"If you liked it then you "should have put a"_test on it", Beyonce rule

UT / μt

| Motivation | Quick Start | Overview | Tutorial | Examples | User Guide | FAQ | Benchmarks |

C++ single header/single module, macro-free μ(micro)/Unit Testing Framework

#include <boost/ut.hpp> // import boost.ut;

constexpr auto sum(auto... values) { return (values + ...); }

int main() {
  using namespace boost::ut;

  "sum"_test = [] {
    expect(sum(0) == 0_i);
    expect(sum(1, 2) == 3_i);
    expect(sum(1, 2) > 0_i and 41_i == sum(40, 2));
  };
}

Running "sum"...
  sum.cpp:11:FAILED [(3 > 0 and 41 == 42)]
FAILED

===============================================================================
tests:   1 | 1 failed
asserts: 3 | 2 passed | 1 failed

https://godbolt.org/z/f4jEcv9vo

Motivation

Testing is a very important part of the Software Development, however, C++ doesn't provide any good testing facilities out of the box, which often leads into a poor testing experience for develops and/or lack of tests/coverage in general.

One should treat testing code as production code!

Additionally, well established testing practises such as Test Driven Development (TDD)/Behaviour Driven Development (BDD) are often not followed due to the same reasons.

The following snippet is a common example of testing with projects in C++.

int main() {
  // should sum numbers
  {
    assert(3 == sum(1, 2));
  }
}

There are quite a few problems with the approach above

• No names for tests (Hard to follow intentions by further readers)
• No automatic registration of tests (No way to run specific tests)
• Hard to debug (Assertions don't provide any information why it failed)
• Hard to scale (No easy path forward for parameterized tests, multiple suites, parallel execution, etc...)
• Hard to integrate (No easy way to have a custom output such as XML for CI integration)
• Easy to make mistakes (With implicit casting, floating point comparison, pointer comparison for strings, etc...)
• Hard to follow good practises such as TDD/BDD (Lack of support for sections and declarative expressions)
• ...

UT is trying to address these issues by simplifying testing experience with a few simple steps:

• Just get a single header/module
• Integrate it into your project
• Learn a few simple concepts (expect, test, suite)

And you good to go!

Okay, great, but why I would use UT over other/similar testing frameworks already available in C++?

• Boost.Test
• GoogleTest
• Catch
• ...

Great question! There are a few unique features which makes UT worth trying

• Firstly, it supports all the basic Unit Testing Framework features (automatic registration of tests, assertions, suites, etc...)
• It's easy to integrate (it's just one header/module)
• It's macro free which makes testing experience that much nicer (it uses modern C++ features instead, macros are opt-in rather than being compulsory - Can I still use macros?)
• It's flexible (all parts of the framework such as: runner, reporter, printer can be customized, basically most other Unit Testing Frameworks can be implemented on top of UT primitives)
• It has smaller learning curve (just a few simple concepts (expect, test, suite))
• It leverages C++ features to support more complex testing (parameterized)
• It's faster to compile and execute than similar frameworks which makes it suitable for bigger projects without additional hassle (Benchmarks)
• It supports TDD/BDD workflows
• It supports Gherkin specification
• It supports Spec
• ...

Sounds intriguing/interesting? Learn more at

• Tutorial
• Examples
• User-Guide

Quick Start

https://bit.ly/ut-quick-start (slides)

Overview

• No dependencies (C++20, Tested Compilers: GCC-9+, Clang-9.0+, Apple Clang-11.0.0+, MSVC-2019+*, Clang-cl-9.0+
• Single header/module (boost/ut.hpp)
• Macro-free (How does it work?)
• Easy to use (Minimal API - test, suite, operators, literals, [expect])
• Fast to compile/execute (Benchmarks)
• Features (Assertions, Suites, Tests, Sections, Parameterized, BDD, Gherkin, Spec, Matchers, Logging, Runners, Reporters, ...)
• Integrations (ApprovalTests.cpp)

Tutorial

    Step 0: Get it...

Get the latest latest header/module from here!

Include/Import

// #include <boost/ut.hpp> // single header
// import boost.ut;        // single module (C++20)

int main() { }

Compile & Run

$CXX main.cpp && ./a.out

All tests passed (0 assert in 0 test)

[Optional] Install it

cmake -Bbuild -H.
cd build && make         # run tests
cd build && make install # install

[Optional] CMake integration

This project provides a CMake config and target. Just load ut with find_package to import the Boost::ut target. Linking against this target will add the necessary include directory for the single header file. This is demonstrated in the following example.

find_package(ut REQUIRED)
add_library(my_test my_test.cpp)
target_link_libraries(my_test PRIVATE Boost::ut)

[Optional] Conan integration

The boost-ext-ut package is available from Conan Center. Just include it in your project's Conanfile with boost-ext-ut/2.1.1.

    Step 1: Expect it...

Let's write our first assertion, shall we?

int main() {
  boost::ut::expect(true);
}

All tests passed (1 asserts in 0 test)

https://godbolt.org/z/vfx-eB

Okay, let's make it fail now?

int main() {
  boost::ut::expect(1 == 2);
}

main.cpp:4:FAILED [false]
===============================================================================
tests:   0 | 0 failed
asserts: 1 | 0 passed | 1 failed

https://godbolt.org/z/7qTePx

Notice that expression 1 == 2 hasn't been printed. Instead we got false?

