Meta LWG issue: 2022-11 meeting

[StephanTLavavej]

(Previous meta-issue: #2872)

At the November 2022 meeting, the following LWG issues were resolved in the C++ Working Paper.

❔ Not yet analyzed

• Remaining issues:

❌ Not applicable

If an issue requires no action from implementers, we mark it as N/A. Categories:

• Pure wording clarifications with nothing to implement (these can be changes to non-normative text like examples and informative notes, or wording cleanups to normative text that don't impact observable behavior)
  • LWG-3600 Making istream_iterator copy constructor trivial is an ABI break
  • LWG-3753 Clarify entity vs. freestanding entity
  • LWG-3781 The exposition-only alias templates cont-key-type and cont-mapped-type should be removed
  • LWG-3818 Exposition-only concepts are not described in library intro
  • LWG-3822 Avoiding normalization in filesystem::weakly_canonical
  • LWG-3824 Number of bind placeholders is underspecified
• Something that increases the restrictions placed on users, but implementers aren't expected to enforce those restrictions
• Fixes for obviously broken wording, where implementers would have done the right thing anyways
  • LWG-3028 Container requirements tables should distinguish const and non-const variables
  • LWG-3177 Limit permission to specialize variable templates to program-defined types
  • LWG-3597 Unsigned integer types don't model advanceable
  • LWG-3732 prepend_range and append_range can't be amortized constant time
  • LWG-3738 Missing preconditions for take_view constructor
  • LWG-3775 Broken dependencies in the Cpp17Allocator requirements
  • LWG-3817 Missing preconditions on forward_list modifiers

😸 Already implemented

Sometimes we cite LWG issues in product code comments as we're implementing their proposed resolutions. When the resolutions are officially accepted, we should remove the citations (as the default assumption is that we're implementing what the Standard says). If something is especially subtle, we can convert the citation to mention the relevant Standard section. Sometimes we should add test coverage - e.g. when the Standard begins requiring something that we were already doing, but weren't explicitly testing for.

• Already implemented, comments need to be removed and messages need to cite the Standard
  • Done by Update comments for approved LWG issues #3238:
    • LWG-3569 join_view fails to support ranges of ranges with non-default_initializable iterators
    • LWG-3745 std::atomic_wait and its friends lack noexcept
    • LWG-3764 reference_wrapper::operator() should propagate noexcept
    • LWG-3766 view_interface::cbegin is underconstrained
    • LWG-3788 jthread::operator=(jthread&&) postconditions are unimplementable under self-assignment
• Implemented without comments
  • LWG-3118 fpos equality comparison unspecified
  • LWG-3636 formatter<T>::format should be const-qualified
  • LWG-3747 ranges::uninitialized_copy_n, ranges::uninitialized_move_n, and ranges::destroy_n should use std::move
  • LWG-3754 Class template expected synopsis contains declarations that do not match the detailed description
  • LWG-3755 tuple-for-each can call user-defined operator,
  • LWG-3757 What's the effect of std::forward_like<void>(x)?
  • LWG-3759 ranges::rotate_copy should use std::move
  • LWG-3765 const_sentinel should be constrained
  • LWG-3770 const_sentinel_t is missing
  • LWG-3782 Should <math.h> declare ::lerp?
  • LWG-3784 std.compat should not provide ::byte and its friends
  • LWG-3792 __cpp_lib_constexpr_algorithms should also be defined in <utility>
  • LWG-3795 Self-move-assignment of std::future and std::shared_future have unimplementable postconditions
  • LWG-3796 movable-box as member should use default-initialization instead of copy-initialization

🩹 Patches an unimplemented feature

We should record this LWG issue in the GitHub issue tracking the feature. That way, we'll remember to verify it, but it doesn't represent net new work.

