P1018R9
C++ Language Evolution status 🦠 pandemic edition 🦠 2021/01–2021/03

From

#

Title

Notes

Resolution

Core

[CWG2261]

Explicit instantiation of in-class friend definition

struct S { 
   template <class T> friend void f(T) { } 
  }; 
  template void f(int);  // Well-formed?

A friend is not found by ordinary name lookup until it is explicitly declared in the containing namespace, but declaration matching does not use ordinary name lookup. There is implementation divergence on the handling of this example.

Note 2020-04-29 Tentative agreement: This should be well-formed.

SF 1 F 10 N 2 A 1 SA 0 

JF emailed EWG / Core about this.

Davis: the current name lookup approach which Core is taking in p1787 would disallow this. Supporting this is possible, it would be inconsistent, but would also be a feature.

Notes 2020-10-22: wait until p1787 is voted into the working draft, because it’s making this behavior intentional. At that point, we can vote on marking the issue as Not a Defect. No objection to unanimous consent.

♻️

Core

[CWG2270]

Non-inline functions and explicit instantiation declarations

[Detailed description pending.]

Hubert: question over the role of the inline keyword in relation to explicit instantiation declarations.

For inline functions, explicit instantiation declarations do not have the effect of suppressing implicit instantiation.

A user’s desire for wanting to suppress implicit instantiation can arise for different reasons:

To reduce space in object files, executables, etc. and similarly to reduce the number of input symbols to the linker

To reduce compile time in performing semantic analysis for instantiations whose definitions are provided elsewhere

To control point-of-instantiation, avoiding contexts where the requisite declarations are not declared

The special rule around inline functions allows inline-ness to be used to indicate that the first reason is the intent and that instantiation-for-inlining is okay.

Consider the following as a translation unit:

template <typename T>
//inline
void f(T t) { g(t); }
enum E : int;
extern template void f(E);
void h(E e) { f(e); }

Marking the template definition inline would mean that the intended declaration for g would need to be provided as the best candidate at the points-of-instantiation for f<E>.

The issue initially points out that this use of the inline keyword does not match a view that inline is essentially an ODR tool to allow multiple definitions (as opposed to a way to indicate desire for inlining) and proposed that extern template merely has the effect of suppressing definitions in terms of linkage (regardless of the inline keyword). Such a change would affect the usability of the feature for user intent that falls within the latter two options above.

I am not sure if CWG is asking EWG a specific question other than the general "we do not believe this is a wording or obvious consistency issue; is this an issue in terms of design?"

Meeting:

Hubert forked the thread on the reflector. Might want the education SG to take a look, or might want a paper.

Meeting 2020-10-28: Inbal will try to put together the wording, to prove / disprove whether this is a defect.

♻️

Lib

[LWG2432]

initializer_list assignability

std::initializer_list::operator= [support.initlist] is horribly broken and it needs deprecation:

std::initializer_list<foo> a = {{1}, {2}, {3}};

a = {{4}, {5}, {6}};

// New sequence is already destroyed.

Assignability of initializer_list isn’t explicitly specified, but most implementations supply a default assignment operator. I’m not sure what [description] says, but it probably doesn’t matter.

Proposed resolution:

Edit [support.initlist] p1, class template initializer_list synopsis, as indicated:

namespace std {

  template<class E> class initializer_list {

  public:

    […]

    constexpr initializer_list() noexcept;

    initializer_list(const initializer_list&) = default;

    initializer_list(initializer_list&&) = default;

    initializer_list& operator=(const initializer_list&) = delete;

    initializer_list& operator=(initializer_list&&) = delete;

    constexpr size_t size() const noexcept;

    […]

  };

  […]

}

LWG telecon appears to want a language change to disallow assigning a braced-init-list to an std::initializer_list but still permit move assignment of std::initializer_list objects. That is,

auto il1 = {1,2,3};

auto il2 = {4,5,6};

il1 = {7,8,9}; // currently well-formed but dangles immediately; should be ill-formed

il1 = std::move(il2); // currently well-formed and should remain so

Meeting: Proposed resolution:

initializer_list(const initializer_list&) = default;
initializer_list(initializer_list&&) = default;
[[deprecated]] initializer_list& operator=(const initializer_list&) = default;
[[deprecated]] initializer_list& operator=(initializer_list&&) = default;

SF F N A SA

0 3 12 0 0

JF emailed LEWG, to see if they have an opinion, no feedback. Asked LEWG chairs to schedule for a telecon.