Let's print it then?

int main() {
  using namespace boost::ut;
  expect(1_i == 2);
}

main.cpp:4:FAILED [1 == 2]
===============================================================================
tests:   0 | 0 failed
asserts: 1 | 0 passed | 1 failed

https://godbolt.org/z/7MXVzu

Okay, now we have it! 1 == 2 has been printed as expected. Notice the User Defined Literal (UDL) 1_i was used. _i is a compile-time constant integer value

• It allows to override comparison operators 👍
• It disallow comparison of different types 👍

See the User-guide for more details.

Alternatively, a terse notation (no expect required) can be used.

int main() {
  using namespace boost::ut::literals;
  using namespace boost::ut::operators::terse;

  1_i == 2; // terse notation
}

main.cpp:7:FAILED [1 == 2]
===============================================================================
tests:   0 | 0 failed
asserts: 1 | 0 passed | 1 failed

https://godbolt.org/z/s77GSm

Other expression syntaxes are also available.

expect(1_i == 2);       // UDL syntax
expect(1 == 2_i);       // UDL syntax
expect(that % 1 == 2);  // Matcher syntax
expect(eq(1, 2));       // eq/neq/gt/ge/lt/le

main.cpp:6:FAILED [1 == 2]
main.cpp:7:FAILED [1 == 2]
main.cpp:8:FAILED [1 == 2]
main.cpp:9:FAILED [1 == 2]
===============================================================================
tests:   0 | 0 failed
asserts: 4 | 0 passed | 4 failed

https://godbolt.org/z/QbgGtc

Okay, but what about the case if my assertion is fatal. Meaning that the program will crash unless the processing will be terminated. Nothing easier, let's just add fatal call to make the test fail immediately.

expect(fatal(1 == 2_i)); // fatal assertion
expect(1_i == 2);        // not executed

main.cpp:6:FAILED [1 == 2]
===============================================================================
tests:   1 | 1 failed
asserts: 2 | 0 passed | 2 failed

https://godbolt.org/z/WMe8Y1

But my expression is more complex than just simple comparisons. Not a problem, logic operators are also supported in the expect 👍.

expect(42l == 42_l and 1 == 2_i); // compound expression

main.cpp:5:FAILED [(42 == 42 and 1 == 2)]
===============================================================================
tests:   0 | 0 failed
asserts: 1 | 0 passed | 1 failed

https://godbolt.org/z/aEhX4t

Can I add a custom message though? Sure, expect calls are streamable!

int main() {
  expect(42l == 42_l and 1 == 2_i) << "additional info";
}

main.cpp:5:FAILED [(42 == 42 and 1 == 2)] additional info
===============================================================================
tests:   0 | 0 failed
asserts: 1 | 0 passed | 1 failed

That's nice, can I use custom messages and fatal assertions? Yes, stream the fatal!

expect(fatal(1 == 2_i)) << "fatal assertion";
expect(1_i == 2);

FAILED
in: main.cpp:6 - test condition:  [1 == 2]

 fatal assertion
===============================================================================
tests:   0 | 2 failed
asserts: 0 | 0 passed | 2 failed

I use std::expected, can I stream its error() upon failure? Yes, since std::expected's error() can only be called when there is no value it requires lazy evaluation.

"lazy log"_test = [] {
  std::expected<bool, std::string> e = std::unexpected("lazy evaluated");
  expect(e.has_value()) << [&] { return e.error(); } << fatal;
  expect(e.value() == true);
};

Running test "lazy log"... FAILED
in: main.cpp:12 - test condition:  [false]

 lazy evaluated
===============================================================================
tests:   1 | 2 failed
asserts: 0 | 0 passed | 2 failed

https://godbolt.org/z/v2PDuU

    Step 2: Group it...

Assertions are great, but how to combine them into more cohesive units? Test cases are the way to go! They allow to group expectations for the same functionality into coherent units.

"hello world"_test = [] { };

Alternatively test("hello world") = [] {} can be used.

All tests passed (0 asserts in 1 tests)

https://godbolt.org/z/Bh-EmY

Notice 1 tests but 0 asserts.

Let's make our first end-2-end test case, shall we?

int main() {
  "hello world"_test = [] {
    int i = 43;
    expect(42_i == i);
  };
}

Running "hello world"...
  main.cpp:8:FAILED [42 == 43]
FAILED
===============================================================================
tests:   1 | 1 failed
asserts: 1 | 0 passed | 1 failed

https://godbolt.org/z/Y43mXz

👍 We are done here!

I'd like to nest my tests, though and share setup/tear-down. With lambdas used to represents tests/sections we can easily achieve that. Let's just take a look at the following example.

int main() {
  "[vector]"_test = [] {
    std::vector<int> v(5);

    expect(fatal(5_ul == std::size(v)));

    should("resize bigger") = [v] { // or "resize bigger"_test
      mut(v).resize(10);
      expect(10_ul == std::size(v));
    };

    expect(fatal(5_ul == std::size(v)));

    should("resize smaller") = [=]() mutable { // or "resize smaller"_test
      v.resize(0);
      expect(0_ul == std::size(v));
    };
  }
}

All tests passed (4 asserts in 1 tests)

https://godbolt.org/z/XWAdYt

Nice! That was easy, but I'm a believer into Behaviour Driven Development (BDD). Is there a support for that? Yes! Same example as above just with the BDD syntax.

