1.77.0
New Libraries
Describe:
A C++14 reflection library, from Peter Dimov. Provides macros for describing enumerators and struct/class members, and primitives for querying this information.
Lambda2:
A C++14, dependency-free, single header lambda library, from Peter Dimov. Allows simple function objects to be constructed via expressions such as _1 + 5, _1 % 2 == 0, _1 > _2, or _1 == ' ' || _1 == '\t'.
Updated Libraries
Any:
Added boost::anys::basic_any - a data type with customizable Small Object Optimization whose instances can hold instances of any type that satisfies ValueType requirements (many thanks to Ruslan Arutyunyan @rarutyun). If you are not sure that you really need it, then use boost::any instead.
Started using GithubActions CI for additional testing
Asio:
Added support for cancellation of individual asynchronous operations.
Added the associator trait, used to generically forward associators.
Changed all asynchronous operations implemented in Asio to invoke their completion handlers as rvalue references.
Added support for asynchronous operations with multiple completion signatures.
Added operator&& and operator|| for awaitable<>, to allow coroutines to be trivially awaited in parallel.
Added the experimental::as_tuple completion token adapter.
Added the experimental::append completion token adapter.
Added the experimental::prepend completion token adapter.
Added the experimental::deferred completion token, whicih allows deferred execution of asynchronous operations.
Added the experimental::parallel_group class and experimental::make_parallel_group function.
Added experimental::promise, which allows eager execution and synchronisation of asynchronous operations.
Added the experimental::coro class template, a C++20 coroutine primitive with the ability to combine both asynchronous waiting (co_await) and yielding (co_yield).
Added move assignment to ssl::stream<>.
Changed co_spawn to dispatch the coroutine's initial step to the executor.
Enabled additional optimisations for any_executor and any_io_executor when used with asynchronous operations.
Added the nodiscard attribute to awaitable<>.
Increased the number of cached slots in the default recycling allocator.
Changed the default allocator behaviour to respect alignment requirements, to support over-aligned types.
Ensured the result strings are always initialised in reverse name resolution.
Fixed recursive template instantiation issues in use_awaitable_t::executor_with_default.
Fixed the any_io_executor equality operators to correctly return a result based on the target executor.
Fixed strand<> to avoid using a potentially moved-from executor.
Ensured gcc tests are not used for clang when detecting compiler features.
Disabled coroutines support for the clang shipped with MSVC.
Fixed compatibility with recent LibreSSL when OPENSSL_NO_SSL_INTERN is defined.
Fixed posix::basic_stream_descriptor move operations to work with non-defaulted executors.
Consult the Revision History for further details.
Atomic:
Added make_atomic_ref and make_ipc_atomic_ref factory functions for constructing atomic reference objects.
Added C++17 template argument deduction guides for atomic_ref and ipc_atomic_ref to allow omitting template arguments when they can be deduced from constructor arguments.
Beast:
This maintenance update brings minor bug fixes and updated CI reporting.
We'd love to know how you or your company use Beast, consider adding an entry to the Companies and Individuals Using Beast list.
See the full Release Notes for a complete list of changes.
Conversion:
boost::implicit_cast is now constexpr.
Fixed broken links.
Started using GithubActions CI for additional testing
Core:
boost/core/uncaught_exceptions.hpp has been modified for compatibility with Mac OS 10.4 and older.
