[libcxx-dev] charconv.to.chars/integral.bool.fail.cpp is incorrect

[Marshall Clow via libcxx-dev]

On Thu, Jul 18, 2019 at 9:50 PM Stephan T. Lavavej via libcxx-dev <
libcxx-dev at lists.llvm.org> wrote:

> Hi,
>
> test/std/utilities/charconv/charconv.to.chars/integral.bool.fail.cpp
> claims that "Integral cannot be bool." and expects that calling to_chars()
> with a bool should fail to compile. This is an incorrect interpretation of
> the Standardese. MSVC's STL implements the overload set exactly as depicted
> in the Standard, so this test case compiles successfully.
>
> The Standardese is N4820 20.19.2 [charconv.to.chars]/7 "Remarks: The
> implementation shall provide overloads for all signed and unsigned integer
> types and char as the type of the parameter value." Observe that these are
> depicted as overloads - not as a template.
>

I suspect that that was not Jens' intention when he wrote the wording.
cc'ing him.

-- Marshall



> If implemented as overloads, here's what happens:
>
> C:\Temp>type woof.cpp
> #include <stdio.h>
>
> void ToChars(char) { puts("char"); }
> void ToChars(signed char) { puts("signed char"); }
> void ToChars(unsigned char) { puts("unsigned char"); }
> void ToChars(short) { puts("short"); }
> void ToChars(unsigned short) { puts("unsigned short"); }
> void ToChars(int) { puts("int"); }
> void ToChars(unsigned int) { puts("unsigned int"); }
> void ToChars(long) { puts("long"); }
> void ToChars(unsigned long) { puts("unsigned long"); }
> void ToChars(long long) { puts("long long"); }
> void ToChars(unsigned long long) { puts("unsigned long long"); }
> void ToChars(float) { puts("float"); }
> void ToChars(double) { puts("double"); }
> void ToChars(long double) { puts("long double"); }
>
> int main() {
>     ToChars(false);
> }
>
> C:\Temp>cl /EHsc /nologo /W4 /std:c++17 /permissive- woof.cpp && woof
> woof.cpp
> int
>
> C:\Temp>clang-cl -m64 /EHsc /nologo /W4 /std:c++17 /permissive-
> -fno-ms-compatibility woof.cpp && woof
> int
>
> C1XX and Clang agree: the int overload is called. This is because bool
> undergoes a Promotion to int, and that's an exact match for int, while all
> other types require a Conversion from int to that type.
>
> This test case should be reworked. I mention it, instead of simply fixing
> the test case, because it indicates that libcxx's implementation of
> to_chars() for integral types is also incorrect.
>
> Thanks,
> STL
> _______________________________________________
> libcxx-dev mailing list
> libcxx-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/libcxx-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/libcxx-dev/attachments/20190723/d137453f/attachment.html>