int main() {
  "vector"_test = [] {
    given("I have a vector") = [] {
      std::vector<int> v(5);
      expect(fatal(5_ul == std::size(v)));

      when("I resize bigger") = [=] {
        mut(v).resize(10);

        then("The size should increase") = [=] {
          expect(10_ul == std::size(v));
        };
      };
    };
  };
}

All tests passed (2 asserts in 1 tests)

https://godbolt.org/z/dnvxsE

On top of that, feature/scenario aliases can be leveraged.

int main() {
  feature("vector") = [] {
    scenario("size") = [] {
      given("I have a vector") = [] {
        std::vector<int> v(5);
        expect(fatal(5_ul == std::size(v)));

        when("I resize bigger") = [=] {
          mut(v).resize(10);

          then("The size should increase") = [=] {
            expect(10_ul == std::size(v));
          };
        };
      };
    };
  };
}

All tests passed (2 asserts in 1 tests)

https://godbolt.org/z/T4cWss

Can I use Gherkin? Yeah, let's rewrite the example using Gherkin specification

int main() {
  bdd::gherkin::steps steps = [](auto& steps) {
    steps.feature("Vector") = [&] {
      steps.scenario("*") = [&] {
        steps.given("I have a vector") = [&] {
          std::vector<int> v(5);
          expect(fatal(5_ul == std::size(v)));

          steps.when("I resize bigger") = [&] {
            v.resize(10);
          };

          steps.then("The size should increase") = [&] {
            expect(10_ul == std::size(v));
          };
        };
      };
    };
  };

  "Vector"_test = steps |
    R"(
      Feature: Vector
        Scenario: Resize
          Given I have a vector
           When I resize bigger
           Then The size should increase
    )";
}

All tests passed (2 asserts in 1 tests)

https://godbolt.org/z/jb1d8P

Nice, is Spec notation supported as well?

int main() {
  describe("vector") = [] {
    std::vector<int> v(5);
    expect(fatal(5_ul == std::size(v)));

    it("should resize bigger") = [v] {
      mut(v).resize(10);
      expect(10_ul == std::size(v));
    };
  };
}

All tests passed (2 asserts in 1 tests)

https://godbolt.org/z/6jKKzT

That's great, but how can call the same tests with different arguments/types to be DRY (Don't Repeat Yourself)? Parameterized tests to the rescue!

int main() {
  for (auto i : std::vector{1, 2, 3}) {
    test("parameterized " + std::to_string(i)) = [i] { // 3 tests
      expect(that % i > 0); // 3 asserts
    };
  }
}

All tests passed (3 asserts in 3 tests)

https://godbolt.org/z/Utnd6X

That's it 😮! Alternatively, a convenient test syntax is also provided 👍

int main() {
  "args"_test = [](const auto& arg) {
    expect(arg > 0_i) << "all values greater than 0";
  } | std::vector{1, 2, 3};
}

All tests passed (3 asserts in 3 tests)

https://godbolt.org/z/6FHtpq

Check Examples for further reading.

    Step 3: Scale it...

Okay, but my project is more complex than that. How can I scale? Test suites will make that possible. By using suite in translation units tests defined inside will be automatically registered 👍

suite errors = [] { // or suite<"nameofsuite">
  "exception"_test = [] {
    expect(throws([] { throw 0; })) << "throws any exception";
  };

  "failure"_test = [] {
    expect(aborts([] { assert(false); }));
  };
};

int main() { }

All tests passed (2 asserts in 2 tests)

https://godbolt.org/z/_ccGwZ

---

What's next?

• Examples
• User-Guide

Examples

    Assertions

// operators
expect(0_i == sum());
expect(2_i != sum(1, 2));
expect(sum(1) >= 0_i);
expect(sum(1) <= 1_i);

// message
expect(3_i == sum(1, 2)) << "wrong sum";

// expressions
expect(0_i == sum() and 42_i == sum(40, 2));
expect(0_i == sum() or 1_i == sum()) << "compound";

// matchers
expect(that % 0 == sum());
expect(that % 42 == sum(40, 2) and that % (1 + 2) == sum(1, 2));
expect(that % 1 != 2 or 2_i > 3);

// eq/neq/gt/ge/lt/le
expect(eq(42, sum(40, 2)));
expect(neq(1, 2));
expect(eq(sum(1), 1) and neq(sum(1, 2), 2));
expect(eq(1, 1) and that % 1 == 1 and 1_i == 1);

// floating points
expect(42.1_d == 42.101) << "epsilon=0.1";
expect(42.10_d == 42.101) << "epsilon=0.01";
expect(42.10000001 == 42.1_d) << "epsilon=0.1";

// constant
constexpr auto compile_time_v = 42;
auto run_time_v = 99;
expect(constant<42_i == compile_time_v> and run_time_v == 99_i);

// failure
expect(1_i == 2) << "should fail";
expect(sum() == 1_i or 2_i == sum()) << "sum?";

assertions.cpp:53:FAILED [1 == 2] should fail
assertions.cpp:54:FAILED [(0 == 1 or 2 == 0)] sum?
===============================================================================
tests:   0  | 0 failed
asserts: 20 | 18 passed | 2 failed

https://godbolt.org/z/E1c7G5

    Tests

        Run/Skip/Tag

"run UDL"_test = [] {
  expect(42_i == 42);
};

skip / "don't run UDL"_test = [] {
  expect(42_i == 43) << "should not fire!";
};