LWG discussed priority.

♻️

Lib

[LWG2813]

std::function should not return dangling references

If a std::function has a reference as a return type, and that reference binds to a prvalue returned by the callable that it wraps, then the reference is always dangling. Because any use of such a reference results in undefined behaviour, the std::function should not be allowed to be initialized with such a callable. Instead, the program should be ill-formed.

A minimal example of well-formed code under the current standard that exhibits this issue:

int main()  {

  std::function<const int&()> F([]{ return 42; });

  int x = F(); // oops!

}

Proposed resolution:

Add a second paragraph to the remarks section of 20.14.16.2.1 [func.wrap.func.con]:

template<class F> function(F f);

-7- Requires: F shall be CopyConstructible.

-8- Remarks: This constructor template shall not participate in overload resolution unless

• F is Lvalue-Callable (20.14.16.2 [func.wrap.func]) for argument types ArgTypes... and return type R, and
  
• If R is type "reference to T" and INVOKE(ArgTypes...) has value category V and type U:
  
  • V is a prvalue, U is a class type, and T is not reference-related (9.4.3 [dcl.init.ref]) to U, and
    
  • V is an lvalue or xvalue, and either U is a class type or T is reference-related to U.
    

[…]

Tim: LWG in Batavia 2018 would like a way to detect when the initialization of a reference would bind it to a temporary. This requires compiler support, since there’s no known way in the current language to do so reliably in the presence of user-defined conversions (see thread starting at https://lists.isocpp.org/lib/2017/07/3256.php).

Meeting: 

Tim wrote p2255 to address this.

Ville thinks there should be an analysis of alternative approaches.

Also see P0932.

♻️

Core

[CWG728]

Restrictions on local classes

Now that the restriction against local classes being used as template arguments has been lifted, they are more useful, yet they are still crippled. For some reason or oversight, the restriction against local classes being templates or having member templates was not lifted. Allowing local classes to have member templates facilitates generic programming (the reason for lifting the other restriction), especially when it comes to the visitor-pattern (see the boost::variant documentation and the following example) as implemented in boost and the boost::MPL library (since functors have to be template classes in mpl, and higher-order functors have to have member templates to be useful). A local class with a member template would allow this desirable solution:

    #include <boost/variant.hpp>

    int main() {

      struct times_two_generic: public boost::static_visitor<> {

        template <typename T> void operator()(T& operand) const {

            operand += operand;

        }

      };

      std::vector<boost::variant<int, std::string>> vec;

      vec.push_back(21);

      vec.push_back("hello ");

      times_two_generic visitor;

      std::for_each(vec.begin(), vec.end(), boost::apply_visitor(visitor));

    }

Is there any compelling reason not to allow this code? Is there any compelling reason not to allow local classes to be templates, have friends, or be able to define their static data members at function scope? Wouldn’t this symmetry amongst local and non-local classes make the language more appealing and less embarrassing?

Also see p2044, seen by EWGI in Prague and requested to be merged with P1988, and implementors queried.

The current paper draft addresses all of this.

Meeting: (notes)

Some folks unsure that we should do anything here. Revisit the resolution with P2044.

Extension issues are a closed status on the issues list, like NAD Future was for LWG I find it helpful to have a status that says EWG may still be considering it.

Suggest closing as Not A Defect because we have significant implementation concerns, but we’ll explore the design space in P2044. All in favor.

NAD

Core

[CWG794]

Base-derived conversion in member type of pointer-to-member conversion

Related to CWG 170, drafting by Clark seems unlikely. This is section 2.1 of Jeff Snyder’s P0149R0, which was approved by EWG, 4 years ago, waiting for wording.

JF reached out to Jeff. Did wording with Jens, main blocker is lack of implementation.

Meeting: (notes) Suggest closing as Not A Defect because we have implementation uncertainties, but we’ll explore the design space in P0149. ABI group will discuss. All in favor.

