1 Introduction

When [P0941R2] was adopted both into the standard and as policy, that paper thoroughly went through all the relevant features that existed at the time. At the Cologne meeting in July 2019, we added signs in both the Core and Library Working Group rooms to ensure that papers in flight did not forget - they had to either explicitly adopt a new feature-test macro or explicitly state why they did not need one.

In the intervening time-frame, several meetings’ worth of papers were adopted which did not have a feature-test macro. The goal is this paper is to go through all of those papers and add the missing ones.

As a general note, [SD6] has been updated and is current through Cologne and through a few LWG issues following that.

2 2017

2.1 Toronto (201707)

[P0683R1] (Default member initializers for bit-fields): no macro necessary.

[P0704R1] (Fixing const-qualified pointers to members): no macro necessary.

[P0409R2] (Allow lambda capture [=, this]): no macro necessary.

[P0306R4] (Comma omission and comma deletion): no macro necessary.

[P0329R4] (Designated Initialization Wording): no macro necessary. It’s possible that you could write something like:

struct X { int first, int second; };

#if __cpp_designated_initializer
#define INIT_EQ(name) .name =
#else
#define INIT_EQ(name)
#endif

X{INIT_EQ(first) 7, INIT_EQ(second) 10};

That does give some benefit, in that if you get the names wrong, you’d get a diagnostic. But would anybody actually do that?

[P0428R2] (Familiar template syntax for generic lambdas): this paper proposes to bump the macro __cpp_generic_lambdas. One of the things this feature allows for is, for instance, defaulting a template parameter on a lambda, which is arguably a feature enhancement:

struct X { int i, j; };

auto f = 
#if __cpp_generic_lambdas >= 201707
    []<class T=X>(T&& var)
#else
    [](auto&& var)
#endif
    {
    };
    
// having the macro allows this usage
f({1, 2});

[P0702R1] (Language support for Constructor Template Argument Deduction): no macro necessary.

[P0734R0] (Wording Paper, C++ extensions for Concepts): this paper proposes the macro __cpp_concepts, which will be bumped later, but for SD-6 purposes will start here with the value 201707.

[P0463R1] (Endian just Endian) to the C++ working paper: Already have a macro from [P1612R1].

[P0682R1] (Repairing elementary string conversions): Already have a macro from [LWG3137].

[P0674R1] (Extending make_shared to Support Arrays): this paper proposes to bump __cpp_lib_shared_ptr_arrays, since make_shared can save an allocation.

2.2 Albuquerque (201711)

[P0614R1] (Range-based for statements with initializer): no macro necessary.

[P0588R1] (Simplifying implicit lambda capture): no macro necessary.

[P0846R0] (ADL and Function Templates that are not Visible): no macro necessary.

[P0641R2] (Resolving Core Issue #1331 (const mismatch with defaulted copy constructor)): no macro necessary.

[P0859R0] (Core Issue 1581: When are constexpr member functions defined?): this paper proposes the macro __cpp_impl_constexpr_members_defined. This issue is a blocker for library being able to implement [P1065R2], so the library needs to know when to be able to do that. It’s unclear if users outside of standard library implementers will need this.

[P0515R3] (Consistent comparison) and [P0768R1] (Library Support for the Spaceship (Comparison) Operator): already have a macro.

[P0857R0] (Wording for “functionality gaps in constraints”): no macro necessary.

[P0692R1] (Access Checking on Specializations): no macro necessary.

[P0624R2] (Default constructible and assignable stateless lambdas): no macro necessary.

[P0767R1] (Deprecate POD): no macro necessary.

[P0315R4] (Wording for lambdas in unevaluated contexts): no macro necessary.

[P0550R2] (Transformation Trait remove_cvref): this paper proposes the macro __cpp_lib_remove_cvref.

[P0777R1] (Treating Unnecessary decay): no macro necessary.

[P0600R1] (nodiscard in the Library): no macro necessary.