All tests passed (1 asserts in 1 tests)
1 tests skipped

test("run function") = [] {
  expect(42_i == 42);
};

skip / test("don't run function") = [] {
  expect(42_i == 43) << "should not fire!";
};

All tests passed (1 asserts in 1 tests)
1 tests skipped

tag("nightly") / tag("slow") /
"performance"_test= [] {
  expect(42_i == 42);
};

tag("slow") /
"run slowly"_test= [] {
  expect(42_i == 43) << "should not fire!";
};

cfg<override> = {.tag = {"nightly"}};

All tests passed (1 asserts in 1 tests)
1 tests skipped

https://godbolt.org/z/X3_kG4

        Sections

"[vector]"_test = [] {
  std::vector<int> v(5);

  expect(fatal(5_ul == std::size(v)));

  should("resize bigger") = [=] { // or "resize bigger"_test
    mut(v).resize(10);
    expect(10_ul == std::size(v));
  };

  expect(fatal(5_ul == std::size(v)));

  should("resize smaller") = [=]() mutable { // or "resize smaller"_test
    v.resize(0);
    expect(0_ul == std::size(v));
  };
};

All tests passed (4 asserts in 1 tests)

https://godbolt.org/z/cE91bj

        Behavior Driven Development (BDD)

"Scenario"_test = [] {
  given("I have...") = [] {
    when("I run...") = [] {
      then("I expect...") = [] { expect(1_i == 1); };
      then("I expect...") = [] { expect(1 == 1_i); };
    };
  };
};

All tests passed (2 asserts in 1 tests)

https://godbolt.org/z/mNBySr

        Gherkin

int main() {
  bdd::gherkin::steps steps = [](auto& steps) {
    steps.feature("*") = [&] {
      steps.scenario("*") = [&] {
        steps.given("I have a number {value}") = [&](int value) {
          auto number = value;
          steps.when("I add {value} to it") = [&](int value) {
            number += value;
          };
          steps.then("I expect number to be {value}") = [&](int value) {
            expect(that % number == value);
          };
        };
      };
    };
  };

  "Gherkin"_test = steps |
    R"(
      Feature: Number
        Scenario: Addition
          Given I have a number 40
           When I add 2 to it
           Then I expect number to be 42
    )";
}

All tests passed (1 asserts in 1 tests)

https://godbolt.org/z/BP3hyt

        Spec

int main() {
  describe("equality") = [] {
    it("should be equal")     = [] { expect(0_i == 0); };
    it("should not be equal") = [] { expect(1_i != 0); };
  };
}

All tests passed (2 asserts in 1 tests)

https://godbolt.org/z/BXYJ3a

        Parameterized

for (auto i : std::vector{1, 2, 3}) {
  test("parameterized " + std::to_string(i)) = [i] {
    expect(that % i > 0);
  };
}

"args"_test =
   [](auto arg) {
      expect(arg >= 1_i);
    }
  | std::vector{1, 2, 3};

"types"_test =
    []<class T> {
      expect(std::is_integral_v<T>) << "all types are integrals";
    }
  | std::tuple<bool, int>{};

"args and types"_test =
    []<class TArg>(TArg arg) {
      expect(fatal(std::is_integral_v<TArg>));
      expect(42_i == arg or "is true"_b == arg);
      expect(type<TArg> == type<int> or type<TArg> == type<bool>);
    }
  | std::tuple{true, 42};

All tests passed (14 asserts in 10 tests)

https://godbolt.org/z/4xGGdo

And whenever I need to know the specific type for which the test failed, I can use reflection::type_name<T>(), like this:

"types with type name"_test =
    []<class T>() {
      expect(std::is_unsigned_v<T>) << reflection::type_name<T>() << "is unsigned";
    }
  | std::tuple<unsigned int, float>{};

Running "types with type name"...PASSED
Running "types with type name"...
  <source>:10:FAILED [false] float is unsigned
FAILED

https://godbolt.org/z/MEnGnbTY4

    Suites

namespace ut = boost::ut;

ut::suite errors = [] {
  using namespace ut;

  "throws"_test = [] {
    expect(throws([] { throw 0; }));
  };

  "doesn't throw"_test = [] {
    expect(nothrow([]{}));
  };
};

int main() { }

All tests passed (2 asserts in 2 tests)

https://godbolt.org/z/CFbTP9

    Misc

        Logging using streams

"logging"_test = [] {
  log << "pre";
  expect(42_i == 43) << "message on failure";
  log << "post";
};

Running "logging"...
pre
  logging.cpp:8:FAILED [42 == 43] message on failure
post
FAILED

===============================================================================

tests:   1 | 1 failed
asserts: 1 | 0 passed | 1 failed

https://godbolt.org/z/26fPSY

        Logging using formatting

This requires using C++20 with a standard library with std::format support.

