is /RTCc, which is, as @sylveon indicated, not supported.

I defer to @StephanTLavavej on anything <charconv> related, but given that we don't support /RTCc I don't think we should merge this.

[StephanTLavavej]

Thanks for trying <charconv>! I've filed #172 to track moving the lookup tables into the static/import lib. (As a workaround, you can use /Zm to increase the PCH size; this should be necessary for x86 only, as x64 uses a massively larger default.)

We blocked /RTCc for exactly this reason - it rejects perfectly conformant code, and there are other places in the STL that are incompatible with it. I agree with @BillyONeal that this change shouldn't be merged, so I'm closing this PR.

If you're curious as to what's happening here: we are truncating uint64_t to uint32_t because we have a mathematical guarantee that we don't need the upper bits. We're starting with uint64_t vr and calculating vr % 100 as vr - 100 * (vr / 100). (This benefits earlier versions of Clang which had difficulty seeing the relationship between modulo and division.) To benefit 32-bit platforms, we don't actually need to use 64-bit math for this. By definition, vr % 100 is within [0, 99]. That means that in the subtraction, vr and 100 * (vr / 100) must match in all of their high bits (only the lowest 7 can differ because 2⁷ - 1 = 128). Therefore we can simply discard the upper 32 bits and perform the math in uint32_t.