NAD

Core

[CWG900]

Lifetime of temporaries in range-based for

// some function

  std::vector<int> foo();

  // correct usage

  auto v = foo();

  for( auto i : reverse(v) ) { std::cout << i << std::endl; }

  // problematic usage

  for( auto i : reverse(foo()) ) { std::cout << i << std::endl; }

Meeting: (notes, also discussed in Rapperswil 2014) Suggest closing as Not A Defect because it’s a change which might have effects on existing code (might cause bugs), and might need to change more than just range-based loops. See p0614, p0577, p0936, p1179. We’ll explore the design space in a separate paper, Herb circled back with Nico on this, might write a paper. All in favor.

NAD

Core

[CWG944]

reinterpret_cast for all types with the same size and alignment

During the discussion of issue 799, which specified the result of using reinterpret_cast to convert an operand to its own type, it was observed that it is probably reasonable to allow reinterpret_cast between any two types that have the same size and alignment.

Additional note, April, 2015: It has been suggested that this question may more properly be the province of EWG, especially in light of discussions during the resolution of issue 330. The priority has been deleted to allow CWG to reconsider its status.

Rationale (May, 2015): CWG agreed that this question should be considered from a language design perspective and is thus being referred to EWG.

Suggest rejection: we chose to add std::bit_cast to provide exactly this functionality instead of extending reinterpret_cast.

Meeting: (notes) Suggest closing as Not A Defect. bit_cast addresses some of the usecases this issue might cover. We’d want separate papers to support other usescases, but have nothing outstanding at the moment. All in favor.

NAD

Core

[CWG1008]

Querying the alignment of an object

https://godbolt.org/z/_SN8iF

• GCC already implements this extension without issuing a warning
• Clang and EDG implement this extension for gcc compatibility, with a warning
• MSVC does not yet implement this feature

Quick example using ‘auto’ illustrates why we might want this capability for objects as well as types.

Principle of least astonishment suggests it is surprising for sizeof and alignof to behave differently in this regard.

Recommend shipping this straight to core as soon as we can find a wording champion.

We need to discuss with WG14.

Questions for EWG to answer before forwarding:

• Should this be a unary operator, like sizeof, so "alignof x" is valid? Or should it be like typeid, where parens are required?
• What would this mean? Is it the alignment of the type of the object? Or the compiler’s best guess of the alignment of the expression itself? Should it take into account any facts that are known about the expression other than its type? If so, which ones? (For example, if applied to an expression x or e.x and x is declared with an alignas attribute, is that value returned?)

This needs a design paper rather than going straight to core.

https://godbolt.org/z/TeVA9T GCC seems to report the alignment of the object not just of decltype(object).

Meeting: (notes, also discussed in Rapperswil 2014) Suggest closing as Not A Defect, the design is complex especially around alignment of object versus type. Invite a paper, Inbal will pitch in, Alidair can collaborate. All in favor.

Inbal’s paper: P2152R0

NAD

Core

[CWG1077]

Explicit specializations in non-containing namespaces

The current wording of 9.8.1.2 [namespace.memdef] and 13.9.3 [temp.expl.spec] requires that an explicit specialization be declared either in the same namespace as the template or in an enclosing namespace. It would be convenient to relax that requirement and allow the specialization to be declared in a non-enclosing namespace to which one or more if the template arguments belongs.

Additional note, April, 2015: See EWG issue 48.

Might allow us to revert DR2061 and all the horribleness that created and described in p1701.  The problem 1077 addresses is the motivating factor in dr2061.

Also see CWG 2370.

Meeting: (notes) Suggest closing as Not A Defect. See p0665, minutes. Continue under p0665 or a forked version of it. All in favor.

NAD

Core

[CWG1433]

trailing-return-type and point of declaration

template <class T> T list(T x);

    template <class H, class ...T>

    auto list(H h, T ...args) -> decltype(list(args...)); // list isn’t in scope in its own trailing-return-type

    auto list3 = list(1, 2, 3);

Meeting: (notes, also discussed in Rapperswil 2014) there might be compiler divergence according to Daveed. Suggest closing as Not A Defect, it would be tricky to change behavior without ambiguity. "Fixing" this would break existing code that relies on seeing only previous declarations. No objection to unanimous consent.