[P0439R0] (Make std::memory_order a scoped enumeration): no macro necessary

[P0053R7] (C++ Synchronized Buffered Ostream): this paper proposes the macro __cpp_lib_syncbuf.

[P0653R2] (Utility to convert a pointer to a raw pointer): this paper proposes the macro __cpp_lib_to_address.

[P0202R3] (Add constexpr modifiers to functions in <algorithm> and <utility> Headers): a macro was already added.

[P0415R1] (Constexpr for std::complex): this paper proposes to bump the macro __cpp_lib_constexpr. The current position is [P1424R1], option A, which states that every constexpr addition to the library should simply increment the version of __cpp_lib_constexpr. User code should then check against the number that they expect.

[P0718R2] (Atomic shared_ptr): this paper proposes the macro __cpp_lib_atomic_shared_ptr.

[P0020R6] (Floating Point Atomic): this paper proposes the macro __cpp_lib_atomic_float.

[P0616R0] (de-pessimize legacy <numeric> algorithms with std::move): no macro necessary.

[P0457R2] (String Prefix and Suffix Checking): this paper proposes the macro __cpp_lib_starts_ends_with.

3 2018

3.1 Jacksonville (201803)

[P0840R2] (Language support for empty objects: already has a macro.

[P0962R1] (Relaxing the range-for loop customization point finding rules): no macro necessary.

[P0969R0] (Allow structured bindings to accessible members): no macro necessary.

[P0961R1] (Relaxing the structured bindings customization point finding rules): no macro necessary.

[P0634R3] (Down with typename!): no macro necessary.

[P0780R2] (Allow pack expansion in lambda init-capture): this paper proposes the macro __cpp_lambda_init_capture_pack. Having such a macro would allow you to avoid using tuple where necessary. The motivating example in that paper could thus conditionally improve compile throughput:

template <class... Args>
auto delay_invoke_foo(Args... args) {
#if __cpp_lambda_init_capture_pack
    return [...args=std::move(args)]() -> decltype(auto) {
        return foo(args...);
    };
#else
    return [tup=std::make_tuple(std::move(args)...)]() -> decltype(auto) {
        return std::apply([](auto const&... args) -> decltype(auto) {
            return foo(args...);
        }, tup);
    };
#endif
}

[P0479R5] (Proposed wording for likely and unlikely attributes (Revision 5)): already have a macro.

[P0905R1] (Symmetry for spaceship): in a vacuum, maybe, but since there isn’t an implementation of <=> that includes just up to this point, it’s probably not worth it.

[P0754R2] (<version>): already checkable with __has_include.

[P0809R0] (Comparing Unordered Containers): no macro necessary.

[P0355R7] (Extending chrono to Calendars and Time Zones): this paper proposes the macro __cpp_lib_chrono_date.

[P0966R1] (string::reserve Should Not Shrink): no macro necessary.

[P0551R3] (Thou Shalt Not Specialize std Function Templates!): no macro necessary.

[P0753R2] (Manipulators for C++ Synchronized Buffered Ostream): this paper proposes to bump the macro __cpp_lib_syncbuf, which is also added by this paper.

[P0122R7] (<span>): already has a macro by way of [LWG3274].

[P0858R0] (Constexpr iterator requirements): already has a macro.

3.2 Rapperswil (201806)

[P0806R2] (Deprecate implicit capture of this via [=]): no macro necessary.

[P1042R1] (__VA_OPT__ wording clarifications): no macro necessary.

[P0929R2] (Checking for abstract class types): no macro necessary.

[P0732R2] (Class Types in Non-Type Template Parameters): already has a macro.

[P1025R1] (Update The Reference To The Unicode Standard): no macro necessary.

[P0528R3] (The Curious Case of Padding Bits, Featuring Atomic Compare-and-Exchange): no macro necessary.

[P0722R3] (Efficient sized delete for variable sized classes): already has a macro.

