[V4] Cancellation 2.0

.codespellrc

@@ -1,5 +1,5 @@
 [codespell]
-builtin = clear,rare,en-GB_to_en-US,names,informal,code
+builtin = clear,rare,names,informal,code
 check-filenames =
 check-hidden =
 ignore-words-list = deque,warmup,stdio,copyable,combinate

AGENTS.md

@@ -103,34 +103,6 @@ All tests should pass. If tests fail, check that:
 - Build completed without errors
 - Any changes you have made are correct
 
-## Linting & Validation
-
-The CI runs two linting tools that you should run before committing:
-
-### codespell (spelling)
-
-```bash
-codespell
-```
-
-Config: `.codespellrc` (ignores: build/, .git/, etc.)
-Should produce no output if passing.
-
-### clang-format (code formatting)
-
-```bash
-find src include test benchmark/src -name "*.cpp" -o -name "*.hpp" -o -name "*.cxx" | xargs clang-format --dry-run --Werror
-```
-
-Config: `.clang-format` (110 column limit, specific style)
-Should produce no output if passing.
-
-**To auto-fix formatting**:
-
-```bash
-find src include test benchmark/src -name "*.cpp" -o -name "*.hpp" -o -name "*.cxx" | xargs clang-format -i
-```
-
 ## Project Structure
 
 ### Source Layout
@@ -150,10 +122,9 @@ libfork/
 │   ├── batteries/            # libfork.batteries — stacks, contexts, adaptors
 │   │   ├── batteries.cxx     #   aggregator
 │   │   └── *.cxx             #   :partitions
-│   ├── schedulers/           # libfork.schedulers — concrete schedulers
+│   └── schedulers/           # libfork.schedulers — concrete schedulers
 │   │   ├── schedulers.cxx    #   aggregator
 │   │   └── *.cxx             #   :partitions
-│   └── exception.cpp         # terminate_with() implementation
 ├── test/src/**/              # Test suite (Catch2) — uses `import libfork;`
 │   └── *.cpp
 ├── benchmark/src/            # Benchmarking suite (google-benchmark)
@@ -189,7 +160,6 @@ All workflows follow this pattern:
 1. **Modify source files** in `src/`, `include/`, `test/`, or `benchmark/`
 2. **Rebuild**: `cmake --build --preset <your-preset>`
 3. **Test**: `ctest --preset <your-preset>`
-4. **Lint**: Run codespell and clang-format checks
 
 #### Adding/removing files from `src/` or `include/`
 
@@ -230,11 +200,3 @@ rm -rf build/
 
 **Problem**: "Could not automatically find libc++.modules.json"
 **Solution**: Ensure LLVM is installed via Homebrew; toolchain auto-detects the path
-
-### Linting Failures
-
-**Problem**: clang-format errors
-**Solution**: Run fix command above to auto-format code
-
-**Problem**: codespell errors
-**Solution**: Fix typos or add to ignore list in `.codespellrc` if false positive

CMakeLists.txt

@@ -31,7 +31,7 @@ set(CMAKE_CXX_MODULE_STD 1)
 add_library(libfork_libfork)
 add_library(libfork::libfork ALIAS libfork_libfork)
 
-# target_link_libraries(libfork_libfork PRIVATE Threads::Threads)
+target_link_libraries(libfork_libfork PUBLIC Threads::Threads)
 
 set_property(TARGET libfork_libfork PROPERTY EXPORT_NAME libfork)
 
@@ -84,6 +84,7 @@ target_sources(libfork_libfork
       src/core/execute.cxx
       src/core/receiver.cxx
       src/core/promise.cxx
+      src/core/stop.cxx
       # libfork.batteries
       src/batteries/batteries.cxx
       src/batteries/deque.cxx

benchmark/CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 4.2.1 FATAL_ERROR)
 
-project(libfork_benchmark LANGUAGES C CXX)
+project(libfork_benchmark LANGUAGES CXX)
 
 if(NOT CMAKE_BUILD_TYPE STREQUAL "Release")
   message(WARNING "It is recommended to build benchmarks in Release mode for accurate results.")
@@ -17,7 +17,8 @@ target_link_libraries(libfork_benchmark
     benchmark::benchmark_main
 )
 
-# Common headers
+# Common components
+
 target_sources(libfork_benchmark
   PRIVATE
     FILE_SET HEADERS FILES
@@ -28,18 +29,13 @@ target_sources(libfork_benchmark
       src
 )
 
-# Common sources
 target_sources(libfork_benchmark
   PRIVATE
     src/libfork_benchmark/uts/uts.cpp
 )
 