"logging"_test = [] {
  log("\npre  {} == {}\n", 42, 43);
  expect(42_i == 43) << "message on failure";
  log("\npost {} == {} -> {}\n", 42, 43, 42 == 43);
};

Running "logging"...
pre  42 == 43
  logging.cpp:8:FAILED [42 == 43] message on failure
post 42 == 43 -> false
FAILED

===============================================================================

tests:   1 | 1 failed
asserts: 1 | 0 passed | 1 failed

https://godbolt.org/z/26fPSY

        Matchers

"matchers"_test = [] {
  constexpr auto is_between = [](auto lhs, auto rhs) {
    return [=](auto value) {
      return that % value >= lhs and that % value <= rhs;
    };
  };

  expect(is_between(1, 100)(42));
  expect(not is_between(1, 100)(0));
};

All tests passed (2 asserts in 1 tests)

https://godbolt.org/z/4qwrCi

        Exceptions/Aborts

"exceptions/aborts"_test = [] {
  expect(throws<std::runtime_error>([] { throw std::runtime_error{""}; }))
    << "throws runtime_error";
  expect(throws([] { throw 0; })) << "throws any exception";
  expect(nothrow([]{})) << "doesn't throw";
  expect(aborts([] { assert(false); }));
};

All tests passed (4 asserts in 1 tests)

https://godbolt.org/z/A2EehK

    Config

        Runner

namespace ut = boost::ut;

namespace cfg {
  class runner {
   public:
    template <class... Ts> auto on(ut::events::test<Ts...> test) { test(); }
    template <class... Ts> auto on(ut::events::skip<Ts...>) {}
    template <class TExpr>
    auto on(ut::events::assertion<TExpr>) -> bool { return true; }
    auto on(ut::events::fatal_assertion) {}
    template <class TMsg> auto on(ut::events::log<TMsg>) {}
  };
} // namespace cfg

template<> auto ut::cfg<ut::override> = cfg::runner{};

https://godbolt.org/z/jdg687

        Reporter

namespace ut = boost::ut;

namespace cfg {
  class reporter {
   public:
    auto on(ut::events::test_begin) -> void {}
    auto on(ut::events::test_run) -> void {}
    auto on(ut::events::test_skip) -> void {}
    auto on(ut::events::test_end) -> void {}
    template <class TMsg> auto on(ut::events::log<TMsg>) -> void {}
    template <class TExpr>
    auto on(ut::events::assertion_pass<TExpr>) -> void {}
    template <class TExpr>
    auto on(ut::events::assertion_fail<TExpr>) -> void {}
    auto on(ut::events::fatal_assertion) -> void {}
    auto on(ut::events::exception) -> void {}
    auto on(ut::events::summary) -> void {}
  };
}  // namespace cfg

template <>
auto ut::cfg<ut::override> = ut::runner<cfg::reporter>{};

https://godbolt.org/z/gsAPKg

        Printer

namespace ut = boost::ut;

namespace cfg {
struct printer : ut::printer {
  template <class T>
  auto& operator<<(T&& t) {
    std::cerr << std::forward<T>(t);
    return *this;
  }
};
}  // namespace cfg

template <>
auto ut::cfg<ut::override> = ut::runner<ut::reporter<cfg::printer>>{};

int main() {
  using namespace ut;
  "printer"_test = [] {};
}

https://godbolt.org/z/XCscF9

User Guide

    API

export module boost.ut; /// __cpp_modules

namespace boost::inline ext::ut::inline v2_1_1 {
  /**
   * Represents test suite object
   */
  struct suite final {
    /**
     * Creates and executes test suite
     * @example suite _ = [] {};
     * @param suite test suite function
     */
    constexpr explicit(false) suite(auto suite);
  };

  /**
   * Creates a test
   * @example "name"_test = [] {};
   * @return test object to be executed
   */
  constexpr auto operator""_test;

  /**
   * Creates a test
   * @example test("name") = [] {};
   * @return test object to be executed
   */
  constexpr auto test = [](const auto name);

  /**
   * Creates a test
   * @example should("name") = [] {};
   * @return test object to be executed
   */
  constexpr auto should = [](const auto name);

  /**
   * Behaviour Driven Development (BDD) helper functions
   * @param name step name
   * @return test object to be executed
   */
  constexpr auto given = [](const auto name);
  constexpr auto when  = [](const auto name);
  constexpr auto then  = [](const auto name);

  /**
   * Evaluates an expression
   * @example expect(42 == 42_i and 1 != 2_i);
   * @param expr expression to be evaluated
   * @param source location https://en.cppreference.com/w/cpp/utility/source_location
   * @return stream
   */
  constexpr OStream& expect(
    Expression expr,
    const std::source_location& location = std::source_location::current()
  );

  struct {
    /**
     * @example (that % 42 == 42);
     * @param expr expression to be evaluated
     */
    [[nodiscard]] constexpr auto operator%(Expression expr) const;
  } that{};

  inline namespace literals {
    /**
     * User defined literals to represent constant values
     * @example 42_i, 0_uc, 1.23_d
     */
    constexpr auto operator""_i;  /// int
    constexpr auto operator""_s;  /// short
    constexpr auto operator""_c;  /// char
    constexpr auto operator""_l;  /// long
    constexpr auto operator""_ll; /// long long
    constexpr auto operator""_u;  /// unsigned
    constexpr auto operator""_uc; /// unsigned char
    constexpr auto operator""_us; /// unsigned short
    constexpr auto operator""_ul; /// unsigned long
    constexpr auto operator""_f;  /// float
    constexpr auto operator""_d;  /// double
    constexpr auto operator""_ld; /// long double

    /**
     * Represents dynamic values
     * @example _i(42), _f(42.)
     */
    constexpr auto _b(bool);
    constexpr auto _c(char);
    constexpr auto _s(short);
    constexpr auto _i(int);
    constexpr auto _l(long);
    constexpr auto _ll(long long);
    constexpr auto _u(unsigned);
    constexpr auto _uc(unsigned char);
    constexpr auto _us(unsigned short);
    constexpr auto _ul(unsigned long);
    constexpr auto _f(float);
    constexpr auto _d(double);
    constexpr auto _ld(long double);

