[Bug c++/97665] New: constexpr union array member incorrectly rejected as non-constexpr

[Bug c++/97665] New: constexpr union array member incorrectly rejected as non-constexpr

[ldalessandro at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

            Bug ID: 97665
           Summary: constexpr union array member incorrectly rejected as
                    non-constexpr
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ldalessandro at gmail dot com
  Target Milestone: ---

GCC currently rejects the following code snippet, where it infers the constexpr
definition of V as non-constexpr. The monostate member provides a temporary
workaround.

https://godbolt.org/z/Ph1n1P

```
struct Foo {
    constexpr Foo() {}
};

union U {
   // struct {} monostate = {};
    Foo foo;
    constexpr U() {}
};

struct V {
    U storage[1];
};

constexpr V v;
```

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[ldalessandro at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

--- Comment #1 from Luke Dalessandro <ldalessandro at gmail dot com> ---
(In reply to Luke Dalessandro from comment #0)
> GCC currently rejects the following code snippet, where it infers the
> constexpr definition of V as non-constexpr. 

The definition of `v`, not the type `V`. I apologize.

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
You can do Foo foo = Foo(); and it compiles.  clang++ rejects it too:
error: constexpr union constructor that does not initialize any member is a
C++20 extension [-Werror,-Wc++20-extensions]
though only with -pedantic-errors.

But g++ rejects it even in -std=c++20 mode:
pr97665.C:15:13: error: ‘V{U [1]{U()}}’ is not a constant expression
   15 | constexpr V v;
      |             ^
pr97665.C:15:13: error: ‘V()’ is not a constant expression because it refers to
an incompletely initialized variable

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mpolacek at gcc dot gnu.org

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
P1331R2 support is there since
r10-5194-g7906797ebec6881d7d90165340f51efcf447d716 (so without [1] it is
accepted for -std=c++2a since that revision).

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[ldalessandro at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

--- Comment #4 from Luke Dalessandro <ldalessandro at gmail dot com> ---
Hi Jakob,

Thank you for looking at this. I restructured the code sample according to your
suggestions and it is available here https://godbolt.org/z/P1bMEz. I don't
understand a couple of things that you said, and ultimately I'm not sure if you
are agreeing that this could be a bug.

> You can do Foo foo = Foo(); and it compiles.
1. I can't do `Foo foo = Foo();` because the purpose of the union is to
allocate uninitialized storage for the `Foo` during `constexpr` execution when
`Foo` has no default constructor. I realize now I meant to write `constexpr
Foo() = delete;` in the original code. I _can_ use the monostate to have an
active member at initialization, but would prefer not to as it complicates the
union. 

> clang++ rejects it too:
> error: constexpr union constructor that does not initialize any member is a 
> C++20 extension [-Werror,-Wc++20-extensions] though only with -pedantic-errors.
2. I can't get clang to emit the warning you describe even with the provided
flags, perhaps I am doing it wrong or misinterpreting your comment.

> P1331R2 support is there since 
> r10-5194-g7906797ebec6881d7d90165340f51efcf447d716 
> (so without [1] it is accepted for -std=c++2a since that revision)
3. I think this means that, if the member is not an array, then it is accepted
with `-std=c++2a`. I do observe this, however it's not my use case. Technically
I have an array of some class template parameter `N`.

The updated test case with the deleted constructor is.

```
struct Foo {
    constexpr Foo() = delete;
};

union U {
    // struct {} monostate = {};
    Foo foo;
    constexpr U() {}
};

struct V {
    U storage[1];
};

constexpr V v;
```

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[ldalessandro at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

--- Comment #5 from Luke Dalessandro <ldalessandro at gmail dot com> ---
Ugh... replying to myself.

> You can do Foo foo = Foo(); and it compiles.

>> 1. I can't do `Foo foo = Foo();` because the purpose of the union is to allocate uninitialized storage for the `Foo` during `constexpr` execution when `Foo` has no default constructor. I realize now I meant to write `constexpr Foo() = delete;`.

This means to say "trivial default constructor," not just "default
constructor". The empty default constructor was an adequate test case for this,
but is misleading. The `= delete` should be more effective as an example.

The context for the use is to create an array that provides uninitialized
vector storage during `constexpr` evaluation in the same way a
`std::byte[T*sizeof(T)]` array combined with `reinterptret_cast<T*>` might be
used during normal execution.

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
All I'm saying is that I believe your code is not valid C++17, but is valid
C++20 and that for C++20 we have a compiler bug we need to fix.

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[ldalessandro at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

--- Comment #7 from Luke Dalessandro <ldalessandro at gmail dot com> ---
Thank you, sorry for the confusion.

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2021-08-04

--- Comment #8 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Confirmed.

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The array is not needed to reproduce this though:
struct Foo {
    constexpr Foo() {}
};

union U {
   // struct {} monostate = {};
    Foo foo;
    constexpr U() {}
};
constexpr U s;

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[dysnomia.eris at online dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

dysnomia.eris at online dot de changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dysnomia.eris at online dot de

--- Comment #10 from dysnomia.eris at online dot de ---
Confirmed.

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[de34 at live dot cn]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

Jiang An <de34 at live dot cn> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |de34 at live dot cn

--- Comment #11 from Jiang An <de34 at live dot cn> ---
This looks like CWG issue 2558 (currently unresolved).

https://cplusplus.github.io/CWG/issues/2558.html

[Bug c++/97665] constexpr union array member incorrectly rejected as non-constexpr

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97665

--- Comment #12 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Jiang An from comment #11)
> This looks like CWG issue 2558 (currently unresolved).
> 
> https://cplusplus.github.io/CWG/issues/2558.html

It was voted in a few months after this comment was added.

But the original testcase has no scalars ...