A library for programming with effect handlers in C++

The C++Eff logo

Effect handlers allow for programming with user-defined computational effects, with applications including custom lightweight concurrency (threads, async-await, actors, generators), error handling, dependency injection, etc. Effect handlers originate from the realm of functional programming, and the main goal of this highly experimental library is to explore how they fit in the more object-oriented setting of C++.

The library relies on modern C++ features (move semantics, variadic templates, compile-time evaluation) to achieve elegant programmer-level interface, memory management of handlers, and relative type-safety. Internally, it uses the boost::context library for call-stack manipulation, and so it implements one-shot handlers only.


  • Reference – A detailed explanation of the library’s API and a short discussion about the overall design of the library. It is a suitable introduction for a reader already familiar with effect handlers (for example, as they are usually discussed in the context of functional programming).

  • Coming soon: Tutorial – An introduction to programming with effect handlers. Suitable for readers not familiar with handlers.

A quick example: lightweight cooperative threads

As a sneak preview, we can use effect handlers to define our own tiny library for cooperative lightweight threads. The programmer’s interface will consist of two functions, yield and fork, together with a class that implements a scheduler:

void yield();                          // Used by a thread to give up control
void fork(std::function<void()> proc); // Start a new thread

class Scheduler {
  static void Start(std::function<void()> f);

The static member function Start initiates the scheduler with f as the body of the first thread. It returns when all threads finish their jobs.

To implement this interface, we first define two commands, which are data structures used for transferring control from the client code to the handler. We implement yield and fork to invoke these commands. (The name of the class OneShot is supposed to remind the programmer that we’re dealing with one-shot handlers only, meaning you cannot resume the same resumption twice).

#include "cpp-effects/cpp-effects.h"
using namespace CppEffects;

struct Yield : Command<void> { };

struct Fork : Command<void> {
  std::function<void()> proc;

void yield()

void fork(std::function<void()> proc)
  OneShot::InvokeCmd(Fork{{}, proc});

We define the scheduler, which is a handler that can interpret the two commands by pushing the resumptions (i.e., captured continuations) to the queue.

// Res is the type of suspended threads
using Res = std::unique_ptr<Resumption<void, void>>;

class Scheduler : public Handler<void, void, Yield, Fork> {
  static void Start(std::function<void()> f)
    while (!queue.empty()) { // Round-robin scheduling
      auto resumption = std::move(queue.front());
  static std::list<Res> queue;
  static void Run(std::function<void()>)
  void CommandClause(Yield, Res r) override
  void CommandClause(Fork f, Res r) override
    queue.push_back(OneShot::MakeResumption<void>(std::bind(Run, f.proc)));
  void ReturnClause() override { }

std::list<Res> Scheduler::queue;

And that’s all it takes! We can now test our library by starting a few threads:

void worker(int k)
  for (int i = 0; i < 10; ++i) {
    std::cout << k;

void starter()
  for (int i = 0; i < 5; ++i) {
    fork(std::bind(worker, i));

int main()

  // Output:
  // 01021032104321043210432104321043210432104321432434

Technical summary

  • Language: C++17
  • Handlers: deep, one-shot, parametrised [1]

[1] – In the library handlers are objects, so they can naturally contain any data, auxiliary functions, and additional programmer’s interface.


The easiest way to compile the library and the examples is to use cmake. You will need cmake and boost in any non-ancient versions. For example, the following should do the trick on macOS:

$ brew install cmake
$ brew install boost
$ cmake .
$ make

This will build the library and the examples. You can check that it works by running an example. The following will run the threads example – you can see the interleaving of threads in the output:

$ bin/threads


View Github