    /**
     * Logical representation of constant boolean (true) value
     * @example "is set"_b     : true
     *          not "is set"_b : false
     */
    constexpr auto operator ""_b;
  } // namespace literals

  inline namespace operators {
    /**
     * Comparison functions to be used in expressions
     * @example eq(42, 42), neq(1, 2)
     */
    constexpr auto eq(Operator lhs, Operator rhs);  /// ==
    constexpr auto neq(Operator lhs, Operator rhs); /// !=
    constexpr auto gt(Operator lhs, Operator rhs);  /// >
    constexpr auto ge(Operator lhs, Operator rhs);  /// >=
    constexpr auto lt(Operator lhs, Operator rhs);  /// <
    constexpr auto le(Operator lhs, Operator rhs);  /// <=

    /**
     * Overloaded comparison operators to be used in expressions
     * @example (42_i != 0)
     */
    constexpr auto operator==;
    constexpr auto operator!=;
    constexpr auto operator>;
    constexpr auto operator>=;
    constexpr auto operator<;
    constexpr auto operator<=;

    /**
     * Overloaded logic operators to be used in expressions
     * @example (42_i != 0 and 1 == 2_i)
     */
    constexpr auto operator and;
    constexpr auto operator or;
    constexpr auto operator not;

    /**
     * Executes parameterized tests
     * @example "parameterized"_test = [](auto arg) {} | std::tuple{1, 2, 3};
     */
    constexpr auto operator|;

    /**
     * Creates tags
     * @example tag("slow") / tag("nightly") / "perf"_test = []{};
     */
    constexpr auto operator/;

    /**
     * Creates a `fatal_assertion` from an expression
     * @example (42_i == 0) >> fatal
     */
    constexpr auto operator>>;
  } // namespace operators

  /**
   * Creates skippable test object
   * @example skip / "don't run"_test = [] { };
   */
  constexpr auto skip = tag("skip");

  struct {
    /**
     * @example log << "message!";
     * @param msg stringable message
     */
    auto& operator<<(Msg msg);
  } log{};

  /**
   * Makes object mutable
   * @example mut(object)
   * @param t object to be mutated
   */
  template<class T> auto mut(const T& t) -> T&;

  /**
   * Default execution flow policy
   */
  class runner {
   public:
    /**
     * @example cfg<override> = {
        .filter  = "test.section.*",
        .colors  = { .none = "" },
        .dry__run = true
       };
     * @param options.filter {default: "*"} runs all tests which names
                                            matches test.section.* filter
     * @param options.colors {default: {
                               .none = "\033[0m",
                               .pass = "\033[32m",
                               .fail  = "\033[31m"
              } if specified then overrides default color values
     * @param options.dry_run {default: false} if true then print test names to be
                                               executed without running them
     */
    auto operator=(options);

    /**
     * @example suite _ = [] {};
     * @param suite() executes suite
     */
    template<class TSuite>
    auto on(ut::events::suite<TSuite>);

    /**
     * @example "name"_test = [] {};
     * @param test.type ["test", "given", "when", "then"]
     * @param test.name "name"
     * @param test.arg parameterized argument
     * @param test() executes test
     */
    template<class... Ts>
    auto on(ut::events::test<Ts...>);

    /**
     * @example skip / "don't run"_test = []{};
     * @param skip.type ["test", "given", "when", "then"]
     * @param skip.name "don't run"
     * @param skip.arg parameterized argument
     */
    template<class... Ts>
    auto on(ut::events::skip<Ts...>);

    /**
     * @example file.cpp:42: expect(42_i == 42);
     * @param assertion.expr 42_i == 42
     * @param assertion.location { "file.cpp", 42 }
     * @return true if expr passes, false otherwise
     */
    template <class TExpr>
    auto on(ut::events::assertion<TExpr>) -> bool;

    /**
     * @example expect((2_i == 1) >> fatal)
     * @note triggered by `fatal`
     *       should std::exit
     */
    auto on(ut::events::fatal_assertion);

    /**
     * @example log << "message"
     * @param log.msg "message"
     */
    template<class TMsg>
    auto on(ut::events::log<TMsg>);

    /**
     * Explicitly runs registered test suites
     * If not called directly test suites are executed with run's destructor
     * @example return run({.report_errors = true})
     * @param run_cfg.report_errors {default: false} if true it prints the summary after running
     */
    auto run(run_cfg);

    /**
     * Runs registered test suites if they haven't been explicitly executed already
     */
    ~run();
  };

  /**
   * Default reporter policy
   */
  class reporter {
   public:
    /**
     * @example file.cpp:42: "name"_test = [] {};
     * @param test_begin.type ["test", "given", "when", "then"]
     * @param test_begin.name "name"
     * @param test_begin.location { "file.cpp", 42 }
     */
    auto on(ut::events::test_begin) -> void;

    /**
     * @example "name"_test = [] {};
     * @param test_run.type ["test", "given", "when", "then"]
     * @param test_run.name "name"
     */
    auto on(ut::events::test_run) -> void;

    /**
     * @example "name"_test = [] {};
     * @param test_skip.type ["test", "given", "when", "then"]
     * @param test_skip.name "name"
     */
    auto on(ut::events::test_skip) -> void;

    /**
     * @example "name"_test = [] {};
     * @param test_end.type ["test", "given", "when", "then"]
     * @param test_end.name "name"
     */
    auto on(ut::events::test_end) -> void;