[P1064R0] (Allowing Virtual Function Calls in Constant Expressions): this paper proposes to bump __cpp_constexpr. There are functions that include virtual calls that would be able to be made constexpr. This macro already has a higher value in the standard, so no wording changes necessary - just to track in SD-6.

[P1008R1] (Prohibit aggregates with user-declared constructors): no macro necessary.

[P1120R0] (Consistency improvements for <=> and other comparison operators): no macro necessary.

[P0542R5] (Support for contract based programming in C++): would have been checked by the attribute.

[P0941R2] (Integrating feature-test macros into the C++ WD (rev. 2)): very meta.

[P0892R2] (explicit(bool)): already has a macro.

[P0476R2] (Bit-casting object representations): already has a macro.

[P0788R3] (Standard Library Specification in a Concepts and Contracts World): no macro necessary.

[P0458R2] (Checking for Existence of an Element in Associative Containers): no macro necessary. The new algorithms are nicer to use, but if you need to support the old code, then the old code works just as well.

[P0759R1] (fpos Requirements): no macro necessary.

[P1023R0] (constexpr comparison operators for std::array): this paper proposes to bump the macro __cpp_lib_constexpr.

[P0769R2] (Add shift to <algorithm>): this paper proposes the macro __cpp_lib_shift.

[P0887R1] (The identity metafunction): this paper proposes the macro __cpp_lib_type_identity.

[P0879R0] (Constexpr for swap and swap related functions): already has a macro.

[P0758R1] (Implicit conversion traits and utility functions). this paper proposes the macro __cpp_lib_nothrow_convertible.

[P0556R3] (Integral power-of-2 operations): this paper proposes the macro __cpp_lib_int_pow2.

[P0019R8] (Atomic Ref): this paper proposes the macro __cpp_lib_atomic_ref.

[P0935R0] (Eradicating unnecessarily explicit default constructors from the standard library): no macro necessary.

[P0646R1] (Improving the Return Value of Erase-Like Algorithms): already has a macro.

[P0619R4] (Reviewing Deprecated Facilities of C++17 for C++20): no macro necessary.

[P0898R3] (Standard Library Concepts): already has a macro.

3.3 San Diego (201811)

[P0982R1] (Weaken Release Sequences): no macro necessary.

[P1084R2] (Today’s _return-type-requirement_s Are Insufficient): this paper proposes to bump __cpp_concepts, which was also added by this paper.

[P1131R2] (Core Issue 2292: simple-template-id is ambiguous between class-name and type-name): no macro necessary.

[P1289R1] (Access control in contract conditions): no macro necessary

[P1002R1] (Try-catch blocks in constexpr functions): this paper proposes to bump __cpp_constexpr, as per previous arguments to allow more functions to be declared constexpr. This one more important than the rest as actually writing try in a function already makes constexpr ill-formed.

[P1327R1] (Allowing dynamic_cast, polymorphic typeid in Constant Expressions): this paper proposes to bump __cpp_constexpr.

[P1236R1] (Alternative Wording for P0907R4 Signed Integers are Two’s Complement): no macro necessary.

[P0482R6] (char8_t: A type for UTF-8 characters and strings (Revision 6)): already has a macro.

[P1353R0] (Missing Feature Test Macros): already adopted.

[P1073R3] (Immediate functions): this paper proposes the macro __cpp_consteval. There are some functions that you really only want to run at compile time. You cannot enforce this in C++17, but having this macro would allow you to conditionally enforce it as it becomes available.

[P0595R2] (std::is_constant_evaluated()): already has a macro.

[P1141R2] (Yet another approach for constrained declarations): no macro necessary.

[P1094R2] (Nested Inline Namespaces): no macro necessary.

[P1330R0] (Changing the active member of a union inside constexpr): this paper proposes to bump __cpp_constexpr, as per previous arguments. This would allow making std::optional fully constexpr, for instance.

