Last week in github.com/boostorg (9-15 November 2026)
Hi all, Activity on the develop branches of the Boost libraries between 9 and 15 November. This report has been generated automatically with AI. THE NUMBERS - 162 commits on develop (22 CI-related) - 41 pull requests opened against develop, 38 merged - 15 issues opened, 23 closed - 29 libraries with activity, 27 contributors Most active libraries: - Move: 29 commits - Decimal: 27 commits - Geometry: 15 commits - Outcome: 15 commits - Interprocess: 10 commits - Log: 10 commits - Container: 9 commits - Math: 7 commits Most active contributors: - Ion Gaztañaga (51 commits, 4 libraries) - Shen-Ta Hsieh (24 commits, 1 library) - Andrey Semashev (14 commits, 2 libraries) - Niall Douglas (11 commits, 1 library) - Tinko Bartels (10 commits, 1 library) - Matt Borland (6 commits, 4 libraries) - Jean-Louis Leroy (6 commits, 1 library) - Saqib Khan (5 commits, 3 libraries) HIGHLIGHTS Boost 1.93.0 beta 1 Boost 1.93.0 beta 1 was published on 11 November. Move Ion Gaztañaga made 29 commits in Move, nearly all bug fixes. uninitialized_move used the wrong value_type, adl_move_swap swapped only the base part of a derived class, is_convertible was wrong for array, function and abstract-class targets, and a throwing move assignment leaked in uninitialized_merge_with_right_placed. unique_ptr now implements LWG 2899 and rejects at compile time an operator* that would bind to a temporary (LWG 4148). BOOST_MOVE_USE_STANDARD_LIBRARY_MOVE is deprecated. Container and Interprocess Ion Gaztañaga refactored Container's monotonic_buffer_resource, which now passes the requested alignment upstream and wastes less memory on over-alignment, and fixed MSVC 14.0 and 14.2 build errors. In both libraries the private node pools gained move construction and assignment. Interprocess allocators with different value types can now be compared, allocator<void> can be instantiated, and allocate_many no longer throws for a request of zero elements. New issues: #285, interprocess_semaphore waiters on macOS busy-spinning or waking about 2.5 ms late. Decimal Shen-Ta Hsieh made to_chars faster and fixed three defects in it: the hex and cohort-preserving scientific formats wrote past the end of a short buffer (#1550), formats without a precision failed in a buffer that held the full text (#1551), and hex with a precision gave wrong digits (#1552). exp is now computed in fixed point and rounded once (#1557): most results were off in the last place, and some normal results came out as 0. New PRs: #1562, the same for sinh, cosh, tanh, expm1 and exp2. Geometry Tinko Bartels merged the first of four groups of fixes found by generated grid tests: nine relate fixes, among them within giving wrong results for multi-linestrings (#1238). Moritz Fischer fixed undefined behaviour in sin_cos_degrees for negative angles, which on arm64 gave wrong geographic results. Nicola Gastaldon fixed a buffer that made a polygon disappear, and Barend Gehrels an intersection lost for nearly collinear boundaries. New PRs: #1515, #1516 and #1517, Tinko Bartels's remaining groups; #1513, more precise spherical projections near the poles, ported from PROJ (Paul S. Mamakos from Space Ocean Corp, 10 comments). Math Nick Thompson added two multi-objective optimizers, NSGA-II and qLogNEHVI, and a WebAssembly build that now binds nearly the whole library. He also fixed the non-central t tail for small degrees of freedom, where the complemented cdf was a factor of 2 too small (#1637). Matt Borland fixed an overrun in hypergeometric_pFq_checked_series and silenced a GCC 15 false positive (#1641). New PRs: #1648, pareto_polish for Pareto fronts. OpenMethod Jean-Louis Leroy fixed copy-assignment of a smart virtual_ptr from an empty smart pointer, which dereferenced null. The CMake minimum went up to 3.16 (fixes #139) and then down to 3.13, with the tests built without a precompiled header below 3.16. The troubleshooting page now has one section per error, each with an example and its fix. Log Andrey Semashev fixed the async sink losing log records on flush with a bounded blocking queue, and a race on Windows that could close an IPC semaphore while a thread still waited on it. He also merged David Christle's fix for once-blocks, whose fast path could see the initialization as done without seeing its data, observable on ARM64 (#261). MySQL Bruno (@bruno-viva) added pool_params::idle_timeout: connection pools now close connections idle for that long, down to initial_size (#499). New PRs: #546, handing out the most recently returned connection first so the pool also shrinks under steady load (for #545); #547, sans-io test coverage (Vladislav Soulgard). Also last week * Outcome: @ConnorQi01 fixed a test that kept the unit tests from compiling on 32-bit targets, and Niall Douglas fixed the WG14 result headers failing their own layout assertions on i386. * Beast: Mehtab Zafar fixed a segmentation fault when moving an empty saved_handler (#3135), and an out-of-bounds read in chunk_extensions after a parse error. * IOStreams: @RichardCao fixed the LZMA compressor silently dropping input when the multithreaded encoder timed out (#196). * Interval: Sayan Samanta fixed intervals built from integers that could miss the integer or be far too wide (#72). New issues: #74, containment lost with -ffast-math; #75, an x87 exception trapping after interval code returns. * Thread: @Keisawaakira added support for the PE TLS callback in Windows ARM64X images (#421). * Intrusive: Ion Gaztañaga fixed floor_log2 on MSVC ARM64, which truncated its input and returned wrong values. * New issues and PRs: Fiber PRs #355 (resetting const values in fiber specific storage) and #356 (documenting non-owning pointers), Context PR #345 (fcontext on arm64e), Algorithm PR #136 (wrong include guard in transform_inclusive_scan.hpp), ICL PR #67 (for #62), Graph PR #699 (removing a loop macro). CI MAINTENANCE * Boost.CI: Alexander Grund reworked the coverage steps: data files are uploaded as a CI artifact, coveralls uploads log more, and the output is quieter. * Container, ContainerHash, Interprocess, Intrusive and Move: Ion Gaztañaga and Peter Dimov removed the retired macos-14 image; Matt Borland replaced it with macos-26 in Charconv. * Charconv, Decimal and Int128: Saqib Khan added an optional workflow that runs the CUDA tests on a simulated H100 GPU; Matt Borland updated the Ubuntu CMake runners. * Outcome: Niall Douglas added a 32-bit Linux job and fixed the release workflow, whose deploy-key output had been silently skipping the push to the boostorg mirror. * IOStreams: Tobias Stensbeck rewrote the Drone configuration in Jsonnet, with Windows jobs and focused sanitizer coverage. Have a good week!
participants (1)
-
Boost Fiscal Sponsorship Committee