Current LLVM copy of to_chars is wrong for integers with more than 9 digits

[smehringer]

Platform

• SeqAn version: tested with 3.0.1
• Operating system: Linux
• Compiler: tested with GCC-7/GCC-9

Description

Converting a number with 9 digits or more fails with the current implementation in the std/charconv header. If I include the stl <charconv> with GCC-9 directly the error does not occur.

There is also the possibility that we introduced the error while copying..
We should update the code and may want to consider copying the MSVC version that is supposed to be fast on floats.

How to repeat the problem

The following test fails with the current implementation

TEST(to_chars_test, large_number)
{
    uint64_t val{123456789};
    std::array<char, 100> buffer{};

    auto res = std::to_chars(buffer.data(), buffer.data() + buffer.size(), val);

    EXPECT_EQ(res.ptr, &buffer[9]);
    EXPECT_EQ(res.ec, std::errc{});
    EXPECT_EQ((std::string_view{buffer.data(), 9}), std::string_view{"123456789"});
}

[marehr]

Core-Meeting:

@marehr has free reign to do this; main idea is to put the llvm implementation into contrib and add the custom bits, like the float overloads, in std/charconv.

[marehr]

2020-06-30 Core Meeting

Discussed license change in the LLVM project from

This file is dual licensed under the MIT and the University of Illinois Open Source Licenses.
https://github.com/llvm/llvm-project/blob/d27489645bb8829a952f79e3b211202844238391/libcxx/src/charconv.cpp

to

SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
https://github.com/llvm/llvm-project/blob/master/libcxx/src/charconv.cpp