NAD

Core

[CWG1451]

Objects with no linkage in non-type template arguments

According to 13.4.2 [temp.arg.nontype] paragraph 1 bullet 3, only objects with linkage can be used to form non-type template arguments. Is this restriction still needed? It would be convenient to use block-scope objects as template arguments.

Resolved by N4268.

The following is valid:

template <int *> struct A;

auto f() {

  static int x;

  A<&x> *ap = nullptr;

  return ap;

}

There is no longer any linkage restriction. There is a restriction on storage duration, which seems reasonable (you can use the address of a local static variable but not of a local automatic variable as a non-type template argument).

Meeting: (notes, also discussed in Rapperswil 2014) Suggest closing as Not A Defect. Addressed by N4268. No objection to unanimous consent.

NAD

Core

[CWG1469]

Omitted bound in array new-expression

The syntax for noptr-new-declarator in 7.6.2.7 [expr.new] paragraph 1 requires an expression, even though the bound could be inferred from a braced-init-list initializer. It is not clear whether 9.4.1 [dcl.init.aggr] paragraph 4, An array of unknown size initialized with a brace-enclosed initializer-list containing n initializer-clauses, where n shall be greater than zero, is defined as having n elements (9.3.3.4 [dcl.array]).

should be considered to apply to the new-type-id variant, e.g.,

  new (int[]){1, 2, 3}

This was addressed by p1009

Meeting: (notes) Suggest closing as Not A Defect. Addressed by P1009. No objection to unanimous consent.

NAD

Core

[CWG1555]

Language linkage and function type compatibility

Currently function types with different language linkage are not compatible, and 7.6.1.2 [expr.call] paragraph 1 makes it undefined behavior to call a function via a type with a different language linkage. These features are generally not enforced by most current implementations (although some do) between functions with C and C++ language linkage. Should these restrictions be relaxed, perhaps as conditionally-supported behavior?

Somewhat related to CWG1463.

Meeting: (notes) no strong consensus at the moment, Erich Keane brought this up with SG12 Undefined Behavior. Long discussion, will need to revisit, need a paper.

Meeting Oct 22nd 2020: will need a volunteer to write a paper, but the wording in the standard is unambiguous, and we have existence proof of platforms which use different calling conventions between C and C++. However many major compilers have chosen to ignore this. Not a defect, but we welcome a paper to change the status quo. No objection to unanimous consent.

NAD

Core

[CWG1643]

Default arguments for template parameter packs

Although 13.2 [temp.param] paragraph 9 forbids default arguments for template parameter packs, allowing them would make some program patterns easier to write. Should this restriction be removed?

Meeting: (notes) Suggest closing as Not A Defect. Interesting design space, but needs a paper, see N3416. No objection to unanimous consent.

NAD

Core

[CWG1864]

List-initialization of array objects

The resolution of issue 1467 now allows for initialization of aggregate classes from an object of the same type. Similar treatment should be afforded to array aggregates.

See recent discussion on allowing 'auto x[] = {1, 2, 3};' -- this topic came up there. The two questions can be answered independently, but some consider them to be related.

Meeting: (notes) Suggest closing as Not A Defect. Fixing anything in this space would require a paper which considers the entire design space, Timur might be interested in this. No objection to unanimous consent.

NAD

Core

[CWG1871]

Non-identifier characters in ud-suffix

JF forwarded to SG16 Unicode given their work on TR31. Tracked on GitHub. SG16 reviewed D1949R3, still needs wording changes. Discussed at the April 22nd SG16 telecon.

SG16 Poll: Is there any objection to unanimous consent for recommending rejection of this proposal? No objection to unanimous consent.

EWG Meeting: (notes 2020-04-15, notes 2020-05-07) Suggest closing as Not A Defect. No objection to unanimous consent.

NAD

Core

[CWG1876]

Preventing explicit specialization

A desire has been expressed for a mechanism to prevent explicitly specializing a given class template, in particular std::initializer_list and perhaps some others in the standard library. It is not clear whether simply adding a prohibition to the description of the templates in the library clauses would be sufficient or whether a core language mechanism is required.