[P1123R0] (Editorial Guidance for merging P0019r8 and P0528r3): no macro necessary.

[P0487R1] (Fixing operator>>(basic_istream&, CharT*) (LWG 2499)): no macro necessary.

[P0602R4] (variant and optional should propagate copy/move triviality): no macro necessary.

[P0655R1] (visit<R>: Explicit Return Type for visit): no macro necessary.

[P0972R0] (<chrono> zero(), min(), and max() should be noexcept): no macro necessary.

[P1006R1] (Constexpr in std::pointer_traits): this paper proposes to bump the macro __cpp_lib_constexpr.

[P1148R0] (Cleaning up Clause 20): no macro necessary.

[P0318R1] (unwrap_ref_decay and unwrap_reference): this paper proposes the macro __cpp_lib_unwrap_ref.

[P0357R3] (reference_wrapper for incomplete types): no macro necessary.

[P0608R3] (A sane variant converting constructor): no macro necessary.

[P0771R1] (std::function move constructor should be noexcept): no macro necessary.

[P1007R3] (std::assume_aligned): this paper proposes the macro __cpp_lib_assume_aligned.

[P1020R1] (Smart pointer creation with default initialization): this paper proposes the macro __cpp_lib_smart_ptr_default_init.

[P1285R0] (Improving Completeness Requirements for Type Traits): no macro necessary.

[P1248R1] (Remove CommonReference requirement from StrictWeakOrdering (a.k.a Fixing Relations): no macro necessary.

[P0591R4] (Utility functions to implement uses-allocator construction): no macro necessary.

[P0899R1] (LWG 3016 is Not a Defect): no macro necessary.

[P1085R2] (Should Span be Regular?): no macro necessary.

[P1165R1] (Make stateful allocator propagation more consistent for operator+(basic_string)): no macro necessary.

[P0896R4] (The One Ranges Proposal): already has a macro.

[P0356R5] (Simplified partial function application): already has a macro.

[P0919R3] (Heterogeneous lookup for unordered containers): already has a macro.

[P1209R0] (Adopt Consistent Container Erasure from Library Fundamentals 2 for C++20): already has a macro.

4 2019

4.1 Kona (201902)

[P1286R2] (Contra CWG DR1778): no macro necessary.

[P1091R3] (Extending structured bindings to be more like variable declarations): no macro necessary.

[P1381R1] (Reference capture of structured bindings): no macro necessary.

[P1041R4] (Make char16_t/char32_t string literals be UTF-16/32): no macro necessary.

[P1139R2] (Address wording issues related to ISO 10646): no macro necessary.

[P1323R2] (Contract postconditions and return type deduction): contracts were removed.

[P0960R3] (Allow initializing aggregates from a parenthesized list of values): already has a macro.

[P1009R2] (Array size deduction in new-expressions): no macro necessary.

[P1103R3] (Merging Modules): already has a macro

[P1185R2] (<=> != ==): this paper proposes to bump __cpp_impl_three_way_comparison.

[P0339R6] (polymorphic_allocator<> as a vocabulary type): this paper proposes the macro __cpp_lib_polymorphic_allocator.

[P0340R3] (Making std::underlying_type SFINAE-friendly): no macro necessary.

[P0738R2] (I Stream, You Stream, We All Stream for istream_iterator): no macro necssary.

[P1458R1] (Mandating the Standard Library: Clause 16 - Language support library): no macro necessary.

[P1459R1] (Mandating the Standard Library: Clause 18 - Diagnostics library): no macro necessary.

[P1462R1] (Mandating the Standard Library: Clause 20 - Strings library): no macro necessary.

[P1463R1] (Mandating the Standard Library: Clause 21 - Containers library): no macro necessary.

[P1464R1] (Mandating the Standard Library: Clause 22 - Iterators library): no macro necessary.

[P1164R1] (Make create_directory() Intuitive): no macro necessary.

