Skip to content

Releases

v3.1.0 — April 19, 2026 · GitHub ↗

macOS support, lint enforcement, code coverage, packaging, and portability fixes.

Added

  • macOS CI pipeline (macOS 26) with OpenMPI
  • Lint workflow: clang-format-21 check and clang-tidy-18 static analysis on every push and PR
  • scripts/lint.sh — local lint helper with --format-only, --tidy-only, --fix, and --jobs flags
  • Code coverage instrumentation and upload to Codecov
  • CMake install rules, pkg-config file, CPack DEB packaging, and a GPG-signed APT repository
  • MSYS2 PKGBUILD for Windows packaging
  • CI publish jobs for both Ubuntu (.deb) and Windows (.pkg.tar.zst) on version tags

Changed

  • Windows CI runner upgraded from 2022 to 2025
  • Queue renamed to queue and wrapped in the gempba namespace
  • Bundled spdlog replaced with system-installed spdlog
  • CPM.cmake download hardened against silent corruption

Fixed

  • Bug in mpi_centralized_scheduler
  • Portability: removed <bits/stdc++.h>, gated <stacktrace> on __has_include, fixed score type-dispatch across platforms, dropped -fconcepts, gated -rdynamic and stdc++exp for non-Linux targets

Removed

  • Private node_manager method that always returned zero

v3.0.0 — December 8, 2025 · GitHub ↗

Ground-up architectural rewrite. Single-header facade, template-free nodes, fully pluggable subsystems.

Added

  • Single-header facade: #include <gempba/gempba.hpp> replaces multi-step initialization
  • Template-free node API — internal type erasure eliminates Node<...> and ResultHolder<...> from user code entirely
  • Lazy node evaluation: arguments constructed only when should_branch() is called, avoiding allocation for pruned branches
  • Visitor-based statistics system (stats / stats_visitor interfaces + default_mpi_stats_visitor)
  • Pluggable architecture: every subsystem (scheduler, load balancer, node core, serial runnable, stats) replaceable via interface

Changed

  • branch_handler → node_manager
  • mpi_scheduler → scheduler, exposing center_view() and worker_view()
  • Factory functions relocated to gempba::mt:: (multithreading) and gempba::mp:: (multiprocessing) namespaces
  • try_push_mt() / try_push_mp() → try_local_submit() / try_remote_submit()

Removed

  • DLB_Handler singleton
  • branch_handler singleton
  • ResultHolder<> template
  • Manual virtual root linking (now handled automatically)

Requirements

  • C++23

v2.0.0 — August 17, 2025 · GitHub ↗

IPC data model overhaul. Raw-byte transport, new result and score types, naming cleanup.

Added

  • gempba::task_packet — raw byte buffer replaces string-serialized IPC data
  • gempba::result — dedicated type for shipping results across processes
  • gempba::score — type-erased numeric, supports multiple primitive types (replaces reference_value)

Changed

  • Classes, member functions, and member variables renamed per .clang_tidy conventions
  • Schedulers now exchange raw bytes instead of serialized strings
  • Fixed MPI communicator probing in centralized scheduler

Build

  • Improved CMakeLists readability
  • openmpi invocation now correctly bound to intended core count

v1.1.0 — August 4, 2025 · GitHub ↗

MPI scheduler cleanup and citation support.

Added

  • Citation files (CITATION.cff)

Changed

  • MPI scheduler communicator probing reorganized for correctness and readability
  • runCenter and runNode refactored with switch statements for clarity

v1.0.2 — June 8, 2025 · GitHub ↗

Improved external dependency integration and build ergonomics. No compatibility breaks.

Added

  • CPM for external CMake dependency management
  • Custom initial process topology injection via new scheduler method
  • Macros now ship with defaults; consumer projects can override as needed

Changed

  • GemPBA now straightforward to import as an external CMake dependency
  • Macros renamed for clarity

Fixed

  • Compilation errors in several examples

Build

  • Improved CMakeLists readability
  • Files reorganized

v1.0.1 — October 30, 2024 · GitHub ↗

Project structure improvements and CI/CD setup. No compatibility breaks.

Added

  • Common parent class for MPI_Scheduler and MPI_Scheduler_Centralized
  • Common parent class for ResultHolder
  • CI/CD pipeline triggered on merges to main
  • Each example now builds as its own standalone executable, ensuring compile-time coverage on changes

Fixed

  • Minor bugs related to macro usage

Build

  • Project structure reorganized

v1.0.0 — April 9, 2024 · GitHub ↗

Initial stable release.