Meeting: (notes) Suggest closing as Not A Defect. No objection to unanimous consent. This could be a language feature, would need library usecase examples.

Poll: we are interested in such a paper SF 2 F 13 N 6 A 1 SA 0

NAD

Core

[CWG1915]

Potentially-invoked destructors in non-throwing constructors

Since the base class constructor is non-throwing, the deleted base class destructor need not be referenced.

There’s a typo in the issues list here: it should say "Since the derived class constructor is non-throwing, the deleted base class destructor need not be referenced."

Meeting: (notes 2020-04-23) the proposed wording changes the implementation leeway in two-phase unwinding, which breaks some existing ABIs. We would need  a paper to explore this further. Suggest closing as Not A Defect. No objection to unanimous consent.

NAD

Core

[CWG1923]

Lvalues of type void

There does not seem to be any significant technical obstacle to allowing a void* pointer to be dereferenced, and that would avoid having to use weighty circumlocutions when casting to a reference to an object designated by such a pointer.

Might consider this a duplicate of the discussion on "regular void".

JF reached out to Matt. He has an implementation in clang. Needs to update the paper. Might need a volunteer to present. Daveed would be interested in presenting.

Meeting: (notes 2020-04-23) Suggest closing as Not A Defect. Explore under the “regular void” umbrella. No objection to unanimous consent.

NAD

Core

[CWG1934]

Relaxing exception-specification compatibility requirements

According to 14.5 [except.spec] paragraph 4,

If any declaration of a function has an exception-specification that is not a noexcept-specification allowing all exceptions, all declarations, including the definition and any explicit specialization, of that function shall have a compatible exception-specification.

This seems excessive for explicit specializations, considering that paragraph 6 applies a looser requirement for virtual functions:

If a virtual function has an exception-specification, all declarations, including the definition, of any function that overrides that virtual function in any derived class shall only allow exceptions that are allowed by the exception-specification of the base class virtual function.

The rule in paragraph 3 is also problematic in regard to explicit specializations of destructors and defaulted special member functions, as the implicit exception-specification of the template member function cannot be determined.

There is also a related problem with defaulted special member functions and exception-specifications. According to 9.5.2 [dcl.fct.def.default] paragraph 3,

If a function that is explicitly defaulted has an explicit exception-specification that is not compatible (14.5 [except.spec]) with the exception-specification on the implicit declaration, then

• if the function is explicitly defaulted on its first declaration, it is defined as deleted;
• otherwise, the program is ill-formed.
  

This rule precludes defaulting a virtual base class destructor or copy/move functions if the derived class function will throw an exception not allowed by the implicit base class member function.

Meeting: (notes 2020-04-23) JF will work with Mike to update wording to C++20. Will revisit.

From Mike: This issue is NAD since we eliminated typed exception-specifications. The current wording, dealing only with noexcept(true) and noexcept(false), does not have this issue. Will remove “extension” status.

Meeting: no objection to CWG taking it back, and marking it as NAD.

NAD

Core

[CWG1957]

decltype(auto) with direct-list-initialization

Paper N3922 changed the rules for deduction from a braced-init-list containing a single expression in a direct-initialization context. Should a corresponding change be made for decltype(auto)? E.g.,

  auto x8a = { 1 };           // decltype(x8a) is std::initializer_list<int>

  decltype(auto) x8d = { 1 }; // ill-formed, a braced-init-list is not an expression

  auto x9a{ 1 };              // decltype(x9a) is int

  decltype(auto) x9d{ 1 };    // decltype(x9d) is int

See also issue 1467, which also effectively ignores braces around a single expression, this change would be parallel to that one, even though the primary motivation for delctype(auto) is in the return type of a forwarding function, where direct-initialization does not apply.

Meeting: (notes 2020-04-23) Suggest closing as Not A Defect. This would be a language change, it’s unclear that we want to make such a change. It would require a paper. Mike Spertus is writing a paper with some overlap, will cover this as well. No objection to unanimous consent.

