=========================================== Libc++ 21.0.0 (In-Progress) Release Notes =========================================== .. contents:: :local: :depth: 2 Written by the `Libc++ Team `_ .. warning:: These are in-progress notes for the upcoming libc++ 21.0.0 release. Release notes for previous releases can be found on `the Download Page `_. Introduction ============ This document contains the release notes for the libc++ C++ Standard Library, part of the LLVM Compiler Infrastructure, release 21.0.0. Here we describe the status of libc++ in some detail, including major improvements from the previous release and new feature work. For the general LLVM release notes, see `the LLVM documentation `_. All LLVM releases may be downloaded from the `LLVM releases web site `_. For more information about libc++, please see the `Libc++ Web Site `_ or the `LLVM Web Site `_. Note that if you are reading this file from a Git checkout or the main Libc++ web page, this document applies to the *next* release, not the current one. To see the release notes for a specific release, please see the `releases page `_. What's New in Libc++ 21.0.0? ============================== Implemented Papers ------------------ - N4258: Cleaning-up noexcept in the Library (`Github `__) - P0767R1: Deprecate POD (`Github `__) - P1361R2: Integration of chrono with text formatting (`Github `__) - P2255R2: A type trait to detect reference binding to temporary (implemented the type traits only) (`Github `__) - P2372R3: Fixing locale handling in chrono formatters (`Github `__) - P2562R1: ``constexpr`` Stable Sorting (`Github `__) - P0472R3: Put std::monostate in (`Github `__) - P1222R4: A Standard ``flat_set`` (`Github `__) - P2897R7: ``aligned_accessor``: An mdspan accessor expressing pointer over-alignment (`Github `__) - P3247R2: Deprecate the notion of trivial types (`Github `__) - P3372R3: ``constexpr`` containers and adaptors (`Github `__) (``forward_list``, ``list``, ``priority_queue``, ``flat_map``, and ``flat_set`` are implemented) - P2441R2: ``views::join_with`` (`Github `__) - P2711R1: Making multi-param constructors of ``views`` ``explicit`` (`Github `__) - P2770R0: Stashing stashing ``iterators`` for proper flattening (`Github `__) - P2655R3: ``common_reference_t`` of ``reference_wrapper`` Should Be a Reference Type (`Github `__) - P2944R3: Comparisons for ``reference_wrapper`` (`Github `__) - P3379R0: Constrain ``std::expected equality`` operators (`Github `__) Improvements and New Features ----------------------------- - The ``std::ranges::{copy, copy_n, copy_backward, move, move_backward, rotate}`` algorithms have been optimized for ``std::vector::iterator``, resulting in a performance improvement of up to 2000x. - The ``std::ranges::equal`` algorithm has been optimized for ``std::vector::iterator``, resulting in a performance improvement of up to 188x. - The ``std::ranges::swap_ranges`` algorithm has been optimized for ``std::vector::iterator``, resulting in a performance improvement of up to 611x. - Updated formatting library to Unicode 16.0.0. - The ``num_put::do_put`` integral overloads have been optimized, resulting in a performance improvement of up to 2.4x. - The ``std::stable_sort`` algorithm uses radix sort for floating-point types now, which can improve the performance up to 10x, depending on type of sorted elements and the initial state of the sorted array. - The segmented iterator optimization for ``std::for_each`` has been backported to C++11. Previously it was only available in C++23 and later. - The ``std::for_each_n``, ``std::ranges::for_each`` and ``std::ranges::for_each_n`` algorithms have been optimized for segmented iterators, resulting in a performance improvement of up to 17.7x for ``std::deque`` iterators, and up to 13.9x for ``std::join_view>>`` iterators. - The ``bitset::to_string`` function has been optimized, resulting in a performance improvement of up to 8.3x for bitsets with uniformly distributed zeros and ones, and up to 13.5x and 16.1x for sparse and dense bitsets, respectively. - The ``flat_map::insert`` and ``flat_set::insert_range`` have been optimized, resulting in a performance improvement of up to 10x for inserting elements into a ``flat_map`` when the input range is a ``flat_map`` or a ``zip_view``. - ``ctype::tolower`` and ``ctype::toupper`` have been optimized, resulting in a 2x performance improvement. - As an experimental feature, Hardening now supports assertion semantics that allow customizing how a hardening assertion failure is handled. The four available semantics, modeled on C++26 Contracts, are ``ignore``, ``observe``, ``quick-enforce`` and ``enforce``. The ``observe`` semantic is intended to make it easier to adopt Hardening in production but should not be used outside of this scenario. Please refer to the :ref:`Hardening documentation ` for details. Deprecations and Removals ------------------------- - ``std::is_pod`` and ``std::is_pod_v`` are deprecated in C++20 and later. - ``std::is_trivial`` and ``std::is_trivial_v`` are deprecated in C++26 and later. - The ``_LIBCPP_VERBOSE_ABORT_NOT_NOEXCEPT`` has been removed, making ``std::__libcpp_verbose_abort`` unconditionally ``noexcept``. - libc++ no longer adds ``constexpr`` to ``std::hash>::operator()``, as the ``constexpr`` addition since C++20 was an unintended extension. - The non-conforming extension ``packaged_task::result_type`` has been removed in LLVM 21. Potentially breaking changes ---------------------------- - The implementation of ``num_put::do_put`` has been replaced to improve the performance, which can lead to different output when printing pointers. - User-defined specializations of ``std::common_reference`` are diagnosed now. To customize the common reference type, ``std::basic_common_reference`` should be specialized instead. - ``std::function`` used to have allocator support, which was removed from the Standard by `http://wg21.link/p0302r1` due to issues with its design and inconsistent support from implementations. Previously, libc++ would provide allocator-aware APIs in ``std::function`` in C++11 and C++14, but ignores the allocator argument in all places but one. Starting in this release, the allocator argument is always ignored. Announcements About Future Releases ----------------------------------- LLVM 22 ~~~~~~~ - The status of the C++03 implementation will be frozen after the LLVM 21 release. This means that starting in LLVM 22, non-critical bug fixes may not be back-ported to C++03, including LWG issues. C++03 is a legacy platform, where most projects are no longer actively maintained. To reduce the amount of fixes required to keep such legacy projects compiling with up-to-date toolchains, libc++ will aim to freeze the status of the headers in C++03 mode to avoid unintended breaking changes. See https://discourse.llvm.org/t/rfc-freezing-c-03-headers-in-libc for more details. If you are using C++03 in your project, you should consider moving to a newer version of the Standard to get the most out of libc++. ABI Affecting Changes --------------------- - When using GCC, the ``std`` namespace is now annotated with ``[[gnu::visibility("default")]]``. This may cause more symbols to be exported from shared libraries when building with ``-fvisibility=hidden``. This also fixes RTTI comparison between shared libraries, since all RTTI has the correct visibility now. There is no behaviour change on Clang. - LLVM 20 contained an ABI break that can result in the size of ``std::unordered_{map,set,multimap,multiset}`` and ``std::deque`` changing when used with an allocator type that is empty and contains a base class that is the same across rebound allocator instantiations (e.g. ``Allocator`` and ``Allocator`` are both empty and contain the same base class). In addition, the layout of a user-defined type that: - contains one of the following containers: ``std::unordered_{map,set,multimap,multiset}``, ``std::deque``, ``std::map``, ``std::set``, ``std::multimap``, ``std::multiset``, ``std::list`` or ``std::vector``, and - passes an empty allocator, comparator or hasher type to that container, and - has a member of that same empty allocator, comparator or hasher type inside the enclosing struct, and - that member is either marked with ``[[no_unique_address]]`` or optimized out via the EBO (empty base optimization) technique saw its size increase from LLVM 19 to LLVM 20. This was caused by the usage of ``[[no_unique_address]]`` within some of libc++'s containers in a way that allowed subtle interactions with enclosing objects. This is fixed in LLVM 21 when using the Clang compiler (returning to the LLVM 19 ABI), however that implies an ABI break from LLVM 20 to LLVM 21. Furthermore, fixing this causes a slight regression to constant evaluation support in ``std::unique_ptr``. Specifically, constant evaluation will now fail when the deleter relies on being value-initialized for constant-evaluation admissibility. If a default-initialized deleter can be used during constant evaluation, or if the default constructor is non-trivial, the ``unique_ptr`` is not affected by this regression. In particular, this regression does not impact any ``unique_ptr`` using the default deleter. Note that there is currently no way to realistically fix this ABI break on GCC, therefore GCC will remain on the ABI introduced in LLVM 19. That also means that Clang and GCC will have a slightly different ABI for the small subset of types listed above until we are able to apply the same fix we did with Clang on GCC. For more details see https://llvm.org/PR154146. Build System Changes -------------------- - TODO