    /**
     * @example log << "message"
     * @param log.msg "message"
     */
    template<class TMsg>
    auto on(ut::events::log<TMsg>) -> void;

    /**
     * @example file.cpp:42: expect(42_i == 42);
     * @param assertion_pass.expr 42_i == 42
     * @param assertion_pass.location { "file.cpp", 42 }
     */
    template <class TExpr>
    auto on(ut::events::assertion_pass<TExpr>) -> void;

    /**
     * @example file.cpp:42: expect(42_i != 42);
     * @param assertion_fail.expr 42_i != 42
     * @param assertion_fail.location { "file.cpp", 42 }
     */
    template <class TExpr>
    auto on(ut::events::assertion_fail<TExpr>) -> void;

    /**
     * @example expect((2_i == 1) >> fatal)
     * @note triggered by `fatal`
     *       should std::exit
     */
    auto on(ut::events::fatal_assertion) -> void;

    /**
     * @example "exception"_test = [] { throw std::runtime_error{""}; };
     */
    auto on(ut::events::exception) -> void;

    /**
     * @note triggered on destruction of runner
     */
    auto on(ut::events::summary) -> void;
  };

  /**
   * Used to override default running policy
   * @example template <> auto cfg<override> = runner<reporter>{};
   */
  struct override {};

  /**
   * Default UT execution policy
   * Can be overwritten with override
   */
  template <class = override> auto cfg = runner<reporter>{};
}

    Configuration

Option	Description	Example
BOOST_UT_VERSION	Current version	2'1'1

FAQ

    How does it work?

suite

/**
 * Represents suite object
 * @example suite _ = []{};
 */
struct suite final {
  /**
   * Assigns and executes test suite
   */
  [[nodiscard]] constexpr explicit(false) suite(Suite suite) {
    suite();
  }
};

test

/**
 * Creates named test object
 * @example "hello world"_test
 * @return test object
 */
[[nodiscard]] constexpr Test operator ""_test(const char* name, std::size_t size) {
  return test{{name, size}};
}

/**
 * Represents test object
 */
struct test final {
  std::string_view name{}; /// test case name

  /**
   * Assigns and executes test function
   * @param test function
   */
  constexpr auto operator=(const Test& test) {
    std::cout << "Running... " << name << '\n';
    test();
  }
};

expect

/**
 * Evaluates an expression
 * @example expect(42_i == 42);
 * @param expr expression to be evaluated
 * @param source location https://en.cppreference.com/w/cpp/utility/source_location
 * @return stream
 */
constexpr OStream& expect(
  Expression expr,
  const std::source_location& location = std::source_location::current()
) {
  if (not static_cast<bool>(expr) {
    std::cerr << location.file()
              << ':'
              << location.line()
              << ":FAILED: "
              << expr
              << '\n';
  }

  return std::cerr;
}

/**
 * Creates constant object for which operators can be overloaded
 * @example 42_i
 * @return integral constant object
 */
template <char... Cs>
[[nodiscard]] constexpr Operator operator""_i() -> integral_constant<int, value<Cs...>>;

/**
 * Overloads comparison if at least one of {lhs, rhs} is an Operator
 * @example (42_i == 42)
 * @param lhs Left-hand side operator
 * @param rhs Right-hand side operator
 * @return Comparison object
 */
[[nodiscard]] constexpr auto operator==(Operator lhs, Operator rhs) {
  return eq{lhs, rhs};
}

/**
 * Comparison Operator
 */
template <Operator TLhs, Operator TRhs>
struct eq final {
  TLhs lhs{}; // Left-hand side operator
  TRhs rhs{}; // Right-hand side operator

  /**
   * Performs comparison operation
   * @return true if expression is successful
   */
  [[nodiscard]] constexpr explicit operator bool() const {
    return lhs == rhs;
  }

  /**
   * Nicely prints the operation
   */
  friend auto operator<<(OStream& os, const eq& op) -> Ostream& {
    return (os << op.lhs << " == " << op.rhs);
  }
};

Sections

/**
 * Convenient aliases for creating test named object
 * @example should("return true") = [] {};
 */
constexpr auto should = [](const auto name) { return test{name}; };

Behaviour Driven Development (BDD)

/**
 * Convenient aliases for creating BDD tests
 * @example feature("Feature") = [] {};
 * @example scenario("Scenario") = [] {};
 * @example given("I have an object") = [] {};
 * @example when("I call it") = [] {};
 * @example then("I should get") = [] {};
 */
constexpr auto feature  = [](const auto name) { return test{name}; };
constexpr auto scenario = [](const auto name) { return test{name}; };
constexpr auto given    = [](const auto name) { return test{name}; };
constexpr auto when     = [](const auto name) { return test{name}; };
constexpr auto then     = [](const auto name) { return test{name}; };

https://godbolt.org/z/6Nk5Mi

Spec

/**
 * Convenient aliases for creating Spec tests
 * @example describe("test") = [] {};
 * @example it("should...") = [] {};
 */
constexpr auto describe = [](const auto name) { return test{name}; };
constexpr auto it       = [](const auto name) { return test{name}; };

Example implementation

Try it online

• Header - https://godbolt.org/z/x96n8b
• Module - https://wandbox.org/permlink/LrV7WwIgghTP1nrs

    Fast compilation times (Benchmarks)?