NAD

Core

[CWG2111]

Array temporaries in reference binding

The current wording of the Standard appears to permit code like

void f(const char (&)[10]);
void g() {
  f("123");
  f({'a','b','c','\0'});
}

creating a temporary array of ten elements and binding the parameter reference to it. This is controversial and should be reconsidered. (See issues 1058 and 1232.)

Meeting: (notes 2020-04-23) JF digging more, talking to Richard about this.

Somewhat related to P2174 compound literals.

Would be very strange if

  const char (&&x)[10] = {'a', 'b', 'c'};

... were invalid but ...

  const char x[10] = {'a', 'b', 'c'};

  const char (&&x)[3] = {'a', 'b', 'c'};

... were both OK. Maybe either we should allow the trailing elements to be zeroed in general (the status quo), or not (a major breaking change). Which means we should reject the issue on consistency grounds. The general rule is that if

  T &&r = init;

... can’t bind directly, we create a temporary initialized as if with

  T r = init;

... and bind to that. (And similarly for const references.)

Another question: Do we want to support (T){inits} as a synonym for T{inits}?

Meeting Oct 22nd 2020: Note that the issue itself is defective, and f("123") isn’t valid.

CWG2352 references p1358, and was resolved, leading us to believe that this is now mainstream and not controversial. Not a defect. No objection to unanimous consent.

NAD

Core

[CWG2125]

Copy elision and comma operator

Currently, _N4750_.15.8 [class.copy] paragraphs 31-32 apply only to the name of a local variable in determining whether a return expression is a candidate for copy elision or move construction. Would it make sense to extend that to include the right operand of a comma operator?

X f() {

  X x;

  return (0, x);

}

Meeting: (notes 2020-04-23) Consider expanding to other places that expand bit-field-ness such as return b ? throw : x;. Suggest closing as Not A Defect. Will need a paper to address, no current volunteer for this. No objection to unanimous consent.

NAD

Core

[CWG2132]

Deprecated default generated copy constructors

EWG has indicated that they are not currently in favor of removing the implicitly declared defaulted copy constructors and assignment operators that are eprecated in _N4750_.15.8 [class.copy] paragraphs 7 and 18. Should this deprecation be removed?

Related: discussing under p2139 deprecations.

Meeting: (note 2020-04-29) Suggest closing as Not A Defect. No objection to unanimous consent. We either want to remove entirely, or un-deprecate. This will need a paper, Ville will talk with Alisdair about forking the topic from p2139.

NAD

Core

[CWG914]

Value-initialization of array types

Although value-initialization is defined for array types and the () initializer is permitted in a mem-initializer naming an array member of a class, the syntax T() (where is an array type) is explicitly forbidden by 7.6.1.3 [expr.type.conv] paragraph 2. This is inconsistent and the syntax should be permitted.

Rationale (July, 2009): The CWG was not convinced of the utility of this extension, especially in light of questions about handling the lifetime of temporary arrays. This suggestion needs a proposal and analysis by the EWG before it can be considered by the CWG.

This has become a more severe inconsistency after we adopted Ville’s P0960 for C++20. Now it’s not only () that has this weird special-case restriction, it’s (a, b, c) too:

using X = int[];

X x{1, 2, 3}; // ok, int[3]

X y(1, 2, 3); // ok, int[3]

f(X{1, 2, 3}); // ok, int[3] temporary

f(X(1, 2, 3)); // error

Meeting: (notes) it is a defect, need a paper. David Stone trying to find a volunteer to write said paper. All in favor.

Paper needed

Core

[CWG1463]

extern "C" alias templates

Currently 13 [temp] paragraph 4 forbids any template from having C linkage. Should alias templates be exempt from this prohibition, since they do not have any linkage?

Additional note, April, 2013: It was suggested (see messages 23364 through 23367) that relaxing this restriction for alias templates could provide a way of addressing the long-standing lack of a way of specifying a language linkage for a dependent function type (see issue 13). The priority was deleted to allow CWG to consider the implications of that potential application of the facility.

We should either have some way to express a dependent function type with C language linkage (and should accept 1555 below) or we should remove the notion that C language linkage (or not) is part of the function type at all. (Apparently some targets used it; are they still in use? EDG probably knows.)