DLL:
More runtime checks for malformed ELFs in boost::dll::library_info
In boost::dll::library_info use info from .dynsym ELF section if .symtab is empty or missing
List protected ELF symbols in boost::dll::library_info as they are available for import
Remove unneeded convert from wchar_t to char on POSIX environment, thanks to Vladislav Shchapov
Started using GithubAction CI for additional testing
Filesystem:
Introducing Boost.Filesystem v4. This new version of the library removes all deprecated features of v3 and also makes a number of breaking API changes intended to make Boost.Filesystem more compatible with std::filesystem introduced in C++17. The differences are described in the release notes and documentation using v3 and v4 tags and are also summarised in a separate section. Users can select Boost.Filesystem version by defining BOOST_FILESYSTEM_VERSION macro to either 3 or 4 when compiling their code. There is no need to separately compile Boost.Filesystem for each library version - a single binary supports both v3 and v4. Users should avoid using both v3 and v4 in the same application as this can lead to subtle bugs. Currently, v3 is the default. In a future release v4 will become the default, and eventually v3 will be removed. v4 is functional, but it is still a work in progress and there may be breaking API changes in the future.
v4: path::filename, path::stem and path::extension no longer consider root name or root directory of the path as a filename if the path only consists of those components. For example, on Windows path("C:").filename() used to return "C:" and path("C:\\").filename() used to return "\" and both will return an empty path now.
v4: path::stem and path::extension no longer treat a filename that starts with a dot and has no other dots as an extension. Filenames starting with a dot are commonly treated as filenames with an empty extension. The leading dot is used to indicate a hidden file on most UNIX-like systems.
New: Improved support for various path prefixes on Windows. Added support for local device prefix ("\\.\") and experimental support for NT path prefix ("\??\"). The prefixes will be included in the root name of a path. Note that using the prefixes with Boost.Filesystem v3 can lead to surprising results (e.g. path("\\\\.\\").stem() == "\\\\"). It is recommended to use the prefixes only with Boost.Filesystem v4.
Reworked path::lexically_normal implementation to eliminate some cases of duplicate dot (".") elements in the normalized paths.
New: Added runtime detection of the statx and getrandom system calls on Linux. This can be useful if the syscall is present at compile time but fails with ENOSYS at run time (for example, in Docker containers that restrict the syscall, even if available on the host).
New: Added support for disabling usage of various system APIs at library build time. This can be useful when a certain API is detected as present by the library configuration scripts but must not be used for some reason (for example, when runtime detection does not work on the target system). See the description of configuration macros in the library documentationfor for more details.
New: Added copy_options::synchronize_data and copy_options::synchronize options for the copy_file operation. These options allow to synchronize the written data and attributes with the permanent storage. These options are expensive in terms of performance, but allow to ensure reliability of the copied data. Note that copy_file performed implicit data synchronization on POSIX systems since Boost.Filesystem 1.74.0. This release adds support for more platforms and disables data synchronization by default while allowing the caller to explicitly request it.
Added handling of EINTR error code on POSIX systems for some system calls issued internally. In particular, EINTR could have been ignored on close, which on HP-UX would result in a leaked file descriptor.
In the copy_file implementations based on Linux sendfile and copy_file_range system calls, added handling of error codes indicating that a particular filesystem does not support the system call and fall back to the generic read/write loop. This should fix copy_file failing on eCryptFS and possibly other filesystems.
The copy_file_range system call is now used since Linux kernel 4.5, whereas previously it was only enabled since 5.3. The copy_file implementation will fall back to sendfile or read/write loop if copy_file_range fails to copy a given file across filesystems.
The copy_file implementations based on Linux sendfile and copy_file_range system calls will not be used on filesystems that are known to contain files with generated content. These system calls are incompatible with such files, and copying them would result in zero-sized files. The generic read/write loop will be used instead. Currently, the blacklisted filesystems are: procfs, sysfs, tracefs and debugfs.
In the copy_file implementation based on read/write loop, increased the maximum size of the buffer used for temporary storage and take into account the target filesystem block size for more optimal performance.
On Windows CE, calling current_path to obtain the current path for a process will now fail with an error instead of returning successfully with a root path. This platform does not support current directory. Changing the current path was already failing similarly in previous releases of Boost.Filesystem.
In canonical, fixed the check for a symlink referencing a directory above root, if an earlier symlink was resolved to an absolute path with a different root from the original path.
