[Bug c++/96593] New: No "declaration changes meaning" diagnostic for alias templates

[Bug c++/96593] New: No "declaration changes meaning" diagnostic for alias templates

[language.lawyer at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96593

            Bug ID: 96593
           Summary: No "declaration changes meaning" diagnostic for alias
                    templates
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Keywords: diagnostic
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: language.lawyer at gmail dot com
  Target Milestone: ---

"A name N used in a class S shall refer to the same declaration in its context
and when re-evaluated in the completed scope of S. No diagnostic is required
for a violation of this rule."

The diagnostic for the rule is not required, but since GCC diagnoses some of
the cases, there is a bugreport about missing diagnostic cases (bug 46983), and
the check might become mandatory (will it?), I'd like file this bug.

GCC accepts the code with alias templates:

struct S {};

template <typename>
struct T {};

struct U
{
        template<typename>
        using S = S;

        template<typename>
        using T = T<void>;
};

but produces the "declaration changes meaning" error for type aliases.
GCC also diagnoses the case if the "outer" declaration is used before the
corresponding alias template:

struct S {};

struct U
{
        S s;

        template<typename>
        using S = S; // error: declaration of ... changes meaning of 'S'
};

(From https://stackoverflow.com/q/63369264/)

[Bug c++/96593] No "declaration changes meaning" diagnostic for alias templates

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96593

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mpolacek at gcc dot gnu.org
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2020-08-18
     Ever confirmed|0                           |1

--- Comment #1 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Like you say, this might be mandatory in C++23: https://wg21.link/p1697r0.  So
confirmed.

[Bug c++/96593] No "declaration changes meaning" diagnostic for alias templates

[gonzalo.gadeschi at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96593

gnzlbg <gonzalo.gadeschi at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gonzalo.gadeschi at gmail dot com

--- Comment #2 from gnzlbg <gonzalo.gadeschi at gmail dot com> ---
*** Bug 102444 has been marked as a duplicate of this bug. ***

[Bug c++/96593] No "declaration changes meaning" diagnostic for alias templates

[gonzalo.gadeschi at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96593

--- Comment #3 from gnzlbg <gonzalo.gadeschi at gmail dot com> ---
>From 102444 (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102444):

See http://eel.is/c++draft/class.member.lookup#6

  [class.member.lookup]/p6: "If it [the result of the search] differs from the
result of a search in T for N from immediately after the class-specifier of T,
the program is ill-formed, no diagnostic required."  A class definition is not
allowed to change the meaning of a name that was already used earlier in the
class definition.

While no diagnostic is required, I think a quality implementation like gcc
should produce a diagnostic here and reject this program since it is
ill-formed.

Example (found in the wild): https://gcc.godbolt.org/z/jhEj68Kq8

    struct plus {
      template<typename... Args>
      using invoke = void;
    };

    template <typename Fn, typename... Args>
    using invoke = typename Fn::template invoke<Args...>;

    template <typename Fn>
    struct compose
    {
      template <typename X, typename Y>
      using F = invoke<Fn, X, Y>;

      template <typename X>
      using invoke = invoke<Fn, X, X>;
    };

    using Q = compose<plus>::F<int, int>;

Is accepted by gcc.