• Recorded in P0429R9 <flat_map> #2910:
  • LWG-3816 flat_map and flat_multimap should impose sequence container requirements
• Recorded in P1222R4 <flat_set> #2912:
  • LWG-3751 Missing feature macro for flat_set
  • LWG-3774 <flat_set> should include <compare>
• Recorded in P1642R11 Freestanding Library: Easy [utilities], [ranges], And [iterators] #2914:
  • LWG-3814 Add freestanding items requested by NB comments
• Recorded in P2286R8 Formatting Ranges #2919:
  • LWG-3750 Too many papers bump __cpp_lib_format
• Recorded in P2321R2 zip #2252:
  • LWG-3773 views::zip_transform still requires F to be copy_constructible when empty pack
• Recorded in P2374R4 views::cartesian_product #2923:
  • LWG-3760 cartesian_product_view::iterator's parent_ is never valid
  • LWG-3761 cartesian_product_view::iterator::operator- should pass by reference
  • LWG-3801 cartesian_product_view::iterator::distance-from ignores the size of last underlying range
• Recorded in P2502R2 <generator>: Synchronous Coroutine Generator For Ranges #2936:
  • LWG-3762 generator::iterator::operator== should pass by reference
  • LWG-3826 Redundant specification [for overload of yield_value]

🐞 Not yet implemented

• Filed a GitHub issue labeled LWG
  • LWG-3515 [stacktrace.basic.nonmem]: operator<< should be less templatized (filed LWG-3515 [stacktrace.basic.nonmem]: operator<< should be less templatized #3217)
  • LWG-3545 std::pointer_traits should be SFINAE-friendly
  • LWG-3594 inout_ptr - inconsistent release() in destructor (filed LWG-3594 inout_ptr - inconsistent release() in destructor #3218)
  • LWG-3629 make_error_code and make_error_condition are customization points (filed LWG-3629 make_error_code and make_error_condition are customization points #3216)
  • LWG-3646 std::ranges::view_interface::size returns a signed type (filed LWG-3646 std::ranges::view_interface::size returns a signed type #3219)
  • LWG-3677 Is a cv-qualified pair specially handled in uses-allocator construction? (filed LWG-3677 Is a cv-qualified pair specially handled in uses-allocator construction? #3220)
  • LWG-3717 common_view::end should improve random_access_range case (filed LWG-3717 common_view::end should improve random_access_range case #3221)
  • LWG-3736 move_iterator missing disable_sized_sentinel_for specialization (filed LWG-3736 move_iterator missing disable_sized_sentinel_for specialization #3222)
  • LWG-3737 take_view::sentinel should provide operator- (filed LWG-3737 take_view::sentinel should provide operator- #3223)
  • LWG-3743 ranges::to's reserve may be ill-formed (filed LWG-3743 ranges::to's reserve may be ill-formed #3224)
  • LWG-3746 optional's spaceship with U with a type derived from optional causes infinite constraint meta-recursion (filed LWG-3746 optional's spaceship with U with a type derived from optional causes infinite constraint meta-recursion #3225)
  • LWG-3778 vector<bool> missing exception specifications (filed LWG-3778 vector<bool> missing exception specifications #3226)
  • LWG-3785 ranges::to is over-constrained on the destination type being a range (filed LWG-3785 ranges::to is over-constrained on the destination type being a range #3227)
  • LWG-3798 Rvalue reference and iterator_category (filed LWG-3798 Rvalue reference and iterator_category #3228)
  • LWG-3823 Unnecessary precondition for is_aggregate (filed LWG-3823 Unnecessary precondition for is_aggregate #3229)
• PR out for review

[fsb4000]

Implemented without comments:
LWG-3636 formatter<T>::format should be const-qualified
7b2c54b

[cpplearner]

LWG-3750 Too many papers bump __cpp_lib_format

Patches an unimplemented feature: P2286R8 Formatting Ranges #2919

LWG-3754 Class template expected synopsis contains declarations that do not match the detailed description

Implemented without comments

[JMazurkiewicz]

Implemented without comments:
LWG-3765 const_sentinel should be constrained (PR: #3043, commit: ed53e6b).