-
 # C lib for UTS
-add_library(uts_c OBJECT
-  src/libfork_benchmark/uts/external/uts.c
-  src/libfork_benchmark/uts/external/rng/brg_sha1.c
-)
+add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/external/uts external/uts)
 
 target_link_libraries(libfork_benchmark PRIVATE uts_c)
 

benchmark/external/uts/CMakeLists.txt

@@ -0,0 +1,19 @@
+cmake_minimum_required(VERSION 4.2.1 FATAL_ERROR)
+
+project(uts_external LANGUAGES C)
+
+add_library(uts_c)
+
+target_sources(uts_c
+  PRIVATE
+    src/uts.c
+    src/rng/brg_sha1.c
+  PUBLIC
+    FILE_SET HEADERS
+    BASE_DIRS ${CMAKE_CURRENT_SOURCE_DIR}/include
+    FILES
+      include/uts/uts.h
+      include/uts/rng/rng.h
+      include/uts/rng/brg_sha1.h
+      include/uts/rng/brg_types.h
+)

benchmark/src/libfork_benchmark/uts/external/rng/brg_sha1.h → benchmark/external/uts/include/uts/rng/brg_sha1.h

@@ -33,7 +33,7 @@
 #ifndef _SHA1_H
 #define _SHA1_H
 
-#include "brg_types.h"
+#include "uts/rng/brg_types.h"
 
 #define SHA1_BLOCK_SIZE 64
 #define SHA1_DIGEST_SIZE 20

benchmark/src/libfork_benchmark/uts/external/rng/rng.h → benchmark/external/uts/include/uts/rng/rng.h

@@ -1,6 +1,6 @@
 #ifndef _RNG_H
 #define _RNG_H
 
-#include "brg_sha1.h"
+#include "uts/rng/brg_sha1.h"
 
 #endif /* _RNG_H */

benchmark/src/libfork_benchmark/uts/external/uts.h → benchmark/external/uts/include/uts/uts.h

@@ -22,7 +22,7 @@
 extern "C" {
   #endif
 
-  #include "rng/rng.h"
+  #include "uts/rng/rng.h"
 
   #define UTS_VERSION "2.1"
 

benchmark/src/libfork_benchmark/uts/external/rng/brg_sha1.c → benchmark/external/uts/src/rng/brg_sha1.c

@@ -37,7 +37,7 @@
 #include <string.h> /* for memcpy() etc.        */
 
 #include "brg_endian.h"
-#include "brg_sha1.h"
+#include "uts/rng/brg_sha1.h"
 
 #if defined(__cplusplus)
 extern "C" {

benchmark/src/libfork_benchmark/uts/external/uts.c → benchmark/external/uts/src/uts.c

@@ -19,7 +19,7 @@
 #include <string.h>
 #include <time.h>
 
-#include "uts.h"
+#include "uts/uts.h"
 
 /***********************************************************
  *  tree generation and search parameters                  *

benchmark/src/libfork_benchmark/common.hpp

@@ -12,7 +12,15 @@ struct incorrect_result : public std::runtime_error {
   using std::runtime_error::runtime_error;
 };
 
-inline constexpr unsigned bench_max_threads = 12;
+inline void bench_thread_args(benchmark::Benchmark *bench, auto make_args) {
+  unsigned hw = std::max(1U, std::thread::hardware_concurrency());
+  for (unsigned t : {1U, 2U, 4U, 6U, 8U, 12U, 16U, 24U, 32U, 48U, 64U, 96U}) {
+    if (t > hw) {
+      return;
+    }
+    make_args(bench, t);
+  }
+}
 
 #define CHECK_RESULT(result, expected)                                                                       \
   do {                                                                                                       \

benchmark/src/libfork_benchmark/fib/libfork.cpp

@@ -22,12 +22,12 @@ struct fib {
     std::int64_t lhs = 0;
     std::int64_t rhs = 0;
 
-    using scope = lf::scope<Context>;
+    auto sc = co_await lf::scope();
 
-    co_await scope::fork(&rhs, fib{}, n - 2);
-    co_await scope::call(&lhs, fib{}, n - 1);
+    co_await sc.fork(&rhs, fib{}, n - 2);
+    co_await sc.call(&lhs, fib{}, n - 1);
 
-    co_await lf::join();
+    co_await sc.join();
 
     co_return lhs + rhs;
   }
@@ -41,6 +41,7 @@ void run(benchmark::State &state) {
 
   state.counters["n"] = static_cast<double>(n);
   state.counters["p"] = static_cast<double>(thread_count<Sch>(state));
+  state.SetComplexityN(static_cast<benchmark::IterationCount>(thread_count<Sch>(state)));
 
   Sch scheduler = make_scheduler<Sch>(state);
 
@@ -89,9 +90,12 @@ BENCH_ALL(inline_scheduler<poly_context<geometric_stack<>, adapt_deque>>)
   BENCHMARK_TEMPLATE(run, __VA_ARGS__)                                                                       \
       ->Name(#mode "/libfork/fib/" #__VA_ARGS__)                                                             \
       ->Apply([](benchmark::Benchmark *b) -> void {                                                          \
-        for (unsigned t = 1; t <= bench_max_threads; ++t) {                                                  \
+        bench_thread_args(b, [](benchmark::Benchmark *b, unsigned t) {                                       \
           b->Args({fib_##mode, static_cast<std::int64_t>(t)});                                               \
-        }                                                                                                    \
+        });                                                                                                  \
+      })                                                                                                     \
+      ->Complexity([](benchmark::IterationCount n) -> double {                                               \
+        return 1.0 / static_cast<double>(n);                                                                 \
       })                                                                                                     \
       ->UseRealTime();
 

benchmark/src/libfork_benchmark/uts/libfork.cpp

@@ -28,6 +28,8 @@ struct uts_fn {
     if (num_children > 0) {
       std::vector<pair> cs(static_cast<std::size_t>(num_children));
 