[P0811R3] (Well-behaved interpolation for numbers and pointers): already has a macro.

[P1001R2] (Target Vectorization Policies from Parallelism V2 TS to C++20): no macro necessary.

[P1227R2] (Signed ssize() functions, unsigned size() functions ): this paper proposes the macro __cpp_lib_ssize.

[P1252R2] (Ranges Design Cleanup): no macro necessary.

[P1024R3] (Usability Enhancements for std::span): already has a macro.

[P0920R2] (Precalculated hash values in lookup): already had a macro, which was subsequently removed.

[P1357R1] (Traits for [Un]bounded Arrays): no macro necessary

4.2 Cologne (201907)

[P1161R3] (Deprecate uses of the comma operator in subscripting expressions): no macro necessary.

[P1331R2] (Permitting trivial default initialization in constexpr contexts): already has a macro.

[P0735R1] (Interaction of memory_order_consume with release sequences): already has a macro.

[P0848R3] (Conditionally Trivial Special Member Functions): no macro necessary. Would people really write both?

[P1186R3] (When do you actually use <=>?): this paper proposed to bump __cpp_impl_three_way_comparison, but shouldn’t have. This paper doesn’t need a macro.

[P1301R4] ([[nodiscard("should have a reason")]]): already has a macro.

[P1099R5] (Using Enum): already has a macro

[P1630R1] (Spaceship needs a tune-up): this paper proposes that the bump of __cpp_impl_three_way_comparison is associated with this paper instead of P1186R3.

[P1616R1] (Using unconstrained template template parameters with constrained templates): no macro necessary.

[P1816R0] (Wording for class template argument deduction for aggregates): no macro necessary. Would you choose to not provide a deduction guide for an aggregate?

[P1668R1] (Enabling constexpr Intrinsics By Permitting Unevaluated inline-assembly in constexpr Functions): this paper proposes to bump __cpp_constexpr.

[P1766R1] (Mitigating minor modules maladies): already has a macro.

[P1811R0] (Relaxing redefinition restrictions for re-exportation robustness): already has a macro.

[P1814R0] (Wording for Class Template Argument Deduction for Alias Templates): already has a macro.

[P1825R0] (Merged wording for P0527R1 and P1155R3): no macro needed, just write std::move().

[P1703R1] (Recognizing Header Unit Imports Requires Full Preprocessing): no macro needed.

[P0784R7] (More constexpr containers): already had a macro, but erroneously introduced a new one (__cpp_lib_constexpr_dynamic_alloc) instead of following the established guidance from [P1424R1]. This paper proposes to remove __cpp_lib_constexpr_dynamic_alloc and bump the macro __cpp_lib_constexpr. This paper proposes no changes for the language version of this macro, __cpp_constexpr_dynamic_alloc.

[P1355R2] (Exposing a narrow contract for ceil2): no macro needed.

[P0553R4] (Bit operations): already has a macro.

[P1424R1] (constexpr feature macro concerns): already resolved.

[P0645R10] (Text Formatting): already has a macro.

[P1361R2] (Integration of chrono with text formatting): already has a macro.

[P1652R1] (Printf corner cases in std::format): already has a macro.

[P0631R8] (Math Constants): already has a macro.

[P1135R6] (The C++20 Synchronization Library), [P1643R1] (Add wait/notify to atomic_ref), and [P1644R0] (Add wait/notify to atomic<shared_ptr>): already have macros.

[P1466R3] (Miscellaneous minor fixes for chrono): already has a macro.

[P1754R1] (Rename concepts to standard_case for C++20, while we still can): this paper proposes to bump __cpp_lib_concepts, even though nobody implements this yet, because otherwise it’s confusing.

[P1614R2] (The Mothership has Landed): erroneously introduced the new macro __cpp_lib_spaceship, this paper proposes removing that macro and instead bumping __cpp_lib_three_way_comparison.

