Reaction is a blazing-fast, modern C++20 header-only reactive framework that brings React/Vue-style dataflow to native C++ – perfect for UI Dataflow, Game logic, Financial Services, Real-time calculation and more.
- Zero-cost abstractions through template metaprogramming
- Minimal runtime overhead with smart change propagation
- Propagation efficiency at the level of millions per second
- Automatic DAG detection and cycle prevention
- Fine-grained change propagation control
- Configurable caching strategies
- Compile-time type checking with C++20 concepts
- Safe value semantics throughout the framework
- Framework manages object lifetime internally
| Feature / Metric | 🟩 QProperty (Qt6) | 🟨 RxCpp | 🟥 Reaction |
|---|---|---|---|
| Expression Support | ✅ setBinding(), but only single-layer |
✅ Supports chained map, combine_latest, etc. |
✅✅ Fully supports deep nested expressions |
| Expression Nesting Depth | ❌ Limited to one layer | ✅ Unlimited depth with automatic dependency tracking | |
| Update Propagation | Manual propagation per layer | Reactive push chain per layer | Automatic DAG-based propagation with pruning |
| Dependency Tracking | ❌ Manual | ✅ Automatic via lazy evaluation capturing dependencies | |
| Performance (Update Delay) | ✅ Fast (O(1) propagation) | ❌ Slow (heap allocations and nested chaining) | ✅✅ Fast (pruned update, lazy eval, diffing) |
| Memory Usage | ✅ Very low (stack + signals) | ❌ High (many heap-allocated observables) | |
| Syntax Simplicity | ✅ Simple (setBinding, value()) |
❌ Verbose template syntax | ✅ Clean expression templates, close to natural syntax |
| Type Support | ✅ Built-ins and registered custom types | ✅ Template-based, supports any type | ✅ Type-erased or templated support for any combination |
| Container Support | ✅ Can be used in containers | ✅ Can compose multiple observables | ✅ Supports container expressions (e.g. map/filter outputs) |
| Threading Model | UI-thread default, manual safety for signals | ✅ Multi-threaded pipelines | ✅ Main thread default, pluggable lock strategies |
| Error Handling | ❌ None | ✅ Robust error flow (on_error_resume_next) |
✅ Error node propagation, pluggable failure strategy |
| Debuggability | ❌ Difficult due to complex types | ✅ Trackable dependencies, observable IDs, chain tracing | |
| Template Instance Size | ✅ Small | ❌ Huge (template explosion) | ✅ Optimized with type-erasure or instance deduplication |
| Build Time | ✅ Fast | ❌ Very slow for large expressions | ✅ Separated headers, controllable instantiation |
| Learning Curve | ✅ Low (Qt-style usage) | ❌ Steep (functional style) | |
| Use Case Fit | UI property binding, light state syncing | Asynchronous pipelines, stream logic | UI + state modeling + expression trees with complex logic |
- Compiler: C++20 compatible (GCC 10+, Clang 12+, MSVC 19.30+)
- Build System: CMake 3.15+
To build and install the reaction reactive framework, follow the steps below:
git clone https://github.com/lumia431/reaction.git && cd reaction
cmake -B build
cmake --build build/
cmake --install build/ --prefix /your/install/pathAfter installation, you can include and link against reaction in your own CMake-based project:
find_package(reaction REQUIRED)
target_link_libraries(your_target PRIVATE reaction)#include <reaction/reaction.h>
#include <iostream>
#include <iomanip>
#include <cmath>
int main() {
using namespace reaction;
// 1. Reactive variables for stock prices
auto buyPrice = var(100.0).setName("buyPrice"); // Price at which stock was bought
auto currentPrice = var(105.0); // Current market price
// 2. Use 'calc' to compute profit or loss amount
auto profit = calc([&]() {
return currentPrice() - buyPrice();
});
// 3. Use 'expr' to compute percentage gain/loss
auto profitPercent = expr(std::abs(currentPrice - buyPrice) / buyPrice * 100);
// 4. Use 'action' to print the log whenever values change
auto logger = action([&]() {
std::cout << std::fixed << std::setprecision(2);
std::cout << "[Stock Update] Current Price: $" << currentPrice()
<< ", Profit: $" << profit()
<< " (" << profitPercent() << "%)\n";
});
// Simulate price changes
currentPrice.value(110.0).value(95.0); // Stock price increases
buyPrice.value(90.0); // Buy price adjusted
return 0;
}Define reactive state variables with var<T>.
auto a = reaction::var(1); // int variable
auto b = reaction::var(3.14); // double variable- get value:
auto val = a.get();- assignment:
a.value(2);Use calc to create reactive computations based on one or more var instances.
