[cwg1467] Overload resolution not working with `initializer_list` and `bool`

[llvmbot]

Bugzilla Link	22259
Version	3.5
OS	Linux
Depends On	#10531
Reporter	LLVM Bugzilla Contributor
CC	@DougGregor,@zygoloid

Extended Description

Test case.

#include <string>
#include <iostream>

void Test(const bool arg) 
{
    std::cout << "Wrong[" << arg << "]" << std::endl;
}
void Test(const std::initializer_list<std::string> arg)
{
    std::cout << "Right[" << arg.size() << "]" << std::endl;
}
    
int main (int /*argc*/, char * const /*argv*/[]) 
{
    Test({"false"});
    return 0;
}

This should print:

  Right[1]

Instead I get:

  Wrong[1]

[llvmbot]

There's something weird going on with strings, here's another test case.

#include <stdio.h>
#include <string>
#include <initializer_list>
    
void Test1(const bool arg) 
{
    printf("Wrong[%s]\n", (arg ? "True" : "False"));
}
void Test1(const std::initializer_list<std::string>& arg)
{
    printf("Right[%d]\n", (int)arg.size());
}

void Test2(const bool arg) 
{
    printf("Wrong[%s]\n", (arg ? "True" : "False"));
}
void Test2(const std::initializer_list<char*>& arg)
{
    printf("Right[%d]\n", (int)arg.size());
}

void Test3(const bool arg) 
{
    printf("Wrong[%s]\n", (arg ? "True" : "False"));
}
void Test3(const std::initializer_list<const char*>& arg)
{
    printf("Right[%d]\n", (int)arg.size());
}

int main(int, char**) 
{
    Test1({"Hello"});
    Test2({"Hello"});
    Test2({(char*)"Hello"});
    Test3({"Hello"});
    return 0;
}

This should print:

 Right[1]
 Right[1]
 Right[1]
 Right[1]

Instead I get:

 Wrong[True]
 Wrong[True]
 Right[1]
 Right[1]

[llvmbot]

This seems directly related to bug 10159, since all pointer types are also freely assignable to bool.

[llvmbot]

So in fact I believe clang is doing the right thing. I was comparing to GCC behavior. I think GCC has the reverse bug instead.

[llvmbot]

Not a bug

[zygoloid]

Under DR1467, your code should pick the Right function in each case. See

http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#1467

Reopening to track the fact that we don't implement DR1467 yet.

[llvmbot]

I don't think this is related to DR1467, instead I think it's about conversion sequences. I think clang is OK, and GCC has incorrect behavior. I raised an issue for GCC here (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64665). I explain it a little more there.

[zygoloid]

I don't think this is related to DR1467, instead I think it's about
conversion sequences.

The final change made by DR1467 affects the ordering of implicit conversion sequences: in particular, it changes the rules so that the "a list initialization sequence prefers a std::initializer_list parameter over any other parameter" overrules all the other rules (note the addition of "even if one of the other rules in this paragraph would otherwise apply").

[llvmbot]

OK that's good, intuitively that's what you'd expect.

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: None (llvmbot)

| | | | --- | --- | | Bugzilla Link | [22259](https://llvm.org/bz22259) | | Version | 3.5 | | OS | Linux | | Depends On | llvm/llvm-project#10531 | | Reporter | LLVM Bugzilla Contributor | | CC | @DougGregor,@zygoloid |

Extended Description

Test case.

#include <string>
#include <iostream>

void Test(const bool arg) 
{
    std::cout << "Wrong[" << arg << "]" << std::endl;
}
void Test(const std::initializer_list<std::string> arg)
{
    std::cout << "Right[" << arg.size() << "]" << std::endl;
}
    
int main (int /*argc*/, char * const /*argv*/[]) 
{
    Test({"false"});
    return 0;
}

This should print:

  Right[1]

Instead I get:

  Wrong[1]