[P0325R4] (to_array from LFTS with updates): already has a macro.

[P0408R7] (Efficient Access to basic_stringbuf’s Buffer): no macro necessary. Trying to move out of the buffer will compile even without this feature, it just won’t be a move.

[P1423R3] (char8_t backward compatibility remediation): already has a macro.

[P1502R1] (Standard library header units for C++20): no macro necessary.

[P1612R1] (Relocate Endian’s Specification): already has a macro.

[P1661R1] (Remove dedicated precalculated hash lookup interface): already removes a macro.

[P1650R0] (Output std::chrono::days with ‘d’ suffix): no macro necessary.

[P1651R0] (bind_front should not unwrap reference_wrapper): no macro necessary.

[P1065R2] (Constexpr INVOKE): as above, this paper proposes to remove __cpp_lib_constexpr_invoke and bump the macro __cpp_lib_constexpr.

[P1207R4] (Movability of Single-pass Iterators): no macro necessary.

[P1035R7] (Input Range Adaptors): this paper proposes to bump the macro __cpp_lib_ranges.

[P1638R1] (basic_istream_view::iterator should not be copyable): no macro necessary.

[P1522R1] (Iterator Difference Type and Integer Overflow): no macro necessary.

[P1004R2] (Making std::vector constexpr): as above, this paper proposes to remove __cpp_lib_constexpr_vector and bump the macro __cpp_lib_constexpr.

[P0980R1] (Making std::string constexpr): as above, this paper proposes to remove __cpp_lib_constexpr_string and bump the macro __cpp_lib_constexpr.

[P0660R10] (Stop Token and Joining Thread, Rev 10): already has a macro.

[P1474R1] (Helpful pointers for ContiguousIterator): no macro necessary.

[P1523R1] (Views and Size Types): no macro necessary.

[P0466R5] (Layout-compatibility and Pointer-interconvertibility Traits): already has a macro.

[P1208R6] (Adopt source_location for C++20): already has a macro.

5 Fine- or coarse-grained macros for constexpr?

Many of the papers over the last few years slowly extend the amount of things we can write in constexpr. This is a fantastic development. However, all of these things are currently checked with the single macro __cpp_constexpr. This makes it difficult to figure out how to actually conditionally apply constexpr on functions. Additionally, implementations will be gated on bumping this macro themselves only in a prescribed, linear order - which means users will not be able to mark functions constexpr even when their implementation allows it.

As such, we could adopt several new macros (in addition to bumping __cpp_constexpr as described earlier) that are each dedicated to single extensions:

• [P1064R0]: __cpp_constexpr_virtual
• [P1002R1]: __cpp_constexpr_try_catch
• [P1327R1]: __cpp_constexpr_dynamic_cast
• [P1330R0]: __cpp_constexpr_change_active_union
• [P1668R1]: __cpp_constexpr_asm
• [P1331R2]: __cpp_constexpr_default_init

However, this would contrast sharply with decision in [P1424R1] to solely bump __cpp_lib_constexpr and not provide the granular macros. Since the only adopted guidance is that paper, and the goal of this paper is to fill in missing pieces, I wanted to simply point out that this problem exists.

Note also that on the library side, __cpp_lib_constexpr for the value 201907L includes four papers: [P0784R7], [P1065R2], [P0980R1], and [P1004R2]. These papers “just” add constexpr to some things, but some of these are much larger than others - which might gate a standard library advertising support for a constexpr std::invoke on the compiler implementing constexpr dynamic allocation.

The same problem exists with nodiscard, where in Cologne we adopted both the ability to add a reason to [[nodiscard]] ([P1301R4]) and the ability to add them to constructors, conceputally the former could merit its own feature-test.

SG-10 welcomes further papers on this issue.

6 Wording

Modify table 17 in 15.10 [cpp.predefined] with the following added:

Modify table 36 in 17.3.1 [support.limits.general] with the following added:

7 References