[Bug c++/107945] New: GCC accepts invalid program involving constexpr static data member of class template

[Bug c++/107945] New: GCC accepts invalid program involving constexpr static data member of class template

[jlame646 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107945

            Bug ID: 107945
           Summary: GCC accepts invalid program involving constexpr static
                    data member of class template
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jlame646 at gmail dot com
  Target Milestone: ---

The following invalid program is accepted by gcc but rejected by clang and
msvc. https://godbolt.org/z/WGz3f75da

```
struct incomplete;
template<typename T> struct C
{
    static constexpr T t{};
};

template<class T>
struct myClass {
    C<T> new_t() { return {}; }
};

int main() {
    myClass<incomplete> d;
    d.new_t();    
}
```

[Bug c++/107945] GCC accepts invalid program involving constexpr static data member of class template

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107945

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Reduced:
```
struct incomplete;
template<typename T> struct C
{
    static constexpr T t{};
};


int main() {
    C<incomplete> t;
   // t.t;
}
```
If we uncomment t.t, then GCC will reject it.

[Bug c++/107945] GCC accepts invalid program involving constexpr static data member of class template

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107945

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
ICC has the same behavior as GCC, though I don't know if it was EDG marking
this as an extension or not.

[Bug c++/107945] static constexpr incomplete (depedent) data member of a class template and in-member initialized accepted

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107945

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|static incomplete           |static constexpr incomplete
                   |(depedent) data member of a |(depedent) data member of a
                   |class template accepted     |class template and
                   |                            |in-member initialized
                   |                            |accepted
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2022-12-01

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
If we make the field not depedent then GCC rejects it:
```
struct incomplete;
template<typename T>
struct C
{
    static constexpr incomplete t{};
};
```

Confirmed.

[Bug c++/107945] static constexpr incomplete (depedent) data member of a class template and in-member initialized incorrectly accepted

[jlame646 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107945

--- Comment #4 from Jason Liam <jlame646 at gmail dot com> ---
Here is another invalid snippet that gcc accepts but both clang and msvc
rejects correctly. https://godbolt.org/z/sz4rczEaG

```
#include <array>

template<class T, std::size_t N>
    requires std::is_arithmetic_v<T> && (N >= 1)
class Vector
{
    static constexpr std::size_t Dimension = N;
    std::array<T, Dimension> Elements;

public:
    constexpr Vector() noexcept : Elements{} {}
    constexpr ~Vector() = default;
    static constexpr Vector ZeroVector{};
};

int main()
{
    Vector<float, 7> boo = Vector<float, 7>::ZeroVector;
}
```

Related thread
https://stackoverflow.com/questions/78248691/static-data-member-of-template-class-type-constexpr-vs-const-constinit

[Bug c++/107945] static constexpr incomplete (depedent) data member of a class template and in-member initialized incorrectly accepted

[bbi5291 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107945

Brian Bi <bbi5291 at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bbi5291 at gmail dot com

--- Comment #5 from Brian Bi <bbi5291 at gmail dot com> ---
In the first case, I think GCC's behaviour can be argued to be correct.
[temp.inst]/3.1 says that when you implicitly instantiate a class template, you
only instantiate the declarations of static data members, not their
definitions. Which means that until you do something that provokes the
instantiation of the *definition* of `C<incomplete>::t`, the compiler must
treat it as a variable of type `const incomplete` that is
declared-but-not-defined, which means that you haven't yet actually attempted
to define a variable with incomplete type, which means the program is
well-formed.

In the most recent example given by Jason Liam we can reduce it to:

template <class T>
struct S {
    constexpr static S s = {};
};
int main() {
    auto s = S<int>::s;
}

This may be related to bug 70820, which was closed as INVALID. I think perhaps
that decision should be revisited; while I can see the argument for the above
code being well-formed, I think the interpretation of the standard being taken
in this case is wrong.

[Bug c++/107945] static constexpr incomplete (depedent) data member of a class template and in-member initialized incorrectly accepted

[jlame646 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107945

--- Comment #6 from Jason Liam <jlame646 at gmail dot com> ---
Here is the thread that explains why this is ill-formed. 
https://stackoverflow.com/questions/74642250/incomplete-type-works-with-gcc-but-not-with-clang-and-msvc