[Bug c++/93699] [gcc10] Invalid operator== (returning non-bool type) candidate

[Bug c++/93699] [gcc10] Invalid operator== (returning non-bool type) candidate

[laurent.stacul at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93699

--- Comment #1 from Laurent Stacul <laurent.stacul at gmail dot com> ---
Please note that reversing b and d works fine:

    if (*b == *d) { // This does not work
    if (*d == *b) { // This is ok

Regards,
Stac

[Bug c++/93699] Invalid operator== (returning non-bool type) candidate

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93699

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[gcc10] Invalid operator==  |Invalid operator==
                   |(returning non-bool type)   |(returning non-bool type)
                   |candidate                   |candidate

--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Started to be rejected with r277865 which implements operator<=>, and I think
the compiler is correct to reject it.

For (*b == *d) the arguments do not match Base::operator==(const Base&)
exactly, so the compiler tries a synthesized operator with reversed arguments,
which is ill-formed because of the bad return type.

[Bug c++/93699] Invalid operator== (returning non-bool type) candidate

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93699

--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> ---
P.S. it helps to provide a testcase that doesn't have lots of spurious warnings
when compiled with -Wall -Wextra, because one of the first things I'm going to
do when triaging bug reports is compile with those warnings on.

Uninitialized variables and unused parameters that are not relevant to the bug
are just noise.

Reduced testcase:

struct Base {
    virtual int operator==(const Base&) const {
        return 1;
    }
};

struct Derived : public Base {
    virtual int operator==(const Base&) const {
        return 1;
    }
};

int main()
{
    Base* b = nullptr;
    Derived* d = nullptr;
    if (*b == *d) {
        return 1;
    }
}

[Bug c++/93699] Invalid operator== (returning non-bool type) candidate

[laurent.stacul at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93699

--- Comment #4 from Laurent Stacul <laurent.stacul at gmail dot com> ---
Sorry for this, I will follow your recommendations next time.

[Bug c++/93699] Invalid operator== (returning non-bool type) candidate

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93699

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |diagnostic

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
clang does this as a pedwarn:
<source>:20:12: warning: ISO C++20 requires return type of selected
'operator==' function for rewritten '==' comparison to be 'bool', not 'int'
[-Wrewrite-not-bool]
    if (*b == *d) {
        ~~ ^  ~~


I don't know if it makes sense now to change GCC to do a pedwarn instead of an
error.

[Bug c++/93699] Invalid operator== (returning non-bool type) candidate

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93699

--- Comment #6 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Personally, I think it should remain ill-formed.

Maybe -fpermissive could allow it (for compat with previous standards) but
unfortunately that flag is far too permissive and enables all kinds of ancient
legacy junk. If we cleaned it up to be a bit permissive, not ridiculously
permissive of pre-1998 code, it could be useful here.