[frederick-vs-ja]

Pure wording clarifications with nothing to implement:

• LWG-3753 Clarify entity vs. freestanding entity
• LWG-3781 The exposition-only alias templates cont-key-type and cont-mapped-type should be removed
• LWG-3818 Exposition-only concepts are not described in library intro
• LWG-3824 Number of bind placeholders is underspecified (nothing to be implemented in MSVC STL, the number has been documented here)

Fixes for obviously broken wording, where implementers would have done the right thing anyways (checking may be added in some cases):

• LWG-3028 Container requirements tables should distinguish const and non-const variables
• LWG-3177 Limit permission to specialize variable templates to program-defined types
• LWG-3597 Unsigned integer types don't model advanceable (the title is wrong, which is my fault)
• LWG-3732 prepend_range and append_range can't be amortized constant time
• LWG-3738 Missing preconditions for take_view constructor
• LWG-3775 Broken dependencies in the Cpp17Allocator requirements
• LWG-3817 Missing preconditions on forward_list modifiers

Already implemented, comments need to be removed and messages need to cite the Standard:

• LWG-3545 std::pointer_traits should be SFINAE-friendly
• LWG-3745 std::atomic_wait and its friends lack noexcept
• LWG-3764 reference_wrapper::operator() should propagate noexcept

Implemented without comments:

• LWG-3118 fpos equality comparison unspecified
• LWG-3629 make_error_code and make_error_condition are customization points
• LWG-3747 ranges::uninitialized_copy_n, ranges::uninitialized_move_n, and ranges::destroy_n should use std::move (IIUC this got completed in Pass rvalues to _Seek_to in ranges::uninitialized_(copy|move)_n #2964)
• LWG-3755 tuple-for-each can call user-defined operator,
• LWG-3757 What's the effect of std::forward_like<void>(x)?
• LWG-3759 ranges::rotate_copy should use std::move
• LWG-3782 Should <math.h> declare ::lerp?
• LWG-3784 std.compat should not provide ::byte and its friends
• LWG-3792 __cpp_lib_constexpr_algorithms should also be defined in <utility>
• LWG-3795 Self-move-assignment of std::future and std::shared_future have unimplementable postconditions
• LWG-3796 movable-box as member should use default-initialization instead of copy-initialization (the repeat_view part is being implemented in Implement P2474R2 views::repeat #3142, no comment added)

Patches an unimplemented feature:

• Patches for P1642R11 P1642R11 Freestanding Library: Easy [utilities], [ranges], And [iterators] #2914
  • LWG-3814 Add freestanding items requested by NB comments

[cpplearner]

• LWG-3629 make_error_code and make_error_condition are customization points

Note that the resolution makes make_error_code & make_error_condition only found by ADL (not by regular unqualified lookup). I believe that this is not yet implemented.

[StephanTLavavej]

Thanks, I've incorporated all your findings! I was able to go through the remaining list and verify that most of those issues require implementation changes, so I've filed tracking issues for them. That leaves:

• LWG-3600 Making istream_iterator copy constructor trivial is an ABI break
  • I'm not sure if we need to make changes here.
• LWG-3822 Avoiding normalization in filesystem::weakly_canonical
  • This appears to be N/A (obviously broken wording) since it merely removes a suggestion for implementers?

[strega-nil-ms]

LWG-3600 – we don't have to do anything here. We don't explicitly define the copy constructor, so it's constexpr whenever the copy of _Myival and _Myistr is constexpr; since _Myistr is a pointer, that means all we care about is _Myival, and thus we have equivalence to the paragraph in the issue.

This issue is basically only interesting for libstdc++, since their existing copy constructor was non-trivial.

[strega-nil-ms]

LWG-3822 – I'm nearly certain that there is nothing for us to do here, however I'll take a further look tomorrow on the big screens 😸

[CaseyCarter]

LWG-3822 – I'm nearly certain that there is nothing for us to do here, however I'll take a further look tomorrow on the big screens 😸

Yes. This issue is actually "we don't need to tell implementers to avoid doing things that are unnecessary." There's no functional change.

[CaseyCarter]

#3238 removes "LWG-XXXX" and strengthened comments for issues we already implement. With that, we're all done with classification. Thanks everyone!