+      auto sc = co_await lf::scope();
+
       for (std::size_t i = 0; i < static_cast<std::size_t>(num_children); ++i) {
         cs[i].child.type = child_type;
         cs[i].child.height = parent->height + 1;
@@ -37,16 +39,14 @@ struct uts_fn {
           rng_spawn(parent->state.state, cs[i].child.state.state, static_cast<int>(i));
         }
 
-        using scope = lf::scope<Context>;
-
         if (i + 1 == static_cast<std::size_t>(num_children)) {
-          co_await scope::call(&cs[i].res, uts_fn{}, depth + 1, &cs[i].child);
+          co_await sc.call(&cs[i].res, uts_fn{}, depth + 1, &cs[i].child);
         } else {
-          co_await scope::fork(&cs[i].res, uts_fn{}, depth + 1, &cs[i].child);
+          co_await sc.fork(&cs[i].res, uts_fn{}, depth + 1, &cs[i].child);
         }
       }
 
-      co_await lf::join();
+      co_await sc.join();
 
       for (auto &&elem : cs) {
         r.maxdepth = std::max(r.maxdepth, elem.res.maxdepth);
@@ -69,6 +69,7 @@ void run(benchmark::State &state) {
   auto expected = expected_result(tree);
 
   state.counters["p"] = static_cast<double>(thread_count<Sch>(state));
+  state.SetComplexityN(static_cast<benchmark::IterationCount>(thread_count<Sch>(state)));
 
   Sch scheduler = make_scheduler<Sch>(state);
 
@@ -90,9 +91,12 @@ void run(benchmark::State &state) {
   BENCHMARK_TEMPLATE(run, __VA_ARGS__)                                                                       \
       ->Name(#mode "/libfork/uts/" tree_name "/" #__VA_ARGS__)                                               \
       ->Apply([](benchmark::Benchmark *b) -> void {                                                          \
-        for (unsigned t = 1; t <= bench_max_threads; ++t) {                                                  \
+        bench_thread_args(b, [](benchmark::Benchmark *b, unsigned t) {                                       \
           b->Args({tree_id, static_cast<std::int64_t>(t)});                                                  \
-        }                                                                                                    \
+        });                                                                                                  \
+      })                                                                                                     \
+      ->Complexity([](benchmark::IterationCount n) -> double {                                               \
+        return 1.0 / static_cast<double>(n);                                                                 \
       })                                                                                                     \
       ->UseRealTime();
 

benchmark/src/libfork_benchmark/uts/uts.hpp

@@ -2,7 +2,7 @@
 
 // Include the C UTS library header first (it defines max/min macros that would
 // clash with std::max/std::min after import std).
-#include "libfork_benchmark/uts/external/uts.h"
+#include "uts/uts.h"
 
 #undef max
 #undef min

src/core/core.cxx

@@ -22,3 +22,4 @@ export import :schedule;
 export import :root;
 export import :execute;
 export import :receiver;
+export import :stop;

src/core/frame.cxx

@@ -7,20 +7,15 @@ import std;
 
 import libfork.utils;
 
-namespace lf {
-// =================== Cancellation =================== //
-
-struct cancellation {
-  cancellation *parent = nullptr;
-  std::atomic<std::uint32_t> stop = 0;
-};
+import :stop;
 
-// =================== Frame =================== //
+namespace lf {
 
 // TODO: remove this and other exports
 export enum class category : std::uint8_t {
   call = 0,
   fork,
+  root,
 };
 
 export struct frame_base {};
@@ -38,7 +33,7 @@ struct frame_type : frame_base {
   uninitialized<std::exception_ptr> except;
 
   frame_type *parent;
-  cancellation *cancel;
+  stop_source::stop_token stop_token;
 
   [[no_unique_address]]
   Checkpoint stack_ckpt;
@@ -56,16 +51,9 @@ struct frame_type : frame_base {
   constexpr frame_type(Checkpoint &&ckpt) noexcept : stack_ckpt(std::move(ckpt)) { joins = k_u16_max; }
 
   [[nodiscard]]
-  constexpr auto is_cancelled() const noexcept -> bool {
-    // TODO: Should exception trigger cancellation?
-    for (cancellation *ptr = cancel; ptr != nullptr; ptr = ptr->parent) {
-      // TODO: if users can't use cancellation outside of fork-join
-      // then this can be relaxed
-      if (ptr->stop.load(std::memory_order_acquire) == 1) {
-        return true;
-      }
-    }
-    return false;
+  constexpr auto stop_requested() const noexcept -> bool {
+    // TODO: Should exception trigger stop?
+    return stop_token.stop_requested();
   }
 
   [[nodiscard]]