- Lambda Capture Style:
auto a = reaction::var(1);
auto b = reaction::var(3.14);
auto sum = reaction::calc([=]() {
return a() + b(); // Retrieve current values using a() and b()
});- Parameter Binding Style (High Performance):
auto ds = reaction::calc([](auto aa, auto bb) {
return std::to_string(aa) + std::to_string(bb);
}, a, b); // Dependencies: a and bexpr provides a clean and concise syntax to declare reactive expressions. The result automatically updates when any dependent variable changes.
auto a = reaction::var(1);
auto b = reaction::var(2);
auto result = reaction::expr(a + b * 3); // result updates automatically when 'a' or 'b' changeRegister actions to perform side effects whenever the observed variables change.
int val = 10;
auto a = reaction::var(1);
auto dds = reaction::action([&val]() {
val = a();
});Ofcourse, to get high performance can use Parameter Binding Style.
int val = 10;
auto a = reaction::var(1);
auto dds = reaction::action([&val](auto aa) {
val = aa;
}, a);For complex types with reactive fields allow you to define struct-like variables whose members are individually reactive.
Here's an example of a PersonField class:
class PersonField : public reaction::FieldBase {
public:
PersonField(std::string name, int age):
m_name(reaction::field(name)),
m_age(reaction::field(age)){}
std::string getName() const { return m_name.get(); }
void setName(const std::string &name) { m_name.value(name); }
int getAge() const { return m_age.get(); }
void setAge(int age) { m_age.value(age); }
private:
reaction::Var<std::string> m_name;
reaction::Var<int> m_age;
};
auto p = reaction::var(PersonField{"Jack", 18});
auto action = reaction::action(
[]() {
std::cout << "Action Trigger , name = " << p().getName() << " age = " << p().getAge() << '\n';
});
p->setName("Jackson"); // Action Trigger
p->setAge(28); // Action Triggerauto a = reaction::var(1);
auto b = reaction::var(3.14);
auto ds = reaction::calc([]() { return a() + b(); });
auto ds_copy = ds;
auto ds_move = std::move(ds);
EXPECT_FALSE(static_cast<bool>(ds));The reaction framework allows you to reset a computation node by replacing its computation function. This mechanism is useful when the result needs to be recalculated using a different logic or different dependencies after the node has been initially created.
Note: The return value type cannot be changed
Below is an example that demonstrates the reset functionality:
TEST(TestReset, ReactionTest) {
auto a = reaction::var(1);
auto b = reaction::var(std::string{"2"});
auto ds = reaction::calc([]() { return std::to_string(a()); });
ds.reset([=]() { return b() + "set"; });
v.value("3");
EXPECT_EQ(ds.get(), "3set");
EXPECT_THROW(ds.reset([=]() { return a(); }), std::runtime_error); // return type changed
EXPECT_THROW(ds.reset([=]() { return ds(); }), std::runtime_error); // cycle dependent
}The reaction framework supports various triggering mode to control when reactive computations are re-evaluated. This example demonstrates three mode:
- Value Change Trigger: The reactive computation is triggered only when the underlying value actually changes.
- Filter Trigger: The reactive computation is triggered when the value crosses a specified condition.
- Always Trigger: (Not explicitly shown in this example) Always triggers regardless of whether the value has changed.
The trigger Mode can be specified by the type parameter
using namespace reaction;
auto stockPrice = var(100.0);
auto profit = expr(stockPrice() - 100.0); // defalut ChangeTrigger
auto assignAction = action<AlwaysTrig>([=]() {
std::cout << "Checky assign, price = " << stockPrice() <<'\n';
});
auto sellAction = action<FilterTrig>([=]() {
std::cout << "It's time to sell, profit = " << profit() <<'\n';
});
sellAction.filter([=]() {
return profit() > 5.0;
});
*stockPrice = 100.0; // assignAction trigger
*stockPrice = 101.0; // assignAction, profit trigger
*stockPrice = 106.0; // all trigger
You can even define a trigger mode yourself in your code, just include the checkTrig method:
struct MyTrig {
bool checkTrig() {
// do something
return true;
}
};
auto a = var(1);
auto b = expr<MyTrig>(a + 1);In the reaction framework, all data sources obtained by users are actually in the form of weak references, and their actual memory is managed in the observer map.