In canonical, added a limit for the maximum number of symlinks that can be resolved during the call. The limit is currently at least 40 symlinks.
On Windows, canonical and weakly_canonical will now use path::preferred_separator for the root directory separator in the resulting paths. This fixes "file not found" errors caused by Windows API not handling generic separators in UNC paths and paths that start with the Win32 filesystem prefix ("\\?\").
New: Added weakly_canonical overloads taking base path as an argument.
On Windows, weakly_canonical no longer fails with an error if the input path contains elements that do not exist in the filesystem but are cancelled by a subsequent dot-dot ("..") element. For example, weakly_canonical("C:\\a\\..") would previously fail if "C:\a" directory did not exist.
In read_symlink on Windows, corrected reparse point handling. The operation would return an empty path for some mount points (for example, created by Box cloud storage driver) and directory junction points that had empty print names. The new implementation now parses substitute name of the reparse point and attempts to reconstruct a Win32 path from it.
On Windows, file streams provided in boost/filesystem/fstream.hpp will use wide character paths on libc++ versions 7.0 and higher, when the standard library supports opening files with wide character paths.
On Windows, creating symlinks should no longer require elevated privileges, if Windows is configured in Developer mode.
With some compilers, global objects used internally in Boost.Filesystem are now destroyed after user's global destructors are called. This allows to call Boost.Filesystem methods during the program termination stage. In particular, this concerns the path locale that is used for character code conversion and can be installed by calling path::imbue. The supported compilers include MSVC, GCC and Clang, as well as other compilers that support customizing program initialization order through #pragma section (for MSVC-compatible compilers) or __attribute__ ((init_priority)) (for GCC-compatible compilers).
JSON:
value_to supports TupleLike types.
value_to and value_from support std::array and similar types.
Implicit conversion operator from string to std::string_view.
std::hash specializations for json types.
Fixed allocation errors in object and key_value_pair.
Fixed crash when constructing array from a pair of iterators that form an empty range.
LexicalCast:
Fixed assignment to an int of a floating point value that is out of representable range for int.
Started using GithubActions CI for additional testing
Log:
Fixed compilation for riscv32 target.
Mp11:
Added mp_intersperse, mp_split, mp_join
Multi-index Containers:
Maintenance work.
PFR:
A MurMur Hash based implementation of hash_combine() is now used to reduce collisions count and improve quality of boost::pfr::hash_value()
Visual Studio 2017 now supported in C++14 mode (thanks to Denis Mikhailov aka @denzor200)
Issues found by inspect tool were fixed
Fixed some warnings, including removal of extra semicolons in include/boost/pfr/detail/fields_count.hpp
Added a compile time assert for inherited types (thanks to Denis Mikhailov aka @denzor200)
Reflection of aggregates with non movable fields is now possible if guaranteed copy elision is on
Fixed spelling issues
Started using GithubActions CI for additional testing
PropertyTree:
This is a maintenance release.
Extend compiler coverage in CI.
Fix Boost compatibility for CMakeLists.txt.
System:
The conversion operator from error_category to std::error_category has been improved and no longer requires <map> or <mutex>.
The comparison operators of error_category are now inline friends instead of member functions (a side effect of the previous change.)
error_condition now defers calling generic_category() to avoid instantiating the object until it's actually needed.
error_condition::failed and error_condition::message have been undeprecated, and operator bool() now once again returns failed().
The system category now doesn't call generic_category(), to avoid instantiating the object.
The return value of default_error_condition changes in some cases into an error_condition from the generic category, instead of from the system category. This happens on POSIX when the input error_code is from the system category and does not correspond to any errc_t value.
The interoperability of error_code and std::error_code has been improved substantially. It is now possible to construct boost::system::error_code from std::error_code, and it's possible to pass boost::system::error_code to functions taking std::error_code&.
A stream insertion operator for error_condition has been added.
Uuid:
Added to_chars for writing uuid strings without allocating
aff0e22727