Actively discussed on CWG reflector.

Davis’ interpretation of CWG discussion:

I don’t speak for Core, of course, but my (Evolutionary) thoughts are those given (last) in the message to which you replied: neither the wording nor the apparent intent of [temp.pre]/6’s restrictions on C linkage for templates make any sense.  Since templates (and explicit specializations) can’t have C linkage of the name-mangling variety (which the standard calls language linkage of a (function) name) but can have C linkage of the calling-convention variety (which the standard calls language linkage of a (function) type), extern "C" should grant them the latter and not the former with no error.  This has certain obvious applications involving C APIs with callbacks.  (Put differently, CWG13 shouldn’t have been rejected; it appears to have gotten bogged down in questions of syntax, but we have an adequate syntactic workaround that we merely have to permit.)

CWG1463 asks for a (very) proper subset of the above: merely that extern "C" be allowed to apply to alias templates (claiming incorrectly that they have no name with linkage at all).  I consider it more relevant (and more productive) to talk about whether entities have names with language linkage than with the ordinary kind.

The Core reflector discussion (which seems to have finished for the moment) also touched on the case where "the same" function template is declared with two different language linkages for its type, but the only relevant Evolution input there would be a decision to go the opposite way and forbid function templates from having types with C language linkage entirely.  (They currently can, but it’s mostly or entirely useless.)

If (for CWG1555, also on your list) the rules about language linkage of (function) types are sufficiently relaxed, then CWG1463 may be moot (and my extension of it with it), but that seems unlikely given Hubert’s recent identification of a case where they matter.

Meeting: (notes 2020-04-15, notes 2020-10-22, also discussed in Rapperswil 2014) We agree that this is an issue. extern “C” on a template should only affect calling convention, and not the mangling. Davis and Hubert volunteer to write a paper. Unanimous consent.

Paper needed

Core

[CWG1790]

Ellipsis following function parameter pack

Although the current wording permits an ellipsis to immediately follow a function parameter pack, it is not clear that the <cstdarg> facilities permit access to the ellipsis arguments.

The problem here (which is not explained in the issue) is: how do you supply the name of the last parameter before the ellipsis to va_start? You can’t put the name of a pack there (it wouldn’t be expanded) and there’s no way to name the last element of the pack (nor to deal with the case where the pack is empty).

Meeting: (notes) 3 options: fix wording around “last parameter”, remove facility entirely (either va_start or function declarator), try to invent a language facility. JF emailed EWG to see if anyone has a strong preference, or if we should send back to CWG to fix wording, long discussion.

Michael Spertus: I am willing to commit to including and analysis on this in an upcoming paper on parameter packs.

JF followed up with Michael and Barry, no response.

Paper needed

Core

[CWG1962]

Type of __func__

Two questions have arisen regarding the treatment of the type of the __func__ built-in variable. First, some implementations accept

  void f() {

    typedef decltype(__func__) T;

    T x = __func__;

  }

even though T is specified to be an array type.

In a related question, it was noted that __func__ is implicitly required to be unique in each function, and that not only the value but the type of __func__ are implementation-defined; e.g., in something like

  inline auto f() { return &__func__; }

the function type is implementation-specific. These concerns could be addressed by making the value a prvalue of type const char* instead of an array lvalue.

Rationale (November, 2018): See also issue 2362, which asks for the ability to use __func__ in a constexpr function. These two goals are incompatible.

The deep question here is about __func__ and the ODR. Does EWG want implementations to somehow behave as if __func__ is the same in all copies of an inline function (in which case it can have an array type and be usable in constant expressions, but the demangling algorithm used to construct it becomes part of the ABI), or does EWG want implementations to behave as if __func__ may differ between copies, so is in effect not known until runtime (in which case it must have either pointer or incomplete array type, and its value is not usable in constant expressions -- but its address could still be usable).

Note: C++20 has std::source_location::function_name.

Meeting: (notes 2020-04-23) We should discuss this with WG14. This is indeed a language issue. No objection to unanimous consent. We’ll need a paper, potentially considering what Reflection could do, JF brought it up on the mailing list.