Users can manually call the close method, so that all dependent data sources will also be closed.
auto a = reaction::var(1);
auto b = reaction::var(2);
auto dsA = reaction::calc([=]() { return a(); });
auto dsB = reaction::calc([=]() { return dsA() + b(); });
dsA.close(); //dsB will automatically close, cause dsB dependents dsA.
EXPECT_FALSE(static_cast<bool>(dsA));
EXPECT_FALSE(static_cast<bool>(dsB));However, for scenarios where the lifecycle of a weak reference acquired by user ends, the reaction framework makes several strategy for different scenarios.
-
DirectCloseStrategy: The node is immediately closed (made invalid) when any of its dependencies become invalid.
-
KeepCalcStrategy: The node continues to recalculate, its dependencies work normally.
-
LastValStrategy: The node retains the last valid, its dependencies use the value to calculate.
Below is a concise example that illustrates all three strategies:
{
auto a = var(1);
auto b = calc([]() { return a(); });
{
auto temp = calc<AlwaysTrig, CloseStra>([]() { return a(); });
b.set([](auto t) { return t; }, temp);
}
// temp lifecycle ends, b will end too.
EXPECT_FALSE(static_cast<bool>(b));
}
{
auto a = var(1);
auto b = calc([]() { return a(); });
{
auto temp = calc<AlwaysTrig, KeepStra>([]() { return a(); }); // default is KeepStra
b.set([](auto t) { return t; }, temp);
}
// temp lifecycle ends, b not be influenced.
EXPECT_TRUE(static_cast<bool>(b));
EXPECT_EQ(b.get(), 1);
a.value(2);
EXPECT_EQ(b.get(), 2);
}
{
auto a = var(1);
auto b = calc([]() { return a(); });
{
auto temp = calc<AlwaysTrig, LastStra>([]() { return a(); });
b.set([](auto t) { return t; }, temp);
}
// temp lifecycle ends, b use its last val to calculate.
EXPECT_TRUE(static_cast<bool>(b));
EXPECT_EQ(b.get(), 1);
a.value(2);
EXPECT_EQ(b.get(), 1);
}Likewise, you can define a strategy yourself in your code, just include the handleInvalid method:
struct MyStra {
void handleInvalid() {
std::cout << "Invalid" << std::endl;
}
};
auto a = var(1);
auto b = expr<AlwaysTrig, MyStra>(a + 1);Reaction supports reactive versions of standard stl containers (vector, list, set, map, etc.).
using namespace reaction;
constexpr int STUDENT_COUNT = 5;
// 1. Student grades container - using vector to store VarExpr
std::vector<Var<double>> grades;
for (int i = 0; i < STUDENT_COUNT; ++i) {
grades.push_back(make(70.0 + i * 5));
}
// 2. Grade statistics container - using list to store CalcExpr
std::list<Calc<double>> stats;
stats.push_back(make([&] {
double sum = 0;
for (auto &grade : grades)
sum += grade();
return sum / grades.size();
}));
stats.push_back(make([&] {
double max = grades[0].get();
for (auto &grade : grades)
max = std::max(max, grade());
return max;
}));
// 3. Grade change monitors - using set to store Action
std::set<Calc<VoidWrapper>> monitors;
for (int i = 0; i < STUDENT_COUNT; ++i) {
monitors.insert(make([i, &grades] {
std::cout << "[Monitor] Student " << i << " grade updated: " << grades[i]() << "\n";
}));
}
// 4. Grade level mapping - using map to store CalcExpr
std::map<int, Calc<const char *>> gradeLevels;
for (int i = 0; i < STUDENT_COUNT; ++i) {
gradeLevels.insert({i, make([i, &grades] {
double g = grades[i]();
if (g >= 90) return "A";
if (g >= 80) return "B";
if (g >= 70) return "C";
return "D";
})});
}
We welcome all forms of contributions to make Reaction even better:
-
Report Issues 🐛 Found a bug? Open an Issue with detailed reproduction steps.
-
Suggest Features 💡 Have an idea? Propose new features through GitHub Discussions.
-
Submit Pull Requests 👩💻 Follow our workflow:
git clone https://github.com/lumia431/reaction.git cd reaction # Create a feature branch (feat/xxx or fix/xxx) # Submit PR against `dev` branch