Implementation

• Leveraging C++20 features

• Avoiding unique types for lambda expressions

  template <class Test>
    requires not std::convertible_to<Test, void (*)()>>
  constexpr auto operator=(Test test);

vs

  // Compiles 5x faster because it doesn't introduce a new type for each lambda
  constexpr auto operator=(void (*test)());

• Type-name erasure (allows types/function memoization)

  eq<integral_constant<42>, int>{ {}, 42 }

vs

  // Can be memoized - faster to compile
  eq<int, int>{42, 42}

• Limiting preprocessor work

  • Single header/module
  • Minimal number of include files

• Simplified versions of

  • std::function
  • std::string_view

    C++20 features?

• API

  • Source Location

    • Assertions - expect(false) - __FILE__:__LINE__:FAILED [false]

  • Designated initializers

    • Configuration - cfg<override> = {.filter = "test"}

  • Non-Type Template Parameter

    • Constant matchers - constant<42_i == 42>

  • Template Parameter List for generic lambdas

    • Parameterized tests - "types"_test = []<class T>() {};

  • Concepts

    • Operators - Operator @ Operator

  • Modules

    • import boost.ut;

    C++2X integration?

Parameterized tests with Expansion statements (https://wg21.link/P1306r1)

template for (auto arg : std::tuple<int, double>{}) {
  test("types " + std::to_string(arg)) = [arg] {
    expect(type(arg) == type<int> or type(arg) == type<double>);
  };
}

All tests passed (2 asserts in 2 tests)

https://cppx.godbolt.org/z/dMmqmM

    Is standardization an option?

Personally, I believe that C++ standard could benefit from common testing primitives (expect, ""_test) because

• It lowers the entry-level to the language (no need for third-party libraries)
• It improves the education aspect (one standard way of doing it)
• It makes the language more coherent/stable (consistent design with other features, stable API)
• It makes the testing a first class citizen (shows that the community cares about this aspect of the language)
• It allows to publish tests for the Standard Library (STL) in the standard way (coherency, easier to extend)
• It allows to act as additional documentation as a way to verify whether a particular implementation is conforming (quality, self-verification)
• It helps with establishing standard vocabulary for testing (common across STL and other projects)

    Can I still use macros?

Sure, although please notice that there are negatives of using macros such as

• Error messages might be not clear and/or point to the wrong line
• Global scope will be polluted
• Type safety will be ignored

#define EXPECT(...) ::boost::ut::expect(::boost::ut::that % __VA_ARGS__)
#define SUITE       ::boost::ut::suite _ = []
#define TEST(name)  ::boost::ut::detail::test{"test", name} = [=]() mutable

SUITE {
  TEST("suite") {
    EXPECT(42 == 42);
  };
};

int main() {
  TEST("macro") {
    EXPECT(1 != 2);
  };

  TEST("vector") {
    std::vector<int> v(5);

   EXPECT(fatal(5u == std::size(v))) << "fatal";

    TEST("resize bigger") {
      v.resize(10);
      EXPECT(10u == std::size(v));
    };
  };
}

All tests passed (4 asserts in 3 tests)

https://godbolt.org/z/WcEKTr

    What about Mocks/Stubs/Fakes?

Consider using one of the following frameworks

• https://github.com/cpp-testing/GUnit/blob/master/docs/GMock.md
• https://github.com/eranpeer/FakeIt
• https://github.com/dascandy/hippomocks

    What about Microbenchmarking?

Example benchmark

Consider using one of the following frameworks

• https://github.com/google/benchmark
• https://github.com/DigitalInBlue/Celero
• https://github.com/libnonius/nonius
• https://github.com/martinus/nanobench

    Related materials/talks?

• [Boost].UT - Unit Testing Framework - Kris Jusiak
• Future of Testing with C++20 - Kris Jusiak
• Macro-Free Testing with C++20 - Kris Jusiak
• "If you liked it then you "should have put a"_test on it", Beyonce rule - Kris Jusiak
• Principles of Unit Testing With C++ - Dave Steffen and Kris Jusiak
• Empirical Unit Testing - Dave Steffen

    How to contribute?

CONTRIBUTING

Benchmarks

Framework	Version	Standard	License	Linkage	Test configuration
Boost.Test	1.71.0	C++03	Boost 1.0	single header/library	static library
GoogleTest	1.10.0	C++11	BSD-3	library	static library
Catch	2.10.2	C++11	Boost 1.0	single header	CATCH_CONFIG_FAST_COMPILE
Doctest	2.3.5	C++11	MIT	single header	DOCTEST_CONFIG_SUPER_FAST_ASSERTS
UT	1.1.0	C++20	Boost 1.0	single header/module

Include / 0 tests, 0 asserts, 1 cpp file
Assert / 1 test, 1'000'000 asserts, 1 cpp file
Test / 1'000 tests, 0 asserts, 1 cpp file
Suite / 10'000 tests, 0 asserts, 100 cpp files
Suite+Assert / 10'000 tests, 40'000 asserts, 100 cpp files
Suite+Assert+STL / 10'000 tests, 20'000 asserts, 100 cpp files
Incremental Build - Suite+Assert+STL / 1 cpp file change (1'000 tests, 20'000 asserts, 100 cpp files)
Suite+Assert+STL / 10'000 tests, 20'000 asserts, 100 cpp files (Headers vs Precompiled headers vs C++20 Modules)

https://github.com/cpp-testing/ut-benchmark

---

Disclaimer UT is not an official Boost library.