Bug in GCC's resolution of C++20 reversed operator== functions?

Bug in GCC's resolution of C++20 reversed operator== functions?

[Chris Peterson]

[-- Attachment #1: Type: text/plain, Size: 1101 bytes --]

While updating Firefox from -std=c++17 to -std=c++20, I found a case where
GCC's resolution of C++20 reversed operator== functions behaves differently
from the Clang, MSVC, and ICX compilers. I believe this difference was a
regression in GCC 10.1.

Here's a Godbolt test case comparing those compilers' output:

https://godbolt.org/z/qneax5oaW

```
#include <type_traits>

struct Thing {
    template <typename T>
    bool operator==(const T& rhs) const {
        /* This operator== is selected by:
         *   GCC versions >= 10.1 -std=c++17
         *   GCC version 9.5 -std=c++2a
         *   Clang 18.1 -std=c++2a
         *   MSVC 19.38 -std=c++20
         *   Intel's ICX 2024.0.0 -std=c++20
         */
        return false;
    }
};

template <typename T>
bool operator==(T const& lhs, Thing const& rhs) {
    /* This operator== is selected by:
     *   GCC versions >= 10.1 -std=c++2a
     */
    return true;
}

bool test() {
    Thing const v{};
    return v == 3;
}
```

(I have an account on the GCC Bugzilla, but I'm not able to log in or reset
my password to file a bug there.)

chris

Re: Bug in GCC's resolution of C++20 reversed operator== functions?

[Jonathan Wakely]

[-- Attachment #1: Type: text/plain, Size: 1344 bytes --]

On Sun, 31 Mar 2024, 23:02 Chris Peterson, <cpeterso@cpeterso.com> wrote:

> While updating Firefox from -std=c++17 to -std=c++20, I found a case where
> GCC's resolution of C++20 reversed operator== functions behaves differently
> from the Clang, MSVC, and ICX compilers. I believe this difference was a
> regression in GCC 10.1.
>
> Here's a Godbolt test case comparing those compilers' output:
>
> https://godbolt.org/z/qneax5oaW
>
> ```
> #include <type_traits>
>
> struct Thing {
>     template <typename T>
>     bool operator==(const T& rhs) const {
>         /* This operator== is selected by:
>          *   GCC versions >= 10.1 -std=c++17
>          *   GCC version 9.5 -std=c++2a
>          *   Clang 18.1 -std=c++2a
>          *   MSVC 19.38 -std=c++20
>          *   Intel's ICX 2024.0.0 -std=c++20
>          */
>         return false;
>     }
> };
>
> template <typename T>
> bool operator==(T const& lhs, Thing const& rhs) {
>     /* This operator== is selected by:
>      *   GCC versions >= 10.1 -std=c++2a
>      */
>     return true;
> }
>
> bool test() {
>     Thing const v{};
>     return v == 3;
> }
> ```
>
> (I have an account on the GCC Bugzilla, but I'm not able to log in or reset
> my password to file a bug there.)
>


This should be reported to bugzilla so I'll contact you off-list to resolve
the login issues.

Re: Bug in GCC's resolution of C++20 reversed operator== functions?

[Chris Peterson]

[-- Attachment #1: Type: text/plain, Size: 1399 bytes --]

I filed a bug in Bugzilla:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114549


On Sun, Mar 31, 2024 at 3:00 PM Chris Peterson <cpeterso@cpeterso.com>
wrote:

> While updating Firefox from -std=c++17 to -std=c++20, I found a case where
> GCC's resolution of C++20 reversed operator== functions behaves differently
> from the Clang, MSVC, and ICX compilers. I believe this difference was a
> regression in GCC 10.1.
>
> Here's a Godbolt test case comparing those compilers' output:
>
> https://godbolt.org/z/qneax5oaW
>
> ```
> #include <type_traits>
>
> struct Thing {
>     template <typename T>
>     bool operator==(const T& rhs) const {
>         /* This operator== is selected by:
>          *   GCC versions >= 10.1 -std=c++17
>          *   GCC version 9.5 -std=c++2a
>          *   Clang 18.1 -std=c++2a
>          *   MSVC 19.38 -std=c++20
>          *   Intel's ICX 2024.0.0 -std=c++20
>          */
>         return false;
>     }
> };
>
> template <typename T>
> bool operator==(T const& lhs, Thing const& rhs) {
>     /* This operator== is selected by:
>      *   GCC versions >= 10.1 -std=c++2a
>      */
>     return true;
> }
>
> bool test() {
>     Thing const v{};
>     return v == 3;
> }
> ```
>
> (I have an account on the GCC Bugzilla, but I'm not able to log in or
> reset my password to file a bug there.)
>
> chris
>