We probably need a paper to disentangle this.

Paper needed

Core

[CWG2228]

Ambiguity resolution for cast to function type

Proposed resolution.

C++ has a blanket disambiguation rule that says if a sequence of tokens can be interpreted as either a type-name or an expression, the type-name interpretation is chosen. This is unhelpful in cases like

    (T())+3

where the current rules make "(T())" a cast to the type "function with no parameters returning T" rather than a parenthesized value-initialization of a temporary of type T. The former interpretation is always ill-formed - you can’t cast to a function type - while the latter could be useful.

Richard’s proposed resolution, cited above, is to avoid the ambiguity by changing the grammar so that cases like "(T())" cannot be parsed as a cast. The wording in the proposal applies that change to a number of different contexts where the ambiguity can come into play. There are two contexts where the change is _not_ applied, however: 1) the operand of typeid, and 2) a template-argument. During our discussion yesterday, there was some support for the idea of applying the change to typeid, as well, although that would be a breaking change for any programs that rely on the current disambiguation to get type information for such function types. There was general agreement, however, to exclude template arguments, since they are used for things like std::function.

CWG would, therefore, like some guidance from EWG on two questions. First, should we apply the new syntax to the operand of typeid, even though it’s a breaking change?

More generally, the question was raised whether we should make this change at all. Although resolving the ambiguity in the other direction would arguably be more convenient in many cases, there is a tension between that convenience and the complexity of the language. In particular, we would be creating a situation where the exact same sequence of tokens would mean two different things, depending on the context in which they appear. Is the convenience worth the cost in complexity?

Meeting: (note 2020-04-29, notes 2020-03-23, note from Core summer 2020, notes from Core Prague 2020, notes from Core 2019-01-07, notes from Core Kona 2019, notes from Core Cologne 2019) This is an issue, we’d like to change the standard to resolve it: SF 0 F 6 N 4 A 3 SA 2

Davis emailed EWG reflector. Long discussion.

Paper needed

Core

[CWG2296]

Are default argument instantiation failures in the “immediate context”?

Example 1:

template <typename T, typename U = T>

void fun(T v, U u = U()) {}

void fun(...) {}

struct X {

    X(int) {} // no default ctor

};

int main()  {  fun (X(1)); }

Consider the following example (taken from issue 3 of paper P0348R0):

Example 2:

  template <typename U> void fun(U u = U());

  struct X {

    X(int) {}

  };

  template <class T> decltype(fun<T>()) g(int) { }

  template <class> void g(long) { }

  int main() { g<X>(0); }

When is the substitution into the return type done? The current specification makes this example ill-formed because the failure to instantiate the default argument in the decltype operand is not in the immediate context of the substitution, although a plausible argument for making this a SFINAE case can be made.

Meeting: (notes 2020-05-07)

The first example under issue 3 of paper P0348R0 should become well-formed.

SF F N A SA

0 1 4 3 6

The second example under issue 3 of paper P0348R0 should become well-formed.

SF F N A SA

1 5 7 3 0

This is an issue. We’d like to see a paper addressing it. It should explore what “Immediate context” means. No objection to unanimous consent.

Daveed / Hubert might entertain writing a paper explaining this.

Ville emailed EWG about this.

JF contacted Andrzej to see if he’s interested in addressing this. Not sure he is.

Meeting 2020-10-28: Tomasz will write a paper which exposes the issue and what he thinks should be done (but not resolving wording itself). Hubert remembers an email about this, will find it and sync with JF.

Meeting 2021-01-14: Andrzej wrote a paper for this (emailed 2021-01-12 to EWG).

Paper needed

Core

[CWG2362]

__func__ should be constexpr

The definition of __func__ in 9.5.1 [dcl.fct.def.general] paragraph 8 is:

  static const char __func__[] = "function-name";

This prohibits its use in constant expressions, e.g.,

  int main () {

    // error: the value of __func__ is not usable in a constant expression

    constexpr char c = __func__[0];

  }

Rationale (November, 2018): See also issue 1962, which asks that the type of __func__ be const char*. These two goals are incompatible.

Meeting